Richard K. Albert, MD

Two national surveys indicate that 14% to 28% of patients admitted to intensive care units (ICU's) are unplanned transfers (i.e., moving a patient to the ICU from other areas in the hospital providing lower intensity care due to an unanticipated change in the patient's clinical status), and that the most common reason for unplanned transfers is respiratory insufficiency/failure.1, 2 Patients suffering adverse events during a hospitalization are more likely to have an unplanned ICU transfer and patients requiring unplanned transfers have a higher mortality.35 Accordingly, the Joint Commission has identified improved recognition and response to changes in a patient's condition as a national patient safety goal,6 and Rapid Response Teams (RRTs) have been advocated to deal with these changes,7 although recent studies question the effectiveness of RRTs.811

We sought to classify the causes of unplanned, in‐hospital transfers to a medical ICU (MICU) with the idea of identifying common problems in care that might be addressed by process improvement activities. We also sought to determine the fraction of patients requiring an unplanned MICU transfer that had evidence of clinical deterioration prior to the time of transfer and whether, in retrospect, different or earlier interventions might have prevented the transfer. Our hypotheses were that (1) most unplanned MICU transfers occurred as a result of errors in care, (2) most were preceded by clinical deterioration within 12 hours prior to the transfer, and (3) most were preventable.

Methods

We conducted a retrospective cohort study of patients transferring to the MICU from non‐ICU Medicine units at Denver Health, a university‐affiliated, public safety net hospital. All adult patients between 18 to 89 years of age, who were admitted to the Medicine service between June, 2005 and May, 2006 were included in the study. Exclusion criteria included patients who (1) transferred from outside hospitals, (2) transferred from nonMedicine units within Denver Health, (3) were admitted directly to the MICU from the emergency department (ED), (4) were prisoners, (5) were readmitted to the MICU during the same hospitalization, (6) were known to be pregnant, or (7) were planned MICU transfers following invasive procedures (eg, elective cardiac catheterization, defibrillator placement, ablations). Patients readmitted to the MICU were excluded because of the difficulty distinguishing between premature transfer from the MICU or potential problems in care that might have occurred prior to the time of transfer from those occurring during follow‐up care on the Medicine floor services.

Computerized medical records of eligible patients were searched for demographic information and for admitting and transfer diagnoses (with the latter being categorized using a taxonomy we developed for classifying unplanned transfers, Table 1). Three independent observers (all of whom were board certified in Internal Medicine and had been practicing as Hospitalists at our institution for a minimum of three years) retrospectively reviewed each patient's hospital record to determine the cause of the unplanned transfer using this taxonomy. All three also made a judgment as to whether deterioration was evident at any time within the 12 hours preceding the unplanned transfer on the basis of clinical criteria used as our hospital's rapid response triggers (Table 2). When clinical triggers were found, each of the reviewers independently judged whether the unplanned transfer might have been prevented had different or earlier interventions been instituted. Each reviewer was blinded to the results of the other two.

All analyses were done using SAS Enterprise Guide 4.1, SAS Institute, Cary, NC. Data are presented as mean (standard deviation [SD]). Interobserver agreement was measured by calculating a statistic. values were interpreted by using the guidelines suggested by Landis and colleagues.12 A chi‐square test was used to seek associations between baseline characteristics, reasons for MICU transfer and mortality. P < 0.05 was considered to be statistically significant. The Colorado Multiple Institutional Review Board approved the research protocol.

Results

Over the period of the study the Medicine floor services had 4468 admissions of which 152 met the inclusion criteria for having an unplanned MICU transfer (Table 3). The most common admitting diagnoses were heart failure (12%) and community acquired pneumonia (9%). The most common diagnoses to which the unplanned MICU transfers were attributed were respiratory failure (27%) and sepsis (9%) (Table 4). Seven cardiopulmonary arrests were successfully resuscitated and transferred to the MICU. Throughout the period of the study, no patients were admitted to non‐MICU units because the MICU was at full capacity. Additionally the investigators did not find any inordinate delays in transfer to the ICU while waiting for a bed.

A total of 51 patients (34%) were transferred within the first 24 hours of admission. The most common diagnoses resulting in transfer in this group were respiratory failure, hypertensive emergency, hypotension, gastrointestinal bleed, and acute coronary syndrome. The remaining 101 patients (66%) were transferred from two to 15 days following admission for a variety of problems but respiratory failure was most common (34 patients, 22%).

Worsening of the problem for which the patients were initially admitted accounted for the unplanned transfers of 73 patients (48%) (Table 5). Development of a new problem unrelated to the admitting diagnosis accounted for the transfer in 59 patients (39%). Five patients were transferred to the ICU for a critical laboratory value that required a closer monitoring of the patient or needed more frequent lab draws that could not be achieved on the floor.

Errors in care were thought to be present in 29 patients (19% of the unplanned transfers). For 15 of these (52%) the error involved incorrect triage from the ED as 14 of the 15 patients met MICU admission criteria at the time they were triaged to non‐MICU units (Table 6). The remaining patient had a dissecting aortic aneurysm that was not considered while he was being evaluated for acute chest pain. All these patients were transferred to the ICU within 24 hours of their admission and the reviewers agreed that all could have been prevented if existing diagnostic and admission algorithms were followed.

Of the remaining 14 patients thought to have errors in care, nine were classified as the development of a new, iatrogenic problem (ie, opiate or benzodiazepine overdose occurring during treatment for pain and/or anxiety in 3, volume overload in 2, insulin‐induced hypoglycemia, antibiotic associated reaction, ‐blocker overdose and acute renal failure from over‐diuresis in one each) and five occurred because the patient's admitting problem worsened because treatment was thought to be either delayed, incorrect, or inadequate (Table 5). The reviewers all agreed that the unplanned transfers could have been prevented in eight of the 14 patients who developed iatrogenic problems if existing algorithms were followed or if an earlier or different intervention had occurred. The reviewers did not agree about whether the unplanned transfer could have been prevented in one patient who developed an iatrogenic problem and in all five patients whose underlying condition worsened. Accordingly, in sum, the reviewers felt that 23 of the 152 unplanned transfers (15%) could have been prevented.

In addition to trying to determine how many of the unplanned MICU transfers could have been prevented, we also investigated the utility of rapid response triggers in alerting the physicians and nurses of impending deteriorations in status and whether earlier recognition of this deterioration might have prevented the transfers. Of the 152 unplanned transfers, 106 (70%) had one or more rapid response triggers within the preceding 12 hours. All three reviewers agreed and concluded that in 94 (89%) of these, the unplanned transfer could not have been prevented, even with different or earlier interventions. For five patients (5% of the 106) all reviewers agreed and concluded that earlier intervention might have averted the subsequent transfer. For the other seven patients (6%), no consensus was reached. If we assume that, for all of these latter seven, earlier or different intervention might have averted the unplanned transfer, a maximum of 12 unplanned transfers (11% of the 106) might have been prevented by having a system of care that employed regularly assessing rapid response triggers and acting on them when recognized.

The interobserver reliability for the three reviewers was moderate to almost perfect with = 0.60, 95% confidence interval (CI) (0.31, 0.88); = 0.90, 95% CI (0.71, 1); = 0.55, 95% CI (0.26, 0.84).

A total of 27 (18%) of the patients with unplanned transfers died in the MICU. During this same time period 91 of 1511 patients (6%) admitted directly from the ED to the MICU died (P < 0.05). Mortality was lower for patients transferred within 24 hours of admission compared to those transferred > 24 hours after admission (4% vs. 22% mortality, respectively, P < 0.05; 95% CI, 0.09‐0.89). We found no difference in mortality as a function of time of admission or time of transfer implying that differences in staffing, or the availability of various services, did not contribute to the unplanned transfers.

Discussion

The important findings of this study were that (1) 19% of unplanned, in‐hospital transfers from Medicine floor services to the MICU seemed to result from apparent errors in care, (2) 15% of the unplanned transfers were potentially preventable, (3) the majority of the errors in care involved inappropriate triage of patients from the ED to the non‐MICU units, (4) 106 (70%) of the patients requiring unplanned transfers developed rapid response criteria within 12 hours prior to the transfer, but on review of these (5) the transfer was thought to be preventable in only a maximum of 12 (11%).

We designed our study in part to find specific errors that commonly resulted in unplanned MICU transfers with the idea that, if these could be identified, they might be corrected, thereby improving care. Contrary to our hypothesis we found that only 29 (19%) of the unplanned transfers seemed to result from errors in care. Of these, however, half were attributable to overlooking that patients met our own institution's MICU admission criteria at the time they were triaged to non‐MICU units. This result is consistent with Walter et al.13 finding that while 88% of MICUs in academic health centers had written MICU admission criteria, only 25% used these criteria on a regular basis. Hospital mortality is likely lower for patients meeting MICU admission criteria when they are appropriately and expeditiously triaged.1418 Accordingly, developing mechanisms by which patients are routinely screened for meeting MICU admission criteria could and should reduce this source of error and improve patient outcomes.

Nine of the remaining 14 errors in care resulted from what the chart reviewers concluded was overly aggressive treatment; either excess fluid resuscitation or excess treatment of pain or anxiety. It is not clear that these represent correctable errors in care, however, as hypotensive patients require fluid resuscitation, and patients with pain or anxiety should receive analgesics or anxiolytics and it is not reasonable to expect that these interventions will be appropriately titrated in every instance. Nonetheless, our reviewers all agreed that, in eight of these patients, different interventions could have prevented the unplanned transfer.

Since 41 (27%) of the unplanned transfers were for respiratory failure, we reviewed each of these patients' records seeking evidence suggesting that the problem might have resulted from excessive use of fluids, narcotics, or anxiolytics. By retrospective analysis only six such cases could be identified. Most were due to worsening of the problem for which the patient was admitted.

Consistent with our hypothesis the majority of patients requiring unplanned MICU transfers (106/152, 70%) developed rapid response clinical triggers within the 12 hours preceding transfer, as has been previously demonstrated by Hillman et al.7 and others.8‐10, 19 Our reviewers tried to determine whether earlier or different interventions might have prevented the deterioration and the resulting unplanned transfer. Interestingly, in the large majority (94/106, 89%) they concluded that nothing different could have been done and that the transfer could not have been avoided. While this observation contrasts with our hypothesis, it is consistent with two studies questioning the utility of RRTs in preventing unplanned ICU transfers.9, 10 In addition some patients may ultimately need an ICU transfer despite receiving appropriate interventions as it is impossible to prevent an ICU transfer in every patient. Conversely, just because a patient meets a rapid response criteria does not mean that the patient needs a higher level of care or an ICU transfer as some can be safely managed on the floor.

Our study has a number of potential limitations. The data came from a single teaching hospital and we only assessed patients admitted to General Internal Medicine units and transferred to a MICU. Accordingly, our results might not generalize to other hospitals (teaching or nonteaching), to other services or to other types of ICUs. We found, however, that (1) unplanned transfers accounted for 10% of the total admissions to our MICU, a similar fraction as reported by Angus et al.1 in 2006; (2) respiratory failure/emnsufficiency and sepsis were the most common diagnoses leading to unplanned transfers as previously reported by Groeger et al.2 and Hillman et al.5; (3) mortality was increased in patients requiring unplanned transfer, as noted by Escarce and Kelley3 and Hillman et al.5; and (4) patients who were transferred to the MICU within 24 hours of admission had better outcomes than those who were transferred later, as reported by Goldhill et al.4 Accordingly, our patient population seems quite similar to others in the literature.

Since we did not use objective criteria to assign patients to each of the categories itemized in Table 5 we could have misclassified patients with respect to the cause for their unplanned MICU transfer. Despite this shortcoming, however, the scores among our independent reviewers were moderate to almost perfect suggesting misclassification did not occur commonly.

Our retrospective study design may have underestimated the utility of RRTs as we had no way of knowing the outcomes of patients who met rapid response criteria and had interventions that prevented unplanned MICU transfers.

In summary, approximately 15% of unplanned MICU transfers seem to be preventable and approximately one‐fifth seem to result from errors in care, the majority of which are errors in triage from the ED. While the large majority of unplanned transfers were preceded by clinical deterioration within the preceding 12 hours, manifested by the presence of rapid response triggers, the large majority of these do not seem to be preventable. From these findings we suggest that unplanned transfers could be reduced by more closely screening patients for the presence of defined MICU admission criteria at the time of admission from the ED, by recognizing that fluid resuscitation and control of pain and/or anxiety can have adverse effects and by monitoring patients receiving these interventions more closely.

Two national surveys indicate that 14% to 28% of patients admitted to intensive care units (ICU's) are unplanned transfers (i.e., moving a patient to the ICU from other areas in the hospital providing lower intensity care due to an unanticipated change in the patient's clinical status), and that the most common reason for unplanned transfers is respiratory insufficiency/failure.1, 2 Patients suffering adverse events during a hospitalization are more likely to have an unplanned ICU transfer and patients requiring unplanned transfers have a higher mortality.35 Accordingly, the Joint Commission has identified improved recognition and response to changes in a patient's condition as a national patient safety goal,6 and Rapid Response Teams (RRTs) have been advocated to deal with these changes,7 although recent studies question the effectiveness of RRTs.811

We sought to classify the causes of unplanned, in‐hospital transfers to a medical ICU (MICU) with the idea of identifying common problems in care that might be addressed by process improvement activities. We also sought to determine the fraction of patients requiring an unplanned MICU transfer that had evidence of clinical deterioration prior to the time of transfer and whether, in retrospect, different or earlier interventions might have prevented the transfer. Our hypotheses were that (1) most unplanned MICU transfers occurred as a result of errors in care, (2) most were preceded by clinical deterioration within 12 hours prior to the transfer, and (3) most were preventable.

Methods

We conducted a retrospective cohort study of patients transferring to the MICU from non‐ICU Medicine units at Denver Health, a university‐affiliated, public safety net hospital. All adult patients between 18 to 89 years of age, who were admitted to the Medicine service between June, 2005 and May, 2006 were included in the study. Exclusion criteria included patients who (1) transferred from outside hospitals, (2) transferred from nonMedicine units within Denver Health, (3) were admitted directly to the MICU from the emergency department (ED), (4) were prisoners, (5) were readmitted to the MICU during the same hospitalization, (6) were known to be pregnant, or (7) were planned MICU transfers following invasive procedures (eg, elective cardiac catheterization, defibrillator placement, ablations). Patients readmitted to the MICU were excluded because of the difficulty distinguishing between premature transfer from the MICU or potential problems in care that might have occurred prior to the time of transfer from those occurring during follow‐up care on the Medicine floor services.

Computerized medical records of eligible patients were searched for demographic information and for admitting and transfer diagnoses (with the latter being categorized using a taxonomy we developed for classifying unplanned transfers, Table 1). Three independent observers (all of whom were board certified in Internal Medicine and had been practicing as Hospitalists at our institution for a minimum of three years) retrospectively reviewed each patient's hospital record to determine the cause of the unplanned transfer using this taxonomy. All three also made a judgment as to whether deterioration was evident at any time within the 12 hours preceding the unplanned transfer on the basis of clinical criteria used as our hospital's rapid response triggers (Table 2). When clinical triggers were found, each of the reviewers independently judged whether the unplanned transfer might have been prevented had different or earlier interventions been instituted. Each reviewer was blinded to the results of the other two.

All analyses were done using SAS Enterprise Guide 4.1, SAS Institute, Cary, NC. Data are presented as mean (standard deviation [SD]). Interobserver agreement was measured by calculating a statistic. values were interpreted by using the guidelines suggested by Landis and colleagues.12 A chi‐square test was used to seek associations between baseline characteristics, reasons for MICU transfer and mortality. P < 0.05 was considered to be statistically significant. The Colorado Multiple Institutional Review Board approved the research protocol.

Results

Over the period of the study the Medicine floor services had 4468 admissions of which 152 met the inclusion criteria for having an unplanned MICU transfer (Table 3). The most common admitting diagnoses were heart failure (12%) and community acquired pneumonia (9%). The most common diagnoses to which the unplanned MICU transfers were attributed were respiratory failure (27%) and sepsis (9%) (Table 4). Seven cardiopulmonary arrests were successfully resuscitated and transferred to the MICU. Throughout the period of the study, no patients were admitted to non‐MICU units because the MICU was at full capacity. Additionally the investigators did not find any inordinate delays in transfer to the ICU while waiting for a bed.

A total of 51 patients (34%) were transferred within the first 24 hours of admission. The most common diagnoses resulting in transfer in this group were respiratory failure, hypertensive emergency, hypotension, gastrointestinal bleed, and acute coronary syndrome. The remaining 101 patients (66%) were transferred from two to 15 days following admission for a variety of problems but respiratory failure was most common (34 patients, 22%).

Worsening of the problem for which the patients were initially admitted accounted for the unplanned transfers of 73 patients (48%) (Table 5). Development of a new problem unrelated to the admitting diagnosis accounted for the transfer in 59 patients (39%). Five patients were transferred to the ICU for a critical laboratory value that required a closer monitoring of the patient or needed more frequent lab draws that could not be achieved on the floor.

Errors in care were thought to be present in 29 patients (19% of the unplanned transfers). For 15 of these (52%) the error involved incorrect triage from the ED as 14 of the 15 patients met MICU admission criteria at the time they were triaged to non‐MICU units (Table 6). The remaining patient had a dissecting aortic aneurysm that was not considered while he was being evaluated for acute chest pain. All these patients were transferred to the ICU within 24 hours of their admission and the reviewers agreed that all could have been prevented if existing diagnostic and admission algorithms were followed.

Of the remaining 14 patients thought to have errors in care, nine were classified as the development of a new, iatrogenic problem (ie, opiate or benzodiazepine overdose occurring during treatment for pain and/or anxiety in 3, volume overload in 2, insulin‐induced hypoglycemia, antibiotic associated reaction, ‐blocker overdose and acute renal failure from over‐diuresis in one each) and five occurred because the patient's admitting problem worsened because treatment was thought to be either delayed, incorrect, or inadequate (Table 5). The reviewers all agreed that the unplanned transfers could have been prevented in eight of the 14 patients who developed iatrogenic problems if existing algorithms were followed or if an earlier or different intervention had occurred. The reviewers did not agree about whether the unplanned transfer could have been prevented in one patient who developed an iatrogenic problem and in all five patients whose underlying condition worsened. Accordingly, in sum, the reviewers felt that 23 of the 152 unplanned transfers (15%) could have been prevented.

In addition to trying to determine how many of the unplanned MICU transfers could have been prevented, we also investigated the utility of rapid response triggers in alerting the physicians and nurses of impending deteriorations in status and whether earlier recognition of this deterioration might have prevented the transfers. Of the 152 unplanned transfers, 106 (70%) had one or more rapid response triggers within the preceding 12 hours. All three reviewers agreed and concluded that in 94 (89%) of these, the unplanned transfer could not have been prevented, even with different or earlier interventions. For five patients (5% of the 106) all reviewers agreed and concluded that earlier intervention might have averted the subsequent transfer. For the other seven patients (6%), no consensus was reached. If we assume that, for all of these latter seven, earlier or different intervention might have averted the unplanned transfer, a maximum of 12 unplanned transfers (11% of the 106) might have been prevented by having a system of care that employed regularly assessing rapid response triggers and acting on them when recognized.

The interobserver reliability for the three reviewers was moderate to almost perfect with = 0.60, 95% confidence interval (CI) (0.31, 0.88); = 0.90, 95% CI (0.71, 1); = 0.55, 95% CI (0.26, 0.84).

A total of 27 (18%) of the patients with unplanned transfers died in the MICU. During this same time period 91 of 1511 patients (6%) admitted directly from the ED to the MICU died (P < 0.05). Mortality was lower for patients transferred within 24 hours of admission compared to those transferred > 24 hours after admission (4% vs. 22% mortality, respectively, P < 0.05; 95% CI, 0.09‐0.89). We found no difference in mortality as a function of time of admission or time of transfer implying that differences in staffing, or the availability of various services, did not contribute to the unplanned transfers.

Discussion

The important findings of this study were that (1) 19% of unplanned, in‐hospital transfers from Medicine floor services to the MICU seemed to result from apparent errors in care, (2) 15% of the unplanned transfers were potentially preventable, (3) the majority of the errors in care involved inappropriate triage of patients from the ED to the non‐MICU units, (4) 106 (70%) of the patients requiring unplanned transfers developed rapid response criteria within 12 hours prior to the transfer, but on review of these (5) the transfer was thought to be preventable in only a maximum of 12 (11%).

We designed our study in part to find specific errors that commonly resulted in unplanned MICU transfers with the idea that, if these could be identified, they might be corrected, thereby improving care. Contrary to our hypothesis we found that only 29 (19%) of the unplanned transfers seemed to result from errors in care. Of these, however, half were attributable to overlooking that patients met our own institution's MICU admission criteria at the time they were triaged to non‐MICU units. This result is consistent with Walter et al.13 finding that while 88% of MICUs in academic health centers had written MICU admission criteria, only 25% used these criteria on a regular basis. Hospital mortality is likely lower for patients meeting MICU admission criteria when they are appropriately and expeditiously triaged.1418 Accordingly, developing mechanisms by which patients are routinely screened for meeting MICU admission criteria could and should reduce this source of error and improve patient outcomes.

Nine of the remaining 14 errors in care resulted from what the chart reviewers concluded was overly aggressive treatment; either excess fluid resuscitation or excess treatment of pain or anxiety. It is not clear that these represent correctable errors in care, however, as hypotensive patients require fluid resuscitation, and patients with pain or anxiety should receive analgesics or anxiolytics and it is not reasonable to expect that these interventions will be appropriately titrated in every instance. Nonetheless, our reviewers all agreed that, in eight of these patients, different interventions could have prevented the unplanned transfer.

Since 41 (27%) of the unplanned transfers were for respiratory failure, we reviewed each of these patients' records seeking evidence suggesting that the problem might have resulted from excessive use of fluids, narcotics, or anxiolytics. By retrospective analysis only six such cases could be identified. Most were due to worsening of the problem for which the patient was admitted.

Consistent with our hypothesis the majority of patients requiring unplanned MICU transfers (106/152, 70%) developed rapid response clinical triggers within the 12 hours preceding transfer, as has been previously demonstrated by Hillman et al.7 and others.8‐10, 19 Our reviewers tried to determine whether earlier or different interventions might have prevented the deterioration and the resulting unplanned transfer. Interestingly, in the large majority (94/106, 89%) they concluded that nothing different could have been done and that the transfer could not have been avoided. While this observation contrasts with our hypothesis, it is consistent with two studies questioning the utility of RRTs in preventing unplanned ICU transfers.9, 10 In addition some patients may ultimately need an ICU transfer despite receiving appropriate interventions as it is impossible to prevent an ICU transfer in every patient. Conversely, just because a patient meets a rapid response criteria does not mean that the patient needs a higher level of care or an ICU transfer as some can be safely managed on the floor.

Our study has a number of potential limitations. The data came from a single teaching hospital and we only assessed patients admitted to General Internal Medicine units and transferred to a MICU. Accordingly, our results might not generalize to other hospitals (teaching or nonteaching), to other services or to other types of ICUs. We found, however, that (1) unplanned transfers accounted for 10% of the total admissions to our MICU, a similar fraction as reported by Angus et al.1 in 2006; (2) respiratory failure/emnsufficiency and sepsis were the most common diagnoses leading to unplanned transfers as previously reported by Groeger et al.2 and Hillman et al.5; (3) mortality was increased in patients requiring unplanned transfer, as noted by Escarce and Kelley3 and Hillman et al.5; and (4) patients who were transferred to the MICU within 24 hours of admission had better outcomes than those who were transferred later, as reported by Goldhill et al.4 Accordingly, our patient population seems quite similar to others in the literature.

Since we did not use objective criteria to assign patients to each of the categories itemized in Table 5 we could have misclassified patients with respect to the cause for their unplanned MICU transfer. Despite this shortcoming, however, the scores among our independent reviewers were moderate to almost perfect suggesting misclassification did not occur commonly.

Our retrospective study design may have underestimated the utility of RRTs as we had no way of knowing the outcomes of patients who met rapid response criteria and had interventions that prevented unplanned MICU transfers.

In summary, approximately 15% of unplanned MICU transfers seem to be preventable and approximately one‐fifth seem to result from errors in care, the majority of which are errors in triage from the ED. While the large majority of unplanned transfers were preceded by clinical deterioration within the preceding 12 hours, manifested by the presence of rapid response triggers, the large majority of these do not seem to be preventable. From these findings we suggest that unplanned transfers could be reduced by more closely screening patients for the presence of defined MICU admission criteria at the time of admission from the ED, by recognizing that fluid resuscitation and control of pain and/or anxiety can have adverse effects and by monitoring patients receiving these interventions more closely.

Two national surveys indicate that 14% to 28% of patients admitted to intensive care units (ICU's) are unplanned transfers (i.e., moving a patient to the ICU from other areas in the hospital providing lower intensity care due to an unanticipated change in the patient's clinical status), and that the most common reason for unplanned transfers is respiratory insufficiency/failure.1, 2 Patients suffering adverse events during a hospitalization are more likely to have an unplanned ICU transfer and patients requiring unplanned transfers have a higher mortality.35 Accordingly, the Joint Commission has identified improved recognition and response to changes in a patient's condition as a national patient safety goal,6 and Rapid Response Teams (RRTs) have been advocated to deal with these changes,7 although recent studies question the effectiveness of RRTs.811

We sought to classify the causes of unplanned, in‐hospital transfers to a medical ICU (MICU) with the idea of identifying common problems in care that might be addressed by process improvement activities. We also sought to determine the fraction of patients requiring an unplanned MICU transfer that had evidence of clinical deterioration prior to the time of transfer and whether, in retrospect, different or earlier interventions might have prevented the transfer. Our hypotheses were that (1) most unplanned MICU transfers occurred as a result of errors in care, (2) most were preceded by clinical deterioration within 12 hours prior to the transfer, and (3) most were preventable.

Methods

We conducted a retrospective cohort study of patients transferring to the MICU from non‐ICU Medicine units at Denver Health, a university‐affiliated, public safety net hospital. All adult patients between 18 to 89 years of age, who were admitted to the Medicine service between June, 2005 and May, 2006 were included in the study. Exclusion criteria included patients who (1) transferred from outside hospitals, (2) transferred from nonMedicine units within Denver Health, (3) were admitted directly to the MICU from the emergency department (ED), (4) were prisoners, (5) were readmitted to the MICU during the same hospitalization, (6) were known to be pregnant, or (7) were planned MICU transfers following invasive procedures (eg, elective cardiac catheterization, defibrillator placement, ablations). Patients readmitted to the MICU were excluded because of the difficulty distinguishing between premature transfer from the MICU or potential problems in care that might have occurred prior to the time of transfer from those occurring during follow‐up care on the Medicine floor services.

Computerized medical records of eligible patients were searched for demographic information and for admitting and transfer diagnoses (with the latter being categorized using a taxonomy we developed for classifying unplanned transfers, Table 1). Three independent observers (all of whom were board certified in Internal Medicine and had been practicing as Hospitalists at our institution for a minimum of three years) retrospectively reviewed each patient's hospital record to determine the cause of the unplanned transfer using this taxonomy. All three also made a judgment as to whether deterioration was evident at any time within the 12 hours preceding the unplanned transfer on the basis of clinical criteria used as our hospital's rapid response triggers (Table 2). When clinical triggers were found, each of the reviewers independently judged whether the unplanned transfer might have been prevented had different or earlier interventions been instituted. Each reviewer was blinded to the results of the other two.

All analyses were done using SAS Enterprise Guide 4.1, SAS Institute, Cary, NC. Data are presented as mean (standard deviation [SD]). Interobserver agreement was measured by calculating a statistic. values were interpreted by using the guidelines suggested by Landis and colleagues.12 A chi‐square test was used to seek associations between baseline characteristics, reasons for MICU transfer and mortality. P < 0.05 was considered to be statistically significant. The Colorado Multiple Institutional Review Board approved the research protocol.

Results

Over the period of the study the Medicine floor services had 4468 admissions of which 152 met the inclusion criteria for having an unplanned MICU transfer (Table 3). The most common admitting diagnoses were heart failure (12%) and community acquired pneumonia (9%). The most common diagnoses to which the unplanned MICU transfers were attributed were respiratory failure (27%) and sepsis (9%) (Table 4). Seven cardiopulmonary arrests were successfully resuscitated and transferred to the MICU. Throughout the period of the study, no patients were admitted to non‐MICU units because the MICU was at full capacity. Additionally the investigators did not find any inordinate delays in transfer to the ICU while waiting for a bed.

A total of 51 patients (34%) were transferred within the first 24 hours of admission. The most common diagnoses resulting in transfer in this group were respiratory failure, hypertensive emergency, hypotension, gastrointestinal bleed, and acute coronary syndrome. The remaining 101 patients (66%) were transferred from two to 15 days following admission for a variety of problems but respiratory failure was most common (34 patients, 22%).

Worsening of the problem for which the patients were initially admitted accounted for the unplanned transfers of 73 patients (48%) (Table 5). Development of a new problem unrelated to the admitting diagnosis accounted for the transfer in 59 patients (39%). Five patients were transferred to the ICU for a critical laboratory value that required a closer monitoring of the patient or needed more frequent lab draws that could not be achieved on the floor.

Errors in care were thought to be present in 29 patients (19% of the unplanned transfers). For 15 of these (52%) the error involved incorrect triage from the ED as 14 of the 15 patients met MICU admission criteria at the time they were triaged to non‐MICU units (Table 6). The remaining patient had a dissecting aortic aneurysm that was not considered while he was being evaluated for acute chest pain. All these patients were transferred to the ICU within 24 hours of their admission and the reviewers agreed that all could have been prevented if existing diagnostic and admission algorithms were followed.

Of the remaining 14 patients thought to have errors in care, nine were classified as the development of a new, iatrogenic problem (ie, opiate or benzodiazepine overdose occurring during treatment for pain and/or anxiety in 3, volume overload in 2, insulin‐induced hypoglycemia, antibiotic associated reaction, ‐blocker overdose and acute renal failure from over‐diuresis in one each) and five occurred because the patient's admitting problem worsened because treatment was thought to be either delayed, incorrect, or inadequate (Table 5). The reviewers all agreed that the unplanned transfers could have been prevented in eight of the 14 patients who developed iatrogenic problems if existing algorithms were followed or if an earlier or different intervention had occurred. The reviewers did not agree about whether the unplanned transfer could have been prevented in one patient who developed an iatrogenic problem and in all five patients whose underlying condition worsened. Accordingly, in sum, the reviewers felt that 23 of the 152 unplanned transfers (15%) could have been prevented.

In addition to trying to determine how many of the unplanned MICU transfers could have been prevented, we also investigated the utility of rapid response triggers in alerting the physicians and nurses of impending deteriorations in status and whether earlier recognition of this deterioration might have prevented the transfers. Of the 152 unplanned transfers, 106 (70%) had one or more rapid response triggers within the preceding 12 hours. All three reviewers agreed and concluded that in 94 (89%) of these, the unplanned transfer could not have been prevented, even with different or earlier interventions. For five patients (5% of the 106) all reviewers agreed and concluded that earlier intervention might have averted the subsequent transfer. For the other seven patients (6%), no consensus was reached. If we assume that, for all of these latter seven, earlier or different intervention might have averted the unplanned transfer, a maximum of 12 unplanned transfers (11% of the 106) might have been prevented by having a system of care that employed regularly assessing rapid response triggers and acting on them when recognized.

The interobserver reliability for the three reviewers was moderate to almost perfect with = 0.60, 95% confidence interval (CI) (0.31, 0.88); = 0.90, 95% CI (0.71, 1); = 0.55, 95% CI (0.26, 0.84).

A total of 27 (18%) of the patients with unplanned transfers died in the MICU. During this same time period 91 of 1511 patients (6%) admitted directly from the ED to the MICU died (P < 0.05). Mortality was lower for patients transferred within 24 hours of admission compared to those transferred > 24 hours after admission (4% vs. 22% mortality, respectively, P < 0.05; 95% CI, 0.09‐0.89). We found no difference in mortality as a function of time of admission or time of transfer implying that differences in staffing, or the availability of various services, did not contribute to the unplanned transfers.

Discussion

The important findings of this study were that (1) 19% of unplanned, in‐hospital transfers from Medicine floor services to the MICU seemed to result from apparent errors in care, (2) 15% of the unplanned transfers were potentially preventable, (3) the majority of the errors in care involved inappropriate triage of patients from the ED to the non‐MICU units, (4) 106 (70%) of the patients requiring unplanned transfers developed rapid response criteria within 12 hours prior to the transfer, but on review of these (5) the transfer was thought to be preventable in only a maximum of 12 (11%).

We designed our study in part to find specific errors that commonly resulted in unplanned MICU transfers with the idea that, if these could be identified, they might be corrected, thereby improving care. Contrary to our hypothesis we found that only 29 (19%) of the unplanned transfers seemed to result from errors in care. Of these, however, half were attributable to overlooking that patients met our own institution's MICU admission criteria at the time they were triaged to non‐MICU units. This result is consistent with Walter et al.13 finding that while 88% of MICUs in academic health centers had written MICU admission criteria, only 25% used these criteria on a regular basis. Hospital mortality is likely lower for patients meeting MICU admission criteria when they are appropriately and expeditiously triaged.1418 Accordingly, developing mechanisms by which patients are routinely screened for meeting MICU admission criteria could and should reduce this source of error and improve patient outcomes.

Nine of the remaining 14 errors in care resulted from what the chart reviewers concluded was overly aggressive treatment; either excess fluid resuscitation or excess treatment of pain or anxiety. It is not clear that these represent correctable errors in care, however, as hypotensive patients require fluid resuscitation, and patients with pain or anxiety should receive analgesics or anxiolytics and it is not reasonable to expect that these interventions will be appropriately titrated in every instance. Nonetheless, our reviewers all agreed that, in eight of these patients, different interventions could have prevented the unplanned transfer.

Since 41 (27%) of the unplanned transfers were for respiratory failure, we reviewed each of these patients' records seeking evidence suggesting that the problem might have resulted from excessive use of fluids, narcotics, or anxiolytics. By retrospective analysis only six such cases could be identified. Most were due to worsening of the problem for which the patient was admitted.

Consistent with our hypothesis the majority of patients requiring unplanned MICU transfers (106/152, 70%) developed rapid response clinical triggers within the 12 hours preceding transfer, as has been previously demonstrated by Hillman et al.7 and others.8‐10, 19 Our reviewers tried to determine whether earlier or different interventions might have prevented the deterioration and the resulting unplanned transfer. Interestingly, in the large majority (94/106, 89%) they concluded that nothing different could have been done and that the transfer could not have been avoided. While this observation contrasts with our hypothesis, it is consistent with two studies questioning the utility of RRTs in preventing unplanned ICU transfers.9, 10 In addition some patients may ultimately need an ICU transfer despite receiving appropriate interventions as it is impossible to prevent an ICU transfer in every patient. Conversely, just because a patient meets a rapid response criteria does not mean that the patient needs a higher level of care or an ICU transfer as some can be safely managed on the floor.

Our study has a number of potential limitations. The data came from a single teaching hospital and we only assessed patients admitted to General Internal Medicine units and transferred to a MICU. Accordingly, our results might not generalize to other hospitals (teaching or nonteaching), to other services or to other types of ICUs. We found, however, that (1) unplanned transfers accounted for 10% of the total admissions to our MICU, a similar fraction as reported by Angus et al.1 in 2006; (2) respiratory failure/emnsufficiency and sepsis were the most common diagnoses leading to unplanned transfers as previously reported by Groeger et al.2 and Hillman et al.5; (3) mortality was increased in patients requiring unplanned transfer, as noted by Escarce and Kelley3 and Hillman et al.5; and (4) patients who were transferred to the MICU within 24 hours of admission had better outcomes than those who were transferred later, as reported by Goldhill et al.4 Accordingly, our patient population seems quite similar to others in the literature.

Since we did not use objective criteria to assign patients to each of the categories itemized in Table 5 we could have misclassified patients with respect to the cause for their unplanned MICU transfer. Despite this shortcoming, however, the scores among our independent reviewers were moderate to almost perfect suggesting misclassification did not occur commonly.

Our retrospective study design may have underestimated the utility of RRTs as we had no way of knowing the outcomes of patients who met rapid response criteria and had interventions that prevented unplanned MICU transfers.

In summary, approximately 15% of unplanned MICU transfers seem to be preventable and approximately one‐fifth seem to result from errors in care, the majority of which are errors in triage from the ED. While the large majority of unplanned transfers were preceded by clinical deterioration within the preceding 12 hours, manifested by the presence of rapid response triggers, the large majority of these do not seem to be preventable. From these findings we suggest that unplanned transfers could be reduced by more closely screening patients for the presence of defined MICU admission criteria at the time of admission from the ED, by recognizing that fluid resuscitation and control of pain and/or anxiety can have adverse effects and by monitoring patients receiving these interventions more closely.

Communication failures are well‐recognized as causes of medical errors.1, 2 Specifically, handoffs of patient care responsibilities, which are increasingly prevalent in academic medical centers,3 have been cited as the most frequent cause of teamwork breakdown resulting in the harmful medical errors found in malpractice claims.1 The Institute of Medicine has recently identified patient handoffs as the moment where patient care errors are most likely to occur.4 A survey of 125 U.S. medical schools, however, found that only 8% specifically taught students how to hand off patient care.3

In July 2003, the American Council of Graduate Medical Education (ACGME) mandated that residency programs decrease resident work hours to improve patient care and safety by reducing fatigue,5 and a recent Institute of Medicine report suggests that they be decreased even further.4 Studies examining outcomes during the first 2 years after reducing duty hours did not find reductions in risk‐adjusted mortality.68 One proposed explanation for this lack of improvement is that the reduction in fatigue‐related medical errors is being offset by discontinuity of care with due to the increased number of patient handoffs resulting from shortened duty hours,911 one recent study found that omission of key information during patient sign outs frequently resulted in adverse patient care outcomes.12

In 2007, the Joint Commission developed a new National Patient Safety Goal that requires organizations to improve communication between caregivers.13 We recently developed an approach by which Internal Medicine residents hand off patient care using a structured process, written and verbal templates, formal training about handoffs, and direct attending supervision.14 Because fourth‐year medical students perform the duties of interns when working as subinterns, we recognized that education about handoffs should occur prior to the time students became interns. Accordingly, we developed a course designed to teach patient handoffs to medical students at the transition between their third and fourth years of training.

Setting

The Handoff Selective was developed by faculty of Denver Health and the University of Colorado Denver School of Medicine.

Program Description

The Selective was first offered in April 2007 as part of an Integrated Clinician's Course (ICC), a 2‐week course for students beginning their fourth year, which starts in April at the University of Colorado. The ICC includes both mandatory and selective sessions that are focused on developing clinical skills and preparing them for their subinternships. The Handoff Selective was conducted in a computerized teaching laboratory, lasted a total of 2 hours and consisted of 2 parts. Each of the 5 Denver Health Hospital Medicine faculty members versed in handoff education taught 2 sessions of 6 to 8 students.

Part 1: Didactic

During the first hour of class, the faculty presented a lecture that summarized the relevant literature on handoffs and explained the importance of the topic. The objectives of the didactic were to: (1) understand the importance of handoffs; (2) explore different communication elements and structures; (3) gain exposure to handoffs outside of healthcare; and (4) learn a structure for handoffs of patient care in hospitalized patients.

We used 3 video clips of handoffs from 2 football games to demonstrate the importance of practice, training, and 2‐way communications in handoffs. The first video clip showed a runner trying to make a spontaneous handoff while being tackled. The receiver was not expecting the handoff and was preoccupied with blocking another player. This attempted handoff resulted in a fumble, which we related to an adverse patient event.

The next 2 video clips showed 2 complex, seldom used, but well‐known football handoffsthe hook and lateral and the Statue of Liberty. Both handoffs were successfully executed presumably as a result of education, practice and the active participation of both players (handing off and receiving) in the process. We then related the teaching and practicing of complex communication to the Joint Commission on Accreditation of Healthcare Organizations (JCAHO; now simply the Joint Commission) data suggesting that most sentinel events have their root cause in communication and training failures.2

Basic communication elements and process structures were then explored using scenarios from everyday life and evidence from fields outside of medicine. We emphasized that structures for communication (modes, vehicles, and settings) must be chosen according to the occasion and that handoffs are common and important in all occupations. In discussing modes (verbal, written, or nonverbal), vehicles (paper, telephone, or e‐mail), and settings (face‐to face, virtual, or disconnected), we emphasized that the most effective structures for communication (verbal, face‐to face meetings, with written materials and other visual aids at the patient's bedside) were also the most time‐consuming (Figure 1). While our standard for resident handoffs is a face‐to‐face verbal interaction with preprinted written materials as an aid, we also emphasized that for complex patients (eg, mental status changes, concern for an acute abdomen) more robust communication is often needed. Accordingly, a more time‐consuming bedside handoff with simultaneous, focused physical exam and history‐taking by both oncoming and off‐going providers may be most appropriate.

Figure 1

As real‐life examples, we asked our students to communicate a happy birthday wish to their mother, who lives in another state. Almost uniformly, in addition to a written aid (birthday card), they choose the telephone as a vehicle for their verbal mode in a virtual setting with 2‐way communication possible. In contrast, when asked to propose marriage to a significant other in another state, students felt that a face‐to‐face meeting with verbal and nonverbal (ie, ring) modes was appropriate. This time‐consuming mode of communication was felt to be necessary to create a sentiment of importance and avert any possible miscommunication.

The didactic session concluded by demonstrating how to use standardized written and verbal templates for handoffs of the care of a hospitalized patient. We explore the differentiation between written and verbal handoffs in our discussion below.

Program Evaluation

We developed a 2‐part survey to evaluate the effectiveness of the Selective and to solicit feedback about the didactic and practicum portions of the course. The first part of the survey (Table 2) contained 16 items to assess the students' knowledge of, and attitudes toward handing off patient care, along with their comfort with the handoff process. Responses to this section were scored using a 5‐point Likert scale with 1 indicating strongly disagree and 5 indicating strongly agree. This part of the survey was administered both prior to and after the Selective.

The second part (Table 3) contained 12 items and was designed to evaluate the perceived usefulness of the different components of the class. This section was only administered at the end of the Selective. It utilized a 4‐point Likert scale with 1 indicating that the component was not useful at all, and 4 indicating that it was extremely useful. The first 6 items of the second section allowed students to evaluate the didactic portion of the handoff. The second 6 items allowed students to evaluate the practicum. Responses to all 12 items were then combined to determine an overall composite usefulness for the Selective.

The Selective was also evaluated qualitatively through the use of open‐ended, written comments that were solicited at the end of the survey. All surveys were administered anonymously.

Results

More students chose the Selective than we had capacity to accommodate (60 of a class of 150). The pre‐ and postcourse survey response rate was 56 of 60 (93%) and 58 of 60 (97%), respectively. After the Selective, the mean score in response to whether handoffs are well taught in medical school increased from 1.6 to 3.5 (P < 0.003). Our students' self‐perceived skills and knowledge about handoffs improved after the Selective (Table 2). The greatest changes in perceived knowledge occurred in questions regarding the what, how, and when of read‐backs, and the knowledge of standard verbal and written handoff structures. The responses to the survey elements which assessed our students' attitudes regarding the importance of standardization and whether they felt handoffs were safer with faculty supervision did not change after the Selective (Table 2).

A total of 92% of the students felt that the course was extremely useful or useful. The role‐playing activity was thought to be more helpful than the didactic, but 84% of the students still rated the didactic portion as useful or extremely useful (Table 3). The element which was the least well received in the didactic portion was the use of video clips to demonstrate successful and unsuccessful (fumbled) college football handoffs, although the majority (64%) of students still found it useful.

The major theme generated from the comments section of the survey was that the Selective should be a required course.

Discussion

We know of no previously published literature that has addressed teaching handoffs to medical students. Horwitz et al.15 developed a sign‐out curriculum for Internal Medicine residents and found that none of their house‐staff had any previous training in handoffs during medical school, consistent with the finding that only 8% of U.S. medical schools provided formal instruction on handoffs.3 Prior to taking the Selective, our students had no knowledge of verbal or written templates for patient handoffs, although both before and after the course they felt that standardization was an important component of the process.

A number of verbal structures for handing off patient care have been described in the literature and there is not a consensus as to which functions best. Perhaps the most cited verbal communication format is SBAR (ie, situation, background, assessment and recommendation).16, 17 This tool was developed by Leonard et al.18 specifically for use by nurses to provide 1‐way communication to physicians pertaining to a change in patient status. We considered teaching the SBAR approach to the students but felt that it did not provide a suitable structure for handoffs because the transfer of care is not generally an event‐based situation and the literature on handoffs indicates that an optimal verbal system includes 2‐way communication.

Additional mnemonics for handoffs found in the literature include SIGNOUT (ie, Sick or DNR, Identifying information, General hospital course, New events of the day, Overall health status, Upcoming possibilities with plan, and Tasks to complete),14 I PASS the BATON (ie, Introduction, Patient, Assessment, Situation, Safety, Background, Actions, Timing, Ownership, Next)19 and the SAIF‐IR system (see boxed text).14

We developed the SAIF‐IR mnemonic to maximize efficiency and effectiveness while differentiating the verbal portion of the handoff from the written and incorporating 2‐way communication into its structure. In the Summary statement, we emphasize that this is not a history of present illness. We ask our students to summarize, in 1 to 3 sentences, the patient's presentation and working diagnosis. When discussing patient issues, we ask our students to only verbalize Active issues, although the written template has inactive, chronic issues listed. Here, we also ask our students to express their level of concern for the active issues and patient in general. If‐then's and Follow‐ups are usually verbalized together. Based on the offgoing provider's knowledge of the patient, we encourage the offgoing provider to anticipate potential problems and advise the oncoming provider on potential responses. Much of this advice is difficult to express in the written format and thus may not be found on the written handoff when the verbal handoff occurs. We encourage oncoming providers to take notes on the preprinted handoff sheet as part of the handoff process.

Through Interactive questioning and Read‐backs, we train our students and house‐staff to use the active listening techniques used outside of healthcare, in settings such as nuclear power plants and National Aeronautics and Space Administration mission control, where poor handoff communication may also result in safety concerns and adverse events.20 Interactive questioning allows the oncoming provider to correct or clarify any information given by the off‐going provider. Read‐backs are a method of confirming follow‐up activity or contingency plans. Together, the SAIF‐IR mnemonic builds a 2‐way communication structure into the patient handoff with both offgoing and oncoming providers having predefined roles.

Much of the information on our written handoff (patient identifying information, medications, language preference, code status, admission date) is not verbalized unless it is part of the active issues or the if‐then, follow‐ups (ie, medication titration for a patient admitted with an acute coronary syndrome or cor status in a patient newly made comfort care). By not reading extraneous information, we seek to emphasize the Active issues as well as the If‐then, Follow‐ups. We feel this emphasis maximizes the effectiveness of the handoff, while the purposeful nonverbalization of written materials such as identifying information maximizes its efficiency. Future work may examine which verbal and written structures for patient handoffs most benefit patient care and workflow through standard communication.

While our students found the Handoff Selective to be useful and to improve their self‐perceived ability to perform handoffs, we were not able to determine whether our program affected downstream outcomes such as adverse events relating to failures in handoff communication. Additionally, since we only taught and evaluated our Selective at the University of Colorado Denver School of Medicine, the response of our students may not generalize to other medical schools. Multicentered, prospective, randomized controlled trials may determine whether handoff education programs are successful in reducing patient adverse events related to transfers of care.

While handoffs occur frequently and are increasingly recognized as a vulnerable time in patient care, little is known about how to effectively teach handoffs to medical students during their clinical years. We developed a formal course to teach the importance of handoffs and how the process should be conducted. Our students reported that the Handoff Selective we developed improved their knowledge about the process and their perception of their ability to perform handoffs in a time‐appropriate and effective manner. In response to the feedback we received from our students, the Handoff Selective is the only course in the ICC that has been made mandatory for all students.

Communication failures are well‐recognized as causes of medical errors.1, 2 Specifically, handoffs of patient care responsibilities, which are increasingly prevalent in academic medical centers,3 have been cited as the most frequent cause of teamwork breakdown resulting in the harmful medical errors found in malpractice claims.1 The Institute of Medicine has recently identified patient handoffs as the moment where patient care errors are most likely to occur.4 A survey of 125 U.S. medical schools, however, found that only 8% specifically taught students how to hand off patient care.3

In July 2003, the American Council of Graduate Medical Education (ACGME) mandated that residency programs decrease resident work hours to improve patient care and safety by reducing fatigue,5 and a recent Institute of Medicine report suggests that they be decreased even further.4 Studies examining outcomes during the first 2 years after reducing duty hours did not find reductions in risk‐adjusted mortality.68 One proposed explanation for this lack of improvement is that the reduction in fatigue‐related medical errors is being offset by discontinuity of care with due to the increased number of patient handoffs resulting from shortened duty hours,911 one recent study found that omission of key information during patient sign outs frequently resulted in adverse patient care outcomes.12

In 2007, the Joint Commission developed a new National Patient Safety Goal that requires organizations to improve communication between caregivers.13 We recently developed an approach by which Internal Medicine residents hand off patient care using a structured process, written and verbal templates, formal training about handoffs, and direct attending supervision.14 Because fourth‐year medical students perform the duties of interns when working as subinterns, we recognized that education about handoffs should occur prior to the time students became interns. Accordingly, we developed a course designed to teach patient handoffs to medical students at the transition between their third and fourth years of training.

Setting

The Handoff Selective was developed by faculty of Denver Health and the University of Colorado Denver School of Medicine.

Program Description

The Selective was first offered in April 2007 as part of an Integrated Clinician's Course (ICC), a 2‐week course for students beginning their fourth year, which starts in April at the University of Colorado. The ICC includes both mandatory and selective sessions that are focused on developing clinical skills and preparing them for their subinternships. The Handoff Selective was conducted in a computerized teaching laboratory, lasted a total of 2 hours and consisted of 2 parts. Each of the 5 Denver Health Hospital Medicine faculty members versed in handoff education taught 2 sessions of 6 to 8 students.

Part 1: Didactic

During the first hour of class, the faculty presented a lecture that summarized the relevant literature on handoffs and explained the importance of the topic. The objectives of the didactic were to: (1) understand the importance of handoffs; (2) explore different communication elements and structures; (3) gain exposure to handoffs outside of healthcare; and (4) learn a structure for handoffs of patient care in hospitalized patients.

We used 3 video clips of handoffs from 2 football games to demonstrate the importance of practice, training, and 2‐way communications in handoffs. The first video clip showed a runner trying to make a spontaneous handoff while being tackled. The receiver was not expecting the handoff and was preoccupied with blocking another player. This attempted handoff resulted in a fumble, which we related to an adverse patient event.

The next 2 video clips showed 2 complex, seldom used, but well‐known football handoffsthe hook and lateral and the Statue of Liberty. Both handoffs were successfully executed presumably as a result of education, practice and the active participation of both players (handing off and receiving) in the process. We then related the teaching and practicing of complex communication to the Joint Commission on Accreditation of Healthcare Organizations (JCAHO; now simply the Joint Commission) data suggesting that most sentinel events have their root cause in communication and training failures.2

Basic communication elements and process structures were then explored using scenarios from everyday life and evidence from fields outside of medicine. We emphasized that structures for communication (modes, vehicles, and settings) must be chosen according to the occasion and that handoffs are common and important in all occupations. In discussing modes (verbal, written, or nonverbal), vehicles (paper, telephone, or e‐mail), and settings (face‐to face, virtual, or disconnected), we emphasized that the most effective structures for communication (verbal, face‐to face meetings, with written materials and other visual aids at the patient's bedside) were also the most time‐consuming (Figure 1). While our standard for resident handoffs is a face‐to‐face verbal interaction with preprinted written materials as an aid, we also emphasized that for complex patients (eg, mental status changes, concern for an acute abdomen) more robust communication is often needed. Accordingly, a more time‐consuming bedside handoff with simultaneous, focused physical exam and history‐taking by both oncoming and off‐going providers may be most appropriate.

Figure 1

As real‐life examples, we asked our students to communicate a happy birthday wish to their mother, who lives in another state. Almost uniformly, in addition to a written aid (birthday card), they choose the telephone as a vehicle for their verbal mode in a virtual setting with 2‐way communication possible. In contrast, when asked to propose marriage to a significant other in another state, students felt that a face‐to‐face meeting with verbal and nonverbal (ie, ring) modes was appropriate. This time‐consuming mode of communication was felt to be necessary to create a sentiment of importance and avert any possible miscommunication.

The didactic session concluded by demonstrating how to use standardized written and verbal templates for handoffs of the care of a hospitalized patient. We explore the differentiation between written and verbal handoffs in our discussion below.

Program Evaluation

We developed a 2‐part survey to evaluate the effectiveness of the Selective and to solicit feedback about the didactic and practicum portions of the course. The first part of the survey (Table 2) contained 16 items to assess the students' knowledge of, and attitudes toward handing off patient care, along with their comfort with the handoff process. Responses to this section were scored using a 5‐point Likert scale with 1 indicating strongly disagree and 5 indicating strongly agree. This part of the survey was administered both prior to and after the Selective.

The second part (Table 3) contained 12 items and was designed to evaluate the perceived usefulness of the different components of the class. This section was only administered at the end of the Selective. It utilized a 4‐point Likert scale with 1 indicating that the component was not useful at all, and 4 indicating that it was extremely useful. The first 6 items of the second section allowed students to evaluate the didactic portion of the handoff. The second 6 items allowed students to evaluate the practicum. Responses to all 12 items were then combined to determine an overall composite usefulness for the Selective.

The Selective was also evaluated qualitatively through the use of open‐ended, written comments that were solicited at the end of the survey. All surveys were administered anonymously.

Results

More students chose the Selective than we had capacity to accommodate (60 of a class of 150). The pre‐ and postcourse survey response rate was 56 of 60 (93%) and 58 of 60 (97%), respectively. After the Selective, the mean score in response to whether handoffs are well taught in medical school increased from 1.6 to 3.5 (P < 0.003). Our students' self‐perceived skills and knowledge about handoffs improved after the Selective (Table 2). The greatest changes in perceived knowledge occurred in questions regarding the what, how, and when of read‐backs, and the knowledge of standard verbal and written handoff structures. The responses to the survey elements which assessed our students' attitudes regarding the importance of standardization and whether they felt handoffs were safer with faculty supervision did not change after the Selective (Table 2).

A total of 92% of the students felt that the course was extremely useful or useful. The role‐playing activity was thought to be more helpful than the didactic, but 84% of the students still rated the didactic portion as useful or extremely useful (Table 3). The element which was the least well received in the didactic portion was the use of video clips to demonstrate successful and unsuccessful (fumbled) college football handoffs, although the majority (64%) of students still found it useful.

The major theme generated from the comments section of the survey was that the Selective should be a required course.

Discussion

We know of no previously published literature that has addressed teaching handoffs to medical students. Horwitz et al.15 developed a sign‐out curriculum for Internal Medicine residents and found that none of their house‐staff had any previous training in handoffs during medical school, consistent with the finding that only 8% of U.S. medical schools provided formal instruction on handoffs.3 Prior to taking the Selective, our students had no knowledge of verbal or written templates for patient handoffs, although both before and after the course they felt that standardization was an important component of the process.

A number of verbal structures for handing off patient care have been described in the literature and there is not a consensus as to which functions best. Perhaps the most cited verbal communication format is SBAR (ie, situation, background, assessment and recommendation).16, 17 This tool was developed by Leonard et al.18 specifically for use by nurses to provide 1‐way communication to physicians pertaining to a change in patient status. We considered teaching the SBAR approach to the students but felt that it did not provide a suitable structure for handoffs because the transfer of care is not generally an event‐based situation and the literature on handoffs indicates that an optimal verbal system includes 2‐way communication.

Additional mnemonics for handoffs found in the literature include SIGNOUT (ie, Sick or DNR, Identifying information, General hospital course, New events of the day, Overall health status, Upcoming possibilities with plan, and Tasks to complete),14 I PASS the BATON (ie, Introduction, Patient, Assessment, Situation, Safety, Background, Actions, Timing, Ownership, Next)19 and the SAIF‐IR system (see boxed text).14

We developed the SAIF‐IR mnemonic to maximize efficiency and effectiveness while differentiating the verbal portion of the handoff from the written and incorporating 2‐way communication into its structure. In the Summary statement, we emphasize that this is not a history of present illness. We ask our students to summarize, in 1 to 3 sentences, the patient's presentation and working diagnosis. When discussing patient issues, we ask our students to only verbalize Active issues, although the written template has inactive, chronic issues listed. Here, we also ask our students to express their level of concern for the active issues and patient in general. If‐then's and Follow‐ups are usually verbalized together. Based on the offgoing provider's knowledge of the patient, we encourage the offgoing provider to anticipate potential problems and advise the oncoming provider on potential responses. Much of this advice is difficult to express in the written format and thus may not be found on the written handoff when the verbal handoff occurs. We encourage oncoming providers to take notes on the preprinted handoff sheet as part of the handoff process.

Through Interactive questioning and Read‐backs, we train our students and house‐staff to use the active listening techniques used outside of healthcare, in settings such as nuclear power plants and National Aeronautics and Space Administration mission control, where poor handoff communication may also result in safety concerns and adverse events.20 Interactive questioning allows the oncoming provider to correct or clarify any information given by the off‐going provider. Read‐backs are a method of confirming follow‐up activity or contingency plans. Together, the SAIF‐IR mnemonic builds a 2‐way communication structure into the patient handoff with both offgoing and oncoming providers having predefined roles.

Much of the information on our written handoff (patient identifying information, medications, language preference, code status, admission date) is not verbalized unless it is part of the active issues or the if‐then, follow‐ups (ie, medication titration for a patient admitted with an acute coronary syndrome or cor status in a patient newly made comfort care). By not reading extraneous information, we seek to emphasize the Active issues as well as the If‐then, Follow‐ups. We feel this emphasis maximizes the effectiveness of the handoff, while the purposeful nonverbalization of written materials such as identifying information maximizes its efficiency. Future work may examine which verbal and written structures for patient handoffs most benefit patient care and workflow through standard communication.

While our students found the Handoff Selective to be useful and to improve their self‐perceived ability to perform handoffs, we were not able to determine whether our program affected downstream outcomes such as adverse events relating to failures in handoff communication. Additionally, since we only taught and evaluated our Selective at the University of Colorado Denver School of Medicine, the response of our students may not generalize to other medical schools. Multicentered, prospective, randomized controlled trials may determine whether handoff education programs are successful in reducing patient adverse events related to transfers of care.

While handoffs occur frequently and are increasingly recognized as a vulnerable time in patient care, little is known about how to effectively teach handoffs to medical students during their clinical years. We developed a formal course to teach the importance of handoffs and how the process should be conducted. Our students reported that the Handoff Selective we developed improved their knowledge about the process and their perception of their ability to perform handoffs in a time‐appropriate and effective manner. In response to the feedback we received from our students, the Handoff Selective is the only course in the ICC that has been made mandatory for all students.

Communication failures are well‐recognized as causes of medical errors.1, 2 Specifically, handoffs of patient care responsibilities, which are increasingly prevalent in academic medical centers,3 have been cited as the most frequent cause of teamwork breakdown resulting in the harmful medical errors found in malpractice claims.1 The Institute of Medicine has recently identified patient handoffs as the moment where patient care errors are most likely to occur.4 A survey of 125 U.S. medical schools, however, found that only 8% specifically taught students how to hand off patient care.3

In July 2003, the American Council of Graduate Medical Education (ACGME) mandated that residency programs decrease resident work hours to improve patient care and safety by reducing fatigue,5 and a recent Institute of Medicine report suggests that they be decreased even further.4 Studies examining outcomes during the first 2 years after reducing duty hours did not find reductions in risk‐adjusted mortality.68 One proposed explanation for this lack of improvement is that the reduction in fatigue‐related medical errors is being offset by discontinuity of care with due to the increased number of patient handoffs resulting from shortened duty hours,911 one recent study found that omission of key information during patient sign outs frequently resulted in adverse patient care outcomes.12

In 2007, the Joint Commission developed a new National Patient Safety Goal that requires organizations to improve communication between caregivers.13 We recently developed an approach by which Internal Medicine residents hand off patient care using a structured process, written and verbal templates, formal training about handoffs, and direct attending supervision.14 Because fourth‐year medical students perform the duties of interns when working as subinterns, we recognized that education about handoffs should occur prior to the time students became interns. Accordingly, we developed a course designed to teach patient handoffs to medical students at the transition between their third and fourth years of training.

Setting

The Handoff Selective was developed by faculty of Denver Health and the University of Colorado Denver School of Medicine.

Program Description

The Selective was first offered in April 2007 as part of an Integrated Clinician's Course (ICC), a 2‐week course for students beginning their fourth year, which starts in April at the University of Colorado. The ICC includes both mandatory and selective sessions that are focused on developing clinical skills and preparing them for their subinternships. The Handoff Selective was conducted in a computerized teaching laboratory, lasted a total of 2 hours and consisted of 2 parts. Each of the 5 Denver Health Hospital Medicine faculty members versed in handoff education taught 2 sessions of 6 to 8 students.

Part 1: Didactic

During the first hour of class, the faculty presented a lecture that summarized the relevant literature on handoffs and explained the importance of the topic. The objectives of the didactic were to: (1) understand the importance of handoffs; (2) explore different communication elements and structures; (3) gain exposure to handoffs outside of healthcare; and (4) learn a structure for handoffs of patient care in hospitalized patients.

We used 3 video clips of handoffs from 2 football games to demonstrate the importance of practice, training, and 2‐way communications in handoffs. The first video clip showed a runner trying to make a spontaneous handoff while being tackled. The receiver was not expecting the handoff and was preoccupied with blocking another player. This attempted handoff resulted in a fumble, which we related to an adverse patient event.

The next 2 video clips showed 2 complex, seldom used, but well‐known football handoffsthe hook and lateral and the Statue of Liberty. Both handoffs were successfully executed presumably as a result of education, practice and the active participation of both players (handing off and receiving) in the process. We then related the teaching and practicing of complex communication to the Joint Commission on Accreditation of Healthcare Organizations (JCAHO; now simply the Joint Commission) data suggesting that most sentinel events have their root cause in communication and training failures.2

Basic communication elements and process structures were then explored using scenarios from everyday life and evidence from fields outside of medicine. We emphasized that structures for communication (modes, vehicles, and settings) must be chosen according to the occasion and that handoffs are common and important in all occupations. In discussing modes (verbal, written, or nonverbal), vehicles (paper, telephone, or e‐mail), and settings (face‐to face, virtual, or disconnected), we emphasized that the most effective structures for communication (verbal, face‐to face meetings, with written materials and other visual aids at the patient's bedside) were also the most time‐consuming (Figure 1). While our standard for resident handoffs is a face‐to‐face verbal interaction with preprinted written materials as an aid, we also emphasized that for complex patients (eg, mental status changes, concern for an acute abdomen) more robust communication is often needed. Accordingly, a more time‐consuming bedside handoff with simultaneous, focused physical exam and history‐taking by both oncoming and off‐going providers may be most appropriate.

Figure 1

As real‐life examples, we asked our students to communicate a happy birthday wish to their mother, who lives in another state. Almost uniformly, in addition to a written aid (birthday card), they choose the telephone as a vehicle for their verbal mode in a virtual setting with 2‐way communication possible. In contrast, when asked to propose marriage to a significant other in another state, students felt that a face‐to‐face meeting with verbal and nonverbal (ie, ring) modes was appropriate. This time‐consuming mode of communication was felt to be necessary to create a sentiment of importance and avert any possible miscommunication.

The didactic session concluded by demonstrating how to use standardized written and verbal templates for handoffs of the care of a hospitalized patient. We explore the differentiation between written and verbal handoffs in our discussion below.

Program Evaluation

We developed a 2‐part survey to evaluate the effectiveness of the Selective and to solicit feedback about the didactic and practicum portions of the course. The first part of the survey (Table 2) contained 16 items to assess the students' knowledge of, and attitudes toward handing off patient care, along with their comfort with the handoff process. Responses to this section were scored using a 5‐point Likert scale with 1 indicating strongly disagree and 5 indicating strongly agree. This part of the survey was administered both prior to and after the Selective.

The second part (Table 3) contained 12 items and was designed to evaluate the perceived usefulness of the different components of the class. This section was only administered at the end of the Selective. It utilized a 4‐point Likert scale with 1 indicating that the component was not useful at all, and 4 indicating that it was extremely useful. The first 6 items of the second section allowed students to evaluate the didactic portion of the handoff. The second 6 items allowed students to evaluate the practicum. Responses to all 12 items were then combined to determine an overall composite usefulness for the Selective.

The Selective was also evaluated qualitatively through the use of open‐ended, written comments that were solicited at the end of the survey. All surveys were administered anonymously.

Results

More students chose the Selective than we had capacity to accommodate (60 of a class of 150). The pre‐ and postcourse survey response rate was 56 of 60 (93%) and 58 of 60 (97%), respectively. After the Selective, the mean score in response to whether handoffs are well taught in medical school increased from 1.6 to 3.5 (P < 0.003). Our students' self‐perceived skills and knowledge about handoffs improved after the Selective (Table 2). The greatest changes in perceived knowledge occurred in questions regarding the what, how, and when of read‐backs, and the knowledge of standard verbal and written handoff structures. The responses to the survey elements which assessed our students' attitudes regarding the importance of standardization and whether they felt handoffs were safer with faculty supervision did not change after the Selective (Table 2).

A total of 92% of the students felt that the course was extremely useful or useful. The role‐playing activity was thought to be more helpful than the didactic, but 84% of the students still rated the didactic portion as useful or extremely useful (Table 3). The element which was the least well received in the didactic portion was the use of video clips to demonstrate successful and unsuccessful (fumbled) college football handoffs, although the majority (64%) of students still found it useful.

The major theme generated from the comments section of the survey was that the Selective should be a required course.

Discussion

We know of no previously published literature that has addressed teaching handoffs to medical students. Horwitz et al.15 developed a sign‐out curriculum for Internal Medicine residents and found that none of their house‐staff had any previous training in handoffs during medical school, consistent with the finding that only 8% of U.S. medical schools provided formal instruction on handoffs.3 Prior to taking the Selective, our students had no knowledge of verbal or written templates for patient handoffs, although both before and after the course they felt that standardization was an important component of the process.

A number of verbal structures for handing off patient care have been described in the literature and there is not a consensus as to which functions best. Perhaps the most cited verbal communication format is SBAR (ie, situation, background, assessment and recommendation).16, 17 This tool was developed by Leonard et al.18 specifically for use by nurses to provide 1‐way communication to physicians pertaining to a change in patient status. We considered teaching the SBAR approach to the students but felt that it did not provide a suitable structure for handoffs because the transfer of care is not generally an event‐based situation and the literature on handoffs indicates that an optimal verbal system includes 2‐way communication.

Additional mnemonics for handoffs found in the literature include SIGNOUT (ie, Sick or DNR, Identifying information, General hospital course, New events of the day, Overall health status, Upcoming possibilities with plan, and Tasks to complete),14 I PASS the BATON (ie, Introduction, Patient, Assessment, Situation, Safety, Background, Actions, Timing, Ownership, Next)19 and the SAIF‐IR system (see boxed text).14

We developed the SAIF‐IR mnemonic to maximize efficiency and effectiveness while differentiating the verbal portion of the handoff from the written and incorporating 2‐way communication into its structure. In the Summary statement, we emphasize that this is not a history of present illness. We ask our students to summarize, in 1 to 3 sentences, the patient's presentation and working diagnosis. When discussing patient issues, we ask our students to only verbalize Active issues, although the written template has inactive, chronic issues listed. Here, we also ask our students to express their level of concern for the active issues and patient in general. If‐then's and Follow‐ups are usually verbalized together. Based on the offgoing provider's knowledge of the patient, we encourage the offgoing provider to anticipate potential problems and advise the oncoming provider on potential responses. Much of this advice is difficult to express in the written format and thus may not be found on the written handoff when the verbal handoff occurs. We encourage oncoming providers to take notes on the preprinted handoff sheet as part of the handoff process.

Through Interactive questioning and Read‐backs, we train our students and house‐staff to use the active listening techniques used outside of healthcare, in settings such as nuclear power plants and National Aeronautics and Space Administration mission control, where poor handoff communication may also result in safety concerns and adverse events.20 Interactive questioning allows the oncoming provider to correct or clarify any information given by the off‐going provider. Read‐backs are a method of confirming follow‐up activity or contingency plans. Together, the SAIF‐IR mnemonic builds a 2‐way communication structure into the patient handoff with both offgoing and oncoming providers having predefined roles.

Much of the information on our written handoff (patient identifying information, medications, language preference, code status, admission date) is not verbalized unless it is part of the active issues or the if‐then, follow‐ups (ie, medication titration for a patient admitted with an acute coronary syndrome or cor status in a patient newly made comfort care). By not reading extraneous information, we seek to emphasize the Active issues as well as the If‐then, Follow‐ups. We feel this emphasis maximizes the effectiveness of the handoff, while the purposeful nonverbalization of written materials such as identifying information maximizes its efficiency. Future work may examine which verbal and written structures for patient handoffs most benefit patient care and workflow through standard communication.

While our students found the Handoff Selective to be useful and to improve their self‐perceived ability to perform handoffs, we were not able to determine whether our program affected downstream outcomes such as adverse events relating to failures in handoff communication. Additionally, since we only taught and evaluated our Selective at the University of Colorado Denver School of Medicine, the response of our students may not generalize to other medical schools. Multicentered, prospective, randomized controlled trials may determine whether handoff education programs are successful in reducing patient adverse events related to transfers of care.

While handoffs occur frequently and are increasingly recognized as a vulnerable time in patient care, little is known about how to effectively teach handoffs to medical students during their clinical years. We developed a formal course to teach the importance of handoffs and how the process should be conducted. Our students reported that the Handoff Selective we developed improved their knowledge about the process and their perception of their ability to perform handoffs in a time‐appropriate and effective manner. In response to the feedback we received from our students, the Handoff Selective is the only course in the ICC that has been made mandatory for all students.

If clinician‐quality improvers are to gain traction as academicians,1 their first objective should be to bring quality improvement (QI) sandly into the world of scientific method. We believe that Dr. Chakraborti's 2 pointsthat the reasons for afferent limb failure need to be more closely investigated, and that lessons learned from 1 hospital's rapid response system (RRS) may not generalize to other hospitalsreflect the immaturity of QI as a science. In clinical science, 3 well‐defined testing phases bring 1 homogeneous, rigorously tested product to market that is monitored in a fourth phase. While Dr. Chakraborti urges us to examine our afferent limb failures more closely, the monitoring and reporting strategies used in the Josie King Patient Safety Program2 resonate with the postmarketing surveillance of Phase IV trials.

Although necessary and valid, we believe that the majority of the QI conundrum of RRS lies in the lack of premarket, stepwise testing of QI products. QI initiatives are often promulgated before an appropriate evidence base has been established. This lack of scientific rigor has resulted in RRS with calling criteria that have poor operating characteristics,3 undetermined methods for achieving afferent success,4 and efferent response arms of varying sizes and compositions.5 Consequently, a heterogeneous group of RRS have produced equivocal outcomes6 and diminished the applicability of lessons learned across institutions.

Indeed, while it is important to ask, What do we do now?, it may be more informative to answer the question, How did we get here?

If clinician‐quality improvers are to gain traction as academicians,1 their first objective should be to bring quality improvement (QI) sandly into the world of scientific method. We believe that Dr. Chakraborti's 2 pointsthat the reasons for afferent limb failure need to be more closely investigated, and that lessons learned from 1 hospital's rapid response system (RRS) may not generalize to other hospitalsreflect the immaturity of QI as a science. In clinical science, 3 well‐defined testing phases bring 1 homogeneous, rigorously tested product to market that is monitored in a fourth phase. While Dr. Chakraborti urges us to examine our afferent limb failures more closely, the monitoring and reporting strategies used in the Josie King Patient Safety Program2 resonate with the postmarketing surveillance of Phase IV trials.

Although necessary and valid, we believe that the majority of the QI conundrum of RRS lies in the lack of premarket, stepwise testing of QI products. QI initiatives are often promulgated before an appropriate evidence base has been established. This lack of scientific rigor has resulted in RRS with calling criteria that have poor operating characteristics,3 undetermined methods for achieving afferent success,4 and efferent response arms of varying sizes and compositions.5 Consequently, a heterogeneous group of RRS have produced equivocal outcomes6 and diminished the applicability of lessons learned across institutions.

Indeed, while it is important to ask, What do we do now?, it may be more informative to answer the question, How did we get here?

If clinician‐quality improvers are to gain traction as academicians,1 their first objective should be to bring quality improvement (QI) sandly into the world of scientific method. We believe that Dr. Chakraborti's 2 pointsthat the reasons for afferent limb failure need to be more closely investigated, and that lessons learned from 1 hospital's rapid response system (RRS) may not generalize to other hospitalsreflect the immaturity of QI as a science. In clinical science, 3 well‐defined testing phases bring 1 homogeneous, rigorously tested product to market that is monitored in a fourth phase. While Dr. Chakraborti urges us to examine our afferent limb failures more closely, the monitoring and reporting strategies used in the Josie King Patient Safety Program2 resonate with the postmarketing surveillance of Phase IV trials.

Although necessary and valid, we believe that the majority of the QI conundrum of RRS lies in the lack of premarket, stepwise testing of QI products. QI initiatives are often promulgated before an appropriate evidence base has been established. This lack of scientific rigor has resulted in RRS with calling criteria that have poor operating characteristics,3 undetermined methods for achieving afferent success,4 and efferent response arms of varying sizes and compositions.5 Consequently, a heterogeneous group of RRS have produced equivocal outcomes6 and diminished the applicability of lessons learned across institutions.

Indeed, while it is important to ask, What do we do now?, it may be more informative to answer the question, How did we get here?

Many in‐hospital cardiac arrests and other adverse events are heralded by warning signs that are evident in the preceding 6 to 8 hours.1 By promptly intervening before further deterioration occurs, rapid response teams (RRTs) are designed to decrease unexpected intensive care unit (ICU) transfers, cardiac arrests, and inpatient mortality. While implementing RRTs is 1 of the 6 initiatives recommended by the Institute for Healthcare Improvement,2 data supporting their effectiveness is equivocal.3, 4

In October 2006, at Denver Health Medical Center, an academic, safety net hospital, we initiated a rapid response systemclinical triggers program (RRS‐CTP).5 In our RRS‐CTP, an abrupt change in patient status (Figure 1) triggers a mandatory call by the patient's nurse to the primary team, which is then required to perform an immediate bedside evaluation. By incorporating the primary team into the RRT‐CTP, we sought to preserve as much continuity of care as possible. Also, since the same house staff compose our cardiopulmonary arrest or cor team, and staff the ICUs and non‐ICU hospital wards, we did not feel that creating a separate RRT was an efficient use of resources. Our nurses have undergone extensive education about the necessity of a prompt bedside evaluation and have been instructed and empowered to escalate concerns to senior physicians if needed. We present a case that illustrates challenges to both implementing an RRS and measuring its potential benefits.

Figure 1

Case

A 59‐year‐old woman with a history of bipolar mood disorder was admitted for altered mental status. At presentation, she had signs of acute mania with normal vital signs. After initial laboratory workup, her altered mental status was felt to be multifactorial due to urinary tract infection, hypernatremia (attributed to lithium‐induced nephrogenic diabetes insipidus), and acute mania (attributed to medication discontinuation). Because she was slow to recover from the acute mania, her hospital stay was prolonged. From admission, the patient was treated with heparin 5000 units subcutaneously twice daily for venous thromboembolism prophylaxis.

On hospital day 7, at 21:32, the patient was noted to have asymptomatic tachycardia at 149 beats per minute and a new oxygen requirement of 3 L/minute. The cross‐cover team was called; however, although criteria were met, the RRS‐CTP was not activated and a bedside evaluation was not performed. A chest X‐ray was found to be normal and, with the exception of the oxygen requirement, her vital signs normalized by 23:45. No further diagnostic testing was performed at the time.

The next morning, at 11:58, the patient was found to have a blood pressure of 60/40 mmHg and heart rate of 42 beats per minute. The RRS‐CTP was activated. The primary team arrived at the bedside at 12:00 and found the patient to be alert, oriented, and without complaints. Her respiratory rate was 30/minute, and her oxygen saturation was 86% on 3 L/minute. An arterial blood gas analysis demonstrated acute respiratory alkalosis with hypoxemia and an electrocardiogram showed sinus tachycardia with a new S₁Q₃T₃ pattern. A computed tomography angiogram revealed a large, nearly occlusive pulmonary embolus (PE) filling an enlarged right pulmonary artery, as well as thrombus within the left main pulmonary artery. She was transferred to the medical ICU and alteplase was administered. The patient survived and was discharged in good clinical condition.

Discussion

Despite the strong theoretical benefit of the RRT concept, a recent review by Ranji et al.4 concluded that RRTs had not yet been shown to improve patient outcomes. In contrast to dedicated RRTs, this case illustrates a different type of RRS that was designed to address abrupt changes in patient status, while maintaining continuity of care and efficiently utilizing resources.

If one considers an RRS to have both afferent (criteria recognition) and efferent (RRT or primary team response) limbs, the afferent limb must be consistently activated in order to obtain the efferent limb's response.6 The greatest opportunities to improve RRSs are thought to lie in the afferent limb.3 Our RRS‐CTP was not triggered in 1 of 2 instances in which criteria for mandatory initiation of the system were met. This is consistent with the findings of the Medical Early Response Intervention and Therapy (MERIT) trial, in which RRTs were called in only 41% of the patients meeting criteria and subsequently having adverse events,7 and with the ongoing monitoring of the use of the system at our hospital. Had the cross‐covering team seen the patient at the bedside initially, the PE might have been diagnosed while the patient was hemodynamically stable, giving the patient nearly a 3‐fold lower relative mortality.8 When the RRS‐CTP was activated, a prompt bedside evaluation occurred, allowing for lytic therapy to be administered before cardiopulmonary arrest (attendant mortality of 90%).9

While rapid response criteria were originally based upon published sensitivity analyses, more recent studies suggest that these criteria lack diagnostic accuracy. As demonstrated by Cretikos et al,10 to reach a sensitivity of 70%, the corresponding specificity would be only 86%. Given that the prevalence of adverse events in the MERIT trial was only 0.6%, the resulting positive predictive value (PPV) of rapid response call criteria is 3%. Accordingly, 33 calls would be needed to prevent 1 unplanned ICU transfer, cardiac arrest, or death. Nurses' attempts to minimize false‐positive calls may help explain the low call rates for patients meeting RRT criteria. The 2 avenues to increase the PPV of criteria are:

• 1

  Increase the prevalence of disease in the population screened by risk factor stratification.

• 2

  Increase the specificity of the call criteria, which has been limited by the associated decrease in sensitivity.10

Regarding the efferent response limb of RRS, our case demonstrates that the primary team (rather than a separate group of caregivers), when alerted appropriately, can effectively respond to critical changes in patient status. Accordingly, our data show that since the inception of the program, cardiopulmonary arrests have decreased from a mean of 4.1 per month to 2.3 per month (P = 0.03).

Many clinical trials of RRTs would not capture the success demonstrated in this case. For example, due to the low prevalence of events, the MERIT trial used a composite endpoint that included unplanned ICU transfers, cardiac arrests, and mortality. Because our patient still required an unplanned ICU transfer after being evaluated by the responding team, she would have been counted as a system failure.

Conclusion

While local needs should inform the type of RRS implemented, this case illustrates one of the major obstacles ubiquitous to RRS implementation: failure of system activation. With appropriate activation, an RRS‐CTP can meet RRS goals while maintaining continuity of care and maximizing existing resources. This case also illustrates the difficulty of achieving a statistically relevant outcome, while demonstrating the potential benefits of evolving RRSs.

Many in‐hospital cardiac arrests and other adverse events are heralded by warning signs that are evident in the preceding 6 to 8 hours.1 By promptly intervening before further deterioration occurs, rapid response teams (RRTs) are designed to decrease unexpected intensive care unit (ICU) transfers, cardiac arrests, and inpatient mortality. While implementing RRTs is 1 of the 6 initiatives recommended by the Institute for Healthcare Improvement,2 data supporting their effectiveness is equivocal.3, 4

In October 2006, at Denver Health Medical Center, an academic, safety net hospital, we initiated a rapid response systemclinical triggers program (RRS‐CTP).5 In our RRS‐CTP, an abrupt change in patient status (Figure 1) triggers a mandatory call by the patient's nurse to the primary team, which is then required to perform an immediate bedside evaluation. By incorporating the primary team into the RRT‐CTP, we sought to preserve as much continuity of care as possible. Also, since the same house staff compose our cardiopulmonary arrest or cor team, and staff the ICUs and non‐ICU hospital wards, we did not feel that creating a separate RRT was an efficient use of resources. Our nurses have undergone extensive education about the necessity of a prompt bedside evaluation and have been instructed and empowered to escalate concerns to senior physicians if needed. We present a case that illustrates challenges to both implementing an RRS and measuring its potential benefits.

Figure 1

Case

A 59‐year‐old woman with a history of bipolar mood disorder was admitted for altered mental status. At presentation, she had signs of acute mania with normal vital signs. After initial laboratory workup, her altered mental status was felt to be multifactorial due to urinary tract infection, hypernatremia (attributed to lithium‐induced nephrogenic diabetes insipidus), and acute mania (attributed to medication discontinuation). Because she was slow to recover from the acute mania, her hospital stay was prolonged. From admission, the patient was treated with heparin 5000 units subcutaneously twice daily for venous thromboembolism prophylaxis.

On hospital day 7, at 21:32, the patient was noted to have asymptomatic tachycardia at 149 beats per minute and a new oxygen requirement of 3 L/minute. The cross‐cover team was called; however, although criteria were met, the RRS‐CTP was not activated and a bedside evaluation was not performed. A chest X‐ray was found to be normal and, with the exception of the oxygen requirement, her vital signs normalized by 23:45. No further diagnostic testing was performed at the time.

The next morning, at 11:58, the patient was found to have a blood pressure of 60/40 mmHg and heart rate of 42 beats per minute. The RRS‐CTP was activated. The primary team arrived at the bedside at 12:00 and found the patient to be alert, oriented, and without complaints. Her respiratory rate was 30/minute, and her oxygen saturation was 86% on 3 L/minute. An arterial blood gas analysis demonstrated acute respiratory alkalosis with hypoxemia and an electrocardiogram showed sinus tachycardia with a new S₁Q₃T₃ pattern. A computed tomography angiogram revealed a large, nearly occlusive pulmonary embolus (PE) filling an enlarged right pulmonary artery, as well as thrombus within the left main pulmonary artery. She was transferred to the medical ICU and alteplase was administered. The patient survived and was discharged in good clinical condition.

Discussion

Despite the strong theoretical benefit of the RRT concept, a recent review by Ranji et al.4 concluded that RRTs had not yet been shown to improve patient outcomes. In contrast to dedicated RRTs, this case illustrates a different type of RRS that was designed to address abrupt changes in patient status, while maintaining continuity of care and efficiently utilizing resources.

If one considers an RRS to have both afferent (criteria recognition) and efferent (RRT or primary team response) limbs, the afferent limb must be consistently activated in order to obtain the efferent limb's response.6 The greatest opportunities to improve RRSs are thought to lie in the afferent limb.3 Our RRS‐CTP was not triggered in 1 of 2 instances in which criteria for mandatory initiation of the system were met. This is consistent with the findings of the Medical Early Response Intervention and Therapy (MERIT) trial, in which RRTs were called in only 41% of the patients meeting criteria and subsequently having adverse events,7 and with the ongoing monitoring of the use of the system at our hospital. Had the cross‐covering team seen the patient at the bedside initially, the PE might have been diagnosed while the patient was hemodynamically stable, giving the patient nearly a 3‐fold lower relative mortality.8 When the RRS‐CTP was activated, a prompt bedside evaluation occurred, allowing for lytic therapy to be administered before cardiopulmonary arrest (attendant mortality of 90%).9

While rapid response criteria were originally based upon published sensitivity analyses, more recent studies suggest that these criteria lack diagnostic accuracy. As demonstrated by Cretikos et al,10 to reach a sensitivity of 70%, the corresponding specificity would be only 86%. Given that the prevalence of adverse events in the MERIT trial was only 0.6%, the resulting positive predictive value (PPV) of rapid response call criteria is 3%. Accordingly, 33 calls would be needed to prevent 1 unplanned ICU transfer, cardiac arrest, or death. Nurses' attempts to minimize false‐positive calls may help explain the low call rates for patients meeting RRT criteria. The 2 avenues to increase the PPV of criteria are:

• 1

  Increase the prevalence of disease in the population screened by risk factor stratification.

• 2

  Increase the specificity of the call criteria, which has been limited by the associated decrease in sensitivity.10

Regarding the efferent response limb of RRS, our case demonstrates that the primary team (rather than a separate group of caregivers), when alerted appropriately, can effectively respond to critical changes in patient status. Accordingly, our data show that since the inception of the program, cardiopulmonary arrests have decreased from a mean of 4.1 per month to 2.3 per month (P = 0.03).

Many clinical trials of RRTs would not capture the success demonstrated in this case. For example, due to the low prevalence of events, the MERIT trial used a composite endpoint that included unplanned ICU transfers, cardiac arrests, and mortality. Because our patient still required an unplanned ICU transfer after being evaluated by the responding team, she would have been counted as a system failure.

Conclusion

While local needs should inform the type of RRS implemented, this case illustrates one of the major obstacles ubiquitous to RRS implementation: failure of system activation. With appropriate activation, an RRS‐CTP can meet RRS goals while maintaining continuity of care and maximizing existing resources. This case also illustrates the difficulty of achieving a statistically relevant outcome, while demonstrating the potential benefits of evolving RRSs.

Many in‐hospital cardiac arrests and other adverse events are heralded by warning signs that are evident in the preceding 6 to 8 hours.1 By promptly intervening before further deterioration occurs, rapid response teams (RRTs) are designed to decrease unexpected intensive care unit (ICU) transfers, cardiac arrests, and inpatient mortality. While implementing RRTs is 1 of the 6 initiatives recommended by the Institute for Healthcare Improvement,2 data supporting their effectiveness is equivocal.3, 4

In October 2006, at Denver Health Medical Center, an academic, safety net hospital, we initiated a rapid response systemclinical triggers program (RRS‐CTP).5 In our RRS‐CTP, an abrupt change in patient status (Figure 1) triggers a mandatory call by the patient's nurse to the primary team, which is then required to perform an immediate bedside evaluation. By incorporating the primary team into the RRT‐CTP, we sought to preserve as much continuity of care as possible. Also, since the same house staff compose our cardiopulmonary arrest or cor team, and staff the ICUs and non‐ICU hospital wards, we did not feel that creating a separate RRT was an efficient use of resources. Our nurses have undergone extensive education about the necessity of a prompt bedside evaluation and have been instructed and empowered to escalate concerns to senior physicians if needed. We present a case that illustrates challenges to both implementing an RRS and measuring its potential benefits.

Figure 1

Case

A 59‐year‐old woman with a history of bipolar mood disorder was admitted for altered mental status. At presentation, she had signs of acute mania with normal vital signs. After initial laboratory workup, her altered mental status was felt to be multifactorial due to urinary tract infection, hypernatremia (attributed to lithium‐induced nephrogenic diabetes insipidus), and acute mania (attributed to medication discontinuation). Because she was slow to recover from the acute mania, her hospital stay was prolonged. From admission, the patient was treated with heparin 5000 units subcutaneously twice daily for venous thromboembolism prophylaxis.

On hospital day 7, at 21:32, the patient was noted to have asymptomatic tachycardia at 149 beats per minute and a new oxygen requirement of 3 L/minute. The cross‐cover team was called; however, although criteria were met, the RRS‐CTP was not activated and a bedside evaluation was not performed. A chest X‐ray was found to be normal and, with the exception of the oxygen requirement, her vital signs normalized by 23:45. No further diagnostic testing was performed at the time.

The next morning, at 11:58, the patient was found to have a blood pressure of 60/40 mmHg and heart rate of 42 beats per minute. The RRS‐CTP was activated. The primary team arrived at the bedside at 12:00 and found the patient to be alert, oriented, and without complaints. Her respiratory rate was 30/minute, and her oxygen saturation was 86% on 3 L/minute. An arterial blood gas analysis demonstrated acute respiratory alkalosis with hypoxemia and an electrocardiogram showed sinus tachycardia with a new S₁Q₃T₃ pattern. A computed tomography angiogram revealed a large, nearly occlusive pulmonary embolus (PE) filling an enlarged right pulmonary artery, as well as thrombus within the left main pulmonary artery. She was transferred to the medical ICU and alteplase was administered. The patient survived and was discharged in good clinical condition.

Discussion

Despite the strong theoretical benefit of the RRT concept, a recent review by Ranji et al.4 concluded that RRTs had not yet been shown to improve patient outcomes. In contrast to dedicated RRTs, this case illustrates a different type of RRS that was designed to address abrupt changes in patient status, while maintaining continuity of care and efficiently utilizing resources.

If one considers an RRS to have both afferent (criteria recognition) and efferent (RRT or primary team response) limbs, the afferent limb must be consistently activated in order to obtain the efferent limb's response.6 The greatest opportunities to improve RRSs are thought to lie in the afferent limb.3 Our RRS‐CTP was not triggered in 1 of 2 instances in which criteria for mandatory initiation of the system were met. This is consistent with the findings of the Medical Early Response Intervention and Therapy (MERIT) trial, in which RRTs were called in only 41% of the patients meeting criteria and subsequently having adverse events,7 and with the ongoing monitoring of the use of the system at our hospital. Had the cross‐covering team seen the patient at the bedside initially, the PE might have been diagnosed while the patient was hemodynamically stable, giving the patient nearly a 3‐fold lower relative mortality.8 When the RRS‐CTP was activated, a prompt bedside evaluation occurred, allowing for lytic therapy to be administered before cardiopulmonary arrest (attendant mortality of 90%).9

While rapid response criteria were originally based upon published sensitivity analyses, more recent studies suggest that these criteria lack diagnostic accuracy. As demonstrated by Cretikos et al,10 to reach a sensitivity of 70%, the corresponding specificity would be only 86%. Given that the prevalence of adverse events in the MERIT trial was only 0.6%, the resulting positive predictive value (PPV) of rapid response call criteria is 3%. Accordingly, 33 calls would be needed to prevent 1 unplanned ICU transfer, cardiac arrest, or death. Nurses' attempts to minimize false‐positive calls may help explain the low call rates for patients meeting RRT criteria. The 2 avenues to increase the PPV of criteria are:

• 1

  Increase the prevalence of disease in the population screened by risk factor stratification.

• 2

  Increase the specificity of the call criteria, which has been limited by the associated decrease in sensitivity.10

Regarding the efferent response limb of RRS, our case demonstrates that the primary team (rather than a separate group of caregivers), when alerted appropriately, can effectively respond to critical changes in patient status. Accordingly, our data show that since the inception of the program, cardiopulmonary arrests have decreased from a mean of 4.1 per month to 2.3 per month (P = 0.03).

Many clinical trials of RRTs would not capture the success demonstrated in this case. For example, due to the low prevalence of events, the MERIT trial used a composite endpoint that included unplanned ICU transfers, cardiac arrests, and mortality. Because our patient still required an unplanned ICU transfer after being evaluated by the responding team, she would have been counted as a system failure.

Conclusion

While local needs should inform the type of RRS implemented, this case illustrates one of the major obstacles ubiquitous to RRS implementation: failure of system activation. With appropriate activation, an RRS‐CTP can meet RRS goals while maintaining continuity of care and maximizing existing resources. This case also illustrates the difficulty of achieving a statistically relevant outcome, while demonstrating the potential benefits of evolving RRSs.

The prevalence of intimate partner violence (IPV; defined as mental and/or physical violence directed from 1 person in an intimate relationship to the other) varies widely, depending on the population sampled and method of data collection. In the United States, IPV against women, occurring within the year prior to contact with a healthcare professional, ranges from 2% to 15% in surveys done by telephone, in primary care clinics, or in face‐to‐face home interviews19 and from 10% to 30% in surveys of patients visiting urgent care or emergency departments.1012 The prevalence of IPV occurring at any time during the life of the patient ranges from 18% in the aforementioned settings to as high as 88% in women applying for welfare.1, 2, 4, 5, 10, 1214

Although reports indicate that victims of IPV are more likely to be hospitalized,1517 the only study assessing the prevalence of IPV in hospitalized patients included women on medical, surgical, and obstetrical services and reported 1‐year and lifetime prevalences of only 5% and 23%, respectively.18

We hypothesized that the prevalence of IPV in hospitalized patients would be at least as high as that reported from emergency departments and sought to measure the 1‐year and lifetime prevalences of IPV in women admitted to a general internal medicine service. In addition, because studies done in various outpatient settings have reported that victims of IPV have a variety of somatic complaints and an increased prevalence of chronic and functional illnesses,1923 we also sought to determine whether women with a history of IPV and women without a history of IPV had different numbers or types of positive responses to questions asked on the review of systems.

PATIENTS AND METHODS

This study was approved by the Colorado Multiple Institution Review Board, and informed consent was obtained from all participants.

Women between the ages of 18 and 60 who were admitted to the internal medicine floor service of Denver Health Medical Center (a university‐affiliated public safety‐net hospital) between January 1 and February 28, 2004 and between October 1 and October 30, 2004 were approached to participate. These dates were selected on the basis of the availability of our interviewers. Patients older than 60 were excluded to avoid overlap between IPV and the problem of elder abuse. Women were excluded if they were unable to give informed consent, were pregnant, were incarcerated, were on contact precautions, or spoke a language other than English or Spanish. Although IPV is common in pregnant women and may occur in women who are incarcerated, these are considered vulnerable populations with respect to obtaining approval from internal review boards.

The questionnaire consisted of 23 review‐of‐systems questions,24 4 questions adapted from a previously validated screen for IPV11 (Table 1), and 1 question about attempts to seek help (Table 1). Women were considered to have experienced IPV if they gave positive responses to any of the 4 questions targeting IPV. According to patient preference, the combined questionnaire was either read and filled out by each subject independently or was read to her by a female interviewer who then recorded the subject's verbal responses. All interviewers were women with a shared common concern about, and interest in, IPV. Although none had advanced training in psychology, social work, or other formal discipline that involved interviewing skills, all interviews were scripted so that interactions with subjects and completion of the questionnaires would be uniform. Responses indicating sometimes were considered to be positive. Responses that were not answered, left blank, or marked as not applicable were considered to be negative.

Each patient's medical record was reviewed to determine her age, race, number of previous hospital admissions, visits to the emergency department and walk‐in clinic, visits to primary care and subspecialty physicians, and whether the patient had been screened for IPV as recorded on the admission history and physical template. Admission diagnosis was obtained from the history and physical template, and the discharge diagnosis was obtained from the discharge paperwork. Functional diagnoses were considered to be symptoms (eg, shortness of breath) or problems (eg, constipation) that could not clearly be linked to a specific disease process. All participants were offered a card containing a list of resources for victims of IPV.

Data were analyzed with SAS 8.1 (SAS Institute, Cary, NC) and SPSS 11.5 (SPSS, Chicago, IL). The Student t test was used to compare continuous variables. Data are reported as means standard deviation. Chi‐square analysis was used to test associations between race, primary language, level of education, insurance status, admitting diagnosis, discharge diagnosis, number of previous hospital admissions, visit type, and the presence of IPV. For these, P < 0.05 was considered to be significant. The association of positive review‐of‐systems responses with the presence of IPV was also tested by chi‐square analysis, but P < 0.002 was considered to be significant on the basis of a Bonferroni adjustment for multiple comparisons. A receiver operating characteristic curve was used to assess the relationship between the number of positive responses to the questions included in the review of systems and a history of IPV. The odds ratio and confidence intervals were calculated to test the association between the number of positive responses to the review‐of‐systems questions and a lifetime history of IPV.

RESULTS

Throughout the dates of the study, 245 women were admitted to the internal medicine service, and 106 were excluded (Figure 1). Of the 139 eligible women, 78 were available to the interviewers and asked to participate, and 72 (92%) agreed. IPV occurring within the year prior to the interview or at any point in the patient's lifetime was reported by 16 (22%) and 44 (61%) subjects, respectively. No significant differences were seen in women who did or did not experience IPV at anytime in their life with respect to age, race, insurance status, education, number of scheduled outpatient, urgent, or emergent visits, or admission or discharge diagnosis even when the diagnoses were grouped into a functional category (although at best our study was powered to detect only >35% differences in prevalences; Tables 2 and 3). Of women reporting a lifetime history of IPV, 26 of 44 (59%) had previously sought help, and 9 of those 26 (35%) said that they sought help from a physician.

Figure 1

Women with a 1‐year history of IPV and women without a 1‐year history of IPV had 11.4 4.7 and 7.7 5.4 positive responses to the review of systems (P < 0.01), respectively. Women with a lifetime history of IPV and women without a lifetime history of IPV had 10.9 4.4 and 7.7 5.4 positive responses (P < 0.01), respectively. The receiver operating characteristic curve of the number of positive responses versus a lifetime history of IPV is presented in Figure 2. Subjects with 10 or more positive responses were 4.8 times more likely to report a lifetime history of IPV than subjects with 9 or fewer positive responses (confidence interval = 1.614.2, P = 0.003). The c‐statistic indicating the ability of the review of systems to properly classify cases when there were 10 or more positive responses was 0.692.

Figure 2

No differences were observed in the responses to the individual review of systems questions in women who did or did not have a lifetime history of IPV, with the exception that those with a positive history more commonly complained of difficulty sleeping and numbness and tingling in their hands or feet (although at best our study was sufficiently powered to detect only >20% differences in prevalences; Table 4). Although the sensitivity of having problems sleeping or experiencing numbness or tingling in patients with IPV was high, the specificity and positive and negative predictive values were not (Table 5).

The admission history forms filled out by first‐year admitting residents showed that only 18 (25%) of the women were screened for IPV, even though the history and physical examination template used at Denver Health Medical Center includes a prompt in the social history section pertaining to a history of violence as a reminder.

DISCUSSION

The important findings of this study were that women admitted to the internal medicine service of a university‐affiliated public safety‐net hospital had a high prevalence of IPV (22% and 61% 1‐year and lifetime prevalences, respectively), that most women with a history of IPV had previously sought help for the problem, many from physicians, that women were more likely to have a history of IPV if they had >10 positive responses to questions asked in a routine review of systems (particularly problems sleeping and experiencing numbness or tingling in their extremities), and that routine screening for IPV was uncommon at the time of admission.

These conclusions should be interpreted with respect to a number of limitations in our study. First, although our study was designed to be a consecutive series, the interviewers did not have sufficient time to meet with and interview every woman admitted before they were discharged. This occurred in part because the interviewers were available only for a portion of each day, some patients were discharged within 24 hours of admission, and many were out of their rooms for ancillary testing. Within the interviewers' time constraints, however, all hospitalized women meeting entry criteria who were available were approached. Our data could, however, overrepresent the prevalence of IPV if hospitalized women with a history of IPV had longer hospital stays than those who did not or if those experiencing IPV were out of their rooms less frequently (eg, for diagnostic tests). On the other hand, our data could underrepresent the true prevalence of IPV if patients with a history of IPV had shorter hospital stays or if they received more ancillary testing that caused them to be out of their rooms more frequently. Second, none of our interviewers had specific training in interviewing techniques. Accordingly, our data could have underestimated the true prevalence of IPV if interviewers with advanced training in probing sensitive topics had more success in eliciting positive responses. Third, the relationship between a history of IPV and multiple positive responses to the review of systems may be confounded if some of these patients also had a history of adverse childhood experiences or other experiences resulting in posttraumatic stress disorder as these patients also have an increased prevalence of chronic and functional disorders.2527 Finally, as our numbers were small, we were not powered to detect clinically important differences in demographics or specific positive answers on the review of systems.

To the best of our knowledge, the only study presenting IPV prevalence data in patients hospitalized for other than psychiatric problems was performed by McKenzie and colleagues18 in 1997. In their group of 130 patients (61 on internal medicine, 59 on surgery, 7 on obstetrics, and 3 on psychiatry), the 1‐year and lifetime prevalences of IPV were only 5% and 26%, respectively. McKenzie and colleagues used only 1 question to screen for IPV, but that single question incorporated 2 of the 4 questions used in our survey. Forty‐three of our 44 patients (98%) with a history of IPV were discovered on the basis of these 2 questions. The hospitals in which the 2 studies were done were similar, as were the ages and levels of education of the 2 populations studied and the percentage of eligible patients who agreed to participate. The patients in the 2 studies were different with respect to race, language mix, and the percentage who were insured, but neither study found differences in the prevalence of IPV as a function of race or insurance (although others have found an association of IPV with being uninsured1, 3, 4, 12, 23). Our study was conducted in women admitted exclusively to an internal medicine service, whereas nearly half of the patients studied by McKenzie and colleagues were admitted to surgical, gynecologic, or psychiatric services. Although McKenzie and colleagues found no difference in the prevalence of IPV as a function of admitting service, others have suggested that the prevalence of IPV is higher in patients admitted for trauma or psychiatric problems.1517, 28 The percentage of patients who self‐administered the questionnaires was 57% in our study and 77% in the study by McKenzie and colleagues. Neither study, however, found a difference in the percentage of IPV in patients who self‐administered the survey versus those who were interviewed. Women may have become more comfortable discussing this issue in the 10‐year interval between these 2 studies, or the prevalence of IPV may have increased. The only other study of IPV in hospitalized patients of which we are aware reported a 90% 1‐year prevalence in suicidal women admitted to a psychiatric service.28

Several studies have reported that victims of IPV have multiple somatic complaints and an increased prevalence of chronic and functional illnesses.1923 We confirmed that women experiencing IPV have more positive responses to questions posed in a review of systems, but the low specificity and positive and negative predictive values of the responses make this association of little clinical utility.

For only 18 of the 72 patients (25%) in our study was there evidence that they were screened for a history of IPV by the admitting resident. If more women were screened without a response being recorded, or if women were screened only for a current history of violence, our data may not accurately reflect the true rate at which screening occurred; however, the rate of screening that we observed is consistent with a number of other studies.12, 22, 2931 Fourteen of 18 patients who were screened for IPV by the resident gave negative responses. Ten of these, however, gave positive responses to our interviewers. Accordingly, the sensitivity, specificity, and positive and negative predictive values of the information recorded by the admitting resident were 40%, 100%, 100%, and 57%, respectively (assuming that the responses given to the IPV survey represent the gold standard), and this confirms that routine screening underestimates the prevalence of this problem. Accordingly, we identified 2 problems pertaining to screening for IPV: (1) it is not routinely done at the time of hospital admission, and (2) responses reported during routine screening are frequently incorrect. A number of barriers to routine screening have been previously identified, as have interventions designed to increase screening.32 Providing specific screening questions increases the identification of victims of IPV, but simply educating healthcare providers does not.32 Our history and physical templates have a prompt for violence victim to facilitate the screening, but as a result of this study, we are changing our prompting question and indicating what should be done if the response is positive.

The US Preventive Services Task Force and the Canadian Task Force on Preventive Health Care both concluded that there was insufficient evidence to recommend for or against routine screening for IPV.3335 Their rationale was that trials assessing the effectiveness of screening have not been published, that studies designed to assess the effectiveness of any resulting intervention are few in number, focused on pregnant women, and limited by problems in study design, that no studies have determined the accuracy of the screening tools, and that none have addressed the potential harm of screening.3335 The US Preventive Services Task Force did recommend screening if providers were concerned about IPV.34 Our data would suggest that there is little in the admission history that distinguishes women who might be victims of IPV from those who might not. Guidelines published by the American Medical Association, the American Academy of Family Physicians, and the American College of Obstetricians and Gynecologists promote routine screening of all patients.3638 Janssen and colleagues39 support the importance of screening on the basis that IPV is associated with numerous physical and mental health problems (eg, arthritis, migraines and other types of headaches, vaginal bleeding, ulcers, spastic colon, chronic pain, substance abuse, depression, and suicide ideation) and that establishing the link between these conditions and IPV could be important with respect to developing appropriate diagnostic and therapeutic approaches to patients' complaints. Screening also allows physicians to become more knowledgeable about their patients' lives, facilitating their ability to provide a supportive relationship that, in turn, increases women's likelihood of using an intervention method.39 We did not confirm an increased prevalence of any of the complaints noted by Janssen and colleagues in the women experiencing a history of IPV, but we did find an increased prevalence of insomnia and extremity numbness in women admitting to IPV as well as an overall increase in the number of positive responses to the review of systems. Screening identifies women who should receive information about reporting IPV, obtaining available assistance, planning for personal safety, and formal counseling as these have all been shown to reduce the severity of IPV and to improve the quality of life in rather large, randomized controlled trials.4043

As previously observed by others,13, 22, 29, 4446 the large majority of women that we approached welcomed screening for IPV. Over half of those with a history of IPV had previously sought help for the problem, over one‐third of these sought help from physicians, and most took the resource card that we offered, regardless of whether they did or did not have a history of IPV (this suggests either that our data may actually underestimate the true prevalence of IPV or that patients taking the information knew of others experiencing this problem). Accordingly, regardless of whether physicians believe that routine screening is warranted, patients see physicians and other healthcare workers as a resource for this problem.

We have confirmed that a history of IPV is very common in women admitted to an internal medicine service of a university‐affiliated public hospital and that female victims of IPV have more positive responses on the review of systems (particularly difficulty sleeping and extremity numbness or tingling) than those who have not. Although we initially hypothesized that finding numerous somatic complaints might serve as a marker for IPV, thereby identifying patients for whom more careful screening should occur, finding such a high prevalence of IPV argues that screening should be a routine part of the history for all women admitted to internal medicine inpatient services.

The prevalence of intimate partner violence (IPV; defined as mental and/or physical violence directed from 1 person in an intimate relationship to the other) varies widely, depending on the population sampled and method of data collection. In the United States, IPV against women, occurring within the year prior to contact with a healthcare professional, ranges from 2% to 15% in surveys done by telephone, in primary care clinics, or in face‐to‐face home interviews19 and from 10% to 30% in surveys of patients visiting urgent care or emergency departments.1012 The prevalence of IPV occurring at any time during the life of the patient ranges from 18% in the aforementioned settings to as high as 88% in women applying for welfare.1, 2, 4, 5, 10, 1214

Although reports indicate that victims of IPV are more likely to be hospitalized,1517 the only study assessing the prevalence of IPV in hospitalized patients included women on medical, surgical, and obstetrical services and reported 1‐year and lifetime prevalences of only 5% and 23%, respectively.18

We hypothesized that the prevalence of IPV in hospitalized patients would be at least as high as that reported from emergency departments and sought to measure the 1‐year and lifetime prevalences of IPV in women admitted to a general internal medicine service. In addition, because studies done in various outpatient settings have reported that victims of IPV have a variety of somatic complaints and an increased prevalence of chronic and functional illnesses,1923 we also sought to determine whether women with a history of IPV and women without a history of IPV had different numbers or types of positive responses to questions asked on the review of systems.

PATIENTS AND METHODS

This study was approved by the Colorado Multiple Institution Review Board, and informed consent was obtained from all participants.

Women between the ages of 18 and 60 who were admitted to the internal medicine floor service of Denver Health Medical Center (a university‐affiliated public safety‐net hospital) between January 1 and February 28, 2004 and between October 1 and October 30, 2004 were approached to participate. These dates were selected on the basis of the availability of our interviewers. Patients older than 60 were excluded to avoid overlap between IPV and the problem of elder abuse. Women were excluded if they were unable to give informed consent, were pregnant, were incarcerated, were on contact precautions, or spoke a language other than English or Spanish. Although IPV is common in pregnant women and may occur in women who are incarcerated, these are considered vulnerable populations with respect to obtaining approval from internal review boards.

The questionnaire consisted of 23 review‐of‐systems questions,24 4 questions adapted from a previously validated screen for IPV11 (Table 1), and 1 question about attempts to seek help (Table 1). Women were considered to have experienced IPV if they gave positive responses to any of the 4 questions targeting IPV. According to patient preference, the combined questionnaire was either read and filled out by each subject independently or was read to her by a female interviewer who then recorded the subject's verbal responses. All interviewers were women with a shared common concern about, and interest in, IPV. Although none had advanced training in psychology, social work, or other formal discipline that involved interviewing skills, all interviews were scripted so that interactions with subjects and completion of the questionnaires would be uniform. Responses indicating sometimes were considered to be positive. Responses that were not answered, left blank, or marked as not applicable were considered to be negative.

Each patient's medical record was reviewed to determine her age, race, number of previous hospital admissions, visits to the emergency department and walk‐in clinic, visits to primary care and subspecialty physicians, and whether the patient had been screened for IPV as recorded on the admission history and physical template. Admission diagnosis was obtained from the history and physical template, and the discharge diagnosis was obtained from the discharge paperwork. Functional diagnoses were considered to be symptoms (eg, shortness of breath) or problems (eg, constipation) that could not clearly be linked to a specific disease process. All participants were offered a card containing a list of resources for victims of IPV.

Data were analyzed with SAS 8.1 (SAS Institute, Cary, NC) and SPSS 11.5 (SPSS, Chicago, IL). The Student t test was used to compare continuous variables. Data are reported as means standard deviation. Chi‐square analysis was used to test associations between race, primary language, level of education, insurance status, admitting diagnosis, discharge diagnosis, number of previous hospital admissions, visit type, and the presence of IPV. For these, P < 0.05 was considered to be significant. The association of positive review‐of‐systems responses with the presence of IPV was also tested by chi‐square analysis, but P < 0.002 was considered to be significant on the basis of a Bonferroni adjustment for multiple comparisons. A receiver operating characteristic curve was used to assess the relationship between the number of positive responses to the questions included in the review of systems and a history of IPV. The odds ratio and confidence intervals were calculated to test the association between the number of positive responses to the review‐of‐systems questions and a lifetime history of IPV.

RESULTS

Throughout the dates of the study, 245 women were admitted to the internal medicine service, and 106 were excluded (Figure 1). Of the 139 eligible women, 78 were available to the interviewers and asked to participate, and 72 (92%) agreed. IPV occurring within the year prior to the interview or at any point in the patient's lifetime was reported by 16 (22%) and 44 (61%) subjects, respectively. No significant differences were seen in women who did or did not experience IPV at anytime in their life with respect to age, race, insurance status, education, number of scheduled outpatient, urgent, or emergent visits, or admission or discharge diagnosis even when the diagnoses were grouped into a functional category (although at best our study was powered to detect only >35% differences in prevalences; Tables 2 and 3). Of women reporting a lifetime history of IPV, 26 of 44 (59%) had previously sought help, and 9 of those 26 (35%) said that they sought help from a physician.

Figure 1

Women with a 1‐year history of IPV and women without a 1‐year history of IPV had 11.4 4.7 and 7.7 5.4 positive responses to the review of systems (P < 0.01), respectively. Women with a lifetime history of IPV and women without a lifetime history of IPV had 10.9 4.4 and 7.7 5.4 positive responses (P < 0.01), respectively. The receiver operating characteristic curve of the number of positive responses versus a lifetime history of IPV is presented in Figure 2. Subjects with 10 or more positive responses were 4.8 times more likely to report a lifetime history of IPV than subjects with 9 or fewer positive responses (confidence interval = 1.614.2, P = 0.003). The c‐statistic indicating the ability of the review of systems to properly classify cases when there were 10 or more positive responses was 0.692.

Figure 2

No differences were observed in the responses to the individual review of systems questions in women who did or did not have a lifetime history of IPV, with the exception that those with a positive history more commonly complained of difficulty sleeping and numbness and tingling in their hands or feet (although at best our study was sufficiently powered to detect only >20% differences in prevalences; Table 4). Although the sensitivity of having problems sleeping or experiencing numbness or tingling in patients with IPV was high, the specificity and positive and negative predictive values were not (Table 5).

The admission history forms filled out by first‐year admitting residents showed that only 18 (25%) of the women were screened for IPV, even though the history and physical examination template used at Denver Health Medical Center includes a prompt in the social history section pertaining to a history of violence as a reminder.

DISCUSSION

The important findings of this study were that women admitted to the internal medicine service of a university‐affiliated public safety‐net hospital had a high prevalence of IPV (22% and 61% 1‐year and lifetime prevalences, respectively), that most women with a history of IPV had previously sought help for the problem, many from physicians, that women were more likely to have a history of IPV if they had >10 positive responses to questions asked in a routine review of systems (particularly problems sleeping and experiencing numbness or tingling in their extremities), and that routine screening for IPV was uncommon at the time of admission.

These conclusions should be interpreted with respect to a number of limitations in our study. First, although our study was designed to be a consecutive series, the interviewers did not have sufficient time to meet with and interview every woman admitted before they were discharged. This occurred in part because the interviewers were available only for a portion of each day, some patients were discharged within 24 hours of admission, and many were out of their rooms for ancillary testing. Within the interviewers' time constraints, however, all hospitalized women meeting entry criteria who were available were approached. Our data could, however, overrepresent the prevalence of IPV if hospitalized women with a history of IPV had longer hospital stays than those who did not or if those experiencing IPV were out of their rooms less frequently (eg, for diagnostic tests). On the other hand, our data could underrepresent the true prevalence of IPV if patients with a history of IPV had shorter hospital stays or if they received more ancillary testing that caused them to be out of their rooms more frequently. Second, none of our interviewers had specific training in interviewing techniques. Accordingly, our data could have underestimated the true prevalence of IPV if interviewers with advanced training in probing sensitive topics had more success in eliciting positive responses. Third, the relationship between a history of IPV and multiple positive responses to the review of systems may be confounded if some of these patients also had a history of adverse childhood experiences or other experiences resulting in posttraumatic stress disorder as these patients also have an increased prevalence of chronic and functional disorders.2527 Finally, as our numbers were small, we were not powered to detect clinically important differences in demographics or specific positive answers on the review of systems.

To the best of our knowledge, the only study presenting IPV prevalence data in patients hospitalized for other than psychiatric problems was performed by McKenzie and colleagues18 in 1997. In their group of 130 patients (61 on internal medicine, 59 on surgery, 7 on obstetrics, and 3 on psychiatry), the 1‐year and lifetime prevalences of IPV were only 5% and 26%, respectively. McKenzie and colleagues used only 1 question to screen for IPV, but that single question incorporated 2 of the 4 questions used in our survey. Forty‐three of our 44 patients (98%) with a history of IPV were discovered on the basis of these 2 questions. The hospitals in which the 2 studies were done were similar, as were the ages and levels of education of the 2 populations studied and the percentage of eligible patients who agreed to participate. The patients in the 2 studies were different with respect to race, language mix, and the percentage who were insured, but neither study found differences in the prevalence of IPV as a function of race or insurance (although others have found an association of IPV with being uninsured1, 3, 4, 12, 23). Our study was conducted in women admitted exclusively to an internal medicine service, whereas nearly half of the patients studied by McKenzie and colleagues were admitted to surgical, gynecologic, or psychiatric services. Although McKenzie and colleagues found no difference in the prevalence of IPV as a function of admitting service, others have suggested that the prevalence of IPV is higher in patients admitted for trauma or psychiatric problems.1517, 28 The percentage of patients who self‐administered the questionnaires was 57% in our study and 77% in the study by McKenzie and colleagues. Neither study, however, found a difference in the percentage of IPV in patients who self‐administered the survey versus those who were interviewed. Women may have become more comfortable discussing this issue in the 10‐year interval between these 2 studies, or the prevalence of IPV may have increased. The only other study of IPV in hospitalized patients of which we are aware reported a 90% 1‐year prevalence in suicidal women admitted to a psychiatric service.28

Several studies have reported that victims of IPV have multiple somatic complaints and an increased prevalence of chronic and functional illnesses.1923 We confirmed that women experiencing IPV have more positive responses to questions posed in a review of systems, but the low specificity and positive and negative predictive values of the responses make this association of little clinical utility.

For only 18 of the 72 patients (25%) in our study was there evidence that they were screened for a history of IPV by the admitting resident. If more women were screened without a response being recorded, or if women were screened only for a current history of violence, our data may not accurately reflect the true rate at which screening occurred; however, the rate of screening that we observed is consistent with a number of other studies.12, 22, 2931 Fourteen of 18 patients who were screened for IPV by the resident gave negative responses. Ten of these, however, gave positive responses to our interviewers. Accordingly, the sensitivity, specificity, and positive and negative predictive values of the information recorded by the admitting resident were 40%, 100%, 100%, and 57%, respectively (assuming that the responses given to the IPV survey represent the gold standard), and this confirms that routine screening underestimates the prevalence of this problem. Accordingly, we identified 2 problems pertaining to screening for IPV: (1) it is not routinely done at the time of hospital admission, and (2) responses reported during routine screening are frequently incorrect. A number of barriers to routine screening have been previously identified, as have interventions designed to increase screening.32 Providing specific screening questions increases the identification of victims of IPV, but simply educating healthcare providers does not.32 Our history and physical templates have a prompt for violence victim to facilitate the screening, but as a result of this study, we are changing our prompting question and indicating what should be done if the response is positive.

The US Preventive Services Task Force and the Canadian Task Force on Preventive Health Care both concluded that there was insufficient evidence to recommend for or against routine screening for IPV.3335 Their rationale was that trials assessing the effectiveness of screening have not been published, that studies designed to assess the effectiveness of any resulting intervention are few in number, focused on pregnant women, and limited by problems in study design, that no studies have determined the accuracy of the screening tools, and that none have addressed the potential harm of screening.3335 The US Preventive Services Task Force did recommend screening if providers were concerned about IPV.34 Our data would suggest that there is little in the admission history that distinguishes women who might be victims of IPV from those who might not. Guidelines published by the American Medical Association, the American Academy of Family Physicians, and the American College of Obstetricians and Gynecologists promote routine screening of all patients.3638 Janssen and colleagues39 support the importance of screening on the basis that IPV is associated with numerous physical and mental health problems (eg, arthritis, migraines and other types of headaches, vaginal bleeding, ulcers, spastic colon, chronic pain, substance abuse, depression, and suicide ideation) and that establishing the link between these conditions and IPV could be important with respect to developing appropriate diagnostic and therapeutic approaches to patients' complaints. Screening also allows physicians to become more knowledgeable about their patients' lives, facilitating their ability to provide a supportive relationship that, in turn, increases women's likelihood of using an intervention method.39 We did not confirm an increased prevalence of any of the complaints noted by Janssen and colleagues in the women experiencing a history of IPV, but we did find an increased prevalence of insomnia and extremity numbness in women admitting to IPV as well as an overall increase in the number of positive responses to the review of systems. Screening identifies women who should receive information about reporting IPV, obtaining available assistance, planning for personal safety, and formal counseling as these have all been shown to reduce the severity of IPV and to improve the quality of life in rather large, randomized controlled trials.4043

As previously observed by others,13, 22, 29, 4446 the large majority of women that we approached welcomed screening for IPV. Over half of those with a history of IPV had previously sought help for the problem, over one‐third of these sought help from physicians, and most took the resource card that we offered, regardless of whether they did or did not have a history of IPV (this suggests either that our data may actually underestimate the true prevalence of IPV or that patients taking the information knew of others experiencing this problem). Accordingly, regardless of whether physicians believe that routine screening is warranted, patients see physicians and other healthcare workers as a resource for this problem.

We have confirmed that a history of IPV is very common in women admitted to an internal medicine service of a university‐affiliated public hospital and that female victims of IPV have more positive responses on the review of systems (particularly difficulty sleeping and extremity numbness or tingling) than those who have not. Although we initially hypothesized that finding numerous somatic complaints might serve as a marker for IPV, thereby identifying patients for whom more careful screening should occur, finding such a high prevalence of IPV argues that screening should be a routine part of the history for all women admitted to internal medicine inpatient services.

The prevalence of intimate partner violence (IPV; defined as mental and/or physical violence directed from 1 person in an intimate relationship to the other) varies widely, depending on the population sampled and method of data collection. In the United States, IPV against women, occurring within the year prior to contact with a healthcare professional, ranges from 2% to 15% in surveys done by telephone, in primary care clinics, or in face‐to‐face home interviews19 and from 10% to 30% in surveys of patients visiting urgent care or emergency departments.1012 The prevalence of IPV occurring at any time during the life of the patient ranges from 18% in the aforementioned settings to as high as 88% in women applying for welfare.1, 2, 4, 5, 10, 1214

Although reports indicate that victims of IPV are more likely to be hospitalized,1517 the only study assessing the prevalence of IPV in hospitalized patients included women on medical, surgical, and obstetrical services and reported 1‐year and lifetime prevalences of only 5% and 23%, respectively.18

We hypothesized that the prevalence of IPV in hospitalized patients would be at least as high as that reported from emergency departments and sought to measure the 1‐year and lifetime prevalences of IPV in women admitted to a general internal medicine service. In addition, because studies done in various outpatient settings have reported that victims of IPV have a variety of somatic complaints and an increased prevalence of chronic and functional illnesses,1923 we also sought to determine whether women with a history of IPV and women without a history of IPV had different numbers or types of positive responses to questions asked on the review of systems.

PATIENTS AND METHODS

This study was approved by the Colorado Multiple Institution Review Board, and informed consent was obtained from all participants.

Women between the ages of 18 and 60 who were admitted to the internal medicine floor service of Denver Health Medical Center (a university‐affiliated public safety‐net hospital) between January 1 and February 28, 2004 and between October 1 and October 30, 2004 were approached to participate. These dates were selected on the basis of the availability of our interviewers. Patients older than 60 were excluded to avoid overlap between IPV and the problem of elder abuse. Women were excluded if they were unable to give informed consent, were pregnant, were incarcerated, were on contact precautions, or spoke a language other than English or Spanish. Although IPV is common in pregnant women and may occur in women who are incarcerated, these are considered vulnerable populations with respect to obtaining approval from internal review boards.

The questionnaire consisted of 23 review‐of‐systems questions,24 4 questions adapted from a previously validated screen for IPV11 (Table 1), and 1 question about attempts to seek help (Table 1). Women were considered to have experienced IPV if they gave positive responses to any of the 4 questions targeting IPV. According to patient preference, the combined questionnaire was either read and filled out by each subject independently or was read to her by a female interviewer who then recorded the subject's verbal responses. All interviewers were women with a shared common concern about, and interest in, IPV. Although none had advanced training in psychology, social work, or other formal discipline that involved interviewing skills, all interviews were scripted so that interactions with subjects and completion of the questionnaires would be uniform. Responses indicating sometimes were considered to be positive. Responses that were not answered, left blank, or marked as not applicable were considered to be negative.

Each patient's medical record was reviewed to determine her age, race, number of previous hospital admissions, visits to the emergency department and walk‐in clinic, visits to primary care and subspecialty physicians, and whether the patient had been screened for IPV as recorded on the admission history and physical template. Admission diagnosis was obtained from the history and physical template, and the discharge diagnosis was obtained from the discharge paperwork. Functional diagnoses were considered to be symptoms (eg, shortness of breath) or problems (eg, constipation) that could not clearly be linked to a specific disease process. All participants were offered a card containing a list of resources for victims of IPV.

Data were analyzed with SAS 8.1 (SAS Institute, Cary, NC) and SPSS 11.5 (SPSS, Chicago, IL). The Student t test was used to compare continuous variables. Data are reported as means standard deviation. Chi‐square analysis was used to test associations between race, primary language, level of education, insurance status, admitting diagnosis, discharge diagnosis, number of previous hospital admissions, visit type, and the presence of IPV. For these, P < 0.05 was considered to be significant. The association of positive review‐of‐systems responses with the presence of IPV was also tested by chi‐square analysis, but P < 0.002 was considered to be significant on the basis of a Bonferroni adjustment for multiple comparisons. A receiver operating characteristic curve was used to assess the relationship between the number of positive responses to the questions included in the review of systems and a history of IPV. The odds ratio and confidence intervals were calculated to test the association between the number of positive responses to the review‐of‐systems questions and a lifetime history of IPV.

RESULTS

Throughout the dates of the study, 245 women were admitted to the internal medicine service, and 106 were excluded (Figure 1). Of the 139 eligible women, 78 were available to the interviewers and asked to participate, and 72 (92%) agreed. IPV occurring within the year prior to the interview or at any point in the patient's lifetime was reported by 16 (22%) and 44 (61%) subjects, respectively. No significant differences were seen in women who did or did not experience IPV at anytime in their life with respect to age, race, insurance status, education, number of scheduled outpatient, urgent, or emergent visits, or admission or discharge diagnosis even when the diagnoses were grouped into a functional category (although at best our study was powered to detect only >35% differences in prevalences; Tables 2 and 3). Of women reporting a lifetime history of IPV, 26 of 44 (59%) had previously sought help, and 9 of those 26 (35%) said that they sought help from a physician.

Figure 1

Women with a 1‐year history of IPV and women without a 1‐year history of IPV had 11.4 4.7 and 7.7 5.4 positive responses to the review of systems (P < 0.01), respectively. Women with a lifetime history of IPV and women without a lifetime history of IPV had 10.9 4.4 and 7.7 5.4 positive responses (P < 0.01), respectively. The receiver operating characteristic curve of the number of positive responses versus a lifetime history of IPV is presented in Figure 2. Subjects with 10 or more positive responses were 4.8 times more likely to report a lifetime history of IPV than subjects with 9 or fewer positive responses (confidence interval = 1.614.2, P = 0.003). The c‐statistic indicating the ability of the review of systems to properly classify cases when there were 10 or more positive responses was 0.692.

Figure 2

No differences were observed in the responses to the individual review of systems questions in women who did or did not have a lifetime history of IPV, with the exception that those with a positive history more commonly complained of difficulty sleeping and numbness and tingling in their hands or feet (although at best our study was sufficiently powered to detect only >20% differences in prevalences; Table 4). Although the sensitivity of having problems sleeping or experiencing numbness or tingling in patients with IPV was high, the specificity and positive and negative predictive values were not (Table 5).