Original Research

Pneumonia is a common clinical syndrome with well‐described epidemiology and microbiology. Aspiration pneumonia comprises 5% to 15% of patients with pneumonia acquired outside of the hospital,[1] but is less well characterized despite being a major syndrome of pneumonia in the elderly.[2, 3] Difficulties in studying aspiration pneumonia include the lack of a sensitive and specific marker for aspiration as well as the potential overlap between aspiration pneumonia and other forms of pneumonia.[4, 5, 6] Additionally, clinicians have difficulty distinguishing between aspiration pneumonia, which develops after the aspiration of oropharyngeal contents, and aspiration pneumonitis, wherein inhalation of gastric contents causes inflammation without the subsequent development of bacterial infection.[7, 8] Central to the study of aspiration pneumonia is whether it should exist as its own entity, or if aspiration is really a designation used for pneumonia in an older patient with greater comorbidities. The ability to clearly understand how a clinician diagnoses aspiration pneumonia, and whether that method has face validity with expert definitions may allow for improved future research, improved generalizability of current or past research, and possibly better clinical care.

Several validated mortality prediction models exist for community‐acquired pneumonia (CAP) using a variety of clinical predictors, but their performance in patients with aspiration pneumonia is less well characterized. Most studies validating pneumonia severity scoring systems excluded aspiration pneumonia from their study population.[9, 10, 11] Severity scoring systems for CAP may not accurately predict disease severity in patients with aspiration pneumonia. The CURB‐65[9] (confusion, uremia, respiratory rate, blood pressure, age 65 years) and the eCURB[12] scoring systems are poor predictors of mortality in patients with aspiration pneumonia, perhaps because they do not account for patient comorbidities.[13] The pneumonia severity index (PSI)[10] might predict mortality better than CURB‐65 in the aspiration population due to the inclusion of comorbidities.

Previous studies have demonstrated that patients with aspiration pneumonia are older and have greater disease severity and more comorbidities.[13, 14, 15] These single‐center studies also demonstrated greater mortality, more frequent admission to an intensive care unit (ICU), and longer hospital lengths of stay in patients with aspiration pneumonia. These studies identified aspiration pneumonia by the presence of a risk factor for aspiration[15] or by physician billing codes.[13] In practice, however, the bedside clinician diagnoses a patient as having aspiration pneumonia, but the logic is likely vague and inconsistent. Despite the potential for variability with individual judgment, an aggregate estimation from independent judgments may perform better than individual judgments.[16] Because there is no gold standard for defining aspiration pneumonia, all previous research has been limited to definitions created by investigators. This multicenter study seeks to determine what clinical characteristics lead physicians to diagnose a patient as having aspiration pneumonia, and whether or not the clinician‐derived diagnosis is distinct and clinically useful.

Our objectives were to: (1) identify covariates associated with bedside clinicians diagnosing a pneumonia patient as having aspiration pneumonia; (2) compare aspiration pneumonia and nonaspiration pneumonia in regard to disease severity, patient demographics, comorbidities, and clinical outcomes; and (3) measure the performance of the PSI in aspiration pneumonia versus nonaspiration pneumonia.

PATIENTS AND METHODS

RESULTS

Our initial query, after exclusion criteria, yielded a study population of 5185 patients (Figure 1). We compared 451 patients diagnosed with aspiration pneumonia to 4734 with CAP (Figure 1). Patient characteristics are summarized in Table 1. Patients with aspiration pneumonia were older, more likely to live in a nursing home, had greater disease severity, and were more likely to be admitted to an ICU. Patients with aspiration pneumonia had longer adjusted hospital lengths of stay and took more days to achieve clinical stability than patients with nonaspiration pneumonia (Figure 2). After adjusting for all variables in Table 1, the Cox proportional hazards models demonstrated that aspiration pneumonia was associated with ongoing hospitalization (hazard ratio [HR] for discharge: 0.77, 95% confidence interval [CI]: 0.65‐0.91, P=0.002) and clinical instability (HR for attaining clinical stability: 0.72, 95% CI: 0.61‐0.84, P<0.001). Patients with aspiration pneumonia presented with greater disease severity than those with nonaspiration pneumonia. Although there was no difference between groups in regard to temperature, respiratory rate, hyponatremia, or presence of pleural effusions or cavitary lesions, all other measured indices of disease severity were worse in patients with aspiration pneumonia. Patients with aspiration pneumonia were more likely to have cerebrovascular disease than those with nonaspiration pneumonia. Aspiration pneumonia patients also had increased prevalence of congestive heart failure. There was no appreciable difference between groups among other measured comorbidities.

Figure 1

Figure 2

The patient characteristics most associated with a physician diagnosis of aspiration pneumonia, identified using logistic regression, were confusion, residence in nursing home, and presence of cerebrovascular disease (odds ratio [OR]: of 4.4, 2.9, and 2.3, respectively), whereas renal disease was associated with decreased physician diagnosis of aspiration pneumonia over nonaspiration pneumonia (OR: 0.58) (Table 2).

Observed in‐patient mortality of aspiration pneumonia was 23%. This mortality was considerably higher than a mean PSI score of 123 would predict (class IV risk group, with expected 30‐day mortality of 8%9%[25]). The PSI score's ability to predict inpatient mortality in patients with aspiration pneumonia was moderate, with an area under the curve (AUC) of 0.71. This was similar to its performance in patients with nonaspiration pneumonia (AUC of 0.75) (Figure 3). These values are lower than the AUC of 0.81 for the PSI in predicting mortality derived from a meta‐analysis of 31 other studies.[26]

Figure 3

Our regression model after propensity score matching demonstrated that aspiration pneumonia independently confers a 2.3‐fold increased odds for inpatient mortality (95% CI: 1.56‐3.45, P<0.001).

DISCUSSION

Pneumonia patients with confusion, nursing home residence, or cerebrovascular disease are more likely to be diagnosed with aspiration pneumonia by clinicians. Although this is unsurprising, it is notable that these patients are more than twice as likely to die in the inpatient setting, even after accounting for age, comorbidities, and disease severity. These findings are similar to three previously published studies comparing aspiration and nonaspiration pneumonia at single institutions, albeit using different aspiration pneumonia definitions.[13, 14, 15] This study is the first large, multicenter, multinational study to demonstrate these findings.

Central to the interpretation of our results is the method of diagnosing aspiration versus nonaspiration. A bottom‐up method that relies on a clinician to check a box for aspiration may appear poorly reproducible. Because there is no diagnostic gold standard, clinicians may use different criteria to diagnose aspiration, creating potential for idiosyncratic noise. The strength of the wisdom of the crowd method used in this study is that an aggregate estimation from independent judgments may reduce the noise from individual judgments.[16] Although clinicians may vary in why they diagnose a particular patient as having aspiration pneumonia, it appears that the overwhelming reason for diagnosing a patient as having aspiration pneumonia is the presence of confusion, followed by previous nursing home residence or cerebrovascular disease. This finding has some face validity when compared with studies using an investigator definition, as altered mental status, chronic debility, and cerebrovascular disease are either prominent features of the definition of aspiration pneumonia[8] or frequently observed in patients with aspiration pneumonia.[13, 15] The distribution of cerebrovascular disease among our study's aspiration and nonaspiration pneumonia patients was similar to studies that used formal criteria in their definitions.[13, 15] Although nursing home residence was more likely in aspiration pneumonia patients, the majority of aspiration pneumonia patients were residing in the community, suggesting that aspiration is not simply a surrogate for healthcare‐associated pneumonia. Although patients with aspiration pneumonia are typically older than their nonaspiration counterparts, it appears that age is not a key determinant in the diagnosis of aspiration. With aspiration pneumonia, confusion, nursing home residence, and the presence of cerebrovascular disease are the greatest contributors in the clinical diagnosis, more than age.

Our data demonstrate that aspiration pneumonia confers increased odds for mortality, even after adjustment for age, disease severity, and comorbidities. These data suggest that aspiration pneumonia is a distinct entity from nonaspiration pneumonia, and that this disease is worse than nonaspiration CAP. If aspiration pneumonia is distinct from nonaspiration pneumonia, some unrecognized host factor other than age, disease severity, or the captured comorbidities decreases survival in aspiration pneumonia patients. However, it is also possible that aspiration pneumonia is merely a clinical designation for one end of the pneumonia spectrum, and we and others have failed to completely account for all measures of disease severity or all measures of comorbidities. Examples of unmeasured comorbidities would include presence of oropharyngeal dysphagia, which is not assessed in the database but could have a significant effect on clinical diagnosis. Unmeasured covariates can include measures beyond that of disease severity or comorbidity, such as the presence of a do not resuscitate (DNR) order, which could have a significant confounding effect on the observed association. A previous, single‐center study demonstrated that increased 30‐day mortality in aspiration pneumonia was mostly attributable to greater disease severity and comorbidities, although aspiration pneumonia independently conferred greater risk for adverse long‐term outcomes.[15] We propose that aspiration pneumonia represents a clinically distinct entity from nonaspiration pneumonia. Patients with chronic aspiration are often chronically malnourished and may have different oral flora than patients without chronic aspiration.[27, 28] Chronic aspiration has been associated with granulomatous reaction, organizing pneumonia, diffuse alveolar damage, and chronic bronchiolitis.[29] Chronic aspiration may elicit changes in the host physiology, and may render the host more susceptible to the development of secondary bacterial infection with morbid consequences.

The ability of the PSI to predict inpatient mortality was moderate (AUC only 0.7), with no significant additional discrimination between the aspiration and nonaspiration pneumonia groups. Although the PSI had moderate ability to predict inpatient mortality, the observed mortality was considerably higher than predicted. It is possible that the PSI incompletely captures clinically relevant comorbidities (eg, malnutrition). Further study to improve mortality prediction of aspiration pneumonia patients could employ sensitivity analysis to determine optimal thresholds and weighting of the PSI components.

Patients with aspiration pneumonia had longer hospital lengths of stay and took longer to achieve clinical stability than their nonaspiration counterparts. Time to clinical stability has been associated with increased posthospitalization mortality and is associated with time to switch from intravenous to oral antibiotics.[17] Although some component of hospital length‐of‐stay is subject to local practice patterns, time to clinical stability has explicit criteria for clinical improvement and failure, and therefore is less likely to be affected by local practice patterns.

We noted a relatively high (16%21%) incidence of prior antibiotic use among patients in this database. Analysis of antibiotic prescription patterns was limited, given the several different countries from which the database draws its cases. Although we used accepted criteria to define CAP cases, it is possible that this population may have a higher rate of resistant or uncommon pathogens than other studies of CAP that have populations with lower incidence of prior antibiotic use. Although not assessed, we suspect a significant component of the prior antibiotic use represented outpatient pneumonia treatment during the few days prior to visiting the hospital.

This study has several limitations, of which the most important may be that we used clinical determination for defining presence of aspiration pneumonia. This method is susceptible to the subjective perceptions of the treating clinician. We did not account for the effect of individual physicians in our model, although we did adjust for regional differences. The retrospective identification of patients allows for the possibility of selection bias, and therefore we have not attempted to make inferences regarding the relative incidence of pneumonia, nor did we adjust for temporal trends in diagnosis. The ratio of aspiration pneumonia patients to nonaspiration pneumonia patients may not necessarily reflect that observed in reality. Microbiologic and antibiotic data were unavailable for analysis. This study cannot inform on nonhospitalized patients with aspiration pneumonia, as only hospitalized patients were enrolled. The database identified cases of pneumonia, so it is possible for a patient to enter into the database more than once. Detection of mortality was limited to the inpatient setting rather than a set interval of 30 days. Inpatient mortality depends on length‐of‐stay patterns that may bias the mortality endpoint.[30] Also not assessed was the presence of a DNR order. It is possible that an older patient with greater comorbidities and disease severity may have care intentionally limited or withdrawn early by the family or clinicians.

Strengths of the study include its size and its multicenter, multinational population. The CAPO database is a large and well‐described population of patients with CAP.[17, 31] These attributes, as well as the clinician‐determined diagnosis, increase the generalizability of the study compared to a single‐center, single‐country study that employs investigator‐defined criteria.

CONCLUSION

Pneumonia patients with confusion, who are nursing home residence, and have cerebrovascular disease are more likely to be diagnosed with aspiration pneumonia by clinicians. Our clinician‐diagnosed cohort appears similar to those derived using an investigator definition. Patients with aspiration pneumonia are older, and have greater disease severity and more comorbidities than patients with nonaspiration pneumonia. They have greater mortality than their PSI score class would predict. Even after accounting for age, disease severity, and comorbidities, the presence of aspiration pneumonia independently conferred a greater than 2‐fold increase in inpatient mortality. These findings together suggest that aspiration pneumonia should be considered a distinct entity from typical pneumonia, and that additional research should be done in this field.

Pneumonia is a common clinical syndrome with well‐described epidemiology and microbiology. Aspiration pneumonia comprises 5% to 15% of patients with pneumonia acquired outside of the hospital,[1] but is less well characterized despite being a major syndrome of pneumonia in the elderly.[2, 3] Difficulties in studying aspiration pneumonia include the lack of a sensitive and specific marker for aspiration as well as the potential overlap between aspiration pneumonia and other forms of pneumonia.[4, 5, 6] Additionally, clinicians have difficulty distinguishing between aspiration pneumonia, which develops after the aspiration of oropharyngeal contents, and aspiration pneumonitis, wherein inhalation of gastric contents causes inflammation without the subsequent development of bacterial infection.[7, 8] Central to the study of aspiration pneumonia is whether it should exist as its own entity, or if aspiration is really a designation used for pneumonia in an older patient with greater comorbidities. The ability to clearly understand how a clinician diagnoses aspiration pneumonia, and whether that method has face validity with expert definitions may allow for improved future research, improved generalizability of current or past research, and possibly better clinical care.

Several validated mortality prediction models exist for community‐acquired pneumonia (CAP) using a variety of clinical predictors, but their performance in patients with aspiration pneumonia is less well characterized. Most studies validating pneumonia severity scoring systems excluded aspiration pneumonia from their study population.[9, 10, 11] Severity scoring systems for CAP may not accurately predict disease severity in patients with aspiration pneumonia. The CURB‐65[9] (confusion, uremia, respiratory rate, blood pressure, age 65 years) and the eCURB[12] scoring systems are poor predictors of mortality in patients with aspiration pneumonia, perhaps because they do not account for patient comorbidities.[13] The pneumonia severity index (PSI)[10] might predict mortality better than CURB‐65 in the aspiration population due to the inclusion of comorbidities.

Previous studies have demonstrated that patients with aspiration pneumonia are older and have greater disease severity and more comorbidities.[13, 14, 15] These single‐center studies also demonstrated greater mortality, more frequent admission to an intensive care unit (ICU), and longer hospital lengths of stay in patients with aspiration pneumonia. These studies identified aspiration pneumonia by the presence of a risk factor for aspiration[15] or by physician billing codes.[13] In practice, however, the bedside clinician diagnoses a patient as having aspiration pneumonia, but the logic is likely vague and inconsistent. Despite the potential for variability with individual judgment, an aggregate estimation from independent judgments may perform better than individual judgments.[16] Because there is no gold standard for defining aspiration pneumonia, all previous research has been limited to definitions created by investigators. This multicenter study seeks to determine what clinical characteristics lead physicians to diagnose a patient as having aspiration pneumonia, and whether or not the clinician‐derived diagnosis is distinct and clinically useful.

Our objectives were to: (1) identify covariates associated with bedside clinicians diagnosing a pneumonia patient as having aspiration pneumonia; (2) compare aspiration pneumonia and nonaspiration pneumonia in regard to disease severity, patient demographics, comorbidities, and clinical outcomes; and (3) measure the performance of the PSI in aspiration pneumonia versus nonaspiration pneumonia.

PATIENTS AND METHODS

RESULTS

Our initial query, after exclusion criteria, yielded a study population of 5185 patients (Figure 1). We compared 451 patients diagnosed with aspiration pneumonia to 4734 with CAP (Figure 1). Patient characteristics are summarized in Table 1. Patients with aspiration pneumonia were older, more likely to live in a nursing home, had greater disease severity, and were more likely to be admitted to an ICU. Patients with aspiration pneumonia had longer adjusted hospital lengths of stay and took more days to achieve clinical stability than patients with nonaspiration pneumonia (Figure 2). After adjusting for all variables in Table 1, the Cox proportional hazards models demonstrated that aspiration pneumonia was associated with ongoing hospitalization (hazard ratio [HR] for discharge: 0.77, 95% confidence interval [CI]: 0.65‐0.91, P=0.002) and clinical instability (HR for attaining clinical stability: 0.72, 95% CI: 0.61‐0.84, P<0.001). Patients with aspiration pneumonia presented with greater disease severity than those with nonaspiration pneumonia. Although there was no difference between groups in regard to temperature, respiratory rate, hyponatremia, or presence of pleural effusions or cavitary lesions, all other measured indices of disease severity were worse in patients with aspiration pneumonia. Patients with aspiration pneumonia were more likely to have cerebrovascular disease than those with nonaspiration pneumonia. Aspiration pneumonia patients also had increased prevalence of congestive heart failure. There was no appreciable difference between groups among other measured comorbidities.

Figure 1

Figure 2

The patient characteristics most associated with a physician diagnosis of aspiration pneumonia, identified using logistic regression, were confusion, residence in nursing home, and presence of cerebrovascular disease (odds ratio [OR]: of 4.4, 2.9, and 2.3, respectively), whereas renal disease was associated with decreased physician diagnosis of aspiration pneumonia over nonaspiration pneumonia (OR: 0.58) (Table 2).

Observed in‐patient mortality of aspiration pneumonia was 23%. This mortality was considerably higher than a mean PSI score of 123 would predict (class IV risk group, with expected 30‐day mortality of 8%9%[25]). The PSI score's ability to predict inpatient mortality in patients with aspiration pneumonia was moderate, with an area under the curve (AUC) of 0.71. This was similar to its performance in patients with nonaspiration pneumonia (AUC of 0.75) (Figure 3). These values are lower than the AUC of 0.81 for the PSI in predicting mortality derived from a meta‐analysis of 31 other studies.[26]

Figure 3

Our regression model after propensity score matching demonstrated that aspiration pneumonia independently confers a 2.3‐fold increased odds for inpatient mortality (95% CI: 1.56‐3.45, P<0.001).

DISCUSSION

Pneumonia patients with confusion, nursing home residence, or cerebrovascular disease are more likely to be diagnosed with aspiration pneumonia by clinicians. Although this is unsurprising, it is notable that these patients are more than twice as likely to die in the inpatient setting, even after accounting for age, comorbidities, and disease severity. These findings are similar to three previously published studies comparing aspiration and nonaspiration pneumonia at single institutions, albeit using different aspiration pneumonia definitions.[13, 14, 15] This study is the first large, multicenter, multinational study to demonstrate these findings.

Central to the interpretation of our results is the method of diagnosing aspiration versus nonaspiration. A bottom‐up method that relies on a clinician to check a box for aspiration may appear poorly reproducible. Because there is no diagnostic gold standard, clinicians may use different criteria to diagnose aspiration, creating potential for idiosyncratic noise. The strength of the wisdom of the crowd method used in this study is that an aggregate estimation from independent judgments may reduce the noise from individual judgments.[16] Although clinicians may vary in why they diagnose a particular patient as having aspiration pneumonia, it appears that the overwhelming reason for diagnosing a patient as having aspiration pneumonia is the presence of confusion, followed by previous nursing home residence or cerebrovascular disease. This finding has some face validity when compared with studies using an investigator definition, as altered mental status, chronic debility, and cerebrovascular disease are either prominent features of the definition of aspiration pneumonia[8] or frequently observed in patients with aspiration pneumonia.[13, 15] The distribution of cerebrovascular disease among our study's aspiration and nonaspiration pneumonia patients was similar to studies that used formal criteria in their definitions.[13, 15] Although nursing home residence was more likely in aspiration pneumonia patients, the majority of aspiration pneumonia patients were residing in the community, suggesting that aspiration is not simply a surrogate for healthcare‐associated pneumonia. Although patients with aspiration pneumonia are typically older than their nonaspiration counterparts, it appears that age is not a key determinant in the diagnosis of aspiration. With aspiration pneumonia, confusion, nursing home residence, and the presence of cerebrovascular disease are the greatest contributors in the clinical diagnosis, more than age.

Our data demonstrate that aspiration pneumonia confers increased odds for mortality, even after adjustment for age, disease severity, and comorbidities. These data suggest that aspiration pneumonia is a distinct entity from nonaspiration pneumonia, and that this disease is worse than nonaspiration CAP. If aspiration pneumonia is distinct from nonaspiration pneumonia, some unrecognized host factor other than age, disease severity, or the captured comorbidities decreases survival in aspiration pneumonia patients. However, it is also possible that aspiration pneumonia is merely a clinical designation for one end of the pneumonia spectrum, and we and others have failed to completely account for all measures of disease severity or all measures of comorbidities. Examples of unmeasured comorbidities would include presence of oropharyngeal dysphagia, which is not assessed in the database but could have a significant effect on clinical diagnosis. Unmeasured covariates can include measures beyond that of disease severity or comorbidity, such as the presence of a do not resuscitate (DNR) order, which could have a significant confounding effect on the observed association. A previous, single‐center study demonstrated that increased 30‐day mortality in aspiration pneumonia was mostly attributable to greater disease severity and comorbidities, although aspiration pneumonia independently conferred greater risk for adverse long‐term outcomes.[15] We propose that aspiration pneumonia represents a clinically distinct entity from nonaspiration pneumonia. Patients with chronic aspiration are often chronically malnourished and may have different oral flora than patients without chronic aspiration.[27, 28] Chronic aspiration has been associated with granulomatous reaction, organizing pneumonia, diffuse alveolar damage, and chronic bronchiolitis.[29] Chronic aspiration may elicit changes in the host physiology, and may render the host more susceptible to the development of secondary bacterial infection with morbid consequences.

The ability of the PSI to predict inpatient mortality was moderate (AUC only 0.7), with no significant additional discrimination between the aspiration and nonaspiration pneumonia groups. Although the PSI had moderate ability to predict inpatient mortality, the observed mortality was considerably higher than predicted. It is possible that the PSI incompletely captures clinically relevant comorbidities (eg, malnutrition). Further study to improve mortality prediction of aspiration pneumonia patients could employ sensitivity analysis to determine optimal thresholds and weighting of the PSI components.

Patients with aspiration pneumonia had longer hospital lengths of stay and took longer to achieve clinical stability than their nonaspiration counterparts. Time to clinical stability has been associated with increased posthospitalization mortality and is associated with time to switch from intravenous to oral antibiotics.[17] Although some component of hospital length‐of‐stay is subject to local practice patterns, time to clinical stability has explicit criteria for clinical improvement and failure, and therefore is less likely to be affected by local practice patterns.

We noted a relatively high (16%21%) incidence of prior antibiotic use among patients in this database. Analysis of antibiotic prescription patterns was limited, given the several different countries from which the database draws its cases. Although we used accepted criteria to define CAP cases, it is possible that this population may have a higher rate of resistant or uncommon pathogens than other studies of CAP that have populations with lower incidence of prior antibiotic use. Although not assessed, we suspect a significant component of the prior antibiotic use represented outpatient pneumonia treatment during the few days prior to visiting the hospital.

This study has several limitations, of which the most important may be that we used clinical determination for defining presence of aspiration pneumonia. This method is susceptible to the subjective perceptions of the treating clinician. We did not account for the effect of individual physicians in our model, although we did adjust for regional differences. The retrospective identification of patients allows for the possibility of selection bias, and therefore we have not attempted to make inferences regarding the relative incidence of pneumonia, nor did we adjust for temporal trends in diagnosis. The ratio of aspiration pneumonia patients to nonaspiration pneumonia patients may not necessarily reflect that observed in reality. Microbiologic and antibiotic data were unavailable for analysis. This study cannot inform on nonhospitalized patients with aspiration pneumonia, as only hospitalized patients were enrolled. The database identified cases of pneumonia, so it is possible for a patient to enter into the database more than once. Detection of mortality was limited to the inpatient setting rather than a set interval of 30 days. Inpatient mortality depends on length‐of‐stay patterns that may bias the mortality endpoint.[30] Also not assessed was the presence of a DNR order. It is possible that an older patient with greater comorbidities and disease severity may have care intentionally limited or withdrawn early by the family or clinicians.

Strengths of the study include its size and its multicenter, multinational population. The CAPO database is a large and well‐described population of patients with CAP.[17, 31] These attributes, as well as the clinician‐determined diagnosis, increase the generalizability of the study compared to a single‐center, single‐country study that employs investigator‐defined criteria.

CONCLUSION

Pneumonia patients with confusion, who are nursing home residence, and have cerebrovascular disease are more likely to be diagnosed with aspiration pneumonia by clinicians. Our clinician‐diagnosed cohort appears similar to those derived using an investigator definition. Patients with aspiration pneumonia are older, and have greater disease severity and more comorbidities than patients with nonaspiration pneumonia. They have greater mortality than their PSI score class would predict. Even after accounting for age, disease severity, and comorbidities, the presence of aspiration pneumonia independently conferred a greater than 2‐fold increase in inpatient mortality. These findings together suggest that aspiration pneumonia should be considered a distinct entity from typical pneumonia, and that additional research should be done in this field.

Pneumonia is a common clinical syndrome with well‐described epidemiology and microbiology. Aspiration pneumonia comprises 5% to 15% of patients with pneumonia acquired outside of the hospital,[1] but is less well characterized despite being a major syndrome of pneumonia in the elderly.[2, 3] Difficulties in studying aspiration pneumonia include the lack of a sensitive and specific marker for aspiration as well as the potential overlap between aspiration pneumonia and other forms of pneumonia.[4, 5, 6] Additionally, clinicians have difficulty distinguishing between aspiration pneumonia, which develops after the aspiration of oropharyngeal contents, and aspiration pneumonitis, wherein inhalation of gastric contents causes inflammation without the subsequent development of bacterial infection.[7, 8] Central to the study of aspiration pneumonia is whether it should exist as its own entity, or if aspiration is really a designation used for pneumonia in an older patient with greater comorbidities. The ability to clearly understand how a clinician diagnoses aspiration pneumonia, and whether that method has face validity with expert definitions may allow for improved future research, improved generalizability of current or past research, and possibly better clinical care.

Several validated mortality prediction models exist for community‐acquired pneumonia (CAP) using a variety of clinical predictors, but their performance in patients with aspiration pneumonia is less well characterized. Most studies validating pneumonia severity scoring systems excluded aspiration pneumonia from their study population.[9, 10, 11] Severity scoring systems for CAP may not accurately predict disease severity in patients with aspiration pneumonia. The CURB‐65[9] (confusion, uremia, respiratory rate, blood pressure, age 65 years) and the eCURB[12] scoring systems are poor predictors of mortality in patients with aspiration pneumonia, perhaps because they do not account for patient comorbidities.[13] The pneumonia severity index (PSI)[10] might predict mortality better than CURB‐65 in the aspiration population due to the inclusion of comorbidities.

Previous studies have demonstrated that patients with aspiration pneumonia are older and have greater disease severity and more comorbidities.[13, 14, 15] These single‐center studies also demonstrated greater mortality, more frequent admission to an intensive care unit (ICU), and longer hospital lengths of stay in patients with aspiration pneumonia. These studies identified aspiration pneumonia by the presence of a risk factor for aspiration[15] or by physician billing codes.[13] In practice, however, the bedside clinician diagnoses a patient as having aspiration pneumonia, but the logic is likely vague and inconsistent. Despite the potential for variability with individual judgment, an aggregate estimation from independent judgments may perform better than individual judgments.[16] Because there is no gold standard for defining aspiration pneumonia, all previous research has been limited to definitions created by investigators. This multicenter study seeks to determine what clinical characteristics lead physicians to diagnose a patient as having aspiration pneumonia, and whether or not the clinician‐derived diagnosis is distinct and clinically useful.

Our objectives were to: (1) identify covariates associated with bedside clinicians diagnosing a pneumonia patient as having aspiration pneumonia; (2) compare aspiration pneumonia and nonaspiration pneumonia in regard to disease severity, patient demographics, comorbidities, and clinical outcomes; and (3) measure the performance of the PSI in aspiration pneumonia versus nonaspiration pneumonia.

PATIENTS AND METHODS

RESULTS

Our initial query, after exclusion criteria, yielded a study population of 5185 patients (Figure 1). We compared 451 patients diagnosed with aspiration pneumonia to 4734 with CAP (Figure 1). Patient characteristics are summarized in Table 1. Patients with aspiration pneumonia were older, more likely to live in a nursing home, had greater disease severity, and were more likely to be admitted to an ICU. Patients with aspiration pneumonia had longer adjusted hospital lengths of stay and took more days to achieve clinical stability than patients with nonaspiration pneumonia (Figure 2). After adjusting for all variables in Table 1, the Cox proportional hazards models demonstrated that aspiration pneumonia was associated with ongoing hospitalization (hazard ratio [HR] for discharge: 0.77, 95% confidence interval [CI]: 0.65‐0.91, P=0.002) and clinical instability (HR for attaining clinical stability: 0.72, 95% CI: 0.61‐0.84, P<0.001). Patients with aspiration pneumonia presented with greater disease severity than those with nonaspiration pneumonia. Although there was no difference between groups in regard to temperature, respiratory rate, hyponatremia, or presence of pleural effusions or cavitary lesions, all other measured indices of disease severity were worse in patients with aspiration pneumonia. Patients with aspiration pneumonia were more likely to have cerebrovascular disease than those with nonaspiration pneumonia. Aspiration pneumonia patients also had increased prevalence of congestive heart failure. There was no appreciable difference between groups among other measured comorbidities.

Figure 1

Figure 2

The patient characteristics most associated with a physician diagnosis of aspiration pneumonia, identified using logistic regression, were confusion, residence in nursing home, and presence of cerebrovascular disease (odds ratio [OR]: of 4.4, 2.9, and 2.3, respectively), whereas renal disease was associated with decreased physician diagnosis of aspiration pneumonia over nonaspiration pneumonia (OR: 0.58) (Table 2).

Observed in‐patient mortality of aspiration pneumonia was 23%. This mortality was considerably higher than a mean PSI score of 123 would predict (class IV risk group, with expected 30‐day mortality of 8%9%[25]). The PSI score's ability to predict inpatient mortality in patients with aspiration pneumonia was moderate, with an area under the curve (AUC) of 0.71. This was similar to its performance in patients with nonaspiration pneumonia (AUC of 0.75) (Figure 3). These values are lower than the AUC of 0.81 for the PSI in predicting mortality derived from a meta‐analysis of 31 other studies.[26]

Figure 3

Our regression model after propensity score matching demonstrated that aspiration pneumonia independently confers a 2.3‐fold increased odds for inpatient mortality (95% CI: 1.56‐3.45, P<0.001).

DISCUSSION

Pneumonia patients with confusion, nursing home residence, or cerebrovascular disease are more likely to be diagnosed with aspiration pneumonia by clinicians. Although this is unsurprising, it is notable that these patients are more than twice as likely to die in the inpatient setting, even after accounting for age, comorbidities, and disease severity. These findings are similar to three previously published studies comparing aspiration and nonaspiration pneumonia at single institutions, albeit using different aspiration pneumonia definitions.[13, 14, 15] This study is the first large, multicenter, multinational study to demonstrate these findings.

Central to the interpretation of our results is the method of diagnosing aspiration versus nonaspiration. A bottom‐up method that relies on a clinician to check a box for aspiration may appear poorly reproducible. Because there is no diagnostic gold standard, clinicians may use different criteria to diagnose aspiration, creating potential for idiosyncratic noise. The strength of the wisdom of the crowd method used in this study is that an aggregate estimation from independent judgments may reduce the noise from individual judgments.[16] Although clinicians may vary in why they diagnose a particular patient as having aspiration pneumonia, it appears that the overwhelming reason for diagnosing a patient as having aspiration pneumonia is the presence of confusion, followed by previous nursing home residence or cerebrovascular disease. This finding has some face validity when compared with studies using an investigator definition, as altered mental status, chronic debility, and cerebrovascular disease are either prominent features of the definition of aspiration pneumonia[8] or frequently observed in patients with aspiration pneumonia.[13, 15] The distribution of cerebrovascular disease among our study's aspiration and nonaspiration pneumonia patients was similar to studies that used formal criteria in their definitions.[13, 15] Although nursing home residence was more likely in aspiration pneumonia patients, the majority of aspiration pneumonia patients were residing in the community, suggesting that aspiration is not simply a surrogate for healthcare‐associated pneumonia. Although patients with aspiration pneumonia are typically older than their nonaspiration counterparts, it appears that age is not a key determinant in the diagnosis of aspiration. With aspiration pneumonia, confusion, nursing home residence, and the presence of cerebrovascular disease are the greatest contributors in the clinical diagnosis, more than age.

Our data demonstrate that aspiration pneumonia confers increased odds for mortality, even after adjustment for age, disease severity, and comorbidities. These data suggest that aspiration pneumonia is a distinct entity from nonaspiration pneumonia, and that this disease is worse than nonaspiration CAP. If aspiration pneumonia is distinct from nonaspiration pneumonia, some unrecognized host factor other than age, disease severity, or the captured comorbidities decreases survival in aspiration pneumonia patients. However, it is also possible that aspiration pneumonia is merely a clinical designation for one end of the pneumonia spectrum, and we and others have failed to completely account for all measures of disease severity or all measures of comorbidities. Examples of unmeasured comorbidities would include presence of oropharyngeal dysphagia, which is not assessed in the database but could have a significant effect on clinical diagnosis. Unmeasured covariates can include measures beyond that of disease severity or comorbidity, such as the presence of a do not resuscitate (DNR) order, which could have a significant confounding effect on the observed association. A previous, single‐center study demonstrated that increased 30‐day mortality in aspiration pneumonia was mostly attributable to greater disease severity and comorbidities, although aspiration pneumonia independently conferred greater risk for adverse long‐term outcomes.[15] We propose that aspiration pneumonia represents a clinically distinct entity from nonaspiration pneumonia. Patients with chronic aspiration are often chronically malnourished and may have different oral flora than patients without chronic aspiration.[27, 28] Chronic aspiration has been associated with granulomatous reaction, organizing pneumonia, diffuse alveolar damage, and chronic bronchiolitis.[29] Chronic aspiration may elicit changes in the host physiology, and may render the host more susceptible to the development of secondary bacterial infection with morbid consequences.

The ability of the PSI to predict inpatient mortality was moderate (AUC only 0.7), with no significant additional discrimination between the aspiration and nonaspiration pneumonia groups. Although the PSI had moderate ability to predict inpatient mortality, the observed mortality was considerably higher than predicted. It is possible that the PSI incompletely captures clinically relevant comorbidities (eg, malnutrition). Further study to improve mortality prediction of aspiration pneumonia patients could employ sensitivity analysis to determine optimal thresholds and weighting of the PSI components.

Patients with aspiration pneumonia had longer hospital lengths of stay and took longer to achieve clinical stability than their nonaspiration counterparts. Time to clinical stability has been associated with increased posthospitalization mortality and is associated with time to switch from intravenous to oral antibiotics.[17] Although some component of hospital length‐of‐stay is subject to local practice patterns, time to clinical stability has explicit criteria for clinical improvement and failure, and therefore is less likely to be affected by local practice patterns.

We noted a relatively high (16%21%) incidence of prior antibiotic use among patients in this database. Analysis of antibiotic prescription patterns was limited, given the several different countries from which the database draws its cases. Although we used accepted criteria to define CAP cases, it is possible that this population may have a higher rate of resistant or uncommon pathogens than other studies of CAP that have populations with lower incidence of prior antibiotic use. Although not assessed, we suspect a significant component of the prior antibiotic use represented outpatient pneumonia treatment during the few days prior to visiting the hospital.

This study has several limitations, of which the most important may be that we used clinical determination for defining presence of aspiration pneumonia. This method is susceptible to the subjective perceptions of the treating clinician. We did not account for the effect of individual physicians in our model, although we did adjust for regional differences. The retrospective identification of patients allows for the possibility of selection bias, and therefore we have not attempted to make inferences regarding the relative incidence of pneumonia, nor did we adjust for temporal trends in diagnosis. The ratio of aspiration pneumonia patients to nonaspiration pneumonia patients may not necessarily reflect that observed in reality. Microbiologic and antibiotic data were unavailable for analysis. This study cannot inform on nonhospitalized patients with aspiration pneumonia, as only hospitalized patients were enrolled. The database identified cases of pneumonia, so it is possible for a patient to enter into the database more than once. Detection of mortality was limited to the inpatient setting rather than a set interval of 30 days. Inpatient mortality depends on length‐of‐stay patterns that may bias the mortality endpoint.[30] Also not assessed was the presence of a DNR order. It is possible that an older patient with greater comorbidities and disease severity may have care intentionally limited or withdrawn early by the family or clinicians.

Strengths of the study include its size and its multicenter, multinational population. The CAPO database is a large and well‐described population of patients with CAP.[17, 31] These attributes, as well as the clinician‐determined diagnosis, increase the generalizability of the study compared to a single‐center, single‐country study that employs investigator‐defined criteria.

CONCLUSION

Pneumonia patients with confusion, who are nursing home residence, and have cerebrovascular disease are more likely to be diagnosed with aspiration pneumonia by clinicians. Our clinician‐diagnosed cohort appears similar to those derived using an investigator definition. Patients with aspiration pneumonia are older, and have greater disease severity and more comorbidities than patients with nonaspiration pneumonia. They have greater mortality than their PSI score class would predict. Even after accounting for age, disease severity, and comorbidities, the presence of aspiration pneumonia independently conferred a greater than 2‐fold increase in inpatient mortality. These findings together suggest that aspiration pneumonia should be considered a distinct entity from typical pneumonia, and that additional research should be done in this field.

Since the Institute of Medicine Report To Err is Human, increased attention has been paid to improving the care of hospitalized patients.[1] Strategies include utilization of guidelines and pathways, and the application of quality improvement techniques to improve or standardize processes. Despite improvements in focused areas such as prevention of hospital‐acquired infections, evidence suggests that outcomes for hospitalized patients remain suboptimal.[2] Rates of errors and hospital‐related complications such as falls, decubitus ulcers, and infections remain high,[3, 4, 5] and not all patients receive what is known to be appropriate care.[6]

Many attempts to improve inpatient care have used process‐improvement approaches, focusing on impacting individuals' behaviors, or on breaking down processes into component parts. Examples include central line bundles or checklists.[7, 8] These approaches attempt to ensure that providers do things in a standardized way, but are implicitly based on the reductionist assumption that we can break processes down into predictable parts to improve the system. An alternative way to understand clinical systems is based on interdependencies between individuals in the system, or the ways in which parts of the system interact with each other, which may be unpredictable over time.[1, 9] Whereas these interdependencies include care processes, they also encompass the providers who care for patients. Providers working together vary in terms of the kinds of relationships they have with each other. Those relationships are crucial to system function because they are the foundation for the interactions that lead to effective patient care.

The application of several frameworks or approaches for considering healthcare systems in terms of relationships highlights the importance of this way of understanding system function. The include complexity science,[1, 7] relational coordination (which is grounded in complexity science),[10] high reliability,[11] and the Big Five for teamwork.[12]

Research indicates that interactions among healthcare providers can have important influences on outcomes.[13, 14, 15, 16, 17] Additionally, the initial implementation of checklists to prevent central‐line associated infections appeared to change provider relationships in a way that significantly influenced their success.[18] For example, positive primary care clinic member relationships as assessed by the Lanham framework have been associated with better chronic care model implementation, learning, and patient experience of care.[19, 20] This framework, which we apply here, identifies 7 relationship characteristics: (1) trust; (2) diversity; (3) respect; (4) mindfulness, or being open to new ideas from others; (5) heedfulness, or an understanding of how one's roles influence those of others; (6) use of rich in‐person or verbal communication, particularly for potentially ambiguous information open to multiple interpretations; and (7) having a mixture of social and task relatedness among teams, or relatedness outside of only work‐related tasks.[19] Relationships within surgical teams that are characterized by psychological safety and diversity are associated with successful uptake of new techniques and decreased mortality.[13, 14] Relationships are important because the ability of patients and providers to learn and make sense of their patients' illnesses is grounded in relationships.

We sought to better understand and characterize inpatient physician teams' relationships, and assess the association between team relationships as evaluated by Lanham's framework and outcomes for hospitalized patients. Data on relationships among inpatient medical teams are few, despite the fact that these teams provide a great proportion of inpatient care. Additionally, the care of hospitalized medical patients is complex and uncertain, often involving multiple providers, making provider relationships potentially even more important to outcomes than in other settings.

METHODS

RESULTS

The research team observed 1941 discussions of 576 individual patients. Observations were conducted over 352 hours and 54 minutes, resulting in 741 pages of notes (see Supporting Table 1 in the online version of this article for data regarding individual team members). Teams observed over half‐months are referred to with a and b designations.

Creation of team profiles yielded 13 common behavior characteristics that were inductively identified and that could potentially distinguish teams, including consideration of perspectives outside of the team and team members performing tasks normally outside of their roles. Table 1 provides examples of and summarizes observed behaviors using examples from the field notes, mapping these behavior characteristics onto the Lanham relationship characteristics. The distribution of relationship characteristics and scores for each team are shown in Table 2.

Correlation between relationship characteristics ranged from 0.32 to 0.95 (see Supporting Table 2 in the online version of this article). Mindfulness and trust are more highly correlated with each other than with other variables, as are diversity and respect. We performed a principal components factor analysis. Based on scree plot inspection and eigenvalues >1, we kept 3 factors that explained 85% of the total variance (see Supporting Table 3 in the online version of this article).

Our analyses of LOS and ULOS included 298 of the 576 patients. Two hundred sixty‐seven patients were excluded because their entire LOS did not occur while under the care of the observed teams. Eleven patients were removed from the analysis because their LOS was >12 days. The analysis of complications included 398 patients. In our preliminary general linear modeling approach, only patient workload was significantly associated with outcomes using a cutoff of P=0.05. Charlson‐Deyo score and mental health comorbidities were not associated with outcomes.

The results of the Kruskal‐Wallis test show the patient average ranking on each of the outcome variables by 3 groups (Table 3). Overall, teams with higher relationship scores had lower rank scores on all outcomes measures. However, the only statistically significant comparisons were for complications. Teams having 6 to 7 characteristics had a significantly lower complication rate ranking than teams with 0 to 2 and 3 to 5 (P=0.001). We did not find consistent differences between individual teams or groups of teams with relationship scores from 0 to 2, 3 to 5, and 6 to 7 with regard to Charlson score, mental health issues, or workload. The only significant differences were between Charlson‐Deyo scores for patients admitted to teams with low relationship scores of 0 to 2 versus high relationship scores of 6 to 7 (6.7 vs 5.1); scores for teams with relationship scores of 3 to 5 were not significantly different from the low or high groups.

Table 4 shows the Kruskal‐Wallace rank test results for each group of relationship characteristics identified in the factor analysis based on whether teams displayed all or none of the characteristics in the factor. There were no differences in these groupings for LOS. Teams that exhibited both mindfulness and trust had lower ranks on ULOS than teams that did not have either. Similarly, teams with heedfulness, social‐task relatedness, and more rich communication demonstrated lower ULOS rankings than teams who did not have all 3 characteristics.

DISCUSSION

Relationships are critical to team function because they are the basis for the social interactions that are central to patient care. These interactions include how providers recognize and make sense of what is happening with patients, and how they learn to care for patients more effectively. Additionally, the high task interdependencies among inpatient providers require effective relationships for optimal care. In our study, inpatient medicine physician teams' relationships varied, and these differences were associated with ULOS and complications. Relationship characteristics are not mutually exclusive, and as our factor analysis demonstrates, are intercorrelated. Trust and mindfulness appear to be particularly important. Trust may foster psychological safety that in turn promotes the willingness of individuals to contribute their thoughts and ideas.[13] In low‐trust teams, providers may fear a negative impact for bringing forward a concern based on limited data. Mindful teams may be more likely to notice nuanced changes, or are more likely to talk when things just do not appear to be going in the right direction with the patient. In the case of acutely ill medical patients, trust and mindfulness may lead to an increased likelihood that clinical changes are recognized and discussed quickly. For example, on a team characterized by trust and mindfulness, the entire team was typically involved in care discussions, and the interns and students frequently asked a lot of questions, even regarding the care of patients they were not directly following. We observed that these questions and discussions often led the team to realize that they needed to make a change in management decisions (eg, discontinuing Bactrim, lowering insulin doses, adjusting antihypertensives, premedicating for intravenous contrast) that they had not caught in the assessment and plan portion of the patient care discussion. In another example, a medical student asked a tentative question after a patient needed to go quickly to the bathroom while they were examining her, leading the team to ask more questions that led to a more rapid evaluation of a potential urinary tract infection. This finding is consistent with the description of failure to rescue among surgical patients, in which mortality has been associated with the failure to recognize complications rapidly and act effectively.[33]

Our findings are limited in several ways. First, these data are from a single academic institution. Although we sought diversity among our teams and collected data across 2 hospitals, there may be local contextual factors that influenced our results. Second, our data demonstrate an association, but not causality. Our findings should be tested in studies that assess causality and potential mechanisms through which relationships influence outcomes. Third, the individuals observing the teams had some knowledge of patient outcomes through hearing patient discussions. However, by involving individuals who did not participate in observations and were blinded to outcomes in assessing team relationships, we addressed this potential bias. Fourth, our observations were largely focused on physician teams, not directly including other providers. Our difficulty in observing regular interactions between physicians and other providers underscores the need to increase contact among those caring for hospitalized patients, such as occurs through multidisciplinary rounds. We did include team communication with other disciplines in our assessment of the relationship characteristics of diversity and rich communication. Finally, our analysis was limited by our sample size. We observed a relatively small number of teams. Although we benefitted from seeing the change in team relationships that occurred with attending changes halfway through some of our data collection months, this did limit the number of patients we could include in our analyses. Though we did not observe obvious differences in relationships between the teams observed across the 2 hospitals, the small number of teams and hospitals precluded our ability to perform multilevel modeling analyses, which would have allowed us to assess or account for the influence of team or organizational factors. However, this small sample size did allow for a richer assessment of team behaviors.

Although preliminary, our findings are an important step in understanding the function of inpatient medical teams not only in terms of processes of care, but also in terms of relationships. Patient care is a social activity, requiring effective communication to develop working diagnoses, recognize changes in patients' clinical courses, and formulate effective treatment plans during and after hospitalization. Future work could follow several directions. One would be to assess the causal mechanisms through which relationships influence patient outcomes. These may include sensemaking, learning, and improved coordination. Positive relationships may facilitate interaction of tacit and explicit information, facilitating the creation of understandings that foster more effective patient care.[34] The dynamic nature of relationships and how patient outcomes in turn feed back into relationships could be an area of exploration. This line of research could build on the idea of teaming.[35] Understanding relationships across multidisciplinary teams or with patients and families would be another direction. Finally, our results could point to potential interventions to improve patient outcomes through improving relationships. Better understanding of the nature of effective relationships among providers should enable us to develop more effective strategies to improve the care of hospitalized patients. In the larger context of payment reforms that require greater coordination and communication among and across providers, a greater understanding of how relationships influence patient outcomes will be important.

Since the Institute of Medicine Report To Err is Human, increased attention has been paid to improving the care of hospitalized patients.[1] Strategies include utilization of guidelines and pathways, and the application of quality improvement techniques to improve or standardize processes. Despite improvements in focused areas such as prevention of hospital‐acquired infections, evidence suggests that outcomes for hospitalized patients remain suboptimal.[2] Rates of errors and hospital‐related complications such as falls, decubitus ulcers, and infections remain high,[3, 4, 5] and not all patients receive what is known to be appropriate care.[6]

Many attempts to improve inpatient care have used process‐improvement approaches, focusing on impacting individuals' behaviors, or on breaking down processes into component parts. Examples include central line bundles or checklists.[7, 8] These approaches attempt to ensure that providers do things in a standardized way, but are implicitly based on the reductionist assumption that we can break processes down into predictable parts to improve the system. An alternative way to understand clinical systems is based on interdependencies between individuals in the system, or the ways in which parts of the system interact with each other, which may be unpredictable over time.[1, 9] Whereas these interdependencies include care processes, they also encompass the providers who care for patients. Providers working together vary in terms of the kinds of relationships they have with each other. Those relationships are crucial to system function because they are the foundation for the interactions that lead to effective patient care.

The application of several frameworks or approaches for considering healthcare systems in terms of relationships highlights the importance of this way of understanding system function. The include complexity science,[1, 7] relational coordination (which is grounded in complexity science),[10] high reliability,[11] and the Big Five for teamwork.[12]

Research indicates that interactions among healthcare providers can have important influences on outcomes.[13, 14, 15, 16, 17] Additionally, the initial implementation of checklists to prevent central‐line associated infections appeared to change provider relationships in a way that significantly influenced their success.[18] For example, positive primary care clinic member relationships as assessed by the Lanham framework have been associated with better chronic care model implementation, learning, and patient experience of care.[19, 20] This framework, which we apply here, identifies 7 relationship characteristics: (1) trust; (2) diversity; (3) respect; (4) mindfulness, or being open to new ideas from others; (5) heedfulness, or an understanding of how one's roles influence those of others; (6) use of rich in‐person or verbal communication, particularly for potentially ambiguous information open to multiple interpretations; and (7) having a mixture of social and task relatedness among teams, or relatedness outside of only work‐related tasks.[19] Relationships within surgical teams that are characterized by psychological safety and diversity are associated with successful uptake of new techniques and decreased mortality.[13, 14] Relationships are important because the ability of patients and providers to learn and make sense of their patients' illnesses is grounded in relationships.

We sought to better understand and characterize inpatient physician teams' relationships, and assess the association between team relationships as evaluated by Lanham's framework and outcomes for hospitalized patients. Data on relationships among inpatient medical teams are few, despite the fact that these teams provide a great proportion of inpatient care. Additionally, the care of hospitalized medical patients is complex and uncertain, often involving multiple providers, making provider relationships potentially even more important to outcomes than in other settings.

METHODS

RESULTS

The research team observed 1941 discussions of 576 individual patients. Observations were conducted over 352 hours and 54 minutes, resulting in 741 pages of notes (see Supporting Table 1 in the online version of this article for data regarding individual team members). Teams observed over half‐months are referred to with a and b designations.

Creation of team profiles yielded 13 common behavior characteristics that were inductively identified and that could potentially distinguish teams, including consideration of perspectives outside of the team and team members performing tasks normally outside of their roles. Table 1 provides examples of and summarizes observed behaviors using examples from the field notes, mapping these behavior characteristics onto the Lanham relationship characteristics. The distribution of relationship characteristics and scores for each team are shown in Table 2.

Correlation between relationship characteristics ranged from 0.32 to 0.95 (see Supporting Table 2 in the online version of this article). Mindfulness and trust are more highly correlated with each other than with other variables, as are diversity and respect. We performed a principal components factor analysis. Based on scree plot inspection and eigenvalues >1, we kept 3 factors that explained 85% of the total variance (see Supporting Table 3 in the online version of this article).

Our analyses of LOS and ULOS included 298 of the 576 patients. Two hundred sixty‐seven patients were excluded because their entire LOS did not occur while under the care of the observed teams. Eleven patients were removed from the analysis because their LOS was >12 days. The analysis of complications included 398 patients. In our preliminary general linear modeling approach, only patient workload was significantly associated with outcomes using a cutoff of P=0.05. Charlson‐Deyo score and mental health comorbidities were not associated with outcomes.

The results of the Kruskal‐Wallis test show the patient average ranking on each of the outcome variables by 3 groups (Table 3). Overall, teams with higher relationship scores had lower rank scores on all outcomes measures. However, the only statistically significant comparisons were for complications. Teams having 6 to 7 characteristics had a significantly lower complication rate ranking than teams with 0 to 2 and 3 to 5 (P=0.001). We did not find consistent differences between individual teams or groups of teams with relationship scores from 0 to 2, 3 to 5, and 6 to 7 with regard to Charlson score, mental health issues, or workload. The only significant differences were between Charlson‐Deyo scores for patients admitted to teams with low relationship scores of 0 to 2 versus high relationship scores of 6 to 7 (6.7 vs 5.1); scores for teams with relationship scores of 3 to 5 were not significantly different from the low or high groups.

Table 4 shows the Kruskal‐Wallace rank test results for each group of relationship characteristics identified in the factor analysis based on whether teams displayed all or none of the characteristics in the factor. There were no differences in these groupings for LOS. Teams that exhibited both mindfulness and trust had lower ranks on ULOS than teams that did not have either. Similarly, teams with heedfulness, social‐task relatedness, and more rich communication demonstrated lower ULOS rankings than teams who did not have all 3 characteristics.

DISCUSSION

Relationships are critical to team function because they are the basis for the social interactions that are central to patient care. These interactions include how providers recognize and make sense of what is happening with patients, and how they learn to care for patients more effectively. Additionally, the high task interdependencies among inpatient providers require effective relationships for optimal care. In our study, inpatient medicine physician teams' relationships varied, and these differences were associated with ULOS and complications. Relationship characteristics are not mutually exclusive, and as our factor analysis demonstrates, are intercorrelated. Trust and mindfulness appear to be particularly important. Trust may foster psychological safety that in turn promotes the willingness of individuals to contribute their thoughts and ideas.[13] In low‐trust teams, providers may fear a negative impact for bringing forward a concern based on limited data. Mindful teams may be more likely to notice nuanced changes, or are more likely to talk when things just do not appear to be going in the right direction with the patient. In the case of acutely ill medical patients, trust and mindfulness may lead to an increased likelihood that clinical changes are recognized and discussed quickly. For example, on a team characterized by trust and mindfulness, the entire team was typically involved in care discussions, and the interns and students frequently asked a lot of questions, even regarding the care of patients they were not directly following. We observed that these questions and discussions often led the team to realize that they needed to make a change in management decisions (eg, discontinuing Bactrim, lowering insulin doses, adjusting antihypertensives, premedicating for intravenous contrast) that they had not caught in the assessment and plan portion of the patient care discussion. In another example, a medical student asked a tentative question after a patient needed to go quickly to the bathroom while they were examining her, leading the team to ask more questions that led to a more rapid evaluation of a potential urinary tract infection. This finding is consistent with the description of failure to rescue among surgical patients, in which mortality has been associated with the failure to recognize complications rapidly and act effectively.[33]

Our findings are limited in several ways. First, these data are from a single academic institution. Although we sought diversity among our teams and collected data across 2 hospitals, there may be local contextual factors that influenced our results. Second, our data demonstrate an association, but not causality. Our findings should be tested in studies that assess causality and potential mechanisms through which relationships influence outcomes. Third, the individuals observing the teams had some knowledge of patient outcomes through hearing patient discussions. However, by involving individuals who did not participate in observations and were blinded to outcomes in assessing team relationships, we addressed this potential bias. Fourth, our observations were largely focused on physician teams, not directly including other providers. Our difficulty in observing regular interactions between physicians and other providers underscores the need to increase contact among those caring for hospitalized patients, such as occurs through multidisciplinary rounds. We did include team communication with other disciplines in our assessment of the relationship characteristics of diversity and rich communication. Finally, our analysis was limited by our sample size. We observed a relatively small number of teams. Although we benefitted from seeing the change in team relationships that occurred with attending changes halfway through some of our data collection months, this did limit the number of patients we could include in our analyses. Though we did not observe obvious differences in relationships between the teams observed across the 2 hospitals, the small number of teams and hospitals precluded our ability to perform multilevel modeling analyses, which would have allowed us to assess or account for the influence of team or organizational factors. However, this small sample size did allow for a richer assessment of team behaviors.

Although preliminary, our findings are an important step in understanding the function of inpatient medical teams not only in terms of processes of care, but also in terms of relationships. Patient care is a social activity, requiring effective communication to develop working diagnoses, recognize changes in patients' clinical courses, and formulate effective treatment plans during and after hospitalization. Future work could follow several directions. One would be to assess the causal mechanisms through which relationships influence patient outcomes. These may include sensemaking, learning, and improved coordination. Positive relationships may facilitate interaction of tacit and explicit information, facilitating the creation of understandings that foster more effective patient care.[34] The dynamic nature of relationships and how patient outcomes in turn feed back into relationships could be an area of exploration. This line of research could build on the idea of teaming.[35] Understanding relationships across multidisciplinary teams or with patients and families would be another direction. Finally, our results could point to potential interventions to improve patient outcomes through improving relationships. Better understanding of the nature of effective relationships among providers should enable us to develop more effective strategies to improve the care of hospitalized patients. In the larger context of payment reforms that require greater coordination and communication among and across providers, a greater understanding of how relationships influence patient outcomes will be important.

Since the Institute of Medicine Report To Err is Human, increased attention has been paid to improving the care of hospitalized patients.[1] Strategies include utilization of guidelines and pathways, and the application of quality improvement techniques to improve or standardize processes. Despite improvements in focused areas such as prevention of hospital‐acquired infections, evidence suggests that outcomes for hospitalized patients remain suboptimal.[2] Rates of errors and hospital‐related complications such as falls, decubitus ulcers, and infections remain high,[3, 4, 5] and not all patients receive what is known to be appropriate care.[6]

Many attempts to improve inpatient care have used process‐improvement approaches, focusing on impacting individuals' behaviors, or on breaking down processes into component parts. Examples include central line bundles or checklists.[7, 8] These approaches attempt to ensure that providers do things in a standardized way, but are implicitly based on the reductionist assumption that we can break processes down into predictable parts to improve the system. An alternative way to understand clinical systems is based on interdependencies between individuals in the system, or the ways in which parts of the system interact with each other, which may be unpredictable over time.[1, 9] Whereas these interdependencies include care processes, they also encompass the providers who care for patients. Providers working together vary in terms of the kinds of relationships they have with each other. Those relationships are crucial to system function because they are the foundation for the interactions that lead to effective patient care.

The application of several frameworks or approaches for considering healthcare systems in terms of relationships highlights the importance of this way of understanding system function. The include complexity science,[1, 7] relational coordination (which is grounded in complexity science),[10] high reliability,[11] and the Big Five for teamwork.[12]

Research indicates that interactions among healthcare providers can have important influences on outcomes.[13, 14, 15, 16, 17] Additionally, the initial implementation of checklists to prevent central‐line associated infections appeared to change provider relationships in a way that significantly influenced their success.[18] For example, positive primary care clinic member relationships as assessed by the Lanham framework have been associated with better chronic care model implementation, learning, and patient experience of care.[19, 20] This framework, which we apply here, identifies 7 relationship characteristics: (1) trust; (2) diversity; (3) respect; (4) mindfulness, or being open to new ideas from others; (5) heedfulness, or an understanding of how one's roles influence those of others; (6) use of rich in‐person or verbal communication, particularly for potentially ambiguous information open to multiple interpretations; and (7) having a mixture of social and task relatedness among teams, or relatedness outside of only work‐related tasks.[19] Relationships within surgical teams that are characterized by psychological safety and diversity are associated with successful uptake of new techniques and decreased mortality.[13, 14] Relationships are important because the ability of patients and providers to learn and make sense of their patients' illnesses is grounded in relationships.

We sought to better understand and characterize inpatient physician teams' relationships, and assess the association between team relationships as evaluated by Lanham's framework and outcomes for hospitalized patients. Data on relationships among inpatient medical teams are few, despite the fact that these teams provide a great proportion of inpatient care. Additionally, the care of hospitalized medical patients is complex and uncertain, often involving multiple providers, making provider relationships potentially even more important to outcomes than in other settings.

METHODS

RESULTS

The research team observed 1941 discussions of 576 individual patients. Observations were conducted over 352 hours and 54 minutes, resulting in 741 pages of notes (see Supporting Table 1 in the online version of this article for data regarding individual team members). Teams observed over half‐months are referred to with a and b designations.

Creation of team profiles yielded 13 common behavior characteristics that were inductively identified and that could potentially distinguish teams, including consideration of perspectives outside of the team and team members performing tasks normally outside of their roles. Table 1 provides examples of and summarizes observed behaviors using examples from the field notes, mapping these behavior characteristics onto the Lanham relationship characteristics. The distribution of relationship characteristics and scores for each team are shown in Table 2.

Correlation between relationship characteristics ranged from 0.32 to 0.95 (see Supporting Table 2 in the online version of this article). Mindfulness and trust are more highly correlated with each other than with other variables, as are diversity and respect. We performed a principal components factor analysis. Based on scree plot inspection and eigenvalues >1, we kept 3 factors that explained 85% of the total variance (see Supporting Table 3 in the online version of this article).

Our analyses of LOS and ULOS included 298 of the 576 patients. Two hundred sixty‐seven patients were excluded because their entire LOS did not occur while under the care of the observed teams. Eleven patients were removed from the analysis because their LOS was >12 days. The analysis of complications included 398 patients. In our preliminary general linear modeling approach, only patient workload was significantly associated with outcomes using a cutoff of P=0.05. Charlson‐Deyo score and mental health comorbidities were not associated with outcomes.

The results of the Kruskal‐Wallis test show the patient average ranking on each of the outcome variables by 3 groups (Table 3). Overall, teams with higher relationship scores had lower rank scores on all outcomes measures. However, the only statistically significant comparisons were for complications. Teams having 6 to 7 characteristics had a significantly lower complication rate ranking than teams with 0 to 2 and 3 to 5 (P=0.001). We did not find consistent differences between individual teams or groups of teams with relationship scores from 0 to 2, 3 to 5, and 6 to 7 with regard to Charlson score, mental health issues, or workload. The only significant differences were between Charlson‐Deyo scores for patients admitted to teams with low relationship scores of 0 to 2 versus high relationship scores of 6 to 7 (6.7 vs 5.1); scores for teams with relationship scores of 3 to 5 were not significantly different from the low or high groups.

Table 4 shows the Kruskal‐Wallace rank test results for each group of relationship characteristics identified in the factor analysis based on whether teams displayed all or none of the characteristics in the factor. There were no differences in these groupings for LOS. Teams that exhibited both mindfulness and trust had lower ranks on ULOS than teams that did not have either. Similarly, teams with heedfulness, social‐task relatedness, and more rich communication demonstrated lower ULOS rankings than teams who did not have all 3 characteristics.

DISCUSSION

Relationships are critical to team function because they are the basis for the social interactions that are central to patient care. These interactions include how providers recognize and make sense of what is happening with patients, and how they learn to care for patients more effectively. Additionally, the high task interdependencies among inpatient providers require effective relationships for optimal care. In our study, inpatient medicine physician teams' relationships varied, and these differences were associated with ULOS and complications. Relationship characteristics are not mutually exclusive, and as our factor analysis demonstrates, are intercorrelated. Trust and mindfulness appear to be particularly important. Trust may foster psychological safety that in turn promotes the willingness of individuals to contribute their thoughts and ideas.[13] In low‐trust teams, providers may fear a negative impact for bringing forward a concern based on limited data. Mindful teams may be more likely to notice nuanced changes, or are more likely to talk when things just do not appear to be going in the right direction with the patient. In the case of acutely ill medical patients, trust and mindfulness may lead to an increased likelihood that clinical changes are recognized and discussed quickly. For example, on a team characterized by trust and mindfulness, the entire team was typically involved in care discussions, and the interns and students frequently asked a lot of questions, even regarding the care of patients they were not directly following. We observed that these questions and discussions often led the team to realize that they needed to make a change in management decisions (eg, discontinuing Bactrim, lowering insulin doses, adjusting antihypertensives, premedicating for intravenous contrast) that they had not caught in the assessment and plan portion of the patient care discussion. In another example, a medical student asked a tentative question after a patient needed to go quickly to the bathroom while they were examining her, leading the team to ask more questions that led to a more rapid evaluation of a potential urinary tract infection. This finding is consistent with the description of failure to rescue among surgical patients, in which mortality has been associated with the failure to recognize complications rapidly and act effectively.[33]

Our findings are limited in several ways. First, these data are from a single academic institution. Although we sought diversity among our teams and collected data across 2 hospitals, there may be local contextual factors that influenced our results. Second, our data demonstrate an association, but not causality. Our findings should be tested in studies that assess causality and potential mechanisms through which relationships influence outcomes. Third, the individuals observing the teams had some knowledge of patient outcomes through hearing patient discussions. However, by involving individuals who did not participate in observations and were blinded to outcomes in assessing team relationships, we addressed this potential bias. Fourth, our observations were largely focused on physician teams, not directly including other providers. Our difficulty in observing regular interactions between physicians and other providers underscores the need to increase contact among those caring for hospitalized patients, such as occurs through multidisciplinary rounds. We did include team communication with other disciplines in our assessment of the relationship characteristics of diversity and rich communication. Finally, our analysis was limited by our sample size. We observed a relatively small number of teams. Although we benefitted from seeing the change in team relationships that occurred with attending changes halfway through some of our data collection months, this did limit the number of patients we could include in our analyses. Though we did not observe obvious differences in relationships between the teams observed across the 2 hospitals, the small number of teams and hospitals precluded our ability to perform multilevel modeling analyses, which would have allowed us to assess or account for the influence of team or organizational factors. However, this small sample size did allow for a richer assessment of team behaviors.

Although preliminary, our findings are an important step in understanding the function of inpatient medical teams not only in terms of processes of care, but also in terms of relationships. Patient care is a social activity, requiring effective communication to develop working diagnoses, recognize changes in patients' clinical courses, and formulate effective treatment plans during and after hospitalization. Future work could follow several directions. One would be to assess the causal mechanisms through which relationships influence patient outcomes. These may include sensemaking, learning, and improved coordination. Positive relationships may facilitate interaction of tacit and explicit information, facilitating the creation of understandings that foster more effective patient care.[34] The dynamic nature of relationships and how patient outcomes in turn feed back into relationships could be an area of exploration. This line of research could build on the idea of teaming.[35] Understanding relationships across multidisciplinary teams or with patients and families would be another direction. Finally, our results could point to potential interventions to improve patient outcomes through improving relationships. Better understanding of the nature of effective relationships among providers should enable us to develop more effective strategies to improve the care of hospitalized patients. In the larger context of payment reforms that require greater coordination and communication among and across providers, a greater understanding of how relationships influence patient outcomes will be important.

Previous studies have advocated the importance of effective communication between clinicians as a critical component in the provision of high‐quality patient care.[1, 2, 3, 4] There is increasing interest in the use of information and communication technologies to improve how clinicians communicate in hospital settings. A number of hospitals have implemented different solutions to improve communication. These solutions include alphanumeric pagers,[5] smartphones,[6] e‐mail,[7] secure text messaging,[8] and a Web‐based interdisciplinary communication tool.[9]

These systems have different limitations that render them inefficient and likely inhibit collaborative care. Current systems, such as pagers, rely on the sender to ensure the message was received and are successful in delivering messages approximately 67% of the time.[5, 9, 10] Although alphanumeric pagers and secure text messaging can increase the likelihood of delivery, these messages are often isolated and not easily viewable by the whole care team.[11] Improved systems should also reduce unnecessary interruptions by providing support for both urgent and delayed messages. Finally, messages should be stored and retrievable to enable increased accountability and allow for review for quality improvement initiatives.

It is also important to consider the unintended consequences of technology implementations.[12] Moving communication to text messages and smartphones has the potential to reduce interprofessional relations and can increase confusion if used for complex issues.[10, 13] In this article, we present a system designed to improve interprofessional communication on general internal medicine wards by incorporating these desired features and describe the usage and attitudes toward the system, specifically assessing for effects on multiple domains including efficiency, interprofessional collaboration, and relationships.

METHODS

Intervention

To address issues with communication, we developed a systemClinical Message (CM)that included 2 main components: a physician handover tool and secure messaging module. The focus of CM was to improve communication and information flow among different healthcare providers (physicians, nurses, pharmacists, social workers and therapists) through a secure, shared platform.

Physician Handover

The physician handover tool was designed to facilitate the physician handover process at shift change and is used as a patient rounding tool for day‐to‐day management of patients. It is also accessed by nurses and other clinicians to view the physicians' notes and to stay informed on the overall care plan. The tool contains standard elements including a list of patients with the following information on each patient: demographics, diagnosis, code status, medical history, active issues, and discharge plans (Figure 1).

Figure 1

Secure Messaging

Secure messaging was designed around our dominant communication: nurses sending messages to physicians who would then respond. Nurses and other health professionals sent messages to the medical teams by accessing CM, selecting the appropriate patient, and filling out a message template. The system automatically populated the To field with the team assigned to the selected patient. Messaging for each team was centralized around a single team smartphone that was carried 24 hours a day, 7 days a week by a physician on that team. This removed the guesswork of trying to identify the individual physician responsible for that patient. For each message, a subject or issue and content were entered (Figure 2). Logic was also incorporated to reduce the amount of unnecessary interruptions. Senders would choose to send the message immediately as an interrupt message (urgent) for urgent/time sensitive issues or as an allow time to respond message (delayed). For the latter, the message was posted to the system where physicians could check and answer them. Interrupt messages were sent to the team smartphone using the Short Message Service (SMS) protocol. To try and ensure the communication loop on any issues was closed, when a message requested a response and did not receive it, the system sent another message. For urgent messages, a repeat message was initiated after 15 minutes. For delayed messages, the sender defined when they needed a response, typically within 2 to 8 hours. Senders were also able to select the mode of response that would best meet their needs from a workflow perspective: call back, text reply, or to specify that a reply was not required. Senders were also able to verify if the messages were received by the physician's smartphone. Physicians could view the messages within CM and reply. For messages that went to their team smartphone, physicians could respond from the smartphone through a secure Web link.

Figure 2

Because the messages were linked to the patients, they were visible to the entire care team, not just the message sender and recipient. If the care of the patient was transferred from 1 clinician to the next, the new clinician could easily review prior messages to understand recent patient events. The system was accessible through a browser on the intranet. The system regularly pulled patient demographic details such as name, age, medical record number, and location from our electronic medical record through a 1‐way interface. Information from this communication system was not considered part of the medical record but was retrievable.

The system was introduced as the new standard method of communication for nurses to reach physicians for all of the general internal medicine wards and for all medical teams at site 1 on May 2, 2011 and site 2 on June 6, 2011. The system replaced a text‐based Web‐paging system and supplemented the numeric pager carried by residents. Initial training of a half hour was provided to all nurses and residents.

RESULTS

DISCUSSION

We describe an implementation of a system to improve clinical communication in hospitals. The system was highly used and was perceived to improve communication by both nurses and physicians. Specifically, users found that the system increased efficiency, accountability, timeliness, and collaboration, but that there were issues with message clarity for complex medical issues.

Other systems and approaches have been implemented to improve communication. These included the use of alphanumeric pagers, e‐mail, secure texting, and smartphones. There is evidence that more advanced systems can improve efficiency for senders.[16] A recent randomized trial of secure text messaging found that it was perceived to be more efficient than paging, but overall usage was low and inconsistent.[8] There is also evidence that smartphones may increase interruptions, worsen interprofessional relationships, and cause issues with professional behavior.[10] Unfortunately, there are a limited number of interventions that improve communication, with some improving efficiency but none demonstrating improved patient‐oriented outcomes.[16, 17] This study evaluated a novel system, with functionality to link communication to patients, and created a system that aligned with the workflow of the clinicians. Messages were linked to the patient, not the sender or receiver, so other clinicians in the patient's circle of care could easily view the communication. Moreover, the system was designed to improve message response rates and allow for nonurgent messages.

Our communication system uses standard, commercially available components (smartphones, SMS), and relatively basic functionality (handover, secure messaging). Important findings are that the current system of paging can be transformed to a more efficient system that users will readily adopt. We found positive effects with components of the system. It appeared to improve efficiency and increase accountability. Accountability is crucial and moves from undocumented conversation to fully documented details of interactions. This can be used for both incident review and to review for quality improvement.

Using the system, physicians perceived that they were bothered by low‐value information, whereas nurses perceived a lack of response, and both found that the system was not ideal for complex messages. The mismatch between what physicians and nurses perceive as important has been attributed to their different timeframes and context.[18] For nurses with an upcoming change of shift, they wanted resolution of issues before handover. A physician on a different ward may not appreciate the context of a nurse having to directly interact with an irate family member. These different perceptions likely contributed to the lack of response to 8.6% of text messages. This is still better than other systems, such as paging, which can be as high as 33%.[10] For nonurgent items, clinicians would ideally check and clear items regularly from the system using a desktop computer, responding within the allotted timeframe. Unfortunately, this never became part of routine physician workflow, likely due to their busy workload, so many physicians would only respond when items became overdue. However, having a method to deal with nonurgent messages may have prevented some interruptions during protected educational times of trainees. The system was also not ideal for urgent or complex items. Complex items can be difficult to convey using the rarified communication medium of text messages.[19, 20] Urgent or complex issues are likely best resolved with a face‐to‐face or telephone conversation.

There are several limitations in our study that should be considered when interpreting the results. It is a study of usage and perceptions after implementation. Although more rigorous study is required to evaluate the effects, we see this as a first step in process improvement. Future research should measure the impact on improving patient care of this system and on patient outcomes such as adverse events. The study and intervention was limited to general internal medicine wards in 2 academic hospital settings where there are frequent rotations of medical personnel. The findings may not be generalizable to other hospital settings.

Future directions should be to further improve on the communication system and to educate and train staff on how to effectively communicate. Survey results showed that although users perceived increased efficiency, there was still significant opportunity to improve. One way to improve would be to have a mobile application in which physicians can easily review nonurgent items. Improvements could also be realized by educating clinicians on the use of the system and providing immediate feedback. Providing feedback to physicians on how well they respond could address nurses' issues around lack of timely response. By creating consensus between nurses and physicians on what is of high and low value to communicate could increase satisfaction for all users.

In summary, we present the usage and perceptions of a system designed to improve hospital communication. We found that there was high uptake, and that users perceived it to improve efficiency, collaboration, and accountability, but it may not be useful for communicating complex issues.

Previous studies have advocated the importance of effective communication between clinicians as a critical component in the provision of high‐quality patient care.[1, 2, 3, 4] There is increasing interest in the use of information and communication technologies to improve how clinicians communicate in hospital settings. A number of hospitals have implemented different solutions to improve communication. These solutions include alphanumeric pagers,[5] smartphones,[6] e‐mail,[7] secure text messaging,[8] and a Web‐based interdisciplinary communication tool.[9]

These systems have different limitations that render them inefficient and likely inhibit collaborative care. Current systems, such as pagers, rely on the sender to ensure the message was received and are successful in delivering messages approximately 67% of the time.[5, 9, 10] Although alphanumeric pagers and secure text messaging can increase the likelihood of delivery, these messages are often isolated and not easily viewable by the whole care team.[11] Improved systems should also reduce unnecessary interruptions by providing support for both urgent and delayed messages. Finally, messages should be stored and retrievable to enable increased accountability and allow for review for quality improvement initiatives.

It is also important to consider the unintended consequences of technology implementations.[12] Moving communication to text messages and smartphones has the potential to reduce interprofessional relations and can increase confusion if used for complex issues.[10, 13] In this article, we present a system designed to improve interprofessional communication on general internal medicine wards by incorporating these desired features and describe the usage and attitudes toward the system, specifically assessing for effects on multiple domains including efficiency, interprofessional collaboration, and relationships.

METHODS

Intervention

To address issues with communication, we developed a systemClinical Message (CM)that included 2 main components: a physician handover tool and secure messaging module. The focus of CM was to improve communication and information flow among different healthcare providers (physicians, nurses, pharmacists, social workers and therapists) through a secure, shared platform.

Physician Handover

The physician handover tool was designed to facilitate the physician handover process at shift change and is used as a patient rounding tool for day‐to‐day management of patients. It is also accessed by nurses and other clinicians to view the physicians' notes and to stay informed on the overall care plan. The tool contains standard elements including a list of patients with the following information on each patient: demographics, diagnosis, code status, medical history, active issues, and discharge plans (Figure 1).

Figure 1

Secure Messaging

Secure messaging was designed around our dominant communication: nurses sending messages to physicians who would then respond. Nurses and other health professionals sent messages to the medical teams by accessing CM, selecting the appropriate patient, and filling out a message template. The system automatically populated the To field with the team assigned to the selected patient. Messaging for each team was centralized around a single team smartphone that was carried 24 hours a day, 7 days a week by a physician on that team. This removed the guesswork of trying to identify the individual physician responsible for that patient. For each message, a subject or issue and content were entered (Figure 2). Logic was also incorporated to reduce the amount of unnecessary interruptions. Senders would choose to send the message immediately as an interrupt message (urgent) for urgent/time sensitive issues or as an allow time to respond message (delayed). For the latter, the message was posted to the system where physicians could check and answer them. Interrupt messages were sent to the team smartphone using the Short Message Service (SMS) protocol. To try and ensure the communication loop on any issues was closed, when a message requested a response and did not receive it, the system sent another message. For urgent messages, a repeat message was initiated after 15 minutes. For delayed messages, the sender defined when they needed a response, typically within 2 to 8 hours. Senders were also able to select the mode of response that would best meet their needs from a workflow perspective: call back, text reply, or to specify that a reply was not required. Senders were also able to verify if the messages were received by the physician's smartphone. Physicians could view the messages within CM and reply. For messages that went to their team smartphone, physicians could respond from the smartphone through a secure Web link.

Figure 2

Because the messages were linked to the patients, they were visible to the entire care team, not just the message sender and recipient. If the care of the patient was transferred from 1 clinician to the next, the new clinician could easily review prior messages to understand recent patient events. The system was accessible through a browser on the intranet. The system regularly pulled patient demographic details such as name, age, medical record number, and location from our electronic medical record through a 1‐way interface. Information from this communication system was not considered part of the medical record but was retrievable.

The system was introduced as the new standard method of communication for nurses to reach physicians for all of the general internal medicine wards and for all medical teams at site 1 on May 2, 2011 and site 2 on June 6, 2011. The system replaced a text‐based Web‐paging system and supplemented the numeric pager carried by residents. Initial training of a half hour was provided to all nurses and residents.

RESULTS

DISCUSSION

We describe an implementation of a system to improve clinical communication in hospitals. The system was highly used and was perceived to improve communication by both nurses and physicians. Specifically, users found that the system increased efficiency, accountability, timeliness, and collaboration, but that there were issues with message clarity for complex medical issues.

Other systems and approaches have been implemented to improve communication. These included the use of alphanumeric pagers, e‐mail, secure texting, and smartphones. There is evidence that more advanced systems can improve efficiency for senders.[16] A recent randomized trial of secure text messaging found that it was perceived to be more efficient than paging, but overall usage was low and inconsistent.[8] There is also evidence that smartphones may increase interruptions, worsen interprofessional relationships, and cause issues with professional behavior.[10] Unfortunately, there are a limited number of interventions that improve communication, with some improving efficiency but none demonstrating improved patient‐oriented outcomes.[16, 17] This study evaluated a novel system, with functionality to link communication to patients, and created a system that aligned with the workflow of the clinicians. Messages were linked to the patient, not the sender or receiver, so other clinicians in the patient's circle of care could easily view the communication. Moreover, the system was designed to improve message response rates and allow for nonurgent messages.

Our communication system uses standard, commercially available components (smartphones, SMS), and relatively basic functionality (handover, secure messaging). Important findings are that the current system of paging can be transformed to a more efficient system that users will readily adopt. We found positive effects with components of the system. It appeared to improve efficiency and increase accountability. Accountability is crucial and moves from undocumented conversation to fully documented details of interactions. This can be used for both incident review and to review for quality improvement.

Using the system, physicians perceived that they were bothered by low‐value information, whereas nurses perceived a lack of response, and both found that the system was not ideal for complex messages. The mismatch between what physicians and nurses perceive as important has been attributed to their different timeframes and context.[18] For nurses with an upcoming change of shift, they wanted resolution of issues before handover. A physician on a different ward may not appreciate the context of a nurse having to directly interact with an irate family member. These different perceptions likely contributed to the lack of response to 8.6% of text messages. This is still better than other systems, such as paging, which can be as high as 33%.[10] For nonurgent items, clinicians would ideally check and clear items regularly from the system using a desktop computer, responding within the allotted timeframe. Unfortunately, this never became part of routine physician workflow, likely due to their busy workload, so many physicians would only respond when items became overdue. However, having a method to deal with nonurgent messages may have prevented some interruptions during protected educational times of trainees. The system was also not ideal for urgent or complex items. Complex items can be difficult to convey using the rarified communication medium of text messages.[19, 20] Urgent or complex issues are likely best resolved with a face‐to‐face or telephone conversation.

There are several limitations in our study that should be considered when interpreting the results. It is a study of usage and perceptions after implementation. Although more rigorous study is required to evaluate the effects, we see this as a first step in process improvement. Future research should measure the impact on improving patient care of this system and on patient outcomes such as adverse events. The study and intervention was limited to general internal medicine wards in 2 academic hospital settings where there are frequent rotations of medical personnel. The findings may not be generalizable to other hospital settings.

Future directions should be to further improve on the communication system and to educate and train staff on how to effectively communicate. Survey results showed that although users perceived increased efficiency, there was still significant opportunity to improve. One way to improve would be to have a mobile application in which physicians can easily review nonurgent items. Improvements could also be realized by educating clinicians on the use of the system and providing immediate feedback. Providing feedback to physicians on how well they respond could address nurses' issues around lack of timely response. By creating consensus between nurses and physicians on what is of high and low value to communicate could increase satisfaction for all users.

In summary, we present the usage and perceptions of a system designed to improve hospital communication. We found that there was high uptake, and that users perceived it to improve efficiency, collaboration, and accountability, but it may not be useful for communicating complex issues.

Previous studies have advocated the importance of effective communication between clinicians as a critical component in the provision of high‐quality patient care.[1, 2, 3, 4] There is increasing interest in the use of information and communication technologies to improve how clinicians communicate in hospital settings. A number of hospitals have implemented different solutions to improve communication. These solutions include alphanumeric pagers,[5] smartphones,[6] e‐mail,[7] secure text messaging,[8] and a Web‐based interdisciplinary communication tool.[9]

These systems have different limitations that render them inefficient and likely inhibit collaborative care. Current systems, such as pagers, rely on the sender to ensure the message was received and are successful in delivering messages approximately 67% of the time.[5, 9, 10] Although alphanumeric pagers and secure text messaging can increase the likelihood of delivery, these messages are often isolated and not easily viewable by the whole care team.[11] Improved systems should also reduce unnecessary interruptions by providing support for both urgent and delayed messages. Finally, messages should be stored and retrievable to enable increased accountability and allow for review for quality improvement initiatives.

It is also important to consider the unintended consequences of technology implementations.[12] Moving communication to text messages and smartphones has the potential to reduce interprofessional relations and can increase confusion if used for complex issues.[10, 13] In this article, we present a system designed to improve interprofessional communication on general internal medicine wards by incorporating these desired features and describe the usage and attitudes toward the system, specifically assessing for effects on multiple domains including efficiency, interprofessional collaboration, and relationships.

METHODS

Intervention

To address issues with communication, we developed a systemClinical Message (CM)that included 2 main components: a physician handover tool and secure messaging module. The focus of CM was to improve communication and information flow among different healthcare providers (physicians, nurses, pharmacists, social workers and therapists) through a secure, shared platform.

Physician Handover

The physician handover tool was designed to facilitate the physician handover process at shift change and is used as a patient rounding tool for day‐to‐day management of patients. It is also accessed by nurses and other clinicians to view the physicians' notes and to stay informed on the overall care plan. The tool contains standard elements including a list of patients with the following information on each patient: demographics, diagnosis, code status, medical history, active issues, and discharge plans (Figure 1).

Figure 1

Secure Messaging

Secure messaging was designed around our dominant communication: nurses sending messages to physicians who would then respond. Nurses and other health professionals sent messages to the medical teams by accessing CM, selecting the appropriate patient, and filling out a message template. The system automatically populated the To field with the team assigned to the selected patient. Messaging for each team was centralized around a single team smartphone that was carried 24 hours a day, 7 days a week by a physician on that team. This removed the guesswork of trying to identify the individual physician responsible for that patient. For each message, a subject or issue and content were entered (Figure 2). Logic was also incorporated to reduce the amount of unnecessary interruptions. Senders would choose to send the message immediately as an interrupt message (urgent) for urgent/time sensitive issues or as an allow time to respond message (delayed). For the latter, the message was posted to the system where physicians could check and answer them. Interrupt messages were sent to the team smartphone using the Short Message Service (SMS) protocol. To try and ensure the communication loop on any issues was closed, when a message requested a response and did not receive it, the system sent another message. For urgent messages, a repeat message was initiated after 15 minutes. For delayed messages, the sender defined when they needed a response, typically within 2 to 8 hours. Senders were also able to select the mode of response that would best meet their needs from a workflow perspective: call back, text reply, or to specify that a reply was not required. Senders were also able to verify if the messages were received by the physician's smartphone. Physicians could view the messages within CM and reply. For messages that went to their team smartphone, physicians could respond from the smartphone through a secure Web link.

Figure 2

Because the messages were linked to the patients, they were visible to the entire care team, not just the message sender and recipient. If the care of the patient was transferred from 1 clinician to the next, the new clinician could easily review prior messages to understand recent patient events. The system was accessible through a browser on the intranet. The system regularly pulled patient demographic details such as name, age, medical record number, and location from our electronic medical record through a 1‐way interface. Information from this communication system was not considered part of the medical record but was retrievable.

The system was introduced as the new standard method of communication for nurses to reach physicians for all of the general internal medicine wards and for all medical teams at site 1 on May 2, 2011 and site 2 on June 6, 2011. The system replaced a text‐based Web‐paging system and supplemented the numeric pager carried by residents. Initial training of a half hour was provided to all nurses and residents.

RESULTS

DISCUSSION

We describe an implementation of a system to improve clinical communication in hospitals. The system was highly used and was perceived to improve communication by both nurses and physicians. Specifically, users found that the system increased efficiency, accountability, timeliness, and collaboration, but that there were issues with message clarity for complex medical issues.

Other systems and approaches have been implemented to improve communication. These included the use of alphanumeric pagers, e‐mail, secure texting, and smartphones. There is evidence that more advanced systems can improve efficiency for senders.[16] A recent randomized trial of secure text messaging found that it was perceived to be more efficient than paging, but overall usage was low and inconsistent.[8] There is also evidence that smartphones may increase interruptions, worsen interprofessional relationships, and cause issues with professional behavior.[10] Unfortunately, there are a limited number of interventions that improve communication, with some improving efficiency but none demonstrating improved patient‐oriented outcomes.[16, 17] This study evaluated a novel system, with functionality to link communication to patients, and created a system that aligned with the workflow of the clinicians. Messages were linked to the patient, not the sender or receiver, so other clinicians in the patient's circle of care could easily view the communication. Moreover, the system was designed to improve message response rates and allow for nonurgent messages.

Our communication system uses standard, commercially available components (smartphones, SMS), and relatively basic functionality (handover, secure messaging). Important findings are that the current system of paging can be transformed to a more efficient system that users will readily adopt. We found positive effects with components of the system. It appeared to improve efficiency and increase accountability. Accountability is crucial and moves from undocumented conversation to fully documented details of interactions. This can be used for both incident review and to review for quality improvement.

Using the system, physicians perceived that they were bothered by low‐value information, whereas nurses perceived a lack of response, and both found that the system was not ideal for complex messages. The mismatch between what physicians and nurses perceive as important has been attributed to their different timeframes and context.[18] For nurses with an upcoming change of shift, they wanted resolution of issues before handover. A physician on a different ward may not appreciate the context of a nurse having to directly interact with an irate family member. These different perceptions likely contributed to the lack of response to 8.6% of text messages. This is still better than other systems, such as paging, which can be as high as 33%.[10] For nonurgent items, clinicians would ideally check and clear items regularly from the system using a desktop computer, responding within the allotted timeframe. Unfortunately, this never became part of routine physician workflow, likely due to their busy workload, so many physicians would only respond when items became overdue. However, having a method to deal with nonurgent messages may have prevented some interruptions during protected educational times of trainees. The system was also not ideal for urgent or complex items. Complex items can be difficult to convey using the rarified communication medium of text messages.[19, 20] Urgent or complex issues are likely best resolved with a face‐to‐face or telephone conversation.

There are several limitations in our study that should be considered when interpreting the results. It is a study of usage and perceptions after implementation. Although more rigorous study is required to evaluate the effects, we see this as a first step in process improvement. Future research should measure the impact on improving patient care of this system and on patient outcomes such as adverse events. The study and intervention was limited to general internal medicine wards in 2 academic hospital settings where there are frequent rotations of medical personnel. The findings may not be generalizable to other hospital settings.

Future directions should be to further improve on the communication system and to educate and train staff on how to effectively communicate. Survey results showed that although users perceived increased efficiency, there was still significant opportunity to improve. One way to improve would be to have a mobile application in which physicians can easily review nonurgent items. Improvements could also be realized by educating clinicians on the use of the system and providing immediate feedback. Providing feedback to physicians on how well they respond could address nurses' issues around lack of timely response. By creating consensus between nurses and physicians on what is of high and low value to communicate could increase satisfaction for all users.

In summary, we present the usage and perceptions of a system designed to improve hospital communication. We found that there was high uptake, and that users perceived it to improve efficiency, collaboration, and accountability, but it may not be useful for communicating complex issues.