Original Research

When Medicare beneficiaries seek healthcare, they are increasingly likely to have that care delivered under observation status. From 2006 to 2010, the annual number of observation hours for Medicare beneficiaries rose by nearly 70%.[1] In 2012, the number of observation stays for Medicare beneficiaries reached 1.5 million.[2] One consequence of this trend is a potential change in patient financial liabilitythe amount patients are expected to pay out of pocket for care. Although observation care is usually delivered in a hospital, Medicare classifies it as an outpatient service, covered through Part B rather than inpatient Part A. In two‐thirds of US hospitals, observation care is largely an administrative classification, delivered in the same units and beds as admitted patients rather than in a protocol‐driven observation care unit.[3] Therefore, patients are often unaware of their outpatient observation status and its financial implications until they receive their hospital bill.

Observation has the potential to impact patient financial liability through 4 mechanisms.[4] First, instead of a fixed cost for an inpatient admission (eg, a fixed deductible for a hospital admission), patients pay a percentage of the cost of each service provided. Therefore, patients who have long observation stays or receive expensive services could have higher than expected liability. A recent study using all‐payer data demonstrated that patients with longer observation stays (greater than 24 hours) paid 21% more than for those with shorter stays.[5]

A second consideration is that Medicare does not cover the same hospital services for observation care as it does for inpatient care. For example, self‐administered medications are generally not covered for beneficiaries receiving observation care. However, the Office of the Inspector General (OIG)[2] recently found that the average patient cost per observation stay in 2012even including the cost of self‐administered medicationswas $528. This was significantly lower than the inpatient deductible ($1156 in 2012) that patients would have paid had they been admitted. Although on average patients paid less for observation care, the OIG report found that 6% of observation stays were more costly to patients than inpatient admissions.

Third, there are certain benefits that Medicare beneficiaries are not eligible for unless they are admitted to the hospital. For a beneficiary to receive skilled nursing facility (SNF) benefits, they must be admitted to the hospital for 3 or more days. This was the basis for Bagnall v Sebelius, a class action lawsuit against the Centers for Medicare & Medicaid Services (CMS) filed in 2009 by the Center for Medicare Advocacy.[6] The OIG estimated that in 2012, Medicare beneficiaries had 600,000 observation stays longer than 3 days that failed to qualify them for SNF services. Since then, CMS created the 2‐midnight rule,[7] stating that CMS will assign inpatient status to all medically necessary stays of 2 midnights or longer. This rule was intended, in part, to curb the use of observation stays greater than 48 hours and was a key factor in Judge Michael Shea's decision to dismiss Bagnall v Sebelius.[6]

Finally, Medicare beneficiaries who must revisit the hospital may have greater cumulative costs under observation care versus inpatient care. Medicare beneficiaries are partially protected from accumulating high costs over multiple inpatient admissions by a benefit design known as the benefit period. A benefit period begins the day a beneficiary is admitted to a hospital or SNF, and ends when he or she has not received any inpatient hospital or SNF care for 60 days in a row. Beneficiaries pay the inpatient deductible only once per benefit period, even if they have multiple readmissions during this time. So, for example, if a beneficiary was readmitted to the hospital 59 days after discharge, he or she would not have to pay the inpatient deductible again. In addition, the benefit period would be extended for an additional 60 days. In contrast, beneficiaries who receive observation care are subject to coinsurance at every subsequent visit; therefore, these beneficiaries could accrue high cumulative costs over multiple observation stays.

To our knowledge, there have been no published studies focusing on the potentially vulnerable population of Medicare beneficiaries who frequently use observation care. Our objectives were to determine the financial liability for patients who have multiple observation stays within a 60‐day period, and then compare this to the inpatient deductible they would have paid as inpatients.

METHODS

RESULTS

Of the 7,470,676 unique Medicare beneficiaries in the 20% denominator file, 691,760 (9.3%) had at least 1 observation visit during the 3‐year study period (Table 1). The proportion of beneficiaries using observation care rose in each year of the study; 4.1% of beneficiaries used observation care in 2010, 4.4% in 2011, and 5.0% in 2012.

Of the beneficiaries receiving observation care over the entire study period, 41,385 (6.0%) had multiple visits (2 observation visits in any 60‐day interval). The number of beneficiaries with multiple visits grew by 21.9% from 2010 to 2012. There were racial differences in the use of observation care; patients with multiple visits were more likely to be black than those without multiple visits or those not receiving observation care (14.3% vs 11.4% vs 10.0, P < 0.01). Multiple observation visits were also associated with a higher number of chronic conditions (3.6 vs 2.8 vs 1.7, P < 0.01) (Table 1).

The mean financial liability for the first observation stay for each beneficiary in our study sample was $469.42 (442.43) (Table 2). This is significantly lower than the standard inpatient deductible of $1100 (p<0.01). For 9.2% of beneficiaries, the financial liability was greater than the inpatient deductible.

The cumulative mean financial liability for beneficiaries with 2 stays in a 60‐day interval was $947.40 (803.62) (Table 2). This is significantly lower than the standard inpatient deductible of $1100 (P < 0.01). However, for 26.6% of beneficiaries, cumulative financial liability was greater than the $1100 inpatient deductible, which is what they would have paid had these hospital visits been inpatient admissions (Figure 1).

Figure 1

There were several factors associated with having this excess cumulative liability (Table 3). Higher frequency of observation visits within a 60‐day period was associated with high liability (odds ratio [OR]: 2.0, 95% confidence interval [CI]: 1.9‐2.1). In addition, having an index hospitalization in the Northeast region of the country was associated with lower odds of being in the high‐liability group (OR: 0.51, 95% CI: 0.47‐0.55). High liability was weakly associated with lower CMS‐HCC risk scores, nondual eligibility, nonblack race, and index hospital stay at an academic, urban, or nonprofit hospital.

DISCUSSION

Our findings suggest that for 91% of Medicare beneficiaries, a single observation stay was less costly than an inpatient admission. However, when beneficiaries had to return to observation care within 60 days of a prior stay, on average, their cumulative costs went up to $947. For more than a quarter of beneficiaries with multiple observation visits, the cumulative costs of these observation visits exceeded the inpatient deductible.

The results of this study are consistent with prior studies of observation care. We found that in 2010, 4.1% of Medicare beneficiaries used observation care, consistent with the estimated 4.0% in 2009 reported by the AARP Public Policy Institute.[14] Also, consistent with the growth rate from the AARP report, we found growth in use of observation care from 4.1% in 2010 to 5.0% in 2012. We found that the mean length of stay for observation care was 30 hours, consistent with recent studies estimating mean length of stay in 2009 as 25.9 hours.[15] We found that beneficiaries paid an average of $468.50 per observation care stay, very close to the $401 estimated by the 2013 OIG report (when self‐administered drugs were excluded).[2] The difference may be explained by the fact that OIG included observation stays of <8 hours in their sample; we excluded these stays because they did not meet criteria for Medicare payment. Like the OIG report, we also found that the vast majority (91%) of beneficiaries pay less for any given observation stay than for an inpatient stay.

However, our findings raise the concern that for a significant proportion of beneficiaries who are likely to return to the hospital, cumulative costs of multiple observation stays may be greater than the inpatient deductible. Therefore, although observation care is, on average, less expensive for beneficiaries than inpatient admission, beneficiaries lack the protection from escalating financial liability over multiple visits.

This finding is worrisome for 3 reasons. First, compared with the general beneficiary population, Medicare beneficiaries who return to the hospital frequently are also typically of lower socioeconomic status[16, 17, 18] and may be disproportionately affected by any increased financial liability. Interestingly, our analysis showed that patients with high financial liability incurred from multiple observation stays actually had a lower comorbidity burden than patients in the multiple observation stay group with lower liability, and were less likely to be black or dual eligible. This finding perhaps reflects the fact that very high‐risk patients who returned to the hospital were readmitted rather than being placed under observation status again, potentially depleting the high‐liability group of patients with these high‐risk characteristics. Second, patients have little control over their classification as observation versus inpatients. In many hospitals, observation is simply an administrative classification for care thatfrom the patients' perspectiveis identical to inpatient care.[4] It is problematic to expose patients to varying financial liability based on differences in administrative classification. Finally, we found that the number of patients with multiple observation visits within a 60‐day period rose by 22% between 2010 and 2012. This means that the problem of excess cumulative financial liability is likely to be increasingly common over the coming years. The increased incidence of multiple observation visits may be simply related to overall increases in use of observation care. Alternatively, some authors worry that this trend may be driven by hospital use of observation care for patients who are likely to be readmitted.[14, 19] A recent analysis by Gerhardt et al.[20] did not find evidence of direct substitution of observation care in the 30‐day window after an index admission. This suggests that physicians are not explicitly shifting patients to observation care in order to avert a readmission and the readmissions penalty.[21] However, it does not exclude the possibility of general shifts toward observation care for patients likely to return.

Experts have suggested capping the total out‐of‐pocket expense for observation care at the inpatient‐deductible amount.[4] This deductible cap would prevent the relatively rare case in which a single observation stay costs more than an inpatient admission. Our findings suggest that a benefit period (as in Part A) during which such a deductible would serve as a cap would also protect a small but significantly impacted population from higher than expected cumulative costs for multiple observation care visits.

This study has several limitations. First, we are only able to measure beneficiary financial responsibility and not the amount actually paid. This can differ from financial responsibility when patients do not pay their bill, when patients accrue additional charges (such as self‐administered medications) that are not reflected on outpatient claims, or when patients have additional third‐party payers who cover part or all of the financial responsibility (as with dually eligible patients). For such beneficiaries with supplemental coverage, out‐of‐pocket cost in both scenarios (inpatient or observation care) may be low or zero. However, the use of financial responsibility as an approximation of actual payment amounts is recommended by the Research Data Assistance Center and is consistent with other studies of cost in observation care.[2] Second, our data source only allowed us to assess facilities fees and not professional expenses. Our comparator of the inpatient deductible also only reflects facilities fees, making this a valid comparison. Third, we selected 60 days as the time interval for defining multiple visits. This interval is intended to approximate a Medicare benefit period, which is the time interval following a discharge from a hospital or an SNF until the time when the deductible resets. However, Medicare actually extends the benefit period another 60 days if a patient is readmitted during that 60‐day period. Thus, 60 days is actually the shortest possible benefit period. By conservatively defining the interval for recurrent observation stays in this way, we are likely underestimating the number and cost of observation stays in a true benefit period, and biasing our results toward the null.

In conclusion, our findings suggest that a significant proportion of Medicare beneficiaries who revisit observation care pay more than they would have had they been readmitted. As CMS policies on observation care continue to evolve, it may be helpful to consider measures to cap total out‐of‐pocket expenses within a benefit period to protect beneficiaries from higher than expected costs.

When Medicare beneficiaries seek healthcare, they are increasingly likely to have that care delivered under observation status. From 2006 to 2010, the annual number of observation hours for Medicare beneficiaries rose by nearly 70%.[1] In 2012, the number of observation stays for Medicare beneficiaries reached 1.5 million.[2] One consequence of this trend is a potential change in patient financial liabilitythe amount patients are expected to pay out of pocket for care. Although observation care is usually delivered in a hospital, Medicare classifies it as an outpatient service, covered through Part B rather than inpatient Part A. In two‐thirds of US hospitals, observation care is largely an administrative classification, delivered in the same units and beds as admitted patients rather than in a protocol‐driven observation care unit.[3] Therefore, patients are often unaware of their outpatient observation status and its financial implications until they receive their hospital bill.

Observation has the potential to impact patient financial liability through 4 mechanisms.[4] First, instead of a fixed cost for an inpatient admission (eg, a fixed deductible for a hospital admission), patients pay a percentage of the cost of each service provided. Therefore, patients who have long observation stays or receive expensive services could have higher than expected liability. A recent study using all‐payer data demonstrated that patients with longer observation stays (greater than 24 hours) paid 21% more than for those with shorter stays.[5]

A second consideration is that Medicare does not cover the same hospital services for observation care as it does for inpatient care. For example, self‐administered medications are generally not covered for beneficiaries receiving observation care. However, the Office of the Inspector General (OIG)[2] recently found that the average patient cost per observation stay in 2012even including the cost of self‐administered medicationswas $528. This was significantly lower than the inpatient deductible ($1156 in 2012) that patients would have paid had they been admitted. Although on average patients paid less for observation care, the OIG report found that 6% of observation stays were more costly to patients than inpatient admissions.

Third, there are certain benefits that Medicare beneficiaries are not eligible for unless they are admitted to the hospital. For a beneficiary to receive skilled nursing facility (SNF) benefits, they must be admitted to the hospital for 3 or more days. This was the basis for Bagnall v Sebelius, a class action lawsuit against the Centers for Medicare & Medicaid Services (CMS) filed in 2009 by the Center for Medicare Advocacy.[6] The OIG estimated that in 2012, Medicare beneficiaries had 600,000 observation stays longer than 3 days that failed to qualify them for SNF services. Since then, CMS created the 2‐midnight rule,[7] stating that CMS will assign inpatient status to all medically necessary stays of 2 midnights or longer. This rule was intended, in part, to curb the use of observation stays greater than 48 hours and was a key factor in Judge Michael Shea's decision to dismiss Bagnall v Sebelius.[6]

Finally, Medicare beneficiaries who must revisit the hospital may have greater cumulative costs under observation care versus inpatient care. Medicare beneficiaries are partially protected from accumulating high costs over multiple inpatient admissions by a benefit design known as the benefit period. A benefit period begins the day a beneficiary is admitted to a hospital or SNF, and ends when he or she has not received any inpatient hospital or SNF care for 60 days in a row. Beneficiaries pay the inpatient deductible only once per benefit period, even if they have multiple readmissions during this time. So, for example, if a beneficiary was readmitted to the hospital 59 days after discharge, he or she would not have to pay the inpatient deductible again. In addition, the benefit period would be extended for an additional 60 days. In contrast, beneficiaries who receive observation care are subject to coinsurance at every subsequent visit; therefore, these beneficiaries could accrue high cumulative costs over multiple observation stays.

To our knowledge, there have been no published studies focusing on the potentially vulnerable population of Medicare beneficiaries who frequently use observation care. Our objectives were to determine the financial liability for patients who have multiple observation stays within a 60‐day period, and then compare this to the inpatient deductible they would have paid as inpatients.

METHODS

RESULTS

Of the 7,470,676 unique Medicare beneficiaries in the 20% denominator file, 691,760 (9.3%) had at least 1 observation visit during the 3‐year study period (Table 1). The proportion of beneficiaries using observation care rose in each year of the study; 4.1% of beneficiaries used observation care in 2010, 4.4% in 2011, and 5.0% in 2012.

Of the beneficiaries receiving observation care over the entire study period, 41,385 (6.0%) had multiple visits (2 observation visits in any 60‐day interval). The number of beneficiaries with multiple visits grew by 21.9% from 2010 to 2012. There were racial differences in the use of observation care; patients with multiple visits were more likely to be black than those without multiple visits or those not receiving observation care (14.3% vs 11.4% vs 10.0, P < 0.01). Multiple observation visits were also associated with a higher number of chronic conditions (3.6 vs 2.8 vs 1.7, P < 0.01) (Table 1).

The mean financial liability for the first observation stay for each beneficiary in our study sample was $469.42 (442.43) (Table 2). This is significantly lower than the standard inpatient deductible of $1100 (p<0.01). For 9.2% of beneficiaries, the financial liability was greater than the inpatient deductible.

The cumulative mean financial liability for beneficiaries with 2 stays in a 60‐day interval was $947.40 (803.62) (Table 2). This is significantly lower than the standard inpatient deductible of $1100 (P < 0.01). However, for 26.6% of beneficiaries, cumulative financial liability was greater than the $1100 inpatient deductible, which is what they would have paid had these hospital visits been inpatient admissions (Figure 1).

Figure 1

There were several factors associated with having this excess cumulative liability (Table 3). Higher frequency of observation visits within a 60‐day period was associated with high liability (odds ratio [OR]: 2.0, 95% confidence interval [CI]: 1.9‐2.1). In addition, having an index hospitalization in the Northeast region of the country was associated with lower odds of being in the high‐liability group (OR: 0.51, 95% CI: 0.47‐0.55). High liability was weakly associated with lower CMS‐HCC risk scores, nondual eligibility, nonblack race, and index hospital stay at an academic, urban, or nonprofit hospital.

DISCUSSION

Our findings suggest that for 91% of Medicare beneficiaries, a single observation stay was less costly than an inpatient admission. However, when beneficiaries had to return to observation care within 60 days of a prior stay, on average, their cumulative costs went up to $947. For more than a quarter of beneficiaries with multiple observation visits, the cumulative costs of these observation visits exceeded the inpatient deductible.

The results of this study are consistent with prior studies of observation care. We found that in 2010, 4.1% of Medicare beneficiaries used observation care, consistent with the estimated 4.0% in 2009 reported by the AARP Public Policy Institute.[14] Also, consistent with the growth rate from the AARP report, we found growth in use of observation care from 4.1% in 2010 to 5.0% in 2012. We found that the mean length of stay for observation care was 30 hours, consistent with recent studies estimating mean length of stay in 2009 as 25.9 hours.[15] We found that beneficiaries paid an average of $468.50 per observation care stay, very close to the $401 estimated by the 2013 OIG report (when self‐administered drugs were excluded).[2] The difference may be explained by the fact that OIG included observation stays of <8 hours in their sample; we excluded these stays because they did not meet criteria for Medicare payment. Like the OIG report, we also found that the vast majority (91%) of beneficiaries pay less for any given observation stay than for an inpatient stay.

However, our findings raise the concern that for a significant proportion of beneficiaries who are likely to return to the hospital, cumulative costs of multiple observation stays may be greater than the inpatient deductible. Therefore, although observation care is, on average, less expensive for beneficiaries than inpatient admission, beneficiaries lack the protection from escalating financial liability over multiple visits.

This finding is worrisome for 3 reasons. First, compared with the general beneficiary population, Medicare beneficiaries who return to the hospital frequently are also typically of lower socioeconomic status[16, 17, 18] and may be disproportionately affected by any increased financial liability. Interestingly, our analysis showed that patients with high financial liability incurred from multiple observation stays actually had a lower comorbidity burden than patients in the multiple observation stay group with lower liability, and were less likely to be black or dual eligible. This finding perhaps reflects the fact that very high‐risk patients who returned to the hospital were readmitted rather than being placed under observation status again, potentially depleting the high‐liability group of patients with these high‐risk characteristics. Second, patients have little control over their classification as observation versus inpatients. In many hospitals, observation is simply an administrative classification for care thatfrom the patients' perspectiveis identical to inpatient care.[4] It is problematic to expose patients to varying financial liability based on differences in administrative classification. Finally, we found that the number of patients with multiple observation visits within a 60‐day period rose by 22% between 2010 and 2012. This means that the problem of excess cumulative financial liability is likely to be increasingly common over the coming years. The increased incidence of multiple observation visits may be simply related to overall increases in use of observation care. Alternatively, some authors worry that this trend may be driven by hospital use of observation care for patients who are likely to be readmitted.[14, 19] A recent analysis by Gerhardt et al.[20] did not find evidence of direct substitution of observation care in the 30‐day window after an index admission. This suggests that physicians are not explicitly shifting patients to observation care in order to avert a readmission and the readmissions penalty.[21] However, it does not exclude the possibility of general shifts toward observation care for patients likely to return.

Experts have suggested capping the total out‐of‐pocket expense for observation care at the inpatient‐deductible amount.[4] This deductible cap would prevent the relatively rare case in which a single observation stay costs more than an inpatient admission. Our findings suggest that a benefit period (as in Part A) during which such a deductible would serve as a cap would also protect a small but significantly impacted population from higher than expected cumulative costs for multiple observation care visits.

This study has several limitations. First, we are only able to measure beneficiary financial responsibility and not the amount actually paid. This can differ from financial responsibility when patients do not pay their bill, when patients accrue additional charges (such as self‐administered medications) that are not reflected on outpatient claims, or when patients have additional third‐party payers who cover part or all of the financial responsibility (as with dually eligible patients). For such beneficiaries with supplemental coverage, out‐of‐pocket cost in both scenarios (inpatient or observation care) may be low or zero. However, the use of financial responsibility as an approximation of actual payment amounts is recommended by the Research Data Assistance Center and is consistent with other studies of cost in observation care.[2] Second, our data source only allowed us to assess facilities fees and not professional expenses. Our comparator of the inpatient deductible also only reflects facilities fees, making this a valid comparison. Third, we selected 60 days as the time interval for defining multiple visits. This interval is intended to approximate a Medicare benefit period, which is the time interval following a discharge from a hospital or an SNF until the time when the deductible resets. However, Medicare actually extends the benefit period another 60 days if a patient is readmitted during that 60‐day period. Thus, 60 days is actually the shortest possible benefit period. By conservatively defining the interval for recurrent observation stays in this way, we are likely underestimating the number and cost of observation stays in a true benefit period, and biasing our results toward the null.

In conclusion, our findings suggest that a significant proportion of Medicare beneficiaries who revisit observation care pay more than they would have had they been readmitted. As CMS policies on observation care continue to evolve, it may be helpful to consider measures to cap total out‐of‐pocket expenses within a benefit period to protect beneficiaries from higher than expected costs.

When Medicare beneficiaries seek healthcare, they are increasingly likely to have that care delivered under observation status. From 2006 to 2010, the annual number of observation hours for Medicare beneficiaries rose by nearly 70%.[1] In 2012, the number of observation stays for Medicare beneficiaries reached 1.5 million.[2] One consequence of this trend is a potential change in patient financial liabilitythe amount patients are expected to pay out of pocket for care. Although observation care is usually delivered in a hospital, Medicare classifies it as an outpatient service, covered through Part B rather than inpatient Part A. In two‐thirds of US hospitals, observation care is largely an administrative classification, delivered in the same units and beds as admitted patients rather than in a protocol‐driven observation care unit.[3] Therefore, patients are often unaware of their outpatient observation status and its financial implications until they receive their hospital bill.

Observation has the potential to impact patient financial liability through 4 mechanisms.[4] First, instead of a fixed cost for an inpatient admission (eg, a fixed deductible for a hospital admission), patients pay a percentage of the cost of each service provided. Therefore, patients who have long observation stays or receive expensive services could have higher than expected liability. A recent study using all‐payer data demonstrated that patients with longer observation stays (greater than 24 hours) paid 21% more than for those with shorter stays.[5]

A second consideration is that Medicare does not cover the same hospital services for observation care as it does for inpatient care. For example, self‐administered medications are generally not covered for beneficiaries receiving observation care. However, the Office of the Inspector General (OIG)[2] recently found that the average patient cost per observation stay in 2012even including the cost of self‐administered medicationswas $528. This was significantly lower than the inpatient deductible ($1156 in 2012) that patients would have paid had they been admitted. Although on average patients paid less for observation care, the OIG report found that 6% of observation stays were more costly to patients than inpatient admissions.

Third, there are certain benefits that Medicare beneficiaries are not eligible for unless they are admitted to the hospital. For a beneficiary to receive skilled nursing facility (SNF) benefits, they must be admitted to the hospital for 3 or more days. This was the basis for Bagnall v Sebelius, a class action lawsuit against the Centers for Medicare & Medicaid Services (CMS) filed in 2009 by the Center for Medicare Advocacy.[6] The OIG estimated that in 2012, Medicare beneficiaries had 600,000 observation stays longer than 3 days that failed to qualify them for SNF services. Since then, CMS created the 2‐midnight rule,[7] stating that CMS will assign inpatient status to all medically necessary stays of 2 midnights or longer. This rule was intended, in part, to curb the use of observation stays greater than 48 hours and was a key factor in Judge Michael Shea's decision to dismiss Bagnall v Sebelius.[6]

Finally, Medicare beneficiaries who must revisit the hospital may have greater cumulative costs under observation care versus inpatient care. Medicare beneficiaries are partially protected from accumulating high costs over multiple inpatient admissions by a benefit design known as the benefit period. A benefit period begins the day a beneficiary is admitted to a hospital or SNF, and ends when he or she has not received any inpatient hospital or SNF care for 60 days in a row. Beneficiaries pay the inpatient deductible only once per benefit period, even if they have multiple readmissions during this time. So, for example, if a beneficiary was readmitted to the hospital 59 days after discharge, he or she would not have to pay the inpatient deductible again. In addition, the benefit period would be extended for an additional 60 days. In contrast, beneficiaries who receive observation care are subject to coinsurance at every subsequent visit; therefore, these beneficiaries could accrue high cumulative costs over multiple observation stays.

To our knowledge, there have been no published studies focusing on the potentially vulnerable population of Medicare beneficiaries who frequently use observation care. Our objectives were to determine the financial liability for patients who have multiple observation stays within a 60‐day period, and then compare this to the inpatient deductible they would have paid as inpatients.

METHODS

RESULTS

Of the 7,470,676 unique Medicare beneficiaries in the 20% denominator file, 691,760 (9.3%) had at least 1 observation visit during the 3‐year study period (Table 1). The proportion of beneficiaries using observation care rose in each year of the study; 4.1% of beneficiaries used observation care in 2010, 4.4% in 2011, and 5.0% in 2012.

Of the beneficiaries receiving observation care over the entire study period, 41,385 (6.0%) had multiple visits (2 observation visits in any 60‐day interval). The number of beneficiaries with multiple visits grew by 21.9% from 2010 to 2012. There were racial differences in the use of observation care; patients with multiple visits were more likely to be black than those without multiple visits or those not receiving observation care (14.3% vs 11.4% vs 10.0, P < 0.01). Multiple observation visits were also associated with a higher number of chronic conditions (3.6 vs 2.8 vs 1.7, P < 0.01) (Table 1).

The mean financial liability for the first observation stay for each beneficiary in our study sample was $469.42 (442.43) (Table 2). This is significantly lower than the standard inpatient deductible of $1100 (p<0.01). For 9.2% of beneficiaries, the financial liability was greater than the inpatient deductible.

The cumulative mean financial liability for beneficiaries with 2 stays in a 60‐day interval was $947.40 (803.62) (Table 2). This is significantly lower than the standard inpatient deductible of $1100 (P < 0.01). However, for 26.6% of beneficiaries, cumulative financial liability was greater than the $1100 inpatient deductible, which is what they would have paid had these hospital visits been inpatient admissions (Figure 1).

Figure 1

There were several factors associated with having this excess cumulative liability (Table 3). Higher frequency of observation visits within a 60‐day period was associated with high liability (odds ratio [OR]: 2.0, 95% confidence interval [CI]: 1.9‐2.1). In addition, having an index hospitalization in the Northeast region of the country was associated with lower odds of being in the high‐liability group (OR: 0.51, 95% CI: 0.47‐0.55). High liability was weakly associated with lower CMS‐HCC risk scores, nondual eligibility, nonblack race, and index hospital stay at an academic, urban, or nonprofit hospital.

DISCUSSION

Our findings suggest that for 91% of Medicare beneficiaries, a single observation stay was less costly than an inpatient admission. However, when beneficiaries had to return to observation care within 60 days of a prior stay, on average, their cumulative costs went up to $947. For more than a quarter of beneficiaries with multiple observation visits, the cumulative costs of these observation visits exceeded the inpatient deductible.

The results of this study are consistent with prior studies of observation care. We found that in 2010, 4.1% of Medicare beneficiaries used observation care, consistent with the estimated 4.0% in 2009 reported by the AARP Public Policy Institute.[14] Also, consistent with the growth rate from the AARP report, we found growth in use of observation care from 4.1% in 2010 to 5.0% in 2012. We found that the mean length of stay for observation care was 30 hours, consistent with recent studies estimating mean length of stay in 2009 as 25.9 hours.[15] We found that beneficiaries paid an average of $468.50 per observation care stay, very close to the $401 estimated by the 2013 OIG report (when self‐administered drugs were excluded).[2] The difference may be explained by the fact that OIG included observation stays of <8 hours in their sample; we excluded these stays because they did not meet criteria for Medicare payment. Like the OIG report, we also found that the vast majority (91%) of beneficiaries pay less for any given observation stay than for an inpatient stay.

However, our findings raise the concern that for a significant proportion of beneficiaries who are likely to return to the hospital, cumulative costs of multiple observation stays may be greater than the inpatient deductible. Therefore, although observation care is, on average, less expensive for beneficiaries than inpatient admission, beneficiaries lack the protection from escalating financial liability over multiple visits.

This finding is worrisome for 3 reasons. First, compared with the general beneficiary population, Medicare beneficiaries who return to the hospital frequently are also typically of lower socioeconomic status[16, 17, 18] and may be disproportionately affected by any increased financial liability. Interestingly, our analysis showed that patients with high financial liability incurred from multiple observation stays actually had a lower comorbidity burden than patients in the multiple observation stay group with lower liability, and were less likely to be black or dual eligible. This finding perhaps reflects the fact that very high‐risk patients who returned to the hospital were readmitted rather than being placed under observation status again, potentially depleting the high‐liability group of patients with these high‐risk characteristics. Second, patients have little control over their classification as observation versus inpatients. In many hospitals, observation is simply an administrative classification for care thatfrom the patients' perspectiveis identical to inpatient care.[4] It is problematic to expose patients to varying financial liability based on differences in administrative classification. Finally, we found that the number of patients with multiple observation visits within a 60‐day period rose by 22% between 2010 and 2012. This means that the problem of excess cumulative financial liability is likely to be increasingly common over the coming years. The increased incidence of multiple observation visits may be simply related to overall increases in use of observation care. Alternatively, some authors worry that this trend may be driven by hospital use of observation care for patients who are likely to be readmitted.[14, 19] A recent analysis by Gerhardt et al.[20] did not find evidence of direct substitution of observation care in the 30‐day window after an index admission. This suggests that physicians are not explicitly shifting patients to observation care in order to avert a readmission and the readmissions penalty.[21] However, it does not exclude the possibility of general shifts toward observation care for patients likely to return.

Experts have suggested capping the total out‐of‐pocket expense for observation care at the inpatient‐deductible amount.[4] This deductible cap would prevent the relatively rare case in which a single observation stay costs more than an inpatient admission. Our findings suggest that a benefit period (as in Part A) during which such a deductible would serve as a cap would also protect a small but significantly impacted population from higher than expected cumulative costs for multiple observation care visits.

This study has several limitations. First, we are only able to measure beneficiary financial responsibility and not the amount actually paid. This can differ from financial responsibility when patients do not pay their bill, when patients accrue additional charges (such as self‐administered medications) that are not reflected on outpatient claims, or when patients have additional third‐party payers who cover part or all of the financial responsibility (as with dually eligible patients). For such beneficiaries with supplemental coverage, out‐of‐pocket cost in both scenarios (inpatient or observation care) may be low or zero. However, the use of financial responsibility as an approximation of actual payment amounts is recommended by the Research Data Assistance Center and is consistent with other studies of cost in observation care.[2] Second, our data source only allowed us to assess facilities fees and not professional expenses. Our comparator of the inpatient deductible also only reflects facilities fees, making this a valid comparison. Third, we selected 60 days as the time interval for defining multiple visits. This interval is intended to approximate a Medicare benefit period, which is the time interval following a discharge from a hospital or an SNF until the time when the deductible resets. However, Medicare actually extends the benefit period another 60 days if a patient is readmitted during that 60‐day period. Thus, 60 days is actually the shortest possible benefit period. By conservatively defining the interval for recurrent observation stays in this way, we are likely underestimating the number and cost of observation stays in a true benefit period, and biasing our results toward the null.

In conclusion, our findings suggest that a significant proportion of Medicare beneficiaries who revisit observation care pay more than they would have had they been readmitted. As CMS policies on observation care continue to evolve, it may be helpful to consider measures to cap total out‐of‐pocket expenses within a benefit period to protect beneficiaries from higher than expected costs.

Despite an estimated annual $2.6 trillion expenditure on healthcare, the United States performs poorly on indicators of health and harm during care.[1, 2, 3] Hospitals around the nation are working to improve the care they deliver. We describe a model developed at our institution and report the evaluation of the outcomes associated with its implementation on the general medical and surgical units. The Indiana University Institutional Review Board approved this work.

SETTING AND DEFINITIONS

Indiana University Health Methodist Hospital (MH) is an academic center in Indianapolis, Indiana, serving over 30,000 patients annually.[4] In 2012, responding to the coexisting needs to improve quality and contain costs, the MH leadership team redesigned care in the hospital. The new model centers around accountable care teams (ACTs). Each ACT is a geographically defined set of providers accepting ownership for the clinical, service, and financial outcomes of their respective inpatient unit. The units studied are described in Table 1.

THE ACT MODEL

The model comprises 8 interventions rooted in 3 foundational domains: (1) enhancing interprofessional collaboration (IPC), (2) enabling data‐driven decisions, and (3) providing leadership. Each intervention is briefly described under its main focus (see Supporting Information, Appendix A, in the online version of this article for further details).

Implementation Timelines and ACT Scores

The development of the ACTs started in the spring of 2012. Physician, nursing, and pharmacy support was sought, and a pilot unit was formed in August 2012. The model was cascaded hospital wide by December 2013, with support from the ACT program director (A.N.). The program director observed and scored the uptake of each intervention by each unit monthly. A score of 1 denoted no implementation, whereas 5 denoted complete implementation. The criteria for scoring are presented in Table 2. The monthly scores for all 8 interventions in each of the 11 units were averaged as an overall ACT score, which reflects the implementation dose of the ACT model. Monthly domain scores for enhancing IPC and enabling data‐driven decisions were also calculated as the average score within each domain. This yielded 3 domain scores. Figure 1A plots by month the overall ACT score for the medical and surgical units, and Figure 1B plots the implementation score for the 3 domains between August 2012 and December 2013 for all units. The uptake of the interventions varied between units. This allowed our analysis to explore the dose relationships between the model and outcomes independent of underlying time trends that may be affected by concomitant initiatives.

Table 2.Scoring Grid

	1	2	3	4	5
NOTE: Abbreviations: ACT, accountable care team. *The ACT disciplines used for this scoring include the hospitalists, clinical nurse specialists, pharmacists, case managers, and social workers. Members of the ACT team not included in the scoring scheme include unit nurse managers, nursing, charge nurse, physical therapists, nutrition support, and occupational therapists. The maximum number of specialists on any unit is 3 (eg, cardiothoracic surgery, cardiology, and vascular surgery on the cardiovascular surgery unit). For general medical units, a score of 3 would be the next score possible after 1.
Geographical cohorting of patients and the ACT*	None	At least 1 discipline comprising the ACT is unit based	All disciplines comprising the ACT except the hospitalist unit based	All disciplines including the hospitalist unit based	4 + 80% of hospitalist provider's patients on the unit
Bedside collaborative rounding	None	Occurring 1 day a week on at least 25% of the patients on the unit	Occurring 2 to 3 days a week on at least 50% of the patients on the unit	Occurring 3 to 4 days a week on at least 75% of the patients on the unit	Occurring MondayFriday on all patients on the unit
Daily huddle	None	Occurring daily, 1 out of 4 ACT disciplines represented, at least 25% of patients on the unit discussed	Occurring daily, 2 out of 4 ACT disciplines represented, at least 50% of patients on the unit discussed	Occurring daily, 3 out of 4 ACT disciplines represented, at least 75% of patients on the unit discussed	Occurring daily, all disciplines of the ACT represented, all patients on the unit discussed
Hospitalist and specialty comanagement agreements	None	One out of 3 specialists represented on the unit collaborating with the hospitalists on at least 25% of relevant patients	One out of 3 specialists represented on the unit collaborating with the hospitalists on at least 50% of relevant patients	Two out of 3 specialists on the unit collaborating with the hospitalists on at least 75% of relevant patients	All specialists on the unit collaborating with the hospitalists on all relevant patients on the unit
Unit white board	None	Present but only used by nursing	Present and used by all ACT disciplines except physician providers	Present and used by entire ACT; use inconsistent	Present and used MondayFriday by all disciplines of ACT
Monthly review of unit level data	None	Nurse manager reviewing data with ACT program director	Nurse manager and unit leader reviewing data with ACT program director	Meeting either not consistently occurring monthly or not consistently attended by entire ACT	Monthly meeting with entire ACT
Weekly patient satisfaction rounding	None	Nurse manager performing up to 1 week a month	Nurse manager performing weekly	Nurse and physician leader performing up to 3 times a month	Nurse and physician leader performing weekly
Leadership	None	For units with specialties, either hospitalist or specialist leader identified	Both hospitalist and specialist leader Identified	Both hospitalist and specialist leaders (where applicable) identified and partially engaged in leadership role	Both hospitalist and specialist leaders (where applicable) identified and engaged in leadership role

Figure 1

Outcomes

Monthly data between August 2012 and December 2013 were analyzed.

Measures of Value

MH is a member of the University Health Consortium, which measures outcomes of participants relative to their peers. MH measures LOS index as a ratio of observed LOS to expected LOS that is adjusted for severity of illness.[5]

Variable direct costs (VDCs) are costs that a hospital can save if a service is not provided.[6] A hospital's case‐mix index (CMI) represents the average diagnosis‐related group relative weight for that hospital. We track VDCs adjusted for CMI (CMI‐adjusted VDC).[7]

Thirty‐day readmission rate is the percentage of cases that are readmitted to MH within 30 days of discharge from the index admission.[8]

Measures of Patient Satisfaction

The Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey covers topics relevant to a patient's experience in the hospital.[9] Patient satisfaction scores are tracked by responses to the HCAHPS survey.

Measures of Provider Satisfaction

Hospitalist and specialty providers, leadership, and case management teams were surveyed via email through SurveyMonkey in July 2014. The survey included Likert responses that elicited opinions and comments about the ACT model.

RESULTS

The overall ACT score was associated with LOS index and CMI‐adjusted VDC (both P < 0.001). For every 1‐unit increase in the overall ACT score, LOS index decreased by 0.078 and CMI‐adjusted VDC decreased by $273.99 (Table 3).

Looking at domains, enhancing IPC resulted in statistically significant decreases in both LOS index and CMI‐adjusted VDC, but providing leadership and enabling data‐driven decisions decreased only the LOS index. Most of the 8 individual interventions were associated with at least 1 of these 2 outcomes. (Even where the associations were not significant, they were all in the direction of decreasing LOS and cost). In these models, the covariate of type of units (medical vs surgical) was not associated with LOS or cost. There was no significant time trend in LOS or cost, except in models where an intervention had no association with either outcome. Inclusion of all individual effective interventions in the same statistical model to assess their relative contributions was not possible because they were highly correlated (correlations 0.450.89).

Thirty‐day readmissions and patient satisfaction were not significantly associated with the overall ACT score, but exploratory analyses showed that patient satisfaction increased with the implementation of geographical cohorting (P = 0.007).

DISCUSSION

The serious problems in US healthcare constitute an urgent imperative to innovate and reform.[10] Inpatient care reflects 31% of the expenditure on healthcare, and in 2010, 35.1 million patients were discharged from the hospital after spending an average of 4.8 days as an inpatient.[11] These figures represent an immense opportunity to intervene. Measuring the impact of quality improvement efforts is often complicated by concomitant changes that affect outcomes over the interval studied. Our approach allowed us to detect statistically significant changes in LOS index and CMI‐adjusted VDC associated with the ACT implementation dose that could be separated from the underlying time trends.

The ACT model we describe is rooted in improving 3 foundational domains; quantifying each intervention's compartmentalized contribution, however, proved difficult. Each intervention intertwines with the others to create changes in attitudes, knowledge, and culture that are difficult to measure yet may synergistically affect outcomes. For example, although geographical cohorting appears to have the strongest statistical association with outcomes, this may be mediated by how it enables other processes to take place more effectively. Based on this analysis, therefore, the ACT model may best be considered a bundled intervention.

The team caring for a patient during hospitalization is so complex that fewer than a quarter of patients know their physician's or nurse's name.[12] This complexity impairs communication between patients and providers and between the providers themselves. Communication failures are consistently identified as root causes in sentinel events reported to the Joint Commission.[13] IPC is the process by which different professional groups work together to positively impact health care. IPC overlaps with communication, coordination, and teamwork, and improvements in IPC may improve care.[14] Some elements of the model we describe have been tested previously.[15, 16, 17] Localization of teams may increase productivity and the frequency with which physicians and nurses communicate. Localization also decreases the number of pages received and steps walked by providers during a workday.[15, 16, 17] However, these studies reported a trend toward an increase in the LOS and neutral effects on cost and readmission rates. We found statistically significant decreases in both LOS and cost associated with the geographic cohorting of patients and providers. Notably, our model localized not only the physician providers but also the interdisciplinary team of pharmacists, clinical nurse specialists, case managers, and social workers. This proximity may facilitate IPC between all members that culminates in improved efficiency. The possibility of delays in discharges to avoid new admissions in a geographically structured team has previously been raised to explain the associated increases in LOS.[16, 17] The accountability of each unit for its metrics, the communication between nursing and physicians, and the timely availability of the unit's performance data aligns everyone toward a shared goal and provides some protection from an unintended consequence.

Structured interdisciplinary rounds decrease adverse events and improve teamwork ratings.[18, 19] The huddle in our model is a forum to collaborate between disciplines that proved to be effective in decreasing LOS and costs. Our huddle aims to discuss all the patients on the unit. This allows the team to assist each other in problem solving for the entire unit and not just the patients on the geographically cohorted team. This approach, in addition to the improved IPC fostered by the ACT model, may help explain how benefits in LOS and costs permeated across all 11 diverse units despite the presence of patients who are not directly served by the geographically cohorted team.

High‐performing clinical systems maintain an awareness of their overarching mission and unit‐based leaders can influence the frontline by reiterating the organizational mission and aligning efforts with outcomes.[20] Our leadership model is similar to those described by other institutions in the strong partnerships between physicians and nursing.[21] As outlined by Kim et al., investing in the professional development of the unit leaders may help them fulfill their roles and serve the organization better.[21]

The fragmentation and lack of ownership over the continuum of patient care causes duplication and waste. The proposal in the Accountable Care Act to create accountable care organizations is rooted in the understanding that providers and organizations will seek out new ways of improving quality when held accountable for their outcomes.[22] To foster ownership and accountability, reporting of metrics at the unit level is needed. Furthermore, an informational infrastructure is critical, as improvements cannot occur without the availability of data to both monitor performance and measure the effect of interventions.[10, 23] Even without any other interventions, providing feedback alone is an effective way of changing practices.[24] According to Berwick et al., this phenomenon reflects practitioners' intrinsic motivation to simply want to be better.[25] Our monthly review of each unit's data is an effective way to provide timely feedback to the frontline that sparks pride, ownership, and innovative thinking.

Based on our mean ACT score and CMI‐adjusted VDC reductions alone, we estimate savings of $649.36 per hospitalization (mean increase in ACT implementation of 2.37 times reduction in cost index of $273.99 per unit increase in overall ACT score). This figure does not include savings realized through reductions in LOS. This is a small decrease relative to the mean cost of hospitalization, yet when compounded over the annual MH census, it would result in substantial savings. The model relied on the restructuring of the existing workforce and the only direct additional cost was the early salary support for the ACT program director.

FURTHER DIRECTIONS

Although there is a need to improve our healthcare system, interventions should be deliberate and evidence based wherever possible.[26] Geographic cohorting may decrease the frequency of paging interruptions for physicians and practitioners while increasing face‐to‐face interruptions.[27] The net effect on safety with this trade‐off should be investigated.

The presence of an intervention does not guarantee its success. Despite geographic cohorting and interdisciplinary meetings, communication that influences physician decision making may not improve.[28] Although instruments to measure ratings of team work and collaboration are available, focusing on clinically relevant outcomes of teamwork, such as prevention of harm, may be more empowering feedback for the frontline. Formal cost‐benefit analyses and outcomes related to physician and nursing retention will be equally important for assessing the sustainability of the model. Involving patients and their caregivers and inviting their perspectives as care is redesigned will also be critical in maintaining patient centeredness. Research addressing interventions to mediate preventable readmission risk and understanding the drivers of patient satisfaction is also needed.

The true value of the model may be in its potential to monitor and drive change within itself. Continuously aligning aims, incentives, performance measures, and feedback will help support this innovation and drive. This affects not only patient care but creates microcosms within which research and education can thrive. We hope that our experience will help guide other institutions as we all strive in our journey to improve the care we deliver.

Despite an estimated annual $2.6 trillion expenditure on healthcare, the United States performs poorly on indicators of health and harm during care.[1, 2, 3] Hospitals around the nation are working to improve the care they deliver. We describe a model developed at our institution and report the evaluation of the outcomes associated with its implementation on the general medical and surgical units. The Indiana University Institutional Review Board approved this work.

SETTING AND DEFINITIONS

Indiana University Health Methodist Hospital (MH) is an academic center in Indianapolis, Indiana, serving over 30,000 patients annually.[4] In 2012, responding to the coexisting needs to improve quality and contain costs, the MH leadership team redesigned care in the hospital. The new model centers around accountable care teams (ACTs). Each ACT is a geographically defined set of providers accepting ownership for the clinical, service, and financial outcomes of their respective inpatient unit. The units studied are described in Table 1.

THE ACT MODEL

The model comprises 8 interventions rooted in 3 foundational domains: (1) enhancing interprofessional collaboration (IPC), (2) enabling data‐driven decisions, and (3) providing leadership. Each intervention is briefly described under its main focus (see Supporting Information, Appendix A, in the online version of this article for further details).

Implementation Timelines and ACT Scores

The development of the ACTs started in the spring of 2012. Physician, nursing, and pharmacy support was sought, and a pilot unit was formed in August 2012. The model was cascaded hospital wide by December 2013, with support from the ACT program director (A.N.). The program director observed and scored the uptake of each intervention by each unit monthly. A score of 1 denoted no implementation, whereas 5 denoted complete implementation. The criteria for scoring are presented in Table 2. The monthly scores for all 8 interventions in each of the 11 units were averaged as an overall ACT score, which reflects the implementation dose of the ACT model. Monthly domain scores for enhancing IPC and enabling data‐driven decisions were also calculated as the average score within each domain. This yielded 3 domain scores. Figure 1A plots by month the overall ACT score for the medical and surgical units, and Figure 1B plots the implementation score for the 3 domains between August 2012 and December 2013 for all units. The uptake of the interventions varied between units. This allowed our analysis to explore the dose relationships between the model and outcomes independent of underlying time trends that may be affected by concomitant initiatives.

Table 2.Scoring Grid

	1	2	3	4	5
NOTE: Abbreviations: ACT, accountable care team. *The ACT disciplines used for this scoring include the hospitalists, clinical nurse specialists, pharmacists, case managers, and social workers. Members of the ACT team not included in the scoring scheme include unit nurse managers, nursing, charge nurse, physical therapists, nutrition support, and occupational therapists. The maximum number of specialists on any unit is 3 (eg, cardiothoracic surgery, cardiology, and vascular surgery on the cardiovascular surgery unit). For general medical units, a score of 3 would be the next score possible after 1.
Geographical cohorting of patients and the ACT*	None	At least 1 discipline comprising the ACT is unit based	All disciplines comprising the ACT except the hospitalist unit based	All disciplines including the hospitalist unit based	4 + 80% of hospitalist provider's patients on the unit
Bedside collaborative rounding	None	Occurring 1 day a week on at least 25% of the patients on the unit	Occurring 2 to 3 days a week on at least 50% of the patients on the unit	Occurring 3 to 4 days a week on at least 75% of the patients on the unit	Occurring MondayFriday on all patients on the unit
Daily huddle	None	Occurring daily, 1 out of 4 ACT disciplines represented, at least 25% of patients on the unit discussed	Occurring daily, 2 out of 4 ACT disciplines represented, at least 50% of patients on the unit discussed	Occurring daily, 3 out of 4 ACT disciplines represented, at least 75% of patients on the unit discussed	Occurring daily, all disciplines of the ACT represented, all patients on the unit discussed
Hospitalist and specialty comanagement agreements	None	One out of 3 specialists represented on the unit collaborating with the hospitalists on at least 25% of relevant patients	One out of 3 specialists represented on the unit collaborating with the hospitalists on at least 50% of relevant patients	Two out of 3 specialists on the unit collaborating with the hospitalists on at least 75% of relevant patients	All specialists on the unit collaborating with the hospitalists on all relevant patients on the unit
Unit white board	None	Present but only used by nursing	Present and used by all ACT disciplines except physician providers	Present and used by entire ACT; use inconsistent	Present and used MondayFriday by all disciplines of ACT
Monthly review of unit level data	None	Nurse manager reviewing data with ACT program director	Nurse manager and unit leader reviewing data with ACT program director	Meeting either not consistently occurring monthly or not consistently attended by entire ACT	Monthly meeting with entire ACT
Weekly patient satisfaction rounding	None	Nurse manager performing up to 1 week a month	Nurse manager performing weekly	Nurse and physician leader performing up to 3 times a month	Nurse and physician leader performing weekly
Leadership	None	For units with specialties, either hospitalist or specialist leader identified	Both hospitalist and specialist leader Identified	Both hospitalist and specialist leaders (where applicable) identified and partially engaged in leadership role	Both hospitalist and specialist leaders (where applicable) identified and engaged in leadership role

Figure 1

Outcomes

Monthly data between August 2012 and December 2013 were analyzed.

Measures of Value

MH is a member of the University Health Consortium, which measures outcomes of participants relative to their peers. MH measures LOS index as a ratio of observed LOS to expected LOS that is adjusted for severity of illness.[5]

Variable direct costs (VDCs) are costs that a hospital can save if a service is not provided.[6] A hospital's case‐mix index (CMI) represents the average diagnosis‐related group relative weight for that hospital. We track VDCs adjusted for CMI (CMI‐adjusted VDC).[7]

Thirty‐day readmission rate is the percentage of cases that are readmitted to MH within 30 days of discharge from the index admission.[8]

Measures of Patient Satisfaction

The Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey covers topics relevant to a patient's experience in the hospital.[9] Patient satisfaction scores are tracked by responses to the HCAHPS survey.

Measures of Provider Satisfaction

Hospitalist and specialty providers, leadership, and case management teams were surveyed via email through SurveyMonkey in July 2014. The survey included Likert responses that elicited opinions and comments about the ACT model.

RESULTS

The overall ACT score was associated with LOS index and CMI‐adjusted VDC (both P < 0.001). For every 1‐unit increase in the overall ACT score, LOS index decreased by 0.078 and CMI‐adjusted VDC decreased by $273.99 (Table 3).

Looking at domains, enhancing IPC resulted in statistically significant decreases in both LOS index and CMI‐adjusted VDC, but providing leadership and enabling data‐driven decisions decreased only the LOS index. Most of the 8 individual interventions were associated with at least 1 of these 2 outcomes. (Even where the associations were not significant, they were all in the direction of decreasing LOS and cost). In these models, the covariate of type of units (medical vs surgical) was not associated with LOS or cost. There was no significant time trend in LOS or cost, except in models where an intervention had no association with either outcome. Inclusion of all individual effective interventions in the same statistical model to assess their relative contributions was not possible because they were highly correlated (correlations 0.450.89).

Thirty‐day readmissions and patient satisfaction were not significantly associated with the overall ACT score, but exploratory analyses showed that patient satisfaction increased with the implementation of geographical cohorting (P = 0.007).

DISCUSSION

The serious problems in US healthcare constitute an urgent imperative to innovate and reform.[10] Inpatient care reflects 31% of the expenditure on healthcare, and in 2010, 35.1 million patients were discharged from the hospital after spending an average of 4.8 days as an inpatient.[11] These figures represent an immense opportunity to intervene. Measuring the impact of quality improvement efforts is often complicated by concomitant changes that affect outcomes over the interval studied. Our approach allowed us to detect statistically significant changes in LOS index and CMI‐adjusted VDC associated with the ACT implementation dose that could be separated from the underlying time trends.

The ACT model we describe is rooted in improving 3 foundational domains; quantifying each intervention's compartmentalized contribution, however, proved difficult. Each intervention intertwines with the others to create changes in attitudes, knowledge, and culture that are difficult to measure yet may synergistically affect outcomes. For example, although geographical cohorting appears to have the strongest statistical association with outcomes, this may be mediated by how it enables other processes to take place more effectively. Based on this analysis, therefore, the ACT model may best be considered a bundled intervention.

The team caring for a patient during hospitalization is so complex that fewer than a quarter of patients know their physician's or nurse's name.[12] This complexity impairs communication between patients and providers and between the providers themselves. Communication failures are consistently identified as root causes in sentinel events reported to the Joint Commission.[13] IPC is the process by which different professional groups work together to positively impact health care. IPC overlaps with communication, coordination, and teamwork, and improvements in IPC may improve care.[14] Some elements of the model we describe have been tested previously.[15, 16, 17] Localization of teams may increase productivity and the frequency with which physicians and nurses communicate. Localization also decreases the number of pages received and steps walked by providers during a workday.[15, 16, 17] However, these studies reported a trend toward an increase in the LOS and neutral effects on cost and readmission rates. We found statistically significant decreases in both LOS and cost associated with the geographic cohorting of patients and providers. Notably, our model localized not only the physician providers but also the interdisciplinary team of pharmacists, clinical nurse specialists, case managers, and social workers. This proximity may facilitate IPC between all members that culminates in improved efficiency. The possibility of delays in discharges to avoid new admissions in a geographically structured team has previously been raised to explain the associated increases in LOS.[16, 17] The accountability of each unit for its metrics, the communication between nursing and physicians, and the timely availability of the unit's performance data aligns everyone toward a shared goal and provides some protection from an unintended consequence.

Structured interdisciplinary rounds decrease adverse events and improve teamwork ratings.[18, 19] The huddle in our model is a forum to collaborate between disciplines that proved to be effective in decreasing LOS and costs. Our huddle aims to discuss all the patients on the unit. This allows the team to assist each other in problem solving for the entire unit and not just the patients on the geographically cohorted team. This approach, in addition to the improved IPC fostered by the ACT model, may help explain how benefits in LOS and costs permeated across all 11 diverse units despite the presence of patients who are not directly served by the geographically cohorted team.

High‐performing clinical systems maintain an awareness of their overarching mission and unit‐based leaders can influence the frontline by reiterating the organizational mission and aligning efforts with outcomes.[20] Our leadership model is similar to those described by other institutions in the strong partnerships between physicians and nursing.[21] As outlined by Kim et al., investing in the professional development of the unit leaders may help them fulfill their roles and serve the organization better.[21]

The fragmentation and lack of ownership over the continuum of patient care causes duplication and waste. The proposal in the Accountable Care Act to create accountable care organizations is rooted in the understanding that providers and organizations will seek out new ways of improving quality when held accountable for their outcomes.[22] To foster ownership and accountability, reporting of metrics at the unit level is needed. Furthermore, an informational infrastructure is critical, as improvements cannot occur without the availability of data to both monitor performance and measure the effect of interventions.[10, 23] Even without any other interventions, providing feedback alone is an effective way of changing practices.[24] According to Berwick et al., this phenomenon reflects practitioners' intrinsic motivation to simply want to be better.[25] Our monthly review of each unit's data is an effective way to provide timely feedback to the frontline that sparks pride, ownership, and innovative thinking.

Based on our mean ACT score and CMI‐adjusted VDC reductions alone, we estimate savings of $649.36 per hospitalization (mean increase in ACT implementation of 2.37 times reduction in cost index of $273.99 per unit increase in overall ACT score). This figure does not include savings realized through reductions in LOS. This is a small decrease relative to the mean cost of hospitalization, yet when compounded over the annual MH census, it would result in substantial savings. The model relied on the restructuring of the existing workforce and the only direct additional cost was the early salary support for the ACT program director.

FURTHER DIRECTIONS

Although there is a need to improve our healthcare system, interventions should be deliberate and evidence based wherever possible.[26] Geographic cohorting may decrease the frequency of paging interruptions for physicians and practitioners while increasing face‐to‐face interruptions.[27] The net effect on safety with this trade‐off should be investigated.

The presence of an intervention does not guarantee its success. Despite geographic cohorting and interdisciplinary meetings, communication that influences physician decision making may not improve.[28] Although instruments to measure ratings of team work and collaboration are available, focusing on clinically relevant outcomes of teamwork, such as prevention of harm, may be more empowering feedback for the frontline. Formal cost‐benefit analyses and outcomes related to physician and nursing retention will be equally important for assessing the sustainability of the model. Involving patients and their caregivers and inviting their perspectives as care is redesigned will also be critical in maintaining patient centeredness. Research addressing interventions to mediate preventable readmission risk and understanding the drivers of patient satisfaction is also needed.

The true value of the model may be in its potential to monitor and drive change within itself. Continuously aligning aims, incentives, performance measures, and feedback will help support this innovation and drive. This affects not only patient care but creates microcosms within which research and education can thrive. We hope that our experience will help guide other institutions as we all strive in our journey to improve the care we deliver.

Despite an estimated annual $2.6 trillion expenditure on healthcare, the United States performs poorly on indicators of health and harm during care.[1, 2, 3] Hospitals around the nation are working to improve the care they deliver. We describe a model developed at our institution and report the evaluation of the outcomes associated with its implementation on the general medical and surgical units. The Indiana University Institutional Review Board approved this work.

SETTING AND DEFINITIONS

Indiana University Health Methodist Hospital (MH) is an academic center in Indianapolis, Indiana, serving over 30,000 patients annually.[4] In 2012, responding to the coexisting needs to improve quality and contain costs, the MH leadership team redesigned care in the hospital. The new model centers around accountable care teams (ACTs). Each ACT is a geographically defined set of providers accepting ownership for the clinical, service, and financial outcomes of their respective inpatient unit. The units studied are described in Table 1.

THE ACT MODEL

The model comprises 8 interventions rooted in 3 foundational domains: (1) enhancing interprofessional collaboration (IPC), (2) enabling data‐driven decisions, and (3) providing leadership. Each intervention is briefly described under its main focus (see Supporting Information, Appendix A, in the online version of this article for further details).

Implementation Timelines and ACT Scores

The development of the ACTs started in the spring of 2012. Physician, nursing, and pharmacy support was sought, and a pilot unit was formed in August 2012. The model was cascaded hospital wide by December 2013, with support from the ACT program director (A.N.). The program director observed and scored the uptake of each intervention by each unit monthly. A score of 1 denoted no implementation, whereas 5 denoted complete implementation. The criteria for scoring are presented in Table 2. The monthly scores for all 8 interventions in each of the 11 units were averaged as an overall ACT score, which reflects the implementation dose of the ACT model. Monthly domain scores for enhancing IPC and enabling data‐driven decisions were also calculated as the average score within each domain. This yielded 3 domain scores. Figure 1A plots by month the overall ACT score for the medical and surgical units, and Figure 1B plots the implementation score for the 3 domains between August 2012 and December 2013 for all units. The uptake of the interventions varied between units. This allowed our analysis to explore the dose relationships between the model and outcomes independent of underlying time trends that may be affected by concomitant initiatives.

Table 2.Scoring Grid

	1	2	3	4	5
NOTE: Abbreviations: ACT, accountable care team. *The ACT disciplines used for this scoring include the hospitalists, clinical nurse specialists, pharmacists, case managers, and social workers. Members of the ACT team not included in the scoring scheme include unit nurse managers, nursing, charge nurse, physical therapists, nutrition support, and occupational therapists. The maximum number of specialists on any unit is 3 (eg, cardiothoracic surgery, cardiology, and vascular surgery on the cardiovascular surgery unit). For general medical units, a score of 3 would be the next score possible after 1.
Geographical cohorting of patients and the ACT*	None	At least 1 discipline comprising the ACT is unit based	All disciplines comprising the ACT except the hospitalist unit based	All disciplines including the hospitalist unit based	4 + 80% of hospitalist provider's patients on the unit
Bedside collaborative rounding	None	Occurring 1 day a week on at least 25% of the patients on the unit	Occurring 2 to 3 days a week on at least 50% of the patients on the unit	Occurring 3 to 4 days a week on at least 75% of the patients on the unit	Occurring MondayFriday on all patients on the unit
Daily huddle	None	Occurring daily, 1 out of 4 ACT disciplines represented, at least 25% of patients on the unit discussed	Occurring daily, 2 out of 4 ACT disciplines represented, at least 50% of patients on the unit discussed	Occurring daily, 3 out of 4 ACT disciplines represented, at least 75% of patients on the unit discussed	Occurring daily, all disciplines of the ACT represented, all patients on the unit discussed
Hospitalist and specialty comanagement agreements	None	One out of 3 specialists represented on the unit collaborating with the hospitalists on at least 25% of relevant patients	One out of 3 specialists represented on the unit collaborating with the hospitalists on at least 50% of relevant patients	Two out of 3 specialists on the unit collaborating with the hospitalists on at least 75% of relevant patients	All specialists on the unit collaborating with the hospitalists on all relevant patients on the unit
Unit white board	None	Present but only used by nursing	Present and used by all ACT disciplines except physician providers	Present and used by entire ACT; use inconsistent	Present and used MondayFriday by all disciplines of ACT
Monthly review of unit level data	None	Nurse manager reviewing data with ACT program director	Nurse manager and unit leader reviewing data with ACT program director	Meeting either not consistently occurring monthly or not consistently attended by entire ACT	Monthly meeting with entire ACT
Weekly patient satisfaction rounding	None	Nurse manager performing up to 1 week a month	Nurse manager performing weekly	Nurse and physician leader performing up to 3 times a month	Nurse and physician leader performing weekly
Leadership	None	For units with specialties, either hospitalist or specialist leader identified	Both hospitalist and specialist leader Identified	Both hospitalist and specialist leaders (where applicable) identified and partially engaged in leadership role	Both hospitalist and specialist leaders (where applicable) identified and engaged in leadership role

Figure 1

Outcomes

Monthly data between August 2012 and December 2013 were analyzed.

Measures of Value

MH is a member of the University Health Consortium, which measures outcomes of participants relative to their peers. MH measures LOS index as a ratio of observed LOS to expected LOS that is adjusted for severity of illness.[5]

Variable direct costs (VDCs) are costs that a hospital can save if a service is not provided.[6] A hospital's case‐mix index (CMI) represents the average diagnosis‐related group relative weight for that hospital. We track VDCs adjusted for CMI (CMI‐adjusted VDC).[7]

Thirty‐day readmission rate is the percentage of cases that are readmitted to MH within 30 days of discharge from the index admission.[8]

Measures of Patient Satisfaction

The Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) survey covers topics relevant to a patient's experience in the hospital.[9] Patient satisfaction scores are tracked by responses to the HCAHPS survey.

Measures of Provider Satisfaction

Hospitalist and specialty providers, leadership, and case management teams were surveyed via email through SurveyMonkey in July 2014. The survey included Likert responses that elicited opinions and comments about the ACT model.

RESULTS

The overall ACT score was associated with LOS index and CMI‐adjusted VDC (both P < 0.001). For every 1‐unit increase in the overall ACT score, LOS index decreased by 0.078 and CMI‐adjusted VDC decreased by $273.99 (Table 3).

Looking at domains, enhancing IPC resulted in statistically significant decreases in both LOS index and CMI‐adjusted VDC, but providing leadership and enabling data‐driven decisions decreased only the LOS index. Most of the 8 individual interventions were associated with at least 1 of these 2 outcomes. (Even where the associations were not significant, they were all in the direction of decreasing LOS and cost). In these models, the covariate of type of units (medical vs surgical) was not associated with LOS or cost. There was no significant time trend in LOS or cost, except in models where an intervention had no association with either outcome. Inclusion of all individual effective interventions in the same statistical model to assess their relative contributions was not possible because they were highly correlated (correlations 0.450.89).

Thirty‐day readmissions and patient satisfaction were not significantly associated with the overall ACT score, but exploratory analyses showed that patient satisfaction increased with the implementation of geographical cohorting (P = 0.007).

DISCUSSION

The serious problems in US healthcare constitute an urgent imperative to innovate and reform.[10] Inpatient care reflects 31% of the expenditure on healthcare, and in 2010, 35.1 million patients were discharged from the hospital after spending an average of 4.8 days as an inpatient.[11] These figures represent an immense opportunity to intervene. Measuring the impact of quality improvement efforts is often complicated by concomitant changes that affect outcomes over the interval studied. Our approach allowed us to detect statistically significant changes in LOS index and CMI‐adjusted VDC associated with the ACT implementation dose that could be separated from the underlying time trends.

The ACT model we describe is rooted in improving 3 foundational domains; quantifying each intervention's compartmentalized contribution, however, proved difficult. Each intervention intertwines with the others to create changes in attitudes, knowledge, and culture that are difficult to measure yet may synergistically affect outcomes. For example, although geographical cohorting appears to have the strongest statistical association with outcomes, this may be mediated by how it enables other processes to take place more effectively. Based on this analysis, therefore, the ACT model may best be considered a bundled intervention.

The team caring for a patient during hospitalization is so complex that fewer than a quarter of patients know their physician's or nurse's name.[12] This complexity impairs communication between patients and providers and between the providers themselves. Communication failures are consistently identified as root causes in sentinel events reported to the Joint Commission.[13] IPC is the process by which different professional groups work together to positively impact health care. IPC overlaps with communication, coordination, and teamwork, and improvements in IPC may improve care.[14] Some elements of the model we describe have been tested previously.[15, 16, 17] Localization of teams may increase productivity and the frequency with which physicians and nurses communicate. Localization also decreases the number of pages received and steps walked by providers during a workday.[15, 16, 17] However, these studies reported a trend toward an increase in the LOS and neutral effects on cost and readmission rates. We found statistically significant decreases in both LOS and cost associated with the geographic cohorting of patients and providers. Notably, our model localized not only the physician providers but also the interdisciplinary team of pharmacists, clinical nurse specialists, case managers, and social workers. This proximity may facilitate IPC between all members that culminates in improved efficiency. The possibility of delays in discharges to avoid new admissions in a geographically structured team has previously been raised to explain the associated increases in LOS.[16, 17] The accountability of each unit for its metrics, the communication between nursing and physicians, and the timely availability of the unit's performance data aligns everyone toward a shared goal and provides some protection from an unintended consequence.

Structured interdisciplinary rounds decrease adverse events and improve teamwork ratings.[18, 19] The huddle in our model is a forum to collaborate between disciplines that proved to be effective in decreasing LOS and costs. Our huddle aims to discuss all the patients on the unit. This allows the team to assist each other in problem solving for the entire unit and not just the patients on the geographically cohorted team. This approach, in addition to the improved IPC fostered by the ACT model, may help explain how benefits in LOS and costs permeated across all 11 diverse units despite the presence of patients who are not directly served by the geographically cohorted team.

High‐performing clinical systems maintain an awareness of their overarching mission and unit‐based leaders can influence the frontline by reiterating the organizational mission and aligning efforts with outcomes.[20] Our leadership model is similar to those described by other institutions in the strong partnerships between physicians and nursing.[21] As outlined by Kim et al., investing in the professional development of the unit leaders may help them fulfill their roles and serve the organization better.[21]

The fragmentation and lack of ownership over the continuum of patient care causes duplication and waste. The proposal in the Accountable Care Act to create accountable care organizations is rooted in the understanding that providers and organizations will seek out new ways of improving quality when held accountable for their outcomes.[22] To foster ownership and accountability, reporting of metrics at the unit level is needed. Furthermore, an informational infrastructure is critical, as improvements cannot occur without the availability of data to both monitor performance and measure the effect of interventions.[10, 23] Even without any other interventions, providing feedback alone is an effective way of changing practices.[24] According to Berwick et al., this phenomenon reflects practitioners' intrinsic motivation to simply want to be better.[25] Our monthly review of each unit's data is an effective way to provide timely feedback to the frontline that sparks pride, ownership, and innovative thinking.

Based on our mean ACT score and CMI‐adjusted VDC reductions alone, we estimate savings of $649.36 per hospitalization (mean increase in ACT implementation of 2.37 times reduction in cost index of $273.99 per unit increase in overall ACT score). This figure does not include savings realized through reductions in LOS. This is a small decrease relative to the mean cost of hospitalization, yet when compounded over the annual MH census, it would result in substantial savings. The model relied on the restructuring of the existing workforce and the only direct additional cost was the early salary support for the ACT program director.

FURTHER DIRECTIONS

Although there is a need to improve our healthcare system, interventions should be deliberate and evidence based wherever possible.[26] Geographic cohorting may decrease the frequency of paging interruptions for physicians and practitioners while increasing face‐to‐face interruptions.[27] The net effect on safety with this trade‐off should be investigated.

The presence of an intervention does not guarantee its success. Despite geographic cohorting and interdisciplinary meetings, communication that influences physician decision making may not improve.[28] Although instruments to measure ratings of team work and collaboration are available, focusing on clinically relevant outcomes of teamwork, such as prevention of harm, may be more empowering feedback for the frontline. Formal cost‐benefit analyses and outcomes related to physician and nursing retention will be equally important for assessing the sustainability of the model. Involving patients and their caregivers and inviting their perspectives as care is redesigned will also be critical in maintaining patient centeredness. Research addressing interventions to mediate preventable readmission risk and understanding the drivers of patient satisfaction is also needed.

The true value of the model may be in its potential to monitor and drive change within itself. Continuously aligning aims, incentives, performance measures, and feedback will help support this innovation and drive. This affects not only patient care but creates microcosms within which research and education can thrive. We hope that our experience will help guide other institutions as we all strive in our journey to improve the care we deliver.

Azithromycin, a macrolide antibiotic, received US Food and Drug Administration (FDA) approval in 1991 and is 1 of the most prescribed antibiotics used for a variety of infections, including community‐acquired pneumonia, bacterial sinusitis, urethritis, and cervicitis. In 2011, it was estimated that 40.3 million outpatients received a prescription for azithromycin.[1] In addition to treating acute bacterial infections, recent literature has pointed to using azithromycin for its unlabeled immunomodulatory and anti‐inflammatory effects, particularly in cystic fibrosis, chronic obstructive pulmonary disease (COPD), and lung transplant recipients.[2, 3, 4] Azithromycin decreases bacterial load and virulence, thus reducing airway secretion, as well as decreasing airway neutrophil accumulation through a reduction in proinflammatory cytokine expression.[4]

Cardiac toxicity can occur with macrolide antibiotics, and prolongation of the QT interval with subsequent Torsades de pointes has been documented with azithromycin.[1, 5, 6] In 2012, Ray et al. published data on a cohort of outpatients receiving azithromycin compared to amoxicillin, ciprofloxacin, or no antibiotics, and showed a small but absolute increase in cardiovascular deaths.[7, 8] Subsequent data, however, have not illustrated increased risk of death from cardiovascular causes. Mortensen et al. showed a lower risk of 90‐day mortality in older patients treated for community acquired pneumonia with azithromycin and ceftriaxone, although there was a nonstatistically significant increased risk of myocardial infarction in this group.[8, 9, 10] In March 2013, the FDA released an official statement regarding increased cardiovascular risk with azithromycin, stating that healthcare professionals should consider the risk of fatal heart rhythms with azithromycin when considering treatment options for patients who are at risk for cardiovascular events.[11]

In recent years, the potential for corrected QT (QTc) prolongation and Torsades de pointes has received increased attention due to its catastrophic nature, and it is thought that hospitalized patients are at a greater risk of drug‐induced Torsades de pointes due to the likelihood of having more risk factors.[12, 13] The American Heart Association released a statement in 2010 to raise awareness among healthcare professionals about risk, electrocardiogram (ECG) monitoring, and management of drug‐induced QT interval prolongation in hospitalized patients, although little data exist regarding quantification of risk in this patient population.[13, 14]

Prescribers currently have no standardized practice guidelines related to cardiovascular safety when prescribing QTc prolonging medications. Given the dramatic increase in azithromycin prescriptions and ongoing concern for cardiovascular risk and QTc prolongation, we investigated the prescribing practices with azithromycin within our institution. Our primary aims were 3‐fold. First, we aimed to describe the frequency azithromycin was prescribed with additional QTc prolonging medications. Second, we assessed the relationship between the number of arrhythmogenic drugs prescribed in addition to azithromycin with ordering telemetry. Finally, we assessed the relationship between baseline ECG abnormalities and telemetry monitoring in patients prescribed azithromycin. The purpose of these objectives was to better understand physician prescribing practices and to determine if patients have a potential risk of developing fatal cardiac arrhythmias

METHODS

RESULTS

Azithromycin use within our hospital system has increased from 15 days of therapy per 1000 patient days in 2002 to 40 days of therapy per 1000 patient days in 2013 (Figure 1). At the same time, azithromycin susceptibility in Streptococcus pneumoniae isolates has decreased over the past decade from 65% to 35% in our hospital.

Figure 1

The baseline characteristics of patients included in this study are noted in Table 1. The mean age of patients was 55 years, with a range of 21 to 97 years, and 61% were female. Forty‐five percent of patients were admitted to either the general medicine teaching service or hospitalist service, and 23% were admitted to the pulmonary service, which includes intensive care unit admission. The average length of patient stay was 9.7 days (range, 3115 days; median 6 days).

Seventy‐nine percent of azithromycin use was empiric for the treatment of suspected infection. The second most common use was for anti‐inflammatory effects (20%), as documented by prescribers in the medical record for patients with cystic fibrosis, lung transplant, and chronic obstructive pulmonary disease. Azithromycin was dosed appropriately according to the documented indication in 67% of patients, with the most discrepancy in dosing noted for anti‐inflammatory use. The average duration of azithromycin therapy was 4.5 days (range, 128 days). Duration was appropriate in 63% of patients. Twenty‐one percent of patients received intravenous formulation of azithromycin, 13% received intravenous followed by oral formulation, and 64% of patient received tablet formulation alone.

Thirty‐five medications have been identified in our formulary as having a potential major drug‐drug interaction when prescribed with azithromycin (see Supporting Information, Appendix, in the online version of this article), and of these medications, 20 were prescribed with azithromycin, with an average overlap of therapy of 4.5 days (Table 2). Seventy‐six percent of patients were concomitantly prescribed a QT‐prolonging drug in addition to azithromycin. The most commonly prescribed agents were ondansetron (48%), trazodone (22%), and moxifloxacin (17%).

Telemetry monitoring was assessed for each patient based on inpatient charges during their hospitalization (Table 3). Forty‐three percent of patients were placed on telemetry. Twenty‐four (24%) of the patients were prescribed azithromycin alone, of whom 45.8% were placed on telemetry. Fifty‐seven percent of patients were prescribed azithromycin with 1 to 2 additional QT‐prolonging medications (medium‐risk arm); 38.5% of patients in this group were placed on telemetry. In the high‐risk arm, 19% of patients were prescribed at least 3 QT‐prolonging medications in addition to azithromycin, of which only 52.6% of patients were monitored with telemetry. No statistically significant association was observed between risk level and telemetry placement (P=0.07).

Telemetry charges were further examined by analyzing baseline ECG evaluation within the past 6 months of their hospitalization (Table 3). Sixty‐six patients received baseline ECGs prior to initiation of azithromycin. Telemetry placement was not statistically correlated to abnormal QTc at baseline (P=0.22). Of those who underwent baseline ECG evaluation, 8.3% were noted to have borderline QTc, and 12.5% had abnormal QTc on admission prior to receiving azithromycin in the low‐risk level (Table 4). Within the medium‐risk level, 63.2% had baseline ECG evaluation, with 5.3% with borderline QTc and 35.7% with abnormal QTc. In the high‐risk level, 73.6% received a baseline ECG, with 21% with borderline QTc and 31.6% with abnormal QTc. No statistically significant association was observed between risk level and obtainment of baseline ECG (P=0.7). In 17 out of 66 patients, average repeat ECGs were obtained on day 3 (range, 27 days). Ten of the 17 ECGs showed increase in QTc (range, 397ms; average 27 ms), whereas the other 7 had a decrease in their QTc interval (range, 618 ms; average 13 ms; P=0.17).

As risk level increased, having an abnormal QTc at baseline was statistically different between low‐ and medium‐risk levels (P=0.03), but this association was lost when comparing the low‐risk arm to the high‐risk arm (P=0.11). When the medium‐ and high‐risk categories were combined, there was a noted statistical significance of having an abnormal ECG at baseline (P=0.03).

Of the 9 patients prescribed azithromycin chronically, 3 patients were in the low‐risk category, 4 in the medium‐risk category, and 2 in the high‐risk category. Only 2 had baseline ECGs obtained, 1 of which was noted to have abnormal QTc and was in the high‐risk category. Only 1 patient was placed on telemetry, but was considered low risk based on medications prescribed.

DISCUSSION

In this study, 76% of patients were prescribed azithromycin with 1 or more medications known to affect QT prolongation; 19% received 3 or more QT‐prolonging medications in addition to azithromycin. Of patients who received a baseline ECG, 43% were documented to have borderline or prolonged QTc on admission. Telemetry monitoring was ordered 43% of the time, but there was no significant association between telemetry placement and risk level (P=0.07), suggesting that telemetry was ordered based on symptoms more than risk. Despite more drug‐druginteracting medications prescribed, there was no association to either telemetry orders or baseline ECG evaluation. Furthermore, if an abnormal QTc was documented on admission, there was no relationship to ordering telemetry as an inpatient (P=0.215), suggesting that healthcare providers are not considering risk of QTc medication accumulation. Given increased warnings issued by the FDA for azithromycin, further prospective studies are indicated to fully assess risk of QTc prolongation and arrhythmias in the setting of multiple drug interactions. This study elucidates the potential for drug‐drug interactions and need for increased vigilance and education of providers in the healthcare setting for QTc prolongation and subsequent arrhythmias.

Forty‐eight percent of patients receiving other QTc prolonging medications were prescribed ondansetron, followed by 23% of patients prescribed trazodone. Both of these medications are included on the admission order set in our institution and can be easily ordered for patients. Despite ordering multiple medications that have potential for QTc prolongation, there are no current alerts set up in our electronic medical record. When patients are separated into drug interaction risk levels, there is a trend of having an abnormal QTc on admission, but this is driven by the large number of patients in the medium‐risk category, and the rate does not increase (and is not significant) when comparing high risk to low risk. However, patients who receive any QTc‐prolonging medication are more likely to have an abnormal QTc when compared to azithromycin prescription alone (P=0.03). The small sample size limits the power and generalizability of this study, and further larger studies are indicated to assess if risk of QTc prolongation is additive.

In the 9 patients prescribed azithromycin chronically, dosing was not consistent, and a vast majority of patients were not placed on telemetry nor had baseline ECGs on admission. This further correlates with the idea that risk of arrhythmia is not fully considered in this patient population, as patients prescribed more than 1 QTc‐prolonging medication were not included in prior studies that examined azithromycin for its anti‐inflammatory effects.[2]

Azithromycin was added to our hospital formulary in 1998, and prescription of this agent remained relatively low until 2006, when azithromycin use increased dramatically from 15 days of therapy (DOT) per 1000 patient days to 40 DOT per 1000 patient days. Although numerous factors may have led to this increase, literature was published in 2006 and 2011 citing benefit from the anti‐inflammatory effects of azithromycin.[2, 17] At the same time, azithromycin susceptibility among Streptococcus pneumoniae in patients within our hospital has decreased over the past decade; studies have found a correlation between increasing use of macrolides and the development of resistance in Streptococcus species.[18, 19, 20] In this study, 79% of patients were prescribed azithromycin empirically for treatment of bacterial infections, whereas 20% were given azithromycin for its anti‐inflammatory effects; both dose and frequency varied among patients, raising the concern for development of resistance. Published studies have shown improvement in quality of life and decreased frequency of exacerbation and infection when azithromycin is used as an anti‐inflammatory agent; however, no QTc monitoring was noted.[2] Drug‐induced QTc prolongation>10 ms above baseline suggests the potential for clinical significance, whereas a QTc prolongation >20 milliseconds above baseline has a substantially increased likelihood of being proarrhythmic.[1] Unfortunately, drug‐induced QT prolongation is unpredictable, and additional risk factors play a role in facilitating Torsades de pointes, including female sex, advanced age, electrolyte disturbances, intravenous formulation, and concurrent use of more than 1 drug that can prolong the QT interval.[15] Azithromycin has recently been added to the growing list of medications that can prolong the QT interval, with 12 cases of Torsades de pointes reported in the literature. In March 2013, the FDA released a warning regarding prescribing azithromycin, but there is a lack of guidance for clinicians in identifying risk of cardiovascular events in susceptible patients.

There are some limitations to this study. Given data were acquired retrospectively and telemetry sheets were unable to be reviewed. Some patients were noted to have arrhythmias, but these data were obtained through physician notes and not examined directly from telemetry sheets. Seventeen patients had repeat ECGs, but most were performed serially for chest pain and not QTc monitoring. Four patients died in this study, but cause of death could not be determined through electronic medical records provided for all 4 patients; families pursued withdrawal of care.

Despite the published FDA warning, there are no national guidelines for clinicians in prescribing QTc‐prolonging medications. The American Heart Association published recommendations in 2010 for prescribing these drugs in the inpatient setting, but because hospitals differ in cardiac monitoring, there is no one‐size‐fits‐all strategy in reducing risk of cardiac events.[14] If the benefit of azithromycin outweighs the risk, QTc prolongation should not limit therapy; however, institutional awareness is necessary, whether it be through automatic stop dates on azithromycin, electronic alerts regarding drug‐drug interaction, enhanced prescriber education, or a combination of all of the above.

Disclosure: Nothing to report.

Azithromycin, a macrolide antibiotic, received US Food and Drug Administration (FDA) approval in 1991 and is 1 of the most prescribed antibiotics used for a variety of infections, including community‐acquired pneumonia, bacterial sinusitis, urethritis, and cervicitis. In 2011, it was estimated that 40.3 million outpatients received a prescription for azithromycin.[1] In addition to treating acute bacterial infections, recent literature has pointed to using azithromycin for its unlabeled immunomodulatory and anti‐inflammatory effects, particularly in cystic fibrosis, chronic obstructive pulmonary disease (COPD), and lung transplant recipients.[2, 3, 4] Azithromycin decreases bacterial load and virulence, thus reducing airway secretion, as well as decreasing airway neutrophil accumulation through a reduction in proinflammatory cytokine expression.[4]

Cardiac toxicity can occur with macrolide antibiotics, and prolongation of the QT interval with subsequent Torsades de pointes has been documented with azithromycin.[1, 5, 6] In 2012, Ray et al. published data on a cohort of outpatients receiving azithromycin compared to amoxicillin, ciprofloxacin, or no antibiotics, and showed a small but absolute increase in cardiovascular deaths.[7, 8] Subsequent data, however, have not illustrated increased risk of death from cardiovascular causes. Mortensen et al. showed a lower risk of 90‐day mortality in older patients treated for community acquired pneumonia with azithromycin and ceftriaxone, although there was a nonstatistically significant increased risk of myocardial infarction in this group.[8, 9, 10] In March 2013, the FDA released an official statement regarding increased cardiovascular risk with azithromycin, stating that healthcare professionals should consider the risk of fatal heart rhythms with azithromycin when considering treatment options for patients who are at risk for cardiovascular events.[11]

In recent years, the potential for corrected QT (QTc) prolongation and Torsades de pointes has received increased attention due to its catastrophic nature, and it is thought that hospitalized patients are at a greater risk of drug‐induced Torsades de pointes due to the likelihood of having more risk factors.[12, 13] The American Heart Association released a statement in 2010 to raise awareness among healthcare professionals about risk, electrocardiogram (ECG) monitoring, and management of drug‐induced QT interval prolongation in hospitalized patients, although little data exist regarding quantification of risk in this patient population.[13, 14]

Prescribers currently have no standardized practice guidelines related to cardiovascular safety when prescribing QTc prolonging medications. Given the dramatic increase in azithromycin prescriptions and ongoing concern for cardiovascular risk and QTc prolongation, we investigated the prescribing practices with azithromycin within our institution. Our primary aims were 3‐fold. First, we aimed to describe the frequency azithromycin was prescribed with additional QTc prolonging medications. Second, we assessed the relationship between the number of arrhythmogenic drugs prescribed in addition to azithromycin with ordering telemetry. Finally, we assessed the relationship between baseline ECG abnormalities and telemetry monitoring in patients prescribed azithromycin. The purpose of these objectives was to better understand physician prescribing practices and to determine if patients have a potential risk of developing fatal cardiac arrhythmias

METHODS

RESULTS

Azithromycin use within our hospital system has increased from 15 days of therapy per 1000 patient days in 2002 to 40 days of therapy per 1000 patient days in 2013 (Figure 1). At the same time, azithromycin susceptibility in Streptococcus pneumoniae isolates has decreased over the past decade from 65% to 35% in our hospital.

Figure 1

The baseline characteristics of patients included in this study are noted in Table 1. The mean age of patients was 55 years, with a range of 21 to 97 years, and 61% were female. Forty‐five percent of patients were admitted to either the general medicine teaching service or hospitalist service, and 23% were admitted to the pulmonary service, which includes intensive care unit admission. The average length of patient stay was 9.7 days (range, 3115 days; median 6 days).

Seventy‐nine percent of azithromycin use was empiric for the treatment of suspected infection. The second most common use was for anti‐inflammatory effects (20%), as documented by prescribers in the medical record for patients with cystic fibrosis, lung transplant, and chronic obstructive pulmonary disease. Azithromycin was dosed appropriately according to the documented indication in 67% of patients, with the most discrepancy in dosing noted for anti‐inflammatory use. The average duration of azithromycin therapy was 4.5 days (range, 128 days). Duration was appropriate in 63% of patients. Twenty‐one percent of patients received intravenous formulation of azithromycin, 13% received intravenous followed by oral formulation, and 64% of patient received tablet formulation alone.

Thirty‐five medications have been identified in our formulary as having a potential major drug‐drug interaction when prescribed with azithromycin (see Supporting Information, Appendix, in the online version of this article), and of these medications, 20 were prescribed with azithromycin, with an average overlap of therapy of 4.5 days (Table 2). Seventy‐six percent of patients were concomitantly prescribed a QT‐prolonging drug in addition to azithromycin. The most commonly prescribed agents were ondansetron (48%), trazodone (22%), and moxifloxacin (17%).

Telemetry monitoring was assessed for each patient based on inpatient charges during their hospitalization (Table 3). Forty‐three percent of patients were placed on telemetry. Twenty‐four (24%) of the patients were prescribed azithromycin alone, of whom 45.8% were placed on telemetry. Fifty‐seven percent of patients were prescribed azithromycin with 1 to 2 additional QT‐prolonging medications (medium‐risk arm); 38.5% of patients in this group were placed on telemetry. In the high‐risk arm, 19% of patients were prescribed at least 3 QT‐prolonging medications in addition to azithromycin, of which only 52.6% of patients were monitored with telemetry. No statistically significant association was observed between risk level and telemetry placement (P=0.07).

Telemetry charges were further examined by analyzing baseline ECG evaluation within the past 6 months of their hospitalization (Table 3). Sixty‐six patients received baseline ECGs prior to initiation of azithromycin. Telemetry placement was not statistically correlated to abnormal QTc at baseline (P=0.22). Of those who underwent baseline ECG evaluation, 8.3% were noted to have borderline QTc, and 12.5% had abnormal QTc on admission prior to receiving azithromycin in the low‐risk level (Table 4). Within the medium‐risk level, 63.2% had baseline ECG evaluation, with 5.3% with borderline QTc and 35.7% with abnormal QTc. In the high‐risk level, 73.6% received a baseline ECG, with 21% with borderline QTc and 31.6% with abnormal QTc. No statistically significant association was observed between risk level and obtainment of baseline ECG (P=0.7). In 17 out of 66 patients, average repeat ECGs were obtained on day 3 (range, 27 days). Ten of the 17 ECGs showed increase in QTc (range, 397ms; average 27 ms), whereas the other 7 had a decrease in their QTc interval (range, 618 ms; average 13 ms; P=0.17).

As risk level increased, having an abnormal QTc at baseline was statistically different between low‐ and medium‐risk levels (P=0.03), but this association was lost when comparing the low‐risk arm to the high‐risk arm (P=0.11). When the medium‐ and high‐risk categories were combined, there was a noted statistical significance of having an abnormal ECG at baseline (P=0.03).

Of the 9 patients prescribed azithromycin chronically, 3 patients were in the low‐risk category, 4 in the medium‐risk category, and 2 in the high‐risk category. Only 2 had baseline ECGs obtained, 1 of which was noted to have abnormal QTc and was in the high‐risk category. Only 1 patient was placed on telemetry, but was considered low risk based on medications prescribed.

DISCUSSION

In this study, 76% of patients were prescribed azithromycin with 1 or more medications known to affect QT prolongation; 19% received 3 or more QT‐prolonging medications in addition to azithromycin. Of patients who received a baseline ECG, 43% were documented to have borderline or prolonged QTc on admission. Telemetry monitoring was ordered 43% of the time, but there was no significant association between telemetry placement and risk level (P=0.07), suggesting that telemetry was ordered based on symptoms more than risk. Despite more drug‐druginteracting medications prescribed, there was no association to either telemetry orders or baseline ECG evaluation. Furthermore, if an abnormal QTc was documented on admission, there was no relationship to ordering telemetry as an inpatient (P=0.215), suggesting that healthcare providers are not considering risk of QTc medication accumulation. Given increased warnings issued by the FDA for azithromycin, further prospective studies are indicated to fully assess risk of QTc prolongation and arrhythmias in the setting of multiple drug interactions. This study elucidates the potential for drug‐drug interactions and need for increased vigilance and education of providers in the healthcare setting for QTc prolongation and subsequent arrhythmias.

Forty‐eight percent of patients receiving other QTc prolonging medications were prescribed ondansetron, followed by 23% of patients prescribed trazodone. Both of these medications are included on the admission order set in our institution and can be easily ordered for patients. Despite ordering multiple medications that have potential for QTc prolongation, there are no current alerts set up in our electronic medical record. When patients are separated into drug interaction risk levels, there is a trend of having an abnormal QTc on admission, but this is driven by the large number of patients in the medium‐risk category, and the rate does not increase (and is not significant) when comparing high risk to low risk. However, patients who receive any QTc‐prolonging medication are more likely to have an abnormal QTc when compared to azithromycin prescription alone (P=0.03). The small sample size limits the power and generalizability of this study, and further larger studies are indicated to assess if risk of QTc prolongation is additive.

In the 9 patients prescribed azithromycin chronically, dosing was not consistent, and a vast majority of patients were not placed on telemetry nor had baseline ECGs on admission. This further correlates with the idea that risk of arrhythmia is not fully considered in this patient population, as patients prescribed more than 1 QTc‐prolonging medication were not included in prior studies that examined azithromycin for its anti‐inflammatory effects.[2]

Azithromycin was added to our hospital formulary in 1998, and prescription of this agent remained relatively low until 2006, when azithromycin use increased dramatically from 15 days of therapy (DOT) per 1000 patient days to 40 DOT per 1000 patient days. Although numerous factors may have led to this increase, literature was published in 2006 and 2011 citing benefit from the anti‐inflammatory effects of azithromycin.[2, 17] At the same time, azithromycin susceptibility among Streptococcus pneumoniae in patients within our hospital has decreased over the past decade; studies have found a correlation between increasing use of macrolides and the development of resistance in Streptococcus species.[18, 19, 20] In this study, 79% of patients were prescribed azithromycin empirically for treatment of bacterial infections, whereas 20% were given azithromycin for its anti‐inflammatory effects; both dose and frequency varied among patients, raising the concern for development of resistance. Published studies have shown improvement in quality of life and decreased frequency of exacerbation and infection when azithromycin is used as an anti‐inflammatory agent; however, no QTc monitoring was noted.[2] Drug‐induced QTc prolongation>10 ms above baseline suggests the potential for clinical significance, whereas a QTc prolongation >20 milliseconds above baseline has a substantially increased likelihood of being proarrhythmic.[1] Unfortunately, drug‐induced QT prolongation is unpredictable, and additional risk factors play a role in facilitating Torsades de pointes, including female sex, advanced age, electrolyte disturbances, intravenous formulation, and concurrent use of more than 1 drug that can prolong the QT interval.[15] Azithromycin has recently been added to the growing list of medications that can prolong the QT interval, with 12 cases of Torsades de pointes reported in the literature. In March 2013, the FDA released a warning regarding prescribing azithromycin, but there is a lack of guidance for clinicians in identifying risk of cardiovascular events in susceptible patients.

There are some limitations to this study. Given data were acquired retrospectively and telemetry sheets were unable to be reviewed. Some patients were noted to have arrhythmias, but these data were obtained through physician notes and not examined directly from telemetry sheets. Seventeen patients had repeat ECGs, but most were performed serially for chest pain and not QTc monitoring. Four patients died in this study, but cause of death could not be determined through electronic medical records provided for all 4 patients; families pursued withdrawal of care.

Despite the published FDA warning, there are no national guidelines for clinicians in prescribing QTc‐prolonging medications. The American Heart Association published recommendations in 2010 for prescribing these drugs in the inpatient setting, but because hospitals differ in cardiac monitoring, there is no one‐size‐fits‐all strategy in reducing risk of cardiac events.[14] If the benefit of azithromycin outweighs the risk, QTc prolongation should not limit therapy; however, institutional awareness is necessary, whether it be through automatic stop dates on azithromycin, electronic alerts regarding drug‐drug interaction, enhanced prescriber education, or a combination of all of the above.

Disclosure: Nothing to report.

Azithromycin, a macrolide antibiotic, received US Food and Drug Administration (FDA) approval in 1991 and is 1 of the most prescribed antibiotics used for a variety of infections, including community‐acquired pneumonia, bacterial sinusitis, urethritis, and cervicitis. In 2011, it was estimated that 40.3 million outpatients received a prescription for azithromycin.[1] In addition to treating acute bacterial infections, recent literature has pointed to using azithromycin for its unlabeled immunomodulatory and anti‐inflammatory effects, particularly in cystic fibrosis, chronic obstructive pulmonary disease (COPD), and lung transplant recipients.[2, 3, 4] Azithromycin decreases bacterial load and virulence, thus reducing airway secretion, as well as decreasing airway neutrophil accumulation through a reduction in proinflammatory cytokine expression.[4]

Cardiac toxicity can occur with macrolide antibiotics, and prolongation of the QT interval with subsequent Torsades de pointes has been documented with azithromycin.[1, 5, 6] In 2012, Ray et al. published data on a cohort of outpatients receiving azithromycin compared to amoxicillin, ciprofloxacin, or no antibiotics, and showed a small but absolute increase in cardiovascular deaths.[7, 8] Subsequent data, however, have not illustrated increased risk of death from cardiovascular causes. Mortensen et al. showed a lower risk of 90‐day mortality in older patients treated for community acquired pneumonia with azithromycin and ceftriaxone, although there was a nonstatistically significant increased risk of myocardial infarction in this group.[8, 9, 10] In March 2013, the FDA released an official statement regarding increased cardiovascular risk with azithromycin, stating that healthcare professionals should consider the risk of fatal heart rhythms with azithromycin when considering treatment options for patients who are at risk for cardiovascular events.[11]

In recent years, the potential for corrected QT (QTc) prolongation and Torsades de pointes has received increased attention due to its catastrophic nature, and it is thought that hospitalized patients are at a greater risk of drug‐induced Torsades de pointes due to the likelihood of having more risk factors.[12, 13] The American Heart Association released a statement in 2010 to raise awareness among healthcare professionals about risk, electrocardiogram (ECG) monitoring, and management of drug‐induced QT interval prolongation in hospitalized patients, although little data exist regarding quantification of risk in this patient population.[13, 14]

Prescribers currently have no standardized practice guidelines related to cardiovascular safety when prescribing QTc prolonging medications. Given the dramatic increase in azithromycin prescriptions and ongoing concern for cardiovascular risk and QTc prolongation, we investigated the prescribing practices with azithromycin within our institution. Our primary aims were 3‐fold. First, we aimed to describe the frequency azithromycin was prescribed with additional QTc prolonging medications. Second, we assessed the relationship between the number of arrhythmogenic drugs prescribed in addition to azithromycin with ordering telemetry. Finally, we assessed the relationship between baseline ECG abnormalities and telemetry monitoring in patients prescribed azithromycin. The purpose of these objectives was to better understand physician prescribing practices and to determine if patients have a potential risk of developing fatal cardiac arrhythmias

METHODS

RESULTS

Azithromycin use within our hospital system has increased from 15 days of therapy per 1000 patient days in 2002 to 40 days of therapy per 1000 patient days in 2013 (Figure 1). At the same time, azithromycin susceptibility in Streptococcus pneumoniae isolates has decreased over the past decade from 65% to 35% in our hospital.

Figure 1

The baseline characteristics of patients included in this study are noted in Table 1. The mean age of patients was 55 years, with a range of 21 to 97 years, and 61% were female. Forty‐five percent of patients were admitted to either the general medicine teaching service or hospitalist service, and 23% were admitted to the pulmonary service, which includes intensive care unit admission. The average length of patient stay was 9.7 days (range, 3115 days; median 6 days).

Seventy‐nine percent of azithromycin use was empiric for the treatment of suspected infection. The second most common use was for anti‐inflammatory effects (20%), as documented by prescribers in the medical record for patients with cystic fibrosis, lung transplant, and chronic obstructive pulmonary disease. Azithromycin was dosed appropriately according to the documented indication in 67% of patients, with the most discrepancy in dosing noted for anti‐inflammatory use. The average duration of azithromycin therapy was 4.5 days (range, 128 days). Duration was appropriate in 63% of patients. Twenty‐one percent of patients received intravenous formulation of azithromycin, 13% received intravenous followed by oral formulation, and 64% of patient received tablet formulation alone.

Thirty‐five medications have been identified in our formulary as having a potential major drug‐drug interaction when prescribed with azithromycin (see Supporting Information, Appendix, in the online version of this article), and of these medications, 20 were prescribed with azithromycin, with an average overlap of therapy of 4.5 days (Table 2). Seventy‐six percent of patients were concomitantly prescribed a QT‐prolonging drug in addition to azithromycin. The most commonly prescribed agents were ondansetron (48%), trazodone (22%), and moxifloxacin (17%).

Telemetry monitoring was assessed for each patient based on inpatient charges during their hospitalization (Table 3). Forty‐three percent of patients were placed on telemetry. Twenty‐four (24%) of the patients were prescribed azithromycin alone, of whom 45.8% were placed on telemetry. Fifty‐seven percent of patients were prescribed azithromycin with 1 to 2 additional QT‐prolonging medications (medium‐risk arm); 38.5% of patients in this group were placed on telemetry. In the high‐risk arm, 19% of patients were prescribed at least 3 QT‐prolonging medications in addition to azithromycin, of which only 52.6% of patients were monitored with telemetry. No statistically significant association was observed between risk level and telemetry placement (P=0.07).

Telemetry charges were further examined by analyzing baseline ECG evaluation within the past 6 months of their hospitalization (Table 3). Sixty‐six patients received baseline ECGs prior to initiation of azithromycin. Telemetry placement was not statistically correlated to abnormal QTc at baseline (P=0.22). Of those who underwent baseline ECG evaluation, 8.3% were noted to have borderline QTc, and 12.5% had abnormal QTc on admission prior to receiving azithromycin in the low‐risk level (Table 4). Within the medium‐risk level, 63.2% had baseline ECG evaluation, with 5.3% with borderline QTc and 35.7% with abnormal QTc. In the high‐risk level, 73.6% received a baseline ECG, with 21% with borderline QTc and 31.6% with abnormal QTc. No statistically significant association was observed between risk level and obtainment of baseline ECG (P=0.7). In 17 out of 66 patients, average repeat ECGs were obtained on day 3 (range, 27 days). Ten of the 17 ECGs showed increase in QTc (range, 397ms; average 27 ms), whereas the other 7 had a decrease in their QTc interval (range, 618 ms; average 13 ms; P=0.17).

As risk level increased, having an abnormal QTc at baseline was statistically different between low‐ and medium‐risk levels (P=0.03), but this association was lost when comparing the low‐risk arm to the high‐risk arm (P=0.11). When the medium‐ and high‐risk categories were combined, there was a noted statistical significance of having an abnormal ECG at baseline (P=0.03).

Of the 9 patients prescribed azithromycin chronically, 3 patients were in the low‐risk category, 4 in the medium‐risk category, and 2 in the high‐risk category. Only 2 had baseline ECGs obtained, 1 of which was noted to have abnormal QTc and was in the high‐risk category. Only 1 patient was placed on telemetry, but was considered low risk based on medications prescribed.

DISCUSSION

In this study, 76% of patients were prescribed azithromycin with 1 or more medications known to affect QT prolongation; 19% received 3 or more QT‐prolonging medications in addition to azithromycin. Of patients who received a baseline ECG, 43% were documented to have borderline or prolonged QTc on admission. Telemetry monitoring was ordered 43% of the time, but there was no significant association between telemetry placement and risk level (P=0.07), suggesting that telemetry was ordered based on symptoms more than risk. Despite more drug‐druginteracting medications prescribed, there was no association to either telemetry orders or baseline ECG evaluation. Furthermore, if an abnormal QTc was documented on admission, there was no relationship to ordering telemetry as an inpatient (P=0.215), suggesting that healthcare providers are not considering risk of QTc medication accumulation. Given increased warnings issued by the FDA for azithromycin, further prospective studies are indicated to fully assess risk of QTc prolongation and arrhythmias in the setting of multiple drug interactions. This study elucidates the potential for drug‐drug interactions and need for increased vigilance and education of providers in the healthcare setting for QTc prolongation and subsequent arrhythmias.

Forty‐eight percent of patients receiving other QTc prolonging medications were prescribed ondansetron, followed by 23% of patients prescribed trazodone. Both of these medications are included on the admission order set in our institution and can be easily ordered for patients. Despite ordering multiple medications that have potential for QTc prolongation, there are no current alerts set up in our electronic medical record. When patients are separated into drug interaction risk levels, there is a trend of having an abnormal QTc on admission, but this is driven by the large number of patients in the medium‐risk category, and the rate does not increase (and is not significant) when comparing high risk to low risk. However, patients who receive any QTc‐prolonging medication are more likely to have an abnormal QTc when compared to azithromycin prescription alone (P=0.03). The small sample size limits the power and generalizability of this study, and further larger studies are indicated to assess if risk of QTc prolongation is additive.

In the 9 patients prescribed azithromycin chronically, dosing was not consistent, and a vast majority of patients were not placed on telemetry nor had baseline ECGs on admission. This further correlates with the idea that risk of arrhythmia is not fully considered in this patient population, as patients prescribed more than 1 QTc‐prolonging medication were not included in prior studies that examined azithromycin for its anti‐inflammatory effects.[2]

Azithromycin was added to our hospital formulary in 1998, and prescription of this agent remained relatively low until 2006, when azithromycin use increased dramatically from 15 days of therapy (DOT) per 1000 patient days to 40 DOT per 1000 patient days. Although numerous factors may have led to this increase, literature was published in 2006 and 2011 citing benefit from the anti‐inflammatory effects of azithromycin.[2, 17] At the same time, azithromycin susceptibility among Streptococcus pneumoniae in patients within our hospital has decreased over the past decade; studies have found a correlation between increasing use of macrolides and the development of resistance in Streptococcus species.[18, 19, 20] In this study, 79% of patients were prescribed azithromycin empirically for treatment of bacterial infections, whereas 20% were given azithromycin for its anti‐inflammatory effects; both dose and frequency varied among patients, raising the concern for development of resistance. Published studies have shown improvement in quality of life and decreased frequency of exacerbation and infection when azithromycin is used as an anti‐inflammatory agent; however, no QTc monitoring was noted.[2] Drug‐induced QTc prolongation>10 ms above baseline suggests the potential for clinical significance, whereas a QTc prolongation >20 milliseconds above baseline has a substantially increased likelihood of being proarrhythmic.[1] Unfortunately, drug‐induced QT prolongation is unpredictable, and additional risk factors play a role in facilitating Torsades de pointes, including female sex, advanced age, electrolyte disturbances, intravenous formulation, and concurrent use of more than 1 drug that can prolong the QT interval.[15] Azithromycin has recently been added to the growing list of medications that can prolong the QT interval, with 12 cases of Torsades de pointes reported in the literature. In March 2013, the FDA released a warning regarding prescribing azithromycin, but there is a lack of guidance for clinicians in identifying risk of cardiovascular events in susceptible patients.

There are some limitations to this study. Given data were acquired retrospectively and telemetry sheets were unable to be reviewed. Some patients were noted to have arrhythmias, but these data were obtained through physician notes and not examined directly from telemetry sheets. Seventeen patients had repeat ECGs, but most were performed serially for chest pain and not QTc monitoring. Four patients died in this study, but cause of death could not be determined through electronic medical records provided for all 4 patients; families pursued withdrawal of care.

Despite the published FDA warning, there are no national guidelines for clinicians in prescribing QTc‐prolonging medications. The American Heart Association published recommendations in 2010 for prescribing these drugs in the inpatient setting, but because hospitals differ in cardiac monitoring, there is no one‐size‐fits‐all strategy in reducing risk of cardiac events.[14] If the benefit of azithromycin outweighs the risk, QTc prolongation should not limit therapy; however, institutional awareness is necessary, whether it be through automatic stop dates on azithromycin, electronic alerts regarding drug‐drug interaction, enhanced prescriber education, or a combination of all of the above.

Disclosure: Nothing to report.