Original Research

On January 1, 2014, several major provisions of the Affordable Care Act (ACA) took effect, including introduction of the individual mandate for health insurance coverage, opening of the Health Insurance Marketplace, and expansion of Medicaid eligibility to Americans earning up to 133% of the federal poverty level.[1] Nearly 9 million US adults have enrolled in Medicaid since that time, primarily in the 31 states and Washington, DC that have opted into Medicaid expansion.[2, 3] ACA implementation has also had a significant impact on hospital payer mix, primarily by reducing the volume of uncompensated care in Medicaid‐expansion states.[4, 5]

The differential shift in payer mix in Medicaid‐expansion versus nonexpansion states may be relevant to hospitals beyond reimbursement. Medicaid insurance has historically been associated with longer hospitalizations and higher in‐hospital mortality in diverse patient populations, more so than commercial insurance and often even uninsured payer status.[6, 7, 8, 9, 10, 11, 12, 13, 14, 15] The disparity in outcomes between patients with Medicaid versus other insurance persists even after adjustment for disease severity and baseline comorbidities. Insurance type may influence the delivery of inpatient care through variation in access to invasive procedures and adherence to guideline‐concordant medical therapies.[9, 10, 11, 12] Medicaid patients may be more likely than uninsured patients to remain hospitalized pending postacute care placement rather than be discharged home with family support.[16] Medicaid patients are also less likely to leave against medical advice than uninsured patients.[17]

Currently, little is known about the impact of state Medicaid expansion status on length of stay (LOS) or mortality nationally. It is possible that hospitals in Medicaid‐expansion states have experienced relative worsening in LOS and mortality as their share of Medicaid patients has grown. Determining the impact of ACA implementation on payer mix and patient outcomes is particularly important for academic medical centers (AMCs), as they traditionally care for the greatest percentage of both Medicaid and uninsured patients.[18] We sought to characterize the impact of state Medicaid expansion status on payer mix, LOS, and in‐hospital mortality for general medicine patients at AMCs in the United States.

METHODS

The University HealthSystem Consortium (UHC) is an alliance of 117 AMCs and 310 affiliated hospitals, representing >90% of such institutions in the US. We queried the online UHC Clinical Data Base/Resource Manager (CDB/RM) to obtain hospital‐level insurance, LOS, and mortality data for inpatients discharged from a general medicine service between October 1, 2012 and September 30, 2015. We excluded hospitals that were missing data for any month within the study period. No patient‐level data were accessed.

Our outcomes of interest were the proportion of discharges by primary payer (Medicare, commercial, Medicaid, uninsured, or other [eg, Tri‐Care or Workers' Compensation]), as well as the LOS index and mortality index. Both indices were defined as the ratio of the observed to expected values. To determine the expected LOS and mortality, the UHC 2015 risk adjustment models were applied to all cases, adjusting for variables such as patient demographics, low socioeconomic status, admit source and status, severity of illness, and comorbid conditions, as described by International Classification of Diseases, Ninth Revision codes. These models have been validated and are used for research and quality benchmarking for member institutions.[19]

We next stratified hospitals according to state Medicaid expansion status. We defined Medicaid‐expansion states as those that had expanded Medicaid by the end of the study period: Arizona, Arkansas, California, Colorado, Connecticut, Illinois, Indiana, Iowa, Kentucky, Maryland, Massachusetts, Michigan, Minnesota, Nevada, New Hampshire, New Jersey, New Mexico, New York, Ohio, Oregon, Pennsylvania, Rhode Island, Washington, Washington DC, and West Virginia. Nonexpansion states included Alabama, Florida, Georgia, Kansas, Louisiana, Missouri, Nebraska, North Carolina, South Carolina, Tennessee, Texas, Utah, Virginia, and Wisconsin. We excluded 12 states due to incomplete data: Alaska, Delaware, Hawaii, Idaho, North Dakota, Maine, Mississippi, Montana, Oklahoma, South Dakota, Vermont, and Wyoming.

We then identified our pre‐ and post‐ACA implementation periods. Medicaid coverage expansion took effect in all expansion states on January 1, 2014, with the exception of Michigan (April 1, 2014), New Hampshire (August 15, 2014), Pennsylvania (January 1, 2015), and Indiana (February 1, 2015).[3] We therefore defined October 1, 2012 to December 31, 2013 as the pre‐ACA implementation period and January 1, 2014 to September 30, 2015 as the post‐ACA implementation period for all states except for Michigan, New Hampshire, Pennsylvania, and Indiana. For these 4 states, we customized the pre‐ and post‐ACA implementation periods to their respective dates of Medicaid expansion; for New Hampshire, we designated October 1, 2012 to July 31, 2014 as the pre‐ACA implementation period and September 1, 2014 to September 30, 2015 as the post‐ACA implementation period, as we were unable to distinguish before versus after data in August 2014 based on the midmonth expansion of Medicaid.

After stratifying hospitals into groups based on whether they were located in Medicaid‐expansion or nonexpansion states, the proportion of discharges by payer was compared between pre‐ and post‐ACA implementation periods both graphically by quarter and using linear regression models weighted for the number of cases from each hospital. Next, for both Medicaid‐expansion and nonexpansion hospitals, LOS index and mortality index were compared before and after ACA implementation using linear regression models weighted for the number of cases from each hospital, both overall and by payer. Difference‐in‐differences estimations were then completed to compare the proportion of discharges by payer, LOS index, and mortality index between Medicaid‐expansion and nonexpansion hospitals before and after ACA implementation. Post hoc linear regression analyses were completed to evaluate the effect of clustering by state level strata on payer mix and LOS and mortality indices. A 2‐sided P value of <0.05 was considered statistically significant. Data analyses were performed using Stata 12.0 (StataCorp, College Station, TX).

RESULTS

We identified 4,258,952 discharges among general medicine patients from 211 hospitals in 38 states and Washington, DC between October 1, 2012, and September 30, 2015. This included 3,144,488 discharges from 156 hospitals in 24 Medicaid‐expansion states and Washington, DC and 1,114,464 discharges from 55 hospitals in 14 nonexpansion states.

Figure 1 shows the trends in payer mix over time for hospitals in both Medicaid‐expansion and nonexpansion states. As summarized in Table 1, hospitals in Medicaid‐expansion states experienced a significant 3.7‐percentage point increase in Medicaid discharges (P = 0.013) and 2.9‐percentage point decrease in uninsured discharges (P < 0.001) after ACA implementation. This represented an approximately 19% jump and 60% drop in Medicaid and uninsured discharges, respectively. Hospitals in nonexpansion states saw no significant change in the proportion of discharges by payer after ACA implementation. In the difference‐in‐differences analysis, there was a trend toward a greater change in the proportion of Medicaid discharges pre‐ to post‐ACA implementation among hospitals in Medicaid‐expansion states compared to hospitals in nonexpansion states (mean difference‐in‐differences 4.1%, 95% confidence interval [CI]: 0.3%, 8.6%, P = 0.070).

Figure 1

Table 1 shows that the overall LOS index remained unchanged pre‐ to post‐ACA implementation for both Medicaid‐expansion (1.017 to 1.006, P = 0.488) and nonexpansion hospitals (1.008 to 0.995, P = 0.574). LOS indices for each payer type also remained unchanged. The overall mortality index significantly improved pre‐ to post‐ACA implementation for both Medicaid‐expansion (1.000 to 0.878, P < 0.001) and nonexpansion hospitals (0.997 to 0.850, P = 0.001). Among both Medicaid‐expansion and nonexpansion hospitals, the mortality index significantly improved for Medicare, commercial, and Medicaid discharges but not for uninsured or other discharges. In the difference‐in‐differences analysis, the changes in LOS indices and mortality indices pre‐ to post‐ACA implementation did not differ significantly between hospitals in Medicaid‐expansion versus nonexpansion states.

In post hoc linear regression analyses of payer mix and LOS and mortality indices clustered by state‐level strata, point estimates were minimally changed. Although 95% CIs were slightly wider, statistical significance was unchanged from our primary analyses (data not shown).

DISCUSSION

We found that ACA implementation had a significant impact on payer mix for general medicine patients at AMCs in the United States, primarily by increasing the number of Medicaid beneficiaries and by decreasing the number of uninsured patients in Medicaid‐expansion states. State Medicaid expansion status did not appear to influence either LOS or in‐hospital mortality.

Our study offers some of the longest‐term data currently available on the impact of ACA implementation on payer mix trends and encompasses more states than others have previously. Although we uniquely focused on general medicine patients at AMCs, our results are similar to those seen for US hospitals overall. Nikpay and colleagues evaluated payer mix trends for non‐Medicare adult inpatient stays in 16 states through the second quarter of 2014 using the Healthcare Cost and Utilization Project database through the Agency for Healthcare Research and Quality.[4] They found a relative 20% increase and 50% decrease in Medicaid and uninsured discharges in Medicaid‐expansion states, along with nonsignificant changes in nonexpansion states. Hempstead and Cantor assessed payer mix for non‐Medicare discharges using state hospital association data from 21 states through the fourth quarter of 2014 and found a significant increase in Medicaid patients as well as a nearly significant decrease in uninsured patients in expansion states relative to nonexpansion states.[5] The Department of Health and Human Services also reported that uninsured/self‐pay discharges fell substantially (65%73%) in Medicaid‐expansion states by the end of 2014, with slight decreases in nonexpansion states.[20]

In contrast to our hypothesis, the overall LOS and in‐hospital mortality indices were not influenced by state Medicaid expansion status. From a purely mathematical standpoint, the contribution of Medicaid patients to the overall LOS and mortality indices may have been eclipsed by Medicare and commercially insured patients, who represented a higher proportion of total discharges. The lack of impact of state Medicaid expansion status on overall LOS and mortality indices did not appear to occur as a result of indices for Medicaid patients trending toward the mean. As predicted based on observational studies, Medicaid patients in our study tended to have a higher LOS index than those with other insurance types. Medicaid patients actually tended to have a lower mortality index in our analysis; the reason for this latter finding is unclear and in contrast to other published studies.[6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 21]

To our knowledge, no other studies have evaluated the effect of payer mix changes under the ACA on inpatient outcomes. However, new evidence is emerging on outpatient outcomes. Low‐income adults in Medicaid‐expansion states have reported greater gains in access to primary care services and in the diagnosis of certain chronic health conditions than those in nonexpansion states as a result of ACA implementation.[22, 23] Such improvements in the outpatient setting might be expected to reduce patient acuity on admission. However, they would not necessarily translate to relative improvements in LOS or mortality indices for Medicaid‐expansion hospitals, as the UHC risk adjustment models controlled for disease severity on admission.

Similarly, few studies have assessed the impact of payer mix changes under previous state Medicaid expansions on inpatient outcomes. After Massachusetts expanded Medicaid and enacted near‐universal healthcare coverage in 2006, a minimal LOS reduction of just 0.05 days was observed.[24] New York expanded Medicaid eligibility to nondisabled childless adults with incomes below 100% of the federal poverty level in September 2001, whereas Arizona did so in November 2001 and Maine in October 2002. A study comparing outcomes in these states to 4 neighboring nonexpansion states found a relative reduction in annual all‐cause mortality of 6.1% population wide; however, it did not assess in‐hospital mortality.[25] The Oregon Health Insurance Experiment that randomized low‐income adults to expanded Medicaid coverage or not in 2008 has also reported on outpatient rather than inpatient outcomes.[26]

Our findings have potential implications for health policymakers. That Medicaid expansion status had a neutral effect on both LOS and mortality indices in our analysis should be reassuring for states contemplating Medicaid expansion in the future. Our results also highlight the need for further efforts to reduce disparities in inpatient care based on payer status. For example, although Medicare, commercially insured, and Medicaid patients witnessed significant improvements in mortality indices pre‐ to post‐ACA implementation in hospitals in both Medicaid‐expansion and nonexpansion states, uninsured patients did not.

This study has several limitations. First, our analysis of the impact of ACA implementation on payer mix did not account for concurrent socioeconomic trends that may have influenced insurance coverage across the United States. However, the main goal of this analysis was to demonstrate that changes in payer mix did in fact occur over time, to provide rationale for our subsequent LOS and mortality analyses. Second, we could not control for variation in the design and implementation of Medicaid expansions across states as permitted under the federal Section 1115 waiver process. Third, we only had access to hospital‐level data through the UHC CDB/RM, rather than individual patient data. We attempted to mitigate this limitation by weighting data according to the number of cases per hospital. Lastly, additional patient‐level factors that may influence LOS or mortality may not be included in the UHC risk adjustment models.

In summary, the differential shift in payer mix between Medicaid‐expansion and nonexpansion states did not influence overall LOS or in‐hospital mortality for general medicine patients at AMCs in the United States. Additional research could help to determine the impact of ACA implementation on other patient outcomes that may be dependent on insurance status, such as readmissions or hospital‐acquired complications.

Disclosures: M.E.A. conceived of the study concept and design, assisted with data acquisition, and drafted the manuscript. J.J.G. assisted with study design and made critical revisions to the manuscript. D.A. assisted with study design and made critical revisions to the manuscript. R.P. assisted with study design and made critical revisions to the manuscript. M.L. assisted with study design and data acquisition and made critical revisions to the manuscript. C.D.J. assisted with study design, performed data analyses, and made critical revisions to the manuscript. A modified abstract was presented in poster format at the University HealthSystem Consortium Annual Conference held September 30 to October 2, 2015 in Orlando, Florida, as well as at the Society of Hospital Medicine Research, Innovations, and Vignettes 2016 Annual Meeting held March 69, 2016, in San Diego, California. The authors report no conflicts of interest.

On January 1, 2014, several major provisions of the Affordable Care Act (ACA) took effect, including introduction of the individual mandate for health insurance coverage, opening of the Health Insurance Marketplace, and expansion of Medicaid eligibility to Americans earning up to 133% of the federal poverty level.[1] Nearly 9 million US adults have enrolled in Medicaid since that time, primarily in the 31 states and Washington, DC that have opted into Medicaid expansion.[2, 3] ACA implementation has also had a significant impact on hospital payer mix, primarily by reducing the volume of uncompensated care in Medicaid‐expansion states.[4, 5]

The differential shift in payer mix in Medicaid‐expansion versus nonexpansion states may be relevant to hospitals beyond reimbursement. Medicaid insurance has historically been associated with longer hospitalizations and higher in‐hospital mortality in diverse patient populations, more so than commercial insurance and often even uninsured payer status.[6, 7, 8, 9, 10, 11, 12, 13, 14, 15] The disparity in outcomes between patients with Medicaid versus other insurance persists even after adjustment for disease severity and baseline comorbidities. Insurance type may influence the delivery of inpatient care through variation in access to invasive procedures and adherence to guideline‐concordant medical therapies.[9, 10, 11, 12] Medicaid patients may be more likely than uninsured patients to remain hospitalized pending postacute care placement rather than be discharged home with family support.[16] Medicaid patients are also less likely to leave against medical advice than uninsured patients.[17]

Currently, little is known about the impact of state Medicaid expansion status on length of stay (LOS) or mortality nationally. It is possible that hospitals in Medicaid‐expansion states have experienced relative worsening in LOS and mortality as their share of Medicaid patients has grown. Determining the impact of ACA implementation on payer mix and patient outcomes is particularly important for academic medical centers (AMCs), as they traditionally care for the greatest percentage of both Medicaid and uninsured patients.[18] We sought to characterize the impact of state Medicaid expansion status on payer mix, LOS, and in‐hospital mortality for general medicine patients at AMCs in the United States.

METHODS

The University HealthSystem Consortium (UHC) is an alliance of 117 AMCs and 310 affiliated hospitals, representing >90% of such institutions in the US. We queried the online UHC Clinical Data Base/Resource Manager (CDB/RM) to obtain hospital‐level insurance, LOS, and mortality data for inpatients discharged from a general medicine service between October 1, 2012 and September 30, 2015. We excluded hospitals that were missing data for any month within the study period. No patient‐level data were accessed.

Our outcomes of interest were the proportion of discharges by primary payer (Medicare, commercial, Medicaid, uninsured, or other [eg, Tri‐Care or Workers' Compensation]), as well as the LOS index and mortality index. Both indices were defined as the ratio of the observed to expected values. To determine the expected LOS and mortality, the UHC 2015 risk adjustment models were applied to all cases, adjusting for variables such as patient demographics, low socioeconomic status, admit source and status, severity of illness, and comorbid conditions, as described by International Classification of Diseases, Ninth Revision codes. These models have been validated and are used for research and quality benchmarking for member institutions.[19]

We next stratified hospitals according to state Medicaid expansion status. We defined Medicaid‐expansion states as those that had expanded Medicaid by the end of the study period: Arizona, Arkansas, California, Colorado, Connecticut, Illinois, Indiana, Iowa, Kentucky, Maryland, Massachusetts, Michigan, Minnesota, Nevada, New Hampshire, New Jersey, New Mexico, New York, Ohio, Oregon, Pennsylvania, Rhode Island, Washington, Washington DC, and West Virginia. Nonexpansion states included Alabama, Florida, Georgia, Kansas, Louisiana, Missouri, Nebraska, North Carolina, South Carolina, Tennessee, Texas, Utah, Virginia, and Wisconsin. We excluded 12 states due to incomplete data: Alaska, Delaware, Hawaii, Idaho, North Dakota, Maine, Mississippi, Montana, Oklahoma, South Dakota, Vermont, and Wyoming.

We then identified our pre‐ and post‐ACA implementation periods. Medicaid coverage expansion took effect in all expansion states on January 1, 2014, with the exception of Michigan (April 1, 2014), New Hampshire (August 15, 2014), Pennsylvania (January 1, 2015), and Indiana (February 1, 2015).[3] We therefore defined October 1, 2012 to December 31, 2013 as the pre‐ACA implementation period and January 1, 2014 to September 30, 2015 as the post‐ACA implementation period for all states except for Michigan, New Hampshire, Pennsylvania, and Indiana. For these 4 states, we customized the pre‐ and post‐ACA implementation periods to their respective dates of Medicaid expansion; for New Hampshire, we designated October 1, 2012 to July 31, 2014 as the pre‐ACA implementation period and September 1, 2014 to September 30, 2015 as the post‐ACA implementation period, as we were unable to distinguish before versus after data in August 2014 based on the midmonth expansion of Medicaid.

After stratifying hospitals into groups based on whether they were located in Medicaid‐expansion or nonexpansion states, the proportion of discharges by payer was compared between pre‐ and post‐ACA implementation periods both graphically by quarter and using linear regression models weighted for the number of cases from each hospital. Next, for both Medicaid‐expansion and nonexpansion hospitals, LOS index and mortality index were compared before and after ACA implementation using linear regression models weighted for the number of cases from each hospital, both overall and by payer. Difference‐in‐differences estimations were then completed to compare the proportion of discharges by payer, LOS index, and mortality index between Medicaid‐expansion and nonexpansion hospitals before and after ACA implementation. Post hoc linear regression analyses were completed to evaluate the effect of clustering by state level strata on payer mix and LOS and mortality indices. A 2‐sided P value of <0.05 was considered statistically significant. Data analyses were performed using Stata 12.0 (StataCorp, College Station, TX).

RESULTS

We identified 4,258,952 discharges among general medicine patients from 211 hospitals in 38 states and Washington, DC between October 1, 2012, and September 30, 2015. This included 3,144,488 discharges from 156 hospitals in 24 Medicaid‐expansion states and Washington, DC and 1,114,464 discharges from 55 hospitals in 14 nonexpansion states.

Figure 1 shows the trends in payer mix over time for hospitals in both Medicaid‐expansion and nonexpansion states. As summarized in Table 1, hospitals in Medicaid‐expansion states experienced a significant 3.7‐percentage point increase in Medicaid discharges (P = 0.013) and 2.9‐percentage point decrease in uninsured discharges (P < 0.001) after ACA implementation. This represented an approximately 19% jump and 60% drop in Medicaid and uninsured discharges, respectively. Hospitals in nonexpansion states saw no significant change in the proportion of discharges by payer after ACA implementation. In the difference‐in‐differences analysis, there was a trend toward a greater change in the proportion of Medicaid discharges pre‐ to post‐ACA implementation among hospitals in Medicaid‐expansion states compared to hospitals in nonexpansion states (mean difference‐in‐differences 4.1%, 95% confidence interval [CI]: 0.3%, 8.6%, P = 0.070).

Figure 1

Table 1 shows that the overall LOS index remained unchanged pre‐ to post‐ACA implementation for both Medicaid‐expansion (1.017 to 1.006, P = 0.488) and nonexpansion hospitals (1.008 to 0.995, P = 0.574). LOS indices for each payer type also remained unchanged. The overall mortality index significantly improved pre‐ to post‐ACA implementation for both Medicaid‐expansion (1.000 to 0.878, P < 0.001) and nonexpansion hospitals (0.997 to 0.850, P = 0.001). Among both Medicaid‐expansion and nonexpansion hospitals, the mortality index significantly improved for Medicare, commercial, and Medicaid discharges but not for uninsured or other discharges. In the difference‐in‐differences analysis, the changes in LOS indices and mortality indices pre‐ to post‐ACA implementation did not differ significantly between hospitals in Medicaid‐expansion versus nonexpansion states.

In post hoc linear regression analyses of payer mix and LOS and mortality indices clustered by state‐level strata, point estimates were minimally changed. Although 95% CIs were slightly wider, statistical significance was unchanged from our primary analyses (data not shown).

DISCUSSION

We found that ACA implementation had a significant impact on payer mix for general medicine patients at AMCs in the United States, primarily by increasing the number of Medicaid beneficiaries and by decreasing the number of uninsured patients in Medicaid‐expansion states. State Medicaid expansion status did not appear to influence either LOS or in‐hospital mortality.

Our study offers some of the longest‐term data currently available on the impact of ACA implementation on payer mix trends and encompasses more states than others have previously. Although we uniquely focused on general medicine patients at AMCs, our results are similar to those seen for US hospitals overall. Nikpay and colleagues evaluated payer mix trends for non‐Medicare adult inpatient stays in 16 states through the second quarter of 2014 using the Healthcare Cost and Utilization Project database through the Agency for Healthcare Research and Quality.[4] They found a relative 20% increase and 50% decrease in Medicaid and uninsured discharges in Medicaid‐expansion states, along with nonsignificant changes in nonexpansion states. Hempstead and Cantor assessed payer mix for non‐Medicare discharges using state hospital association data from 21 states through the fourth quarter of 2014 and found a significant increase in Medicaid patients as well as a nearly significant decrease in uninsured patients in expansion states relative to nonexpansion states.[5] The Department of Health and Human Services also reported that uninsured/self‐pay discharges fell substantially (65%73%) in Medicaid‐expansion states by the end of 2014, with slight decreases in nonexpansion states.[20]

In contrast to our hypothesis, the overall LOS and in‐hospital mortality indices were not influenced by state Medicaid expansion status. From a purely mathematical standpoint, the contribution of Medicaid patients to the overall LOS and mortality indices may have been eclipsed by Medicare and commercially insured patients, who represented a higher proportion of total discharges. The lack of impact of state Medicaid expansion status on overall LOS and mortality indices did not appear to occur as a result of indices for Medicaid patients trending toward the mean. As predicted based on observational studies, Medicaid patients in our study tended to have a higher LOS index than those with other insurance types. Medicaid patients actually tended to have a lower mortality index in our analysis; the reason for this latter finding is unclear and in contrast to other published studies.[6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 21]

To our knowledge, no other studies have evaluated the effect of payer mix changes under the ACA on inpatient outcomes. However, new evidence is emerging on outpatient outcomes. Low‐income adults in Medicaid‐expansion states have reported greater gains in access to primary care services and in the diagnosis of certain chronic health conditions than those in nonexpansion states as a result of ACA implementation.[22, 23] Such improvements in the outpatient setting might be expected to reduce patient acuity on admission. However, they would not necessarily translate to relative improvements in LOS or mortality indices for Medicaid‐expansion hospitals, as the UHC risk adjustment models controlled for disease severity on admission.

Similarly, few studies have assessed the impact of payer mix changes under previous state Medicaid expansions on inpatient outcomes. After Massachusetts expanded Medicaid and enacted near‐universal healthcare coverage in 2006, a minimal LOS reduction of just 0.05 days was observed.[24] New York expanded Medicaid eligibility to nondisabled childless adults with incomes below 100% of the federal poverty level in September 2001, whereas Arizona did so in November 2001 and Maine in October 2002. A study comparing outcomes in these states to 4 neighboring nonexpansion states found a relative reduction in annual all‐cause mortality of 6.1% population wide; however, it did not assess in‐hospital mortality.[25] The Oregon Health Insurance Experiment that randomized low‐income adults to expanded Medicaid coverage or not in 2008 has also reported on outpatient rather than inpatient outcomes.[26]

Our findings have potential implications for health policymakers. That Medicaid expansion status had a neutral effect on both LOS and mortality indices in our analysis should be reassuring for states contemplating Medicaid expansion in the future. Our results also highlight the need for further efforts to reduce disparities in inpatient care based on payer status. For example, although Medicare, commercially insured, and Medicaid patients witnessed significant improvements in mortality indices pre‐ to post‐ACA implementation in hospitals in both Medicaid‐expansion and nonexpansion states, uninsured patients did not.

This study has several limitations. First, our analysis of the impact of ACA implementation on payer mix did not account for concurrent socioeconomic trends that may have influenced insurance coverage across the United States. However, the main goal of this analysis was to demonstrate that changes in payer mix did in fact occur over time, to provide rationale for our subsequent LOS and mortality analyses. Second, we could not control for variation in the design and implementation of Medicaid expansions across states as permitted under the federal Section 1115 waiver process. Third, we only had access to hospital‐level data through the UHC CDB/RM, rather than individual patient data. We attempted to mitigate this limitation by weighting data according to the number of cases per hospital. Lastly, additional patient‐level factors that may influence LOS or mortality may not be included in the UHC risk adjustment models.

In summary, the differential shift in payer mix between Medicaid‐expansion and nonexpansion states did not influence overall LOS or in‐hospital mortality for general medicine patients at AMCs in the United States. Additional research could help to determine the impact of ACA implementation on other patient outcomes that may be dependent on insurance status, such as readmissions or hospital‐acquired complications.

Disclosures: M.E.A. conceived of the study concept and design, assisted with data acquisition, and drafted the manuscript. J.J.G. assisted with study design and made critical revisions to the manuscript. D.A. assisted with study design and made critical revisions to the manuscript. R.P. assisted with study design and made critical revisions to the manuscript. M.L. assisted with study design and data acquisition and made critical revisions to the manuscript. C.D.J. assisted with study design, performed data analyses, and made critical revisions to the manuscript. A modified abstract was presented in poster format at the University HealthSystem Consortium Annual Conference held September 30 to October 2, 2015 in Orlando, Florida, as well as at the Society of Hospital Medicine Research, Innovations, and Vignettes 2016 Annual Meeting held March 69, 2016, in San Diego, California. The authors report no conflicts of interest.

On January 1, 2014, several major provisions of the Affordable Care Act (ACA) took effect, including introduction of the individual mandate for health insurance coverage, opening of the Health Insurance Marketplace, and expansion of Medicaid eligibility to Americans earning up to 133% of the federal poverty level.[1] Nearly 9 million US adults have enrolled in Medicaid since that time, primarily in the 31 states and Washington, DC that have opted into Medicaid expansion.[2, 3] ACA implementation has also had a significant impact on hospital payer mix, primarily by reducing the volume of uncompensated care in Medicaid‐expansion states.[4, 5]

The differential shift in payer mix in Medicaid‐expansion versus nonexpansion states may be relevant to hospitals beyond reimbursement. Medicaid insurance has historically been associated with longer hospitalizations and higher in‐hospital mortality in diverse patient populations, more so than commercial insurance and often even uninsured payer status.[6, 7, 8, 9, 10, 11, 12, 13, 14, 15] The disparity in outcomes between patients with Medicaid versus other insurance persists even after adjustment for disease severity and baseline comorbidities. Insurance type may influence the delivery of inpatient care through variation in access to invasive procedures and adherence to guideline‐concordant medical therapies.[9, 10, 11, 12] Medicaid patients may be more likely than uninsured patients to remain hospitalized pending postacute care placement rather than be discharged home with family support.[16] Medicaid patients are also less likely to leave against medical advice than uninsured patients.[17]

Currently, little is known about the impact of state Medicaid expansion status on length of stay (LOS) or mortality nationally. It is possible that hospitals in Medicaid‐expansion states have experienced relative worsening in LOS and mortality as their share of Medicaid patients has grown. Determining the impact of ACA implementation on payer mix and patient outcomes is particularly important for academic medical centers (AMCs), as they traditionally care for the greatest percentage of both Medicaid and uninsured patients.[18] We sought to characterize the impact of state Medicaid expansion status on payer mix, LOS, and in‐hospital mortality for general medicine patients at AMCs in the United States.

METHODS

The University HealthSystem Consortium (UHC) is an alliance of 117 AMCs and 310 affiliated hospitals, representing >90% of such institutions in the US. We queried the online UHC Clinical Data Base/Resource Manager (CDB/RM) to obtain hospital‐level insurance, LOS, and mortality data for inpatients discharged from a general medicine service between October 1, 2012 and September 30, 2015. We excluded hospitals that were missing data for any month within the study period. No patient‐level data were accessed.

Our outcomes of interest were the proportion of discharges by primary payer (Medicare, commercial, Medicaid, uninsured, or other [eg, Tri‐Care or Workers' Compensation]), as well as the LOS index and mortality index. Both indices were defined as the ratio of the observed to expected values. To determine the expected LOS and mortality, the UHC 2015 risk adjustment models were applied to all cases, adjusting for variables such as patient demographics, low socioeconomic status, admit source and status, severity of illness, and comorbid conditions, as described by International Classification of Diseases, Ninth Revision codes. These models have been validated and are used for research and quality benchmarking for member institutions.[19]

We next stratified hospitals according to state Medicaid expansion status. We defined Medicaid‐expansion states as those that had expanded Medicaid by the end of the study period: Arizona, Arkansas, California, Colorado, Connecticut, Illinois, Indiana, Iowa, Kentucky, Maryland, Massachusetts, Michigan, Minnesota, Nevada, New Hampshire, New Jersey, New Mexico, New York, Ohio, Oregon, Pennsylvania, Rhode Island, Washington, Washington DC, and West Virginia. Nonexpansion states included Alabama, Florida, Georgia, Kansas, Louisiana, Missouri, Nebraska, North Carolina, South Carolina, Tennessee, Texas, Utah, Virginia, and Wisconsin. We excluded 12 states due to incomplete data: Alaska, Delaware, Hawaii, Idaho, North Dakota, Maine, Mississippi, Montana, Oklahoma, South Dakota, Vermont, and Wyoming.

We then identified our pre‐ and post‐ACA implementation periods. Medicaid coverage expansion took effect in all expansion states on January 1, 2014, with the exception of Michigan (April 1, 2014), New Hampshire (August 15, 2014), Pennsylvania (January 1, 2015), and Indiana (February 1, 2015).[3] We therefore defined October 1, 2012 to December 31, 2013 as the pre‐ACA implementation period and January 1, 2014 to September 30, 2015 as the post‐ACA implementation period for all states except for Michigan, New Hampshire, Pennsylvania, and Indiana. For these 4 states, we customized the pre‐ and post‐ACA implementation periods to their respective dates of Medicaid expansion; for New Hampshire, we designated October 1, 2012 to July 31, 2014 as the pre‐ACA implementation period and September 1, 2014 to September 30, 2015 as the post‐ACA implementation period, as we were unable to distinguish before versus after data in August 2014 based on the midmonth expansion of Medicaid.

After stratifying hospitals into groups based on whether they were located in Medicaid‐expansion or nonexpansion states, the proportion of discharges by payer was compared between pre‐ and post‐ACA implementation periods both graphically by quarter and using linear regression models weighted for the number of cases from each hospital. Next, for both Medicaid‐expansion and nonexpansion hospitals, LOS index and mortality index were compared before and after ACA implementation using linear regression models weighted for the number of cases from each hospital, both overall and by payer. Difference‐in‐differences estimations were then completed to compare the proportion of discharges by payer, LOS index, and mortality index between Medicaid‐expansion and nonexpansion hospitals before and after ACA implementation. Post hoc linear regression analyses were completed to evaluate the effect of clustering by state level strata on payer mix and LOS and mortality indices. A 2‐sided P value of <0.05 was considered statistically significant. Data analyses were performed using Stata 12.0 (StataCorp, College Station, TX).

RESULTS

We identified 4,258,952 discharges among general medicine patients from 211 hospitals in 38 states and Washington, DC between October 1, 2012, and September 30, 2015. This included 3,144,488 discharges from 156 hospitals in 24 Medicaid‐expansion states and Washington, DC and 1,114,464 discharges from 55 hospitals in 14 nonexpansion states.

Figure 1 shows the trends in payer mix over time for hospitals in both Medicaid‐expansion and nonexpansion states. As summarized in Table 1, hospitals in Medicaid‐expansion states experienced a significant 3.7‐percentage point increase in Medicaid discharges (P = 0.013) and 2.9‐percentage point decrease in uninsured discharges (P < 0.001) after ACA implementation. This represented an approximately 19% jump and 60% drop in Medicaid and uninsured discharges, respectively. Hospitals in nonexpansion states saw no significant change in the proportion of discharges by payer after ACA implementation. In the difference‐in‐differences analysis, there was a trend toward a greater change in the proportion of Medicaid discharges pre‐ to post‐ACA implementation among hospitals in Medicaid‐expansion states compared to hospitals in nonexpansion states (mean difference‐in‐differences 4.1%, 95% confidence interval [CI]: 0.3%, 8.6%, P = 0.070).

Figure 1

Table 1 shows that the overall LOS index remained unchanged pre‐ to post‐ACA implementation for both Medicaid‐expansion (1.017 to 1.006, P = 0.488) and nonexpansion hospitals (1.008 to 0.995, P = 0.574). LOS indices for each payer type also remained unchanged. The overall mortality index significantly improved pre‐ to post‐ACA implementation for both Medicaid‐expansion (1.000 to 0.878, P < 0.001) and nonexpansion hospitals (0.997 to 0.850, P = 0.001). Among both Medicaid‐expansion and nonexpansion hospitals, the mortality index significantly improved for Medicare, commercial, and Medicaid discharges but not for uninsured or other discharges. In the difference‐in‐differences analysis, the changes in LOS indices and mortality indices pre‐ to post‐ACA implementation did not differ significantly between hospitals in Medicaid‐expansion versus nonexpansion states.

In post hoc linear regression analyses of payer mix and LOS and mortality indices clustered by state‐level strata, point estimates were minimally changed. Although 95% CIs were slightly wider, statistical significance was unchanged from our primary analyses (data not shown).

DISCUSSION

We found that ACA implementation had a significant impact on payer mix for general medicine patients at AMCs in the United States, primarily by increasing the number of Medicaid beneficiaries and by decreasing the number of uninsured patients in Medicaid‐expansion states. State Medicaid expansion status did not appear to influence either LOS or in‐hospital mortality.

Our study offers some of the longest‐term data currently available on the impact of ACA implementation on payer mix trends and encompasses more states than others have previously. Although we uniquely focused on general medicine patients at AMCs, our results are similar to those seen for US hospitals overall. Nikpay and colleagues evaluated payer mix trends for non‐Medicare adult inpatient stays in 16 states through the second quarter of 2014 using the Healthcare Cost and Utilization Project database through the Agency for Healthcare Research and Quality.[4] They found a relative 20% increase and 50% decrease in Medicaid and uninsured discharges in Medicaid‐expansion states, along with nonsignificant changes in nonexpansion states. Hempstead and Cantor assessed payer mix for non‐Medicare discharges using state hospital association data from 21 states through the fourth quarter of 2014 and found a significant increase in Medicaid patients as well as a nearly significant decrease in uninsured patients in expansion states relative to nonexpansion states.[5] The Department of Health and Human Services also reported that uninsured/self‐pay discharges fell substantially (65%73%) in Medicaid‐expansion states by the end of 2014, with slight decreases in nonexpansion states.[20]

In contrast to our hypothesis, the overall LOS and in‐hospital mortality indices were not influenced by state Medicaid expansion status. From a purely mathematical standpoint, the contribution of Medicaid patients to the overall LOS and mortality indices may have been eclipsed by Medicare and commercially insured patients, who represented a higher proportion of total discharges. The lack of impact of state Medicaid expansion status on overall LOS and mortality indices did not appear to occur as a result of indices for Medicaid patients trending toward the mean. As predicted based on observational studies, Medicaid patients in our study tended to have a higher LOS index than those with other insurance types. Medicaid patients actually tended to have a lower mortality index in our analysis; the reason for this latter finding is unclear and in contrast to other published studies.[6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 21]

To our knowledge, no other studies have evaluated the effect of payer mix changes under the ACA on inpatient outcomes. However, new evidence is emerging on outpatient outcomes. Low‐income adults in Medicaid‐expansion states have reported greater gains in access to primary care services and in the diagnosis of certain chronic health conditions than those in nonexpansion states as a result of ACA implementation.[22, 23] Such improvements in the outpatient setting might be expected to reduce patient acuity on admission. However, they would not necessarily translate to relative improvements in LOS or mortality indices for Medicaid‐expansion hospitals, as the UHC risk adjustment models controlled for disease severity on admission.

Similarly, few studies have assessed the impact of payer mix changes under previous state Medicaid expansions on inpatient outcomes. After Massachusetts expanded Medicaid and enacted near‐universal healthcare coverage in 2006, a minimal LOS reduction of just 0.05 days was observed.[24] New York expanded Medicaid eligibility to nondisabled childless adults with incomes below 100% of the federal poverty level in September 2001, whereas Arizona did so in November 2001 and Maine in October 2002. A study comparing outcomes in these states to 4 neighboring nonexpansion states found a relative reduction in annual all‐cause mortality of 6.1% population wide; however, it did not assess in‐hospital mortality.[25] The Oregon Health Insurance Experiment that randomized low‐income adults to expanded Medicaid coverage or not in 2008 has also reported on outpatient rather than inpatient outcomes.[26]

Our findings have potential implications for health policymakers. That Medicaid expansion status had a neutral effect on both LOS and mortality indices in our analysis should be reassuring for states contemplating Medicaid expansion in the future. Our results also highlight the need for further efforts to reduce disparities in inpatient care based on payer status. For example, although Medicare, commercially insured, and Medicaid patients witnessed significant improvements in mortality indices pre‐ to post‐ACA implementation in hospitals in both Medicaid‐expansion and nonexpansion states, uninsured patients did not.

This study has several limitations. First, our analysis of the impact of ACA implementation on payer mix did not account for concurrent socioeconomic trends that may have influenced insurance coverage across the United States. However, the main goal of this analysis was to demonstrate that changes in payer mix did in fact occur over time, to provide rationale for our subsequent LOS and mortality analyses. Second, we could not control for variation in the design and implementation of Medicaid expansions across states as permitted under the federal Section 1115 waiver process. Third, we only had access to hospital‐level data through the UHC CDB/RM, rather than individual patient data. We attempted to mitigate this limitation by weighting data according to the number of cases per hospital. Lastly, additional patient‐level factors that may influence LOS or mortality may not be included in the UHC risk adjustment models.

In summary, the differential shift in payer mix between Medicaid‐expansion and nonexpansion states did not influence overall LOS or in‐hospital mortality for general medicine patients at AMCs in the United States. Additional research could help to determine the impact of ACA implementation on other patient outcomes that may be dependent on insurance status, such as readmissions or hospital‐acquired complications.

Disclosures: M.E.A. conceived of the study concept and design, assisted with data acquisition, and drafted the manuscript. J.J.G. assisted with study design and made critical revisions to the manuscript. D.A. assisted with study design and made critical revisions to the manuscript. R.P. assisted with study design and made critical revisions to the manuscript. M.L. assisted with study design and data acquisition and made critical revisions to the manuscript. C.D.J. assisted with study design, performed data analyses, and made critical revisions to the manuscript. A modified abstract was presented in poster format at the University HealthSystem Consortium Annual Conference held September 30 to October 2, 2015 in Orlando, Florida, as well as at the Society of Hospital Medicine Research, Innovations, and Vignettes 2016 Annual Meeting held March 69, 2016, in San Diego, California. The authors report no conflicts of interest.

Maternal postpartum depression occurs in 5% to 25% of all mothers, and up to 40% to 60% in high‐risk populations such as low‐income women.[1, 2, 3, 4] Children of affected mothers suffer negative health consequences such as decreased physical growth, poor maternalchild bond, problem behavior, and child abuse.[5, 6, 7] Timely recognition of symptoms and treatment may improve child outcomes.[8] Published guidelines recommend pediatricians screen for postpartum depression at infant 1‐, 2‐, 4‐, and 6‐month outpatient visits.[9] There are no current guidelines for or studies of screening in general inpatient settings, although emergency rooms[10] and neonatal intensive care units (NICUs)[11] have been examined. Pediatric hospitalization may offer an additional opportunity for expanding screening and intervention.

Augmenting outpatient screening practices with additional inpatient screening would have several benefits. Infant health problems have been associated with postpartum depression, and therefore mothers in the hospital may be at higher risk.[12] Inpatient screening would also improve access to mothers not screened as outpatients. Missed screening could occur due to physician discomfort with screening, time constraints during busy office visits, or noncompliance with recommended visit schedules.[13, 14, 15, 16] Finally, inpatient providers would benefit from understanding the psychosocial milieu of children now under their care. Recent studies note hospital discharges may be improved and readmissions reduced by assessing socioeconomic risk factors during hospitalization.[17] The evidence‐based Peds Effective Discharge: Better Handoff to Home through Safer Transitions Better Outcomes by Optimizing Safe Transitions (Pedi‐BOOST) toolkit specifically recommends an assessment of parental psychiatric issues.[18] Postpartum depression strongly correlates with impaired maternalchild bonding,[19] which in turn negatively affects mothers' engagement with healthcare providers.[20] This could impact patient education and recommendations provided during hospitalization.

Therefore, we sought to perform postpartum depression screening during infant hospitalizations. Our primary goal was to determine rate of postpartum depression in our population and proportion of women previously unscreened who could be captured by inpatient screening. We additionally aimed to determine the proportion of women with poor maternalinfant bond. Our next goal was to identify maternal or infant factors associated with positive postpartum depression screening. Finally, we performed follow‐up calls to determine if in‐hospital interventions resulted in formal postpartum depression diagnosis, use of recommended referrals, improved maternalchild bond, and decreased symptoms of depression over time.

METHODS

RESULTS

Out of 366 motherinfant pairs, 56 (15%) refused, and 310 (85%) mothers were fully enrolled (Figure 1A). Mothers had an average age of 28.17 years, were 68.3% Hispanic/Latina by self‐report, and 45.2% were married. Infants were an average of 4.24 months old, 81.9% were born term (>37 weeks), and 64.8% were previously healthy (Table 1).

Figure 1

Eighty‐seven (28%) mothers were EPDS+; 223 (72%) were EPDS. Only 42 mothers reported previous postpartum depression screening since the birth of their most recent child. However, 30 infants were <1 month in age, thus outside recommended screening range. Eliminating these infants revealed a 14.6% rate of appropriate prior screening. Higher EPDS scores were associated with higher (worse) MIB scores by linear regression ( = 0.11, P < 0.001). The vast majority (77%) of mothers scored a 0 or 1 on the MIB scale, indicating good bonding; further statistical comparison using the MIB scale as a secondary outcome was therefore inappropriate.

On bivariate logistic regression, Hispanic/Latina women were less likely to be EPDS+ (odds ratio [OR]: 0.43; 95% CI: 0.23‐0.84) compared to white/Caucasian women. Mothers who identified Spanish as their primary language and took the Spanish EPDS had lower odds of a positive screen (OR: 0.47; 95% CI: 0.25‐0.88). The racial differences did not persist on multivariate analysis (OR: 0.64; 95% CI: 0.30‐1.38) (Table 2). Maternal characteristics identified as potential risk factors for positive screens were poor social support (OR: 3.58; 95% CI: 1.95‐6.59) and history of a prior psychiatric diagnosis (OR: 5.07; 95% CI: 2.65‐9.72). There were no differences in age, number of children or people living in the home, relationship status, or breastfeeding rates by EPDS score.

Infant characteristics were next examined. Children of EPDS+ and EPDS mothers were similar in age, number of prior hospital admissions, gestational age at birth, and overall use of medical equipment (Table 2). To examine the effect of illness leading to hospitalization on EPDS+ risk, discharge diagnoses were collected and grouped into categories. Infants of EPDS+ mothers were more likely hospitalized for neurologic illness (P = 0.008) (see Supporting Table 1 in the online version of this article), but otherwise similar.

We next compared differences in long‐term infant comorbidities. The rate of having any comorbidity was similar between children of EPDS+ and EPDS mothers (39.1% vs 33.6%; P = 0.551). However, children of EPDS+ mothers were more likely to have mental retardation, hydrocephalus, or require ventriculoperitoneal shunt (VPS); however, the overall number of infants with each comorbidity was low. A neurodevelopmental comorbidity variable was created combining mental retardation, cerebral palsy, epilepsy, hydrocephalus, craniosynostosis, and VPS, resulting in 22 (7.1%) unique infants with 1 or more of these conditions. Having an infant with a neurodevelopmental comorbidity was a risk factor for positive postpartum depression screen (OR: 3.41; 95% CI: 1.41‐8.21). This continued to be significant (OR: 2.78; 95% CI: 1.03‐7.52) (Table 2) when controlling for maternal race/ethnicity, psychiatric history, and social support in multivariate logistic regression.

To determine if women screened followed through with recommendations, participants were called 3 and 6 months postenrollment. We attempted to call all women and successfully reached 120; 19 (16%) refused the call. One hundred one of the original 310 enrolled (33%) completed at least 1 follow‐up call; 47 at 3 months, 40 at 6 months, and only 14 (14%) responded at both time points. Due to this response rate, the first call at either 3 or 6 months was used as a single follow‐up time point for statistical analysis. A slightly higher proportion of EPDS‐ mothers (80/223, 36%) completed calls compared to EPDS+ mothers (21/87, 24%; P = 0.047).

Of 21 mothers initially EPDS+ who completed a follow‐up call, 10 (48%) later screened negative. Seven of these 10 (70%) reported discussing postpartum depression with their physician or using provided referral resources in the interim; 1 woman both spoke to a doctor and used a referral resource. One additional woman used resources, but repeat EPDS was still positive (Table 3). Reasons cited for not seeking evaluation included too busy (n = 4) and lost paperwork (n = 1), or no reason was given (n = 2). Mothers utilizing appropriate follow‐up had reduction in scores compared to those not (F(1,19) = 5.743, P = 0.027), although all scores decreased over time (F(1,19) = 11.54, P = 0.0030) (Figure 1B).

Of 80 women initially EPDS, most stayed negative (73/80, 91%), but 7 (9%) became EPDS+. These mothers received education and referral information over the phone, but none completed a subsequent call. Infants of mothers initially EPDS had a higher frequency of hospitalization postenrollment compared to EPDS+ mothers (P = 0.021) (Table 3). Two (33%) mothers who converted from EPDS to EPDS+ had infants readmitted in the follow‐up period.

DISCUSSION

This study demonstrated almost a third of mothers of hospitalized infants are at risk for postpartum depression and most had not been previously screened. Stress due to hospitalization did not seem to falsely elevate EPDS scores; the proportion of EPDS+ mothers matched our prestudy prediction (28% vs 27.9%). Follow‐up calls indicated that EPDS+ mothers not pursuing further evaluation tended to remain EPDS+. Higher (worse) MIB score was strongly correlated to increased EPDS score as expected, supporting screening accuracy. Our results suggest that postpartum depression screening in hospital settings can be used to complement outpatient practice and capture mothers who would otherwise be missed.

Although we were able to screen, it is difficult to know whether this correctly identified mothers with postpartum depression. Only 2 mothers reported subsequent official diagnosis of postpartum depression, and 1 of these was EPDS originally. This reflects weakness of our survey‐based design; we only know if the mother self‐reported a formal diagnosis of postpartum depression, because we do not have access to their medical charts. We also had higher than expected loss to follow up (67%), leaving 66 initially EPDS+ mothers with unknown eventual diagnoses. The EPDS has been validated in multiple populations and has a positive predictive value ranging from 23% to 93%.[23] Therefore, somewhere between 20 and 80 women in our study should meet diagnostic criteria for postpartum depression. A limitation of children's hospital‐based screening with the EPDS is lack of adult‐trained psychiatrists who could immediately follow screening with diagnosis. Such integration may already be possible at community or hospital‐within‐a‐hospital models, and could be trialed at children's hospitals. Regardless, participation in the study seemed to increase mothers' awareness of postpartum depression. Prior to enrollment, only 14.6% of subjects reported discussing postpartum depression with a physician, although recall bias likely contributed to some mothers not remembering a screen. Promisingly, on follow‐up, 37% of called participants reported they discussed postpartum depression with a doctor following their child's hospital discharge.

Our study identified low social support and history of past psychiatric diagnosis as maternal risk factors for EPDS+ screens, which is consistent with previous reports.[29] There was a slight increase in subsequent infant hospitalizations in the EPDS group, which is contrary to reports stating that increased healthcare utilization is associated with postpartum depression.[30] However, most studies have shown an increase in only acute or emergency room care visits[30, 31] and no association between maternal depression and infant hospitalization.[30, 32] In our study, the median number of hospitalizations for both groups was 0, indicating overall low utilization. Because 2 of the mothers who converted from EPDS to EPDS+ had children readmitted, this underscores the benefit of reassessment at each medical encounter. A large proportion of mothers (36.5%) reported that the infant had been previously hospitalized, adding another potential missed screening opportunity. Our study supports others advocating repeated screenings and suggests mothers should be screened at any medical encounter that occurs in the first postpartum year.

We identified neurodevelopmental illness as the major infant characteristic associated with postpartum depression risk. Conversely, Garfield et al. did not find correlation between poorer Neurobiologic Risk Score and increased maternal depression risk in a NICU setting.[11] Perhaps our population of older and mainly full‐term infants makes consequences of neurologic insult more obvious and affects mothers more significantly. Cheng et al. reported that 26.9% of mothers of children with cognitive delay reported high depressive symptoms, compared with 17.4% of mothers of typically developing children at 4 years of age.[33] Another body of evidence suggests maternal emotional state during pregnancy influences neurodevelopmental outcome in the child. Maternal anxiety or depression has been associated with altered placental function, reduced infant gray matter density, and worse cognitive function.[34, 35] Therefore, future research may focus on mothers of infants with neurodevelopmental disease to better understand this relationship.

There were several limitations to this study. Some data collected by a survey are subject to information bias. Women may report a more supportive social network than actually exists or omit history of mental health diagnoses. We attempted to control for this by using validated measures where possible and performing chart review to verify reported infant characteristics. Our population was overwhelmingly Hispanic/Latina, and a third of infants were not previously healthy, which limits applicability to other settings. We used a convenience method that could introduce sampling bias. Our hospital's overall patient demographic is 65% Hispanic, which is similar to the 68% sampled in our study. In addition, the proportions of infant diagnoses approximate the overall rates at CHLA, so we feel our sample was fairly representative. There is a general consensus that depression studies have recruitment difficulties.[36] In the unlikely event that all 56 of women who declined to participate were EPDS+, overall proportion of at‐risk mothers would rise to 39%. If our study does show slight underestimation of risk, that would only mean more potential for intervention if screening were mandatory. Another weakness was high loss to follow‐up, which led us to combine the 3‐ and 6‐month follow‐up calls into 1 outcome. Sixty percent of calls used in analysis occurred at 3 months, so long‐term maintenance of improved EPDS scores remains unclear. Although conducting repeat EPDS via phone may affect honest answering of sensitive questions, other studies have used this technique successfully.[4]

CONCLUSION

This is the first study evaluating a screening program for maternal postpartum depression during infant hospitalizations. In our population, risk factors for positive postpartum depression screening were low social support, history of maternal psychiatric diagnosis, and having an infant with neurodevelopmental disease. We believe mothers should receive postpartum depression screening at all medical encounters during the child's first year.

Disclosures: Dr. Trost is an Institutional Career Development Program Scholar through the Southern California Clinical and Translational Science Institute (SC‐CTSI) at the University of Southern California Keck School of Medicine. The content is solely the responsibility of the author(s) and does not represent the official view of the SC‐CTSI. Dr. Trost conceptualized and designed the study, drafted the initial manuscript, and approved the final manuscript as submitted. Dr. Molas‐Torreblanca co‐designed the study, reviewed and revised the manuscript, and approved the final manuscript as submitted. Ms. Man coordinated and supervised hospital data collection, critically reviewed the manuscript, and approved the final manuscript as submitted. Mr. Casillas coordinated and supervised the phone call data collection, critically reviewed the manuscript, and approved the final manuscript as submitted. Ms. Sapir coordinated the referral process for enrolled patients, supervised the design of patient handouts, and critically reviewed and approved the final manuscript as submitted. Dr. Schrager guided study design, supervised the statistical analysis of the final data, critically reviewed and revised the manuscript, and approved the final manuscript as submitted. The authors report no conflicts of interest.

Maternal postpartum depression occurs in 5% to 25% of all mothers, and up to 40% to 60% in high‐risk populations such as low‐income women.[1, 2, 3, 4] Children of affected mothers suffer negative health consequences such as decreased physical growth, poor maternalchild bond, problem behavior, and child abuse.[5, 6, 7] Timely recognition of symptoms and treatment may improve child outcomes.[8] Published guidelines recommend pediatricians screen for postpartum depression at infant 1‐, 2‐, 4‐, and 6‐month outpatient visits.[9] There are no current guidelines for or studies of screening in general inpatient settings, although emergency rooms[10] and neonatal intensive care units (NICUs)[11] have been examined. Pediatric hospitalization may offer an additional opportunity for expanding screening and intervention.

Augmenting outpatient screening practices with additional inpatient screening would have several benefits. Infant health problems have been associated with postpartum depression, and therefore mothers in the hospital may be at higher risk.[12] Inpatient screening would also improve access to mothers not screened as outpatients. Missed screening could occur due to physician discomfort with screening, time constraints during busy office visits, or noncompliance with recommended visit schedules.[13, 14, 15, 16] Finally, inpatient providers would benefit from understanding the psychosocial milieu of children now under their care. Recent studies note hospital discharges may be improved and readmissions reduced by assessing socioeconomic risk factors during hospitalization.[17] The evidence‐based Peds Effective Discharge: Better Handoff to Home through Safer Transitions Better Outcomes by Optimizing Safe Transitions (Pedi‐BOOST) toolkit specifically recommends an assessment of parental psychiatric issues.[18] Postpartum depression strongly correlates with impaired maternalchild bonding,[19] which in turn negatively affects mothers' engagement with healthcare providers.[20] This could impact patient education and recommendations provided during hospitalization.

Therefore, we sought to perform postpartum depression screening during infant hospitalizations. Our primary goal was to determine rate of postpartum depression in our population and proportion of women previously unscreened who could be captured by inpatient screening. We additionally aimed to determine the proportion of women with poor maternalinfant bond. Our next goal was to identify maternal or infant factors associated with positive postpartum depression screening. Finally, we performed follow‐up calls to determine if in‐hospital interventions resulted in formal postpartum depression diagnosis, use of recommended referrals, improved maternalchild bond, and decreased symptoms of depression over time.

METHODS

RESULTS

Out of 366 motherinfant pairs, 56 (15%) refused, and 310 (85%) mothers were fully enrolled (Figure 1A). Mothers had an average age of 28.17 years, were 68.3% Hispanic/Latina by self‐report, and 45.2% were married. Infants were an average of 4.24 months old, 81.9% were born term (>37 weeks), and 64.8% were previously healthy (Table 1).

Figure 1

Eighty‐seven (28%) mothers were EPDS+; 223 (72%) were EPDS. Only 42 mothers reported previous postpartum depression screening since the birth of their most recent child. However, 30 infants were <1 month in age, thus outside recommended screening range. Eliminating these infants revealed a 14.6% rate of appropriate prior screening. Higher EPDS scores were associated with higher (worse) MIB scores by linear regression ( = 0.11, P < 0.001). The vast majority (77%) of mothers scored a 0 or 1 on the MIB scale, indicating good bonding; further statistical comparison using the MIB scale as a secondary outcome was therefore inappropriate.

On bivariate logistic regression, Hispanic/Latina women were less likely to be EPDS+ (odds ratio [OR]: 0.43; 95% CI: 0.23‐0.84) compared to white/Caucasian women. Mothers who identified Spanish as their primary language and took the Spanish EPDS had lower odds of a positive screen (OR: 0.47; 95% CI: 0.25‐0.88). The racial differences did not persist on multivariate analysis (OR: 0.64; 95% CI: 0.30‐1.38) (Table 2). Maternal characteristics identified as potential risk factors for positive screens were poor social support (OR: 3.58; 95% CI: 1.95‐6.59) and history of a prior psychiatric diagnosis (OR: 5.07; 95% CI: 2.65‐9.72). There were no differences in age, number of children or people living in the home, relationship status, or breastfeeding rates by EPDS score.

Infant characteristics were next examined. Children of EPDS+ and EPDS mothers were similar in age, number of prior hospital admissions, gestational age at birth, and overall use of medical equipment (Table 2). To examine the effect of illness leading to hospitalization on EPDS+ risk, discharge diagnoses were collected and grouped into categories. Infants of EPDS+ mothers were more likely hospitalized for neurologic illness (P = 0.008) (see Supporting Table 1 in the online version of this article), but otherwise similar.

We next compared differences in long‐term infant comorbidities. The rate of having any comorbidity was similar between children of EPDS+ and EPDS mothers (39.1% vs 33.6%; P = 0.551). However, children of EPDS+ mothers were more likely to have mental retardation, hydrocephalus, or require ventriculoperitoneal shunt (VPS); however, the overall number of infants with each comorbidity was low. A neurodevelopmental comorbidity variable was created combining mental retardation, cerebral palsy, epilepsy, hydrocephalus, craniosynostosis, and VPS, resulting in 22 (7.1%) unique infants with 1 or more of these conditions. Having an infant with a neurodevelopmental comorbidity was a risk factor for positive postpartum depression screen (OR: 3.41; 95% CI: 1.41‐8.21). This continued to be significant (OR: 2.78; 95% CI: 1.03‐7.52) (Table 2) when controlling for maternal race/ethnicity, psychiatric history, and social support in multivariate logistic regression.

To determine if women screened followed through with recommendations, participants were called 3 and 6 months postenrollment. We attempted to call all women and successfully reached 120; 19 (16%) refused the call. One hundred one of the original 310 enrolled (33%) completed at least 1 follow‐up call; 47 at 3 months, 40 at 6 months, and only 14 (14%) responded at both time points. Due to this response rate, the first call at either 3 or 6 months was used as a single follow‐up time point for statistical analysis. A slightly higher proportion of EPDS‐ mothers (80/223, 36%) completed calls compared to EPDS+ mothers (21/87, 24%; P = 0.047).

Of 21 mothers initially EPDS+ who completed a follow‐up call, 10 (48%) later screened negative. Seven of these 10 (70%) reported discussing postpartum depression with their physician or using provided referral resources in the interim; 1 woman both spoke to a doctor and used a referral resource. One additional woman used resources, but repeat EPDS was still positive (Table 3). Reasons cited for not seeking evaluation included too busy (n = 4) and lost paperwork (n = 1), or no reason was given (n = 2). Mothers utilizing appropriate follow‐up had reduction in scores compared to those not (F(1,19) = 5.743, P = 0.027), although all scores decreased over time (F(1,19) = 11.54, P = 0.0030) (Figure 1B).

Of 80 women initially EPDS, most stayed negative (73/80, 91%), but 7 (9%) became EPDS+. These mothers received education and referral information over the phone, but none completed a subsequent call. Infants of mothers initially EPDS had a higher frequency of hospitalization postenrollment compared to EPDS+ mothers (P = 0.021) (Table 3). Two (33%) mothers who converted from EPDS to EPDS+ had infants readmitted in the follow‐up period.

DISCUSSION

This study demonstrated almost a third of mothers of hospitalized infants are at risk for postpartum depression and most had not been previously screened. Stress due to hospitalization did not seem to falsely elevate EPDS scores; the proportion of EPDS+ mothers matched our prestudy prediction (28% vs 27.9%). Follow‐up calls indicated that EPDS+ mothers not pursuing further evaluation tended to remain EPDS+. Higher (worse) MIB score was strongly correlated to increased EPDS score as expected, supporting screening accuracy. Our results suggest that postpartum depression screening in hospital settings can be used to complement outpatient practice and capture mothers who would otherwise be missed.

Although we were able to screen, it is difficult to know whether this correctly identified mothers with postpartum depression. Only 2 mothers reported subsequent official diagnosis of postpartum depression, and 1 of these was EPDS originally. This reflects weakness of our survey‐based design; we only know if the mother self‐reported a formal diagnosis of postpartum depression, because we do not have access to their medical charts. We also had higher than expected loss to follow up (67%), leaving 66 initially EPDS+ mothers with unknown eventual diagnoses. The EPDS has been validated in multiple populations and has a positive predictive value ranging from 23% to 93%.[23] Therefore, somewhere between 20 and 80 women in our study should meet diagnostic criteria for postpartum depression. A limitation of children's hospital‐based screening with the EPDS is lack of adult‐trained psychiatrists who could immediately follow screening with diagnosis. Such integration may already be possible at community or hospital‐within‐a‐hospital models, and could be trialed at children's hospitals. Regardless, participation in the study seemed to increase mothers' awareness of postpartum depression. Prior to enrollment, only 14.6% of subjects reported discussing postpartum depression with a physician, although recall bias likely contributed to some mothers not remembering a screen. Promisingly, on follow‐up, 37% of called participants reported they discussed postpartum depression with a doctor following their child's hospital discharge.

Our study identified low social support and history of past psychiatric diagnosis as maternal risk factors for EPDS+ screens, which is consistent with previous reports.[29] There was a slight increase in subsequent infant hospitalizations in the EPDS group, which is contrary to reports stating that increased healthcare utilization is associated with postpartum depression.[30] However, most studies have shown an increase in only acute or emergency room care visits[30, 31] and no association between maternal depression and infant hospitalization.[30, 32] In our study, the median number of hospitalizations for both groups was 0, indicating overall low utilization. Because 2 of the mothers who converted from EPDS to EPDS+ had children readmitted, this underscores the benefit of reassessment at each medical encounter. A large proportion of mothers (36.5%) reported that the infant had been previously hospitalized, adding another potential missed screening opportunity. Our study supports others advocating repeated screenings and suggests mothers should be screened at any medical encounter that occurs in the first postpartum year.

We identified neurodevelopmental illness as the major infant characteristic associated with postpartum depression risk. Conversely, Garfield et al. did not find correlation between poorer Neurobiologic Risk Score and increased maternal depression risk in a NICU setting.[11] Perhaps our population of older and mainly full‐term infants makes consequences of neurologic insult more obvious and affects mothers more significantly. Cheng et al. reported that 26.9% of mothers of children with cognitive delay reported high depressive symptoms, compared with 17.4% of mothers of typically developing children at 4 years of age.[33] Another body of evidence suggests maternal emotional state during pregnancy influences neurodevelopmental outcome in the child. Maternal anxiety or depression has been associated with altered placental function, reduced infant gray matter density, and worse cognitive function.[34, 35] Therefore, future research may focus on mothers of infants with neurodevelopmental disease to better understand this relationship.

There were several limitations to this study. Some data collected by a survey are subject to information bias. Women may report a more supportive social network than actually exists or omit history of mental health diagnoses. We attempted to control for this by using validated measures where possible and performing chart review to verify reported infant characteristics. Our population was overwhelmingly Hispanic/Latina, and a third of infants were not previously healthy, which limits applicability to other settings. We used a convenience method that could introduce sampling bias. Our hospital's overall patient demographic is 65% Hispanic, which is similar to the 68% sampled in our study. In addition, the proportions of infant diagnoses approximate the overall rates at CHLA, so we feel our sample was fairly representative. There is a general consensus that depression studies have recruitment difficulties.[36] In the unlikely event that all 56 of women who declined to participate were EPDS+, overall proportion of at‐risk mothers would rise to 39%. If our study does show slight underestimation of risk, that would only mean more potential for intervention if screening were mandatory. Another weakness was high loss to follow‐up, which led us to combine the 3‐ and 6‐month follow‐up calls into 1 outcome. Sixty percent of calls used in analysis occurred at 3 months, so long‐term maintenance of improved EPDS scores remains unclear. Although conducting repeat EPDS via phone may affect honest answering of sensitive questions, other studies have used this technique successfully.[4]

CONCLUSION

This is the first study evaluating a screening program for maternal postpartum depression during infant hospitalizations. In our population, risk factors for positive postpartum depression screening were low social support, history of maternal psychiatric diagnosis, and having an infant with neurodevelopmental disease. We believe mothers should receive postpartum depression screening at all medical encounters during the child's first year.

Disclosures: Dr. Trost is an Institutional Career Development Program Scholar through the Southern California Clinical and Translational Science Institute (SC‐CTSI) at the University of Southern California Keck School of Medicine. The content is solely the responsibility of the author(s) and does not represent the official view of the SC‐CTSI. Dr. Trost conceptualized and designed the study, drafted the initial manuscript, and approved the final manuscript as submitted. Dr. Molas‐Torreblanca co‐designed the study, reviewed and revised the manuscript, and approved the final manuscript as submitted. Ms. Man coordinated and supervised hospital data collection, critically reviewed the manuscript, and approved the final manuscript as submitted. Mr. Casillas coordinated and supervised the phone call data collection, critically reviewed the manuscript, and approved the final manuscript as submitted. Ms. Sapir coordinated the referral process for enrolled patients, supervised the design of patient handouts, and critically reviewed and approved the final manuscript as submitted. Dr. Schrager guided study design, supervised the statistical analysis of the final data, critically reviewed and revised the manuscript, and approved the final manuscript as submitted. The authors report no conflicts of interest.

Maternal postpartum depression occurs in 5% to 25% of all mothers, and up to 40% to 60% in high‐risk populations such as low‐income women.[1, 2, 3, 4] Children of affected mothers suffer negative health consequences such as decreased physical growth, poor maternalchild bond, problem behavior, and child abuse.[5, 6, 7] Timely recognition of symptoms and treatment may improve child outcomes.[8] Published guidelines recommend pediatricians screen for postpartum depression at infant 1‐, 2‐, 4‐, and 6‐month outpatient visits.[9] There are no current guidelines for or studies of screening in general inpatient settings, although emergency rooms[10] and neonatal intensive care units (NICUs)[11] have been examined. Pediatric hospitalization may offer an additional opportunity for expanding screening and intervention.

Augmenting outpatient screening practices with additional inpatient screening would have several benefits. Infant health problems have been associated with postpartum depression, and therefore mothers in the hospital may be at higher risk.[12] Inpatient screening would also improve access to mothers not screened as outpatients. Missed screening could occur due to physician discomfort with screening, time constraints during busy office visits, or noncompliance with recommended visit schedules.[13, 14, 15, 16] Finally, inpatient providers would benefit from understanding the psychosocial milieu of children now under their care. Recent studies note hospital discharges may be improved and readmissions reduced by assessing socioeconomic risk factors during hospitalization.[17] The evidence‐based Peds Effective Discharge: Better Handoff to Home through Safer Transitions Better Outcomes by Optimizing Safe Transitions (Pedi‐BOOST) toolkit specifically recommends an assessment of parental psychiatric issues.[18] Postpartum depression strongly correlates with impaired maternalchild bonding,[19] which in turn negatively affects mothers' engagement with healthcare providers.[20] This could impact patient education and recommendations provided during hospitalization.

Therefore, we sought to perform postpartum depression screening during infant hospitalizations. Our primary goal was to determine rate of postpartum depression in our population and proportion of women previously unscreened who could be captured by inpatient screening. We additionally aimed to determine the proportion of women with poor maternalinfant bond. Our next goal was to identify maternal or infant factors associated with positive postpartum depression screening. Finally, we performed follow‐up calls to determine if in‐hospital interventions resulted in formal postpartum depression diagnosis, use of recommended referrals, improved maternalchild bond, and decreased symptoms of depression over time.

METHODS

RESULTS

Out of 366 motherinfant pairs, 56 (15%) refused, and 310 (85%) mothers were fully enrolled (Figure 1A). Mothers had an average age of 28.17 years, were 68.3% Hispanic/Latina by self‐report, and 45.2% were married. Infants were an average of 4.24 months old, 81.9% were born term (>37 weeks), and 64.8% were previously healthy (Table 1).

Figure 1

Eighty‐seven (28%) mothers were EPDS+; 223 (72%) were EPDS. Only 42 mothers reported previous postpartum depression screening since the birth of their most recent child. However, 30 infants were <1 month in age, thus outside recommended screening range. Eliminating these infants revealed a 14.6% rate of appropriate prior screening. Higher EPDS scores were associated with higher (worse) MIB scores by linear regression ( = 0.11, P < 0.001). The vast majority (77%) of mothers scored a 0 or 1 on the MIB scale, indicating good bonding; further statistical comparison using the MIB scale as a secondary outcome was therefore inappropriate.

On bivariate logistic regression, Hispanic/Latina women were less likely to be EPDS+ (odds ratio [OR]: 0.43; 95% CI: 0.23‐0.84) compared to white/Caucasian women. Mothers who identified Spanish as their primary language and took the Spanish EPDS had lower odds of a positive screen (OR: 0.47; 95% CI: 0.25‐0.88). The racial differences did not persist on multivariate analysis (OR: 0.64; 95% CI: 0.30‐1.38) (Table 2). Maternal characteristics identified as potential risk factors for positive screens were poor social support (OR: 3.58; 95% CI: 1.95‐6.59) and history of a prior psychiatric diagnosis (OR: 5.07; 95% CI: 2.65‐9.72). There were no differences in age, number of children or people living in the home, relationship status, or breastfeeding rates by EPDS score.

Infant characteristics were next examined. Children of EPDS+ and EPDS mothers were similar in age, number of prior hospital admissions, gestational age at birth, and overall use of medical equipment (Table 2). To examine the effect of illness leading to hospitalization on EPDS+ risk, discharge diagnoses were collected and grouped into categories. Infants of EPDS+ mothers were more likely hospitalized for neurologic illness (P = 0.008) (see Supporting Table 1 in the online version of this article), but otherwise similar.

We next compared differences in long‐term infant comorbidities. The rate of having any comorbidity was similar between children of EPDS+ and EPDS mothers (39.1% vs 33.6%; P = 0.551). However, children of EPDS+ mothers were more likely to have mental retardation, hydrocephalus, or require ventriculoperitoneal shunt (VPS); however, the overall number of infants with each comorbidity was low. A neurodevelopmental comorbidity variable was created combining mental retardation, cerebral palsy, epilepsy, hydrocephalus, craniosynostosis, and VPS, resulting in 22 (7.1%) unique infants with 1 or more of these conditions. Having an infant with a neurodevelopmental comorbidity was a risk factor for positive postpartum depression screen (OR: 3.41; 95% CI: 1.41‐8.21). This continued to be significant (OR: 2.78; 95% CI: 1.03‐7.52) (Table 2) when controlling for maternal race/ethnicity, psychiatric history, and social support in multivariate logistic regression.

To determine if women screened followed through with recommendations, participants were called 3 and 6 months postenrollment. We attempted to call all women and successfully reached 120; 19 (16%) refused the call. One hundred one of the original 310 enrolled (33%) completed at least 1 follow‐up call; 47 at 3 months, 40 at 6 months, and only 14 (14%) responded at both time points. Due to this response rate, the first call at either 3 or 6 months was used as a single follow‐up time point for statistical analysis. A slightly higher proportion of EPDS‐ mothers (80/223, 36%) completed calls compared to EPDS+ mothers (21/87, 24%; P = 0.047).

Of 21 mothers initially EPDS+ who completed a follow‐up call, 10 (48%) later screened negative. Seven of these 10 (70%) reported discussing postpartum depression with their physician or using provided referral resources in the interim; 1 woman both spoke to a doctor and used a referral resource. One additional woman used resources, but repeat EPDS was still positive (Table 3). Reasons cited for not seeking evaluation included too busy (n = 4) and lost paperwork (n = 1), or no reason was given (n = 2). Mothers utilizing appropriate follow‐up had reduction in scores compared to those not (F(1,19) = 5.743, P = 0.027), although all scores decreased over time (F(1,19) = 11.54, P = 0.0030) (Figure 1B).

Of 80 women initially EPDS, most stayed negative (73/80, 91%), but 7 (9%) became EPDS+. These mothers received education and referral information over the phone, but none completed a subsequent call. Infants of mothers initially EPDS had a higher frequency of hospitalization postenrollment compared to EPDS+ mothers (P = 0.021) (Table 3). Two (33%) mothers who converted from EPDS to EPDS+ had infants readmitted in the follow‐up period.

DISCUSSION

This study demonstrated almost a third of mothers of hospitalized infants are at risk for postpartum depression and most had not been previously screened. Stress due to hospitalization did not seem to falsely elevate EPDS scores; the proportion of EPDS+ mothers matched our prestudy prediction (28% vs 27.9%). Follow‐up calls indicated that EPDS+ mothers not pursuing further evaluation tended to remain EPDS+. Higher (worse) MIB score was strongly correlated to increased EPDS score as expected, supporting screening accuracy. Our results suggest that postpartum depression screening in hospital settings can be used to complement outpatient practice and capture mothers who would otherwise be missed.

Although we were able to screen, it is difficult to know whether this correctly identified mothers with postpartum depression. Only 2 mothers reported subsequent official diagnosis of postpartum depression, and 1 of these was EPDS originally. This reflects weakness of our survey‐based design; we only know if the mother self‐reported a formal diagnosis of postpartum depression, because we do not have access to their medical charts. We also had higher than expected loss to follow up (67%), leaving 66 initially EPDS+ mothers with unknown eventual diagnoses. The EPDS has been validated in multiple populations and has a positive predictive value ranging from 23% to 93%.[23] Therefore, somewhere between 20 and 80 women in our study should meet diagnostic criteria for postpartum depression. A limitation of children's hospital‐based screening with the EPDS is lack of adult‐trained psychiatrists who could immediately follow screening with diagnosis. Such integration may already be possible at community or hospital‐within‐a‐hospital models, and could be trialed at children's hospitals. Regardless, participation in the study seemed to increase mothers' awareness of postpartum depression. Prior to enrollment, only 14.6% of subjects reported discussing postpartum depression with a physician, although recall bias likely contributed to some mothers not remembering a screen. Promisingly, on follow‐up, 37% of called participants reported they discussed postpartum depression with a doctor following their child's hospital discharge.

Our study identified low social support and history of past psychiatric diagnosis as maternal risk factors for EPDS+ screens, which is consistent with previous reports.[29] There was a slight increase in subsequent infant hospitalizations in the EPDS group, which is contrary to reports stating that increased healthcare utilization is associated with postpartum depression.[30] However, most studies have shown an increase in only acute or emergency room care visits[30, 31] and no association between maternal depression and infant hospitalization.[30, 32] In our study, the median number of hospitalizations for both groups was 0, indicating overall low utilization. Because 2 of the mothers who converted from EPDS to EPDS+ had children readmitted, this underscores the benefit of reassessment at each medical encounter. A large proportion of mothers (36.5%) reported that the infant had been previously hospitalized, adding another potential missed screening opportunity. Our study supports others advocating repeated screenings and suggests mothers should be screened at any medical encounter that occurs in the first postpartum year.

We identified neurodevelopmental illness as the major infant characteristic associated with postpartum depression risk. Conversely, Garfield et al. did not find correlation between poorer Neurobiologic Risk Score and increased maternal depression risk in a NICU setting.[11] Perhaps our population of older and mainly full‐term infants makes consequences of neurologic insult more obvious and affects mothers more significantly. Cheng et al. reported that 26.9% of mothers of children with cognitive delay reported high depressive symptoms, compared with 17.4% of mothers of typically developing children at 4 years of age.[33] Another body of evidence suggests maternal emotional state during pregnancy influences neurodevelopmental outcome in the child. Maternal anxiety or depression has been associated with altered placental function, reduced infant gray matter density, and worse cognitive function.[34, 35] Therefore, future research may focus on mothers of infants with neurodevelopmental disease to better understand this relationship.

There were several limitations to this study. Some data collected by a survey are subject to information bias. Women may report a more supportive social network than actually exists or omit history of mental health diagnoses. We attempted to control for this by using validated measures where possible and performing chart review to verify reported infant characteristics. Our population was overwhelmingly Hispanic/Latina, and a third of infants were not previously healthy, which limits applicability to other settings. We used a convenience method that could introduce sampling bias. Our hospital's overall patient demographic is 65% Hispanic, which is similar to the 68% sampled in our study. In addition, the proportions of infant diagnoses approximate the overall rates at CHLA, so we feel our sample was fairly representative. There is a general consensus that depression studies have recruitment difficulties.[36] In the unlikely event that all 56 of women who declined to participate were EPDS+, overall proportion of at‐risk mothers would rise to 39%. If our study does show slight underestimation of risk, that would only mean more potential for intervention if screening were mandatory. Another weakness was high loss to follow‐up, which led us to combine the 3‐ and 6‐month follow‐up calls into 1 outcome. Sixty percent of calls used in analysis occurred at 3 months, so long‐term maintenance of improved EPDS scores remains unclear. Although conducting repeat EPDS via phone may affect honest answering of sensitive questions, other studies have used this technique successfully.[4]

CONCLUSION

This is the first study evaluating a screening program for maternal postpartum depression during infant hospitalizations. In our population, risk factors for positive postpartum depression screening were low social support, history of maternal psychiatric diagnosis, and having an infant with neurodevelopmental disease. We believe mothers should receive postpartum depression screening at all medical encounters during the child's first year.

Disclosures: Dr. Trost is an Institutional Career Development Program Scholar through the Southern California Clinical and Translational Science Institute (SC‐CTSI) at the University of Southern California Keck School of Medicine. The content is solely the responsibility of the author(s) and does not represent the official view of the SC‐CTSI. Dr. Trost conceptualized and designed the study, drafted the initial manuscript, and approved the final manuscript as submitted. Dr. Molas‐Torreblanca co‐designed the study, reviewed and revised the manuscript, and approved the final manuscript as submitted. Ms. Man coordinated and supervised hospital data collection, critically reviewed the manuscript, and approved the final manuscript as submitted. Mr. Casillas coordinated and supervised the phone call data collection, critically reviewed the manuscript, and approved the final manuscript as submitted. Ms. Sapir coordinated the referral process for enrolled patients, supervised the design of patient handouts, and critically reviewed and approved the final manuscript as submitted. Dr. Schrager guided study design, supervised the statistical analysis of the final data, critically reviewed and revised the manuscript, and approved the final manuscript as submitted. The authors report no conflicts of interest.

Pneumonia is the leading inpatient infectious diagnosis for which antimicrobials are prescribed in the United States.[1] Supported by moderate‐ to high‐quality evidence, guidelines produced jointly by the Infectious Diseases Society of America (IDSA) and American Thoracic Society (ATS) recommend treating pneumonia with the shortest appropriate duration of antimicrobial therapy to minimize risk for antimicrobial‐related adverse events.[2, 3, 4]

Evidence supports short duration of therapy for treatment of uncomplicated pneumonia.[3, 4, 5, 6, 7, 8, 9, 10, 11, 12] IDSA/ATS guidelines state, patients with CAP [community‐acquired pneumonia] should be treated for a minimum of 5 days (level 1 evidence), should be afebrile for 4872 hours, and should have no more than 1 CAP‐associated sign of clinical instabilitybefore discontinuation of therapy (level II evidence). (Moderate recommendation.) A longer duration of therapy may be warranted if initial therapy was not active against the identified pathogen or if it was complicated by [abscess, empyema, severe immunosuppression, or] extra‐pulmonary infection such as meningitis or endocarditis. (Weak recommendation; level III evidence).[3] Recommended therapy duration for patients with uncomplicated healthcare‐associated pneumonia (HCAP) who respond to initial therapy is 7 to 8 days unless gram‐negative nonfermenting rods or complications are identified (level I evidence).[4]

Within the Veterans Health Administration (VHA), the Antimicrobial Stewardship Taskforce (ASTF) was created to optimize care by developing, deploying, and monitoring a national‐level strategic plan for antimicrobial therapy management improvements.[13, 14] Although single‐center studies have found antimicrobial therapy for CAP being frequently prescribed for longer than recommended, the reproducibility of this finding across different facilities has not been assessed.[15, 16] The ASTF collaborated with the VHA Center for Medication Safety to assess total duration of antimicrobial therapy prescribed for veterans hospitalized with uncomplicated pneumonia.[17]

METHODS

This retrospective multicenter evaluation was conducted in 30 VHA facilities that volunteered to participate in this project. Inpatients discharged with a primary International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) diagnosis code for pneumonia (or pneumonia diagnosis secondary to primary sepsis diagnosis) during 2013 were evaluated.[18] Diagnoses, admissions, and patient demographics were identified using Veterans Affairs (VA) integrated databases through the Austin Integrated Technology Center. Up to 200 admissions per facility were randomly selected for review. Clinical pharmacists at each facility performed manual record reviews utilizing a standardized protocol and collection form. Completed cases were uploaded to a central database for analysis. Standardized chart abstraction was facilitated by detailed instructions, a data dictionary, and monthly conference calls.

Inclusion criteria required patient admission to any medical ward including intensive care unit (ICU) wards for 48 hours, receipt of >24 hours inpatient antimicrobial therapy (eg, at least 2 doses of a once‐daily antibiotic), documentation of pneumonia discharge diagnosis, and survival until discharge. Exclusion criteria were: complicated pneumonia (lung abscess, necrotizing pneumonia, thoracentesis performed), significant immunosuppression (cancer chemotherapy or absolute neutrophil count <1500 cell/mm³ within 28 days, organ transplantation, human immunodeficiency virus infection); or extrapulmonary infection (eg, meningitis, endocarditis).[3] Patients were also excluded if directly transferred from another inpatient facility, pneumonia occurred >48 hours after admission, index hospitalization was >14 days, previously hospitalized within 28 days prior to index admission, or discharged without documentation of completing a full course of therapy. In addition, patients who received initial therapy discordant with culture and susceptibility findings, were not clinically stable by discharge, or had gram‐negative nonfermentative bacilli cultured were excluded from analysis because according to the guidelines, either data are lacking to support a short duration of therapy such as initial discordant therapy, or a longer duration of therapy may be warranted such as gram‐negative nonfermentative bacilli and clinical instability at discharge.[4] Our intent for these exclusions was to minimize bias against clinician decision making for cases where a longer duration of therapy may have been appropriate.

Patients meeting all criteria had the following abstracted: demographics; prior healthcare exposures, admitting location (ICU or non‐ICU ward), parameters for calculation of Pneumonia Severity Index (PSI), culture results obtained 48 hours of admission, duration of antimicrobials administered during hospitalization and prescribed upon discharge (or recommendations for outpatient duration in the discharge summary for patients receiving medications from non‐VA sources), daily clinical stability assessment, Clostridium difficile infection (CDI) test results, and readmission or death within 28 days of discharge.[19]

Guideline‐similar CAP therapy duration was defined as a minimum of 5 days of antimicrobials, up to a maximum of 3 additional days beginning the first day the patient was afebrile and exhibited 1 sign of clinical instability (heart rate > 100 beats/minute, respiratory rate >24 breaths/minute, systolic blood pressure <90, oxygen saturation <90% or partial pressure of oxygen <60 mm Hg on room air or baseline O₂ requirements, or not returned to baseline mental status).[3] This definition was made by consensus decision of the investigators and was necessary to operationalize the relationship between clinical stability and appropriate duration of therapy. Guideline‐similar HCAP therapy duration was defined as 8 days.[4] CDI was defined in accordance with VA criteria for hospital onset and community‐onset healthcare‐facilityassociated CDI.[20] All‐cause hospital readmission and all‐cause death were defined as inpatient readmission or any death, respectively, within 28 days after discharge for the pneumonia admission.

Demographics, comorbidities, microbiology results, antimicrobial utilization, CDI, readmission, and death rates between guideline‐similar and guideline‐excessive duration of antimicrobial therapy groups were characterized with descriptive statistics, Mann‐Whitney U test, or ² test as indicated (significance defined as P < 0.05). Multivariable logistic regression (SAS version 9.3 [SAS Institute, Cary, NC]) was used to assess association between duration of therapy exceeding recommended guidelines with all‐cause readmission and all‐cause death after adjustment for pertinent covariates. Odds ratios (OR) with 95% confidence intervals ( 95% CI) were reported. This medication utilization evaluation (MUE) was reviewed by the Hines VHA Institutional Review Board for Human Subjects Protection. Based on VHA Policy Handbook 1058.05, which defines operations activities that may constitute research, the board determined that the evaluation constituted quality improvement rather than research, and thus was exempt from VHA Human Subjects Research requirements.

RESULTS

There were 3881 admissions eligible for chart review. After manual chart review of inclusion and exclusion criteria, 1739 (44.8%) patients were available for duration of therapy analysis. (Figure 1). Only 1 admission for each patient was analyzed.

Figure 1

The cohort was comprised primarily of elderly male patients (96.6%) of whom more than two‐thirds were hospitalized for CAP (Table 1). Most patients had significant disease severity as indicated by PSI score; however, only 12% were directly admitted to the ICU. Blood cultures were collected in >95% of cases; lower respiratory cultures were obtained in 39.9% of cases.

Commonly administered antimicrobials during hospitalization and at discharge are summarized in Table 2. Anti‐pseudomonal ‐lactams and antimethicillin‐resistant Staphylococcus aureus antimicrobials were more frequently administered to patients with HCAP, whereas third‐generation cephalosporins and macrolides were more likely to be administered to patients with CAP. Fluoroquinolones were prescribed to 55.3% of patients upon discharge.

Overall, 13.9% of patients with uncomplicated pneumonia received guideline‐similar duration of therapy (Table 3). A greater proportion of HCAP patients (29.0%) received guideline‐similar therapy duration as compared to CAP patients (6.9%) (P < 0.01 (Table 3). Median duration of therapy was 7 days (interquartile range [IQR] = 78 days) for guideline‐similar therapy compared to 10 days (913 days) for therapy duration in excess of guideline recommendations. Overall, 97.1 % of patients met clinical stability criteria before day 4 of therapy, yet 50% received 4 days of intravenous (IV) therapy (median was 4 days, IQR = 36 days). Antimicrobial therapy was generally completed after discharge, as only 17.3% received their entire treatment course during hospitalization. Median duration of outpatient oral (PO) antimicrobial therapy was twice as long for guideline‐excessive therapy compared to guideline‐similar therapy (6 vs 3 days), whereas duration of inpatient IV and PO antimicrobial therapy was similar. Patients discharged on a fluoroquinolone were more likely to receive guideline‐similar duration of therapy. The VHA classifies facilities into 3 levels of complexity, with lower scores indicating more complex facilities.[21] Guideline‐similar therapy duration occurred in 10.4% of cases in lower complexity facilities (levels 2 and 3),and 15.1% in more complex facilities (level 1) (P = 0.01). The median duration of therapy was similar for more and less complex facilities, respectively (10 days, IQR = 812 days vs 10 days, IQR = 813 days).

The 28‐day postdischarge all‐cause readmission rate for patients who received guideline‐similar therapy duration was higher (17.4%) than for patients who received therapy duration in excess of guideline recommendations (12.2%) (P = 0.03). After adjustment for covariates associated with readmission (HCAP, age, prior skilled nursing facility residence, PSI score comorbidity elements), we found no evidence that patients who received guideline‐similar therapy duration were more likely to be readmitted than were patients who received guideline‐excessive duration (OR: 1.1 [95% CI: 0.8, 1.7]) (Table 3). Likewise, no difference in 28‐day all‐cause postdischarge mortality was identified between guideline‐similar and guideline‐excessive duration after adjustment for the same covariates (adjusted OR: 0.5 [95% CI: 0.2, 1.2]) (Table 4).

CDI cases (n = 15) were too sparse to adequately perform multivariable logistic regression analysis; however, a higher percentage of patients who received guideline‐similar duration of therapy developed CDI compared to patients who received guideline‐excessive duration of therapy (40.0% vs 13.6%, P < 0.01). The median duration of therapy for patients who did and did not develop CDI was similar (8 days, IQR = 714 days vs 10 days, IQR = 812 days, P = 0.85, respectively). Patients who developed CDI had a higher rate of HCAP diagnosis (1.5% vs 0.6%; P = 0.06), were more likely to have concomitant non‐pneumonia infection (40.0% vs 9.5%, P < 0.01), have chronic comorbidity (86.7% vs 59.1%, P = 0.03), and to have been admitted to the ICU (26.7% vs 12.1%, P = 0.09).

DISCUSSION

IDSA/ATS guidelines for pneumonia duration of therapy generally agree with other professional society guidelines including the British Thoracic Society and National Institute for Health and Care Excellence.[22, 23] In contrast to existing evidence and guideline recommendations, this multi‐centered evaluation identified prolonged durations of antimicrobial therapy prescribed in 93% and 71% of patients with uncomplicated CAP and HCAP (Table 3), respectively.[3, 4, 5, 6, 7, 8, 9, 10, 11, 12] Almost all (97.1%) uncomplicated CAP and HCAP patients met clinical stability criteria before day 4 of hospitalization, yet the median duration of IV therapy was 4 days. Because criteria for IV to PO conversion and the clinical stability definition utilized in this analysis were similar, many patients may have been eligible for PO therapy earlier, favorably impacting length of stay, cost, and adverse effects.[3, 12, 24, 25, 26] Although median days of inpatient PO therapy administered was 1 day (IQR = 03 days), inpatient observation after PO conversion may not be necessary. The duration of PO therapy was based on calendar days, where if a patient received 1 dose of a once daily antibiotic (ie, levofloxacin), they were considered to have received 1 day of inpatient PO antibiotics even if discharged the same day.

Approximately half of all days of therapy occurred after discharge. Although the median therapy duration for inpatients was similar, the median duration of antimicrobials administered upon hospital discharge was twice as long for patients receiving guideline‐excessive compared to guideline‐similar duration of therapy. The median excess in antibiotic duration is almost entirely accounted for by excess outpatient days of therapy. This is an important consideration for antimicrobial stewardship programs that tend to focus on inpatient antimicrobial use.

Noteworthy observations include the low rate of respiratory tract culture collection (41%) and frequent use of fluoroquinolones upon discharge. Collection of respiratory tract cultures is recommended for all patients with HCAP and patients with CAP who have risk factors for resistant pathogens, characteristics that were common in this cohort.[3, 4] Recently, we identified that respiratory culture collection is associated with increased de‐escalation rates in HCAP, and that culture‐negative patients frequently receive fluoroquinolones.[27] IDSA/ATS CAP guidelines discourage empirically switching to PO fluoroquinolone therapy based on bioavailability considerations alone.[3] Further, fluoroquinolones are considered to be associated with high risk of CDI.[28, 29] Prescription of fluoroquinolone upon discharge was associated with guideline‐similar duration of therapy and was not shown to be associated with CDI; however, power to detect differences between exposures to specific antimicrobials and CDI was low.

CDI was more common in patients with CAP (1.2% vs 0.5%) and HCAP (3.2% vs 0.8%) who received duration of therapy similar with guideline recommendations. This observation is confounded, as patients with CDI had significantly greater comorbidity as well as secondary infections and tended to more frequently receive ICU care. There were no differences in adjusted rates of readmission or death between patients receiving guideline‐similar and guideline‐excessive duration of therapy.

Evaluation strengths included exclusion of patients with complicating conditions possibly requiring prolonged antimicrobial treatment courses, which allowed the evaluation to focus on patients most likely to benefit from shorter course therapy. The definition of appropriate therapy duration was based upon daily assessment of clinical stability criteria that paralleled the CAP guidelines. The definition utilized objective parameters while accounting for patient variability in achieving clinical stability criteria. Finally, the analyses of clinical end points suggest that shorter duration of therapy may be as safe and effective as longer duration of therapy in uncomplicated pneumonia.

Limitations include those common to other analyses conducted within the VHA, including a predominantly elderly male cohort.[30] Only ICD‐9‐CM codes consistent with a discharge diagnosis of pneumonia were used to identify the cohort, and clinical impressions not documented in the medical record may have impacted the clinician's treatment duration decisions. The upper limit of appropriate duration of therapy for CAP was arbitrarily set at up to 3 days beyond meeting clinical stability criteria to provide a reasonable duration of appropriate therapy beyond clinical stability to operationalize the duration of therapy recommendations within the context of the IDSA/ATS guidelines. Additionally, CIs for the ORs of readmission and mortality were broad, and thus too imprecise to determine whether guideline‐similar durations increased or decreased readmission or mortality in comparison with therapy that exceeded guideline recommendations. We could not fully assess the potential for association between guideline‐excessive therapy duration and risk for CDI due to sparse cases. Finally, non‐VA prescription data were not available for all patients, and we relied on intended duration of therapy as recommended by the discharging provider in 4.1% of cases.

Most quality assessments of pneumonia treatment have focused on antimicrobial selection and timely administration or conversion from IV to PO therapy.[31, 32] This evaluation identified potential opportunities for expansion of antimicrobial stewardship activities during the transition of care setting. The efficacy of short‐course ‐lactam, macrolide, or fluoroquinolone therapy for CAP appears equivalent to longer treatment regimens with no difference in adverse event rates, suggesting that optimal duration of therapy may be a rational target for quality improvement.[5, 6, 7, 8, 9, 10, 11, 12, 15, 31] Recommendations for HCAP duration of therapy are extrapolated from a prospective multicentered study, which randomized patients with hospital‐acquired pneumonia to receive 8 versus 15 days of therapy, that identified similar outcomes to ours.[4, 12]

Single‐center studies have identified that antimicrobial therapy for pneumonia is frequently prescribed for longer than recommended by guidelines, which found a similar median duration of therapy as our evaluation.[15, 16] Similar to Jenkins et al., we observed a high rate of fluoroquinolone prescriptions upon discharge.[16]

There are few published examples of interventions designed to limit excessive duration of therapy, particularly for antimicrobials prescribed upon hospital discharge.[15, 33, 34] Serial procalcitonin measurements have been used to guide duration of therapy for pneumonia; however, the costbenefit ratio of procalcitonin measurement is unclear.[35, 36] Procalcitonin use was uncommon, and none of the participating facilities in our evaluation utilized a specific algorithm to guide therapy duration. Limited data suggest that patient‐level prospective audit with feedback may be effective. Advic et al. evaluated management of presumed CAP before and after education and prospective feedback to medical teams concerning antimicrobial selection and duration of therapy.[15] The intervention led to a decrease in median duration of therapy from 10 days (IQR = 813 days) to 7 days (IQR = 78 days) without increasing clinical failure or readmission rates. We recently reported a single‐center evaluation in which pharmacists utilizing a decision support tool while performing discharge medication reconciliation were able to reduce excessive mean duration of therapy from 9.5 days ( 2.4 days) to 8.3 days ( 2.9 days) in patients without complicated pneumonia, with a 19.2% reduction in duration of therapy prescribed at discharge.[37] A similar approach utilizing pharmacists performing discharge review has recently been reported in a community hospital.[38]

Future work should recognize that few patients complete their entire course of therapy as inpatients, and the majority of treatment is prescribed upon discharge. Pivotal time points for antimicrobial stewardship intervention include day 2 to 3 of hospitalization when conveying suggestions for antimicrobial de‐escalation and/or IV to PO conversion, and toward the end of hospitalization during discharge planning. Although it may not be feasible for antimicrobial stewards to review all uncomplicated cases of pneumonia during hospitalization, most facilities have a systematic process for reviewing medications during transitions of care. We believe that interventions intended to assess and recommend shortened courses of therapy are appropriate. These interventions should include a mechanism for support by stewardship personnel or other infectious diseases specialists. Based on our evaluation, the ASTF produced and disseminated clinical guidance documents and tools to triage pneumonia case severity and assess response to therapy. Qualified personnel are encouraged to use this information to make recommendations to providers regarding excessive duration of therapy for uncomplicated cases where appropriate. Other work should include an in‐depth assessment of clinical outcomes related to treatment duration, investigation of provider rationale for prolonged treatment, and duration of antimicrobial therapy prescribed upon discharge for other common disease states. Finally, manual chart review to classify uncomplicated cases and related outcomes was laborious, and automated case identification is technologically plausible and should be explored.[39]

In conclusion, this national VHA MUE found that patients with uncomplicated pneumonia were commonly prescribed antimicrobials for the duration of therapy in excess of guideline recommendations. Patients with uncomplicated pneumonia who received therapy duration consistent with guideline recommendations did not have significantly different all‐cause readmission and death rates compared to patients receiving prolonged treatment. Approximately half of all therapy was prescribed upon hospital discharge, and clinicians as well as antimicrobial stewardship programs should consider these findings to address excessive duration of antimicrobial therapy upon hospital discharge.

Disclosures: Karl Madaras‐Kelly is employed full time by Idaho State University and has a without compensation appointment as a clinical pharmacist at the Boise VA Medical Center. He receives grant support unrelated to this work through the Department of Veterans Affairs subcontracted to Idaho State University. Muriel Burk is employed full time through the Department of Veterans Affairs as clinical pharmacy specialist in outcomes and medication safety evaluation. Christina Caplinger was employed by the Department of Veterans Affairs as an infectious diseases fellow at the time this work was completed. She is currently employed by Micromedex. Jefferson Bohan is employed full time by the Department of Veterans Affairs as an infectious diseases fellow. Melinda Neuhauser is employed full time through the Department of Veterans Affairs as a clinical pharmacy specialistinfectious diseases. Matthew Goetz is employed full time through the Department of Veterans Affairs as an infectious diseases physician. Rhongping Zhang is employed full time through the Department of Veterans Affairs as a data analyst. Francesca Cunningham is employed full time through the Department of Veterans Affairs as the director of the VA Center for Medication Safety. This work was supported with resources and use of the Department of Veterans Affairs healthcare system. The views expressed in this article are solely those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs. The authors report no conflicts of interest.

Pneumonia is the leading inpatient infectious diagnosis for which antimicrobials are prescribed in the United States.[1] Supported by moderate‐ to high‐quality evidence, guidelines produced jointly by the Infectious Diseases Society of America (IDSA) and American Thoracic Society (ATS) recommend treating pneumonia with the shortest appropriate duration of antimicrobial therapy to minimize risk for antimicrobial‐related adverse events.[2, 3, 4]

Evidence supports short duration of therapy for treatment of uncomplicated pneumonia.[3, 4, 5, 6, 7, 8, 9, 10, 11, 12] IDSA/ATS guidelines state, patients with CAP [community‐acquired pneumonia] should be treated for a minimum of 5 days (level 1 evidence), should be afebrile for 4872 hours, and should have no more than 1 CAP‐associated sign of clinical instabilitybefore discontinuation of therapy (level II evidence). (Moderate recommendation.) A longer duration of therapy may be warranted if initial therapy was not active against the identified pathogen or if it was complicated by [abscess, empyema, severe immunosuppression, or] extra‐pulmonary infection such as meningitis or endocarditis. (Weak recommendation; level III evidence).[3] Recommended therapy duration for patients with uncomplicated healthcare‐associated pneumonia (HCAP) who respond to initial therapy is 7 to 8 days unless gram‐negative nonfermenting rods or complications are identified (level I evidence).[4]

Within the Veterans Health Administration (VHA), the Antimicrobial Stewardship Taskforce (ASTF) was created to optimize care by developing, deploying, and monitoring a national‐level strategic plan for antimicrobial therapy management improvements.[13, 14] Although single‐center studies have found antimicrobial therapy for CAP being frequently prescribed for longer than recommended, the reproducibility of this finding across different facilities has not been assessed.[15, 16] The ASTF collaborated with the VHA Center for Medication Safety to assess total duration of antimicrobial therapy prescribed for veterans hospitalized with uncomplicated pneumonia.[17]

METHODS

This retrospective multicenter evaluation was conducted in 30 VHA facilities that volunteered to participate in this project. Inpatients discharged with a primary International Classification of Diseases, Ninth Revision, Clinical Modification (ICD‐9‐CM) diagnosis code for pneumonia (or pneumonia diagnosis secondary to primary sepsis diagnosis) during 2013 were evaluated.[18] Diagnoses, admissions, and patient demographics were identified using Veterans Affairs (VA) integrated databases through the Austin Integrated Technology Center. Up to 200 admissions per facility were randomly selected for review. Clinical pharmacists at each facility performed manual record reviews utilizing a standardized protocol and collection form. Completed cases were uploaded to a central database for analysis. Standardized chart abstraction was facilitated by detailed instructions, a data dictionary, and monthly conference calls.

Inclusion criteria required patient admission to any medical ward including intensive care unit (ICU) wards for 48 hours, receipt of >24 hours inpatient antimicrobial therapy (eg, at least 2 doses of a once‐daily antibiotic), documentation of pneumonia discharge diagnosis, and survival until discharge. Exclusion criteria were: complicated pneumonia (lung abscess, necrotizing pneumonia, thoracentesis performed), significant immunosuppression (cancer chemotherapy or absolute neutrophil count <1500 cell/mm³ within 28 days, organ transplantation, human immunodeficiency virus infection); or extrapulmonary infection (eg, meningitis, endocarditis).[3] Patients were also excluded if directly transferred from another inpatient facility, pneumonia occurred >48 hours after admission, index hospitalization was >14 days, previously hospitalized within 28 days prior to index admission, or discharged without documentation of completing a full course of therapy. In addition, patients who received initial therapy discordant with culture and susceptibility findings, were not clinically stable by discharge, or had gram‐negative nonfermentative bacilli cultured were excluded from analysis because according to the guidelines, either data are lacking to support a short duration of therapy such as initial discordant therapy, or a longer duration of therapy may be warranted such as gram‐negative nonfermentative bacilli and clinical instability at discharge.[4] Our intent for these exclusions was to minimize bias against clinician decision making for cases where a longer duration of therapy may have been appropriate.

Patients meeting all criteria had the following abstracted: demographics; prior healthcare exposures, admitting location (ICU or non‐ICU ward), parameters for calculation of Pneumonia Severity Index (PSI), culture results obtained 48 hours of admission, duration of antimicrobials administered during hospitalization and prescribed upon discharge (or recommendations for outpatient duration in the discharge summary for patients receiving medications from non‐VA sources), daily clinical stability assessment, Clostridium difficile infection (CDI) test results, and readmission or death within 28 days of discharge.[19]

Guideline‐similar CAP therapy duration was defined as a minimum of 5 days of antimicrobials, up to a maximum of 3 additional days beginning the first day the patient was afebrile and exhibited 1 sign of clinical instability (heart rate > 100 beats/minute, respiratory rate >24 breaths/minute, systolic blood pressure <90, oxygen saturation <90% or partial pressure of oxygen <60 mm Hg on room air or baseline O₂ requirements, or not returned to baseline mental status).[3] This definition was made by consensus decision of the investigators and was necessary to operationalize the relationship between clinical stability and appropriate duration of therapy. Guideline‐similar HCAP therapy duration was defined as 8 days.[4] CDI was defined in accordance with VA criteria for hospital onset and community‐onset healthcare‐facilityassociated CDI.[20] All‐cause hospital readmission and all‐cause death were defined as inpatient readmission or any death, respectively, within 28 days after discharge for the pneumonia admission.

Demographics, comorbidities, microbiology results, antimicrobial utilization, CDI, readmission, and death rates between guideline‐similar and guideline‐excessive duration of antimicrobial therapy groups were characterized with descriptive statistics, Mann‐Whitney U test, or ² test as indicated (significance defined as P < 0.05). Multivariable logistic regression (SAS version 9.3 [SAS Institute, Cary, NC]) was used to assess association between duration of therapy exceeding recommended guidelines with all‐cause readmission and all‐cause death after adjustment for pertinent covariates. Odds ratios (OR) with 95% confidence intervals ( 95% CI) were reported. This medication utilization evaluation (MUE) was reviewed by the Hines VHA Institutional Review Board for Human Subjects Protection. Based on VHA Policy Handbook 1058.05, which defines operations activities that may constitute research, the board determined that the evaluation constituted quality improvement rather than research, and thus was exempt from VHA Human Subjects Research requirements.

RESULTS

There were 3881 admissions eligible for chart review. After manual chart review of inclusion and exclusion criteria, 1739 (44.8%) patients were available for duration of therapy analysis. (Figure 1). Only 1 admission for each patient was analyzed.

Figure 1

The cohort was comprised primarily of elderly male patients (96.6%) of whom more than two‐thirds were hospitalized for CAP (Table 1). Most patients had significant disease severity as indicated by PSI score; however, only 12% were directly admitted to the ICU. Blood cultures were collected in >95% of cases; lower respiratory cultures were obtained in 39.9% of cases.

Commonly administered antimicrobials during hospitalization and at discharge are summarized in Table 2. Anti‐pseudomonal ‐lactams and antimethicillin‐resistant Staphylococcus aureus antimicrobials were more frequently administered to patients with HCAP, whereas third‐generation cephalosporins and macrolides were more likely to be administered to patients with CAP. Fluoroquinolones were prescribed to 55.3% of patients upon discharge.

Overall, 13.9% of patients with uncomplicated pneumonia received guideline‐similar duration of therapy (Table 3). A greater proportion of HCAP patients (29.0%) received guideline‐similar therapy duration as compared to CAP patients (6.9%) (P < 0.01 (Table 3). Median duration of therapy was 7 days (interquartile range [IQR] = 78 days) for guideline‐similar therapy compared to 10 days (913 days) for therapy duration in excess of guideline recommendations. Overall, 97.1 % of patients met clinical stability criteria before day 4 of therapy, yet 50% received 4 days of intravenous (IV) therapy (median was 4 days, IQR = 36 days). Antimicrobial therapy was generally completed after discharge, as only 17.3% received their entire treatment course during hospitalization. Median duration of outpatient oral (PO) antimicrobial therapy was twice as long for guideline‐excessive therapy compared to guideline‐similar therapy (6 vs 3 days), whereas duration of inpatient IV and PO antimicrobial therapy was similar. Patients discharged on a fluoroquinolone were more likely to receive guideline‐similar duration of therapy. The VHA classifies facilities into 3 levels of complexity, with lower scores indicating more complex facilities.[21] Guideline‐similar therapy duration occurred in 10.4% of cases in lower complexity facilities (levels 2 and 3),and 15.1% in more complex facilities (level 1) (P = 0.01). The median duration of therapy was similar for more and less complex facilities, respectively (10 days, IQR = 812 days vs 10 days, IQR = 813 days).