Research
Original Investigation
Association Between Skilled Nursing Facility Quality Indicators and Hospital Readmissions Mark D. Neuman, MD, MSc; Christopher Wirtalla, BA; Rachel M. Werner, MD, PhD Editorial page 1517 IMPORTANCE Hospital readmissions are common, costly, and potentially preventable. Little is known about the association between available skilled nursing facility (SNF) performance measures and the risk of hospital readmission. OBJECTIVE To measure the association between SNF performance measures and hospital readmissions among Medicare beneficiaries receiving postacute care at SNFs in the United States.
Related article page 1531 CME Quiz at jamanetworkcme.com and CME Questions page 1585
DESIGN AND PARTICIPANTS Using national Medicare data on fee-for-service Medicare beneficiaries discharged to a SNF after an acute care hospitalization between September 1, 2009, and August 31, 2010, we examined the association between SNF performance on publicly available metrics (SNF staffing intensity, health deficiencies identified through site inspections, and the percentages of SNF patients with delirium, moderate to severe pain, and new or worsening pressure ulcers) and the risk of readmission or death 30 days after discharge to a SNF. Adjusted analyses controlled for patient case mix, SNF facility factors, and the discharging hospital. MAIN OUTCOMES AND MEASURES Readmission to an acute care hospital or death within 30 days of the index hospital discharge. RESULTS Of 1 530 824 patients discharged, 357 752 (23.3%; 99% CI, 23.3%-23.5%) were readmitted or died within 30 days; 72 472 died within 30 days (4.7%; 99% CI, 4.7%-4.8%), and 321 709 were readmitted (21.0%; 99% CI, 20.9%-21.1%). The unadjusted risk of readmission or death was lower at SNFs with better staffing ratings and better facility inspection ratings. No. (%) of SNFs
30-d Risk of Readmission or Death, % (99% CI)
Lowest Rating
Highest Rating
Lowest Rating
Highest Rating
Unadjusted risk, by staffing rating
2742 (19.2)
959 (6.7)
25.5 (25.3-25.8)
19.8 (19.5-20.1)
Unadjusted risk, by facility inspection rating
2867 (20.1)
1396 (9.8)
24.9 (24.7-25.1)
21.5 (21.2-21.7)
23.7 (23.7-23.7)
23.0 (23.0-23.1)
CONCLUSIONS AND RELEVANCE Among fee-for-service Medicare beneficiaries discharged to a SNF after an acute care hospitalization, available performance measures were not consistently associated with differences in the adjusted risk of readmission or death.
Author Affiliations: Department of Anesthesiology and Critical Care, Perelman School of Medicine, University of Pennsylvania, Philadelphia (Neuman); Leonard Davis Institute of Health Economics, University of Pennsylvania (Neuman, Werner); Division of General Internal Medicine, Department of Medicine, Perelman School of Medicine, University of Pennsylvania (Wirtalla, Werner); Center for Health Equity Research and Promotion, Philadelphia VA Medical Center, Philadelphia (Werner).
JAMA. 2014;312(15):1542-1551. doi:10.1001/jama.2014.13513
Corresponding Author: Mark D. Neuman, MD, MSc, Department of Anesthesiology and Critical Care, Perelman School of Medicine, University of Pennsylvania, 423 Guardian Dr, 1119A Blockley Hall, Philadelphia, PA 19104 (neumanm @mail.med.upenn.edu).
Adjusted risk, by facility inspection rating
Adjustment for patient factors, SNF facility factors, and the discharging hospital attenuated these associations; we observed small differences in the adjusted risk of readmission or death according to SNF facility inspection ratings. Other measures did not predict clinically meaningful differences in the adjusted risk of readmission or death.
1542
jama.com
Copyright 2014 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ by a Fudan University User on 05/15/2015
Skilled Nursing Facility Quality and Hospital Readmissions
O
ne in 5 Medicare beneficiaries is readmitted to the hospital within 30 days of discharge.1 Under traditional fee-for-service reimbursement, hospitals had few incentives to invest in reducing readmission rates. However, with Medicare’s Hospital Readmission Reduction Program2 and the increasing prevalence of bundled payments and sharedsavings programs since the passage of the Patient Protection and Affordable Care Act,3-6 hospitals have increased incentives to improve postdischarge management. One commonly discussed way to do so is through more effective use of postacute care.7,8 Skilled nursing facilities (SNFs) represent the most common setting for postacute care in the United States. Rates of readmission from SNFs are high. One in 4 patients discharged to a SNF is readmitted within 30 days,9 and two-thirds of these readmissions may be preventable.10 Because readmission rates vary across SNFs,11 preferential discharge of postacute care patients to high-quality SNFs may be one strategy by which hospitals could decrease the likelihood of readmission among these patients. Information about SNF performance on common quality metrics is widely available through Medicare’s Nursing Home Compare website. However, little is known about whether performance on these metrics is associated with differences in performance that could predict the likelihood of readmission. To address this, we examined the association between available indicators of SNF quality and hospital readmission among Medicare beneficiaries receiving postacute care at US SNFs.
Methods Data This study was approved by the Perelman School of Medicine Institutional Review Board, which waived the requirement for participant informed consent. Data sources included the 2008-2010 100% Medicare Provider Analysis and Review files, which include records of inpatient care for all fee-for-service Medicare beneficiaries; the 2009 and 2010 Nursing Home Minimum Data Set, which includes detailed clinical data on all patients in Medicare-certified SNFs; the 2009 and 2010 Medicare Beneficiary Summary files, which record vital status and health maintenance organization enrollment; the 2009 and 2010 Medicare Online Survey, Certification, and Reporting files, which compile data on SNF facility characteristics; and SNF performance data published on the Nursing Home Compare website in 2009 and 2010.
Study Sample We based our inclusion criteria on methods used to calculate risk-adjusted hospital-wide readmission rates by the Hospital Readmission Reduction Program.12 Our starting sample included all Medicare discharges from nonfederal acute care hospitals between September 1, 2009, and August 31, 2010, to Medicare-certified SNFs for postacute care, as indicated by an appropriate Nursing Home Minimum Data Set admission assessment within 7 days of discharge.
Original Investigation Research
Because we obtained patient comorbidity data from claims filed up to 12 months before the index discharge, we excluded beneficiaries who were younger than 66 years at hospital discharge or who were enrolled in a health maintenance organization during the 12 months before the index because their claims were unavailable in our data. We also excluded patients who were enrolled in a health maintenance organization in the 30 days after hospital discharge because we could not identify readmissions among them; those who were discharged against medical advice or discharged to hospice (as recorded in the Medicare Provider Analysis and Review discharge status field); those for whom the primary reason for hospitalization was a psychiatric condition, rehabilitation, or medical cancer treatment, following Hospital Readmission Reduction Program definitions,12 because readmissions after hospitalizations for these indications are likely to occur for different reasons than readmissions after other acute care hospitalizations; and those who received postacute care at SNFs that were excluded from Nursing Home Compare at the index discharge for 1 or more of the 5 performance measures we examined because of low case volumes or an insufficient duration of participation in the Nursing Home Compare program. For consistency with Hospital Readmission Reduction Program methods, if a patient had more than 1 eligible discharge during the study period, all discharges meeting the above criteria were used in our regression analyses. In other words, our analysis was at the discharge level rather than the patient level.12 However, we also conducted a supplementary patientlevel analysis that included only the first eligible discharge for each patient in our sample. The sample definition used International Classification of Diseases, Ninth Revision (ICD-9) diagnosis and procedure codes for each discharge, grouped by Agency for Healthcare Research and Quality Clinical Classification Software.12
Outcomes Our primary outcome was a composite end point of unplanned readmission or death from any cause within 30 days of hospital discharge. To allow a uniform window for outcomes assessment for each discharge, we did not distinguish between patients who were directly readmitted from a SNF and those who were discharged home from one and subsequently readmitted, as long as this readmission occurred within 30 days of the index discharge. Death within 30 days was included in our primary outcome to prevent inappropriate censoring of observations13,14; however, for purposes of comparison, we conducted a secondary analysis using an end point of readmission at 30 days, rather than a combined end point of readmission or death. We considered a readmission to be unplanned if it involved an admission to an acute care hospital that occurred within 30 days of hospital discharge and the reason for admission was not bone marrow or solid organ transplant, maintenance chemotherapy, rehabilitation, or a potentially planned procedure not performed to treat an acute condition or a complication of previous care.12
jama.com
JAMA October 15, 2014 Volume 312, Number 15
Copyright 2014 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ by a Fudan University User on 05/15/2015
1543
Research Original Investigation
Skilled Nursing Facility Quality and Hospital Readmissions
Table 1. Skilled Nursing Facilities Included in the Study Sample (n=14 251) No. (%) Skilled Nursing Facility Performance Measures Clinical measures of postacute care quality, median (IQR) Percentage of postacute care residents with delirium
0.7 (0-2.1)
Percentage of postacute care residents with moderate to severe pain
17.3 (9.3-27.2)
Percentage of postacute care residents with new or worsening pressure ulcers)
12.1 (8.2-16.7)
Staffing rating, stars 1 (lowest)
2742 (19.2)
2
2722 (19.1)
3
3018 (21.8)
4
4810 (33.8)
5 (highest)
959 (6.7)
Facility inspection rating, stars 1 (lowest)
2867 (20.1)
2
3331 (23.4)
3
3326 (23.3)
4
3331 (23.4)
5 (highest)
1396 (9.8)
Skilled Nursing Facility Characteristics Total beds ≤50
1533 (10.8)
51-100
5323 (37.4)
101-150
4817 (33.8)
≥151
2578 (18.0)
Percentage of beds occupied, median (IQR)
87.4 (75.8-93.5)
Percentage of all residents covered by Medicare, median (IQR)
12.8 (8.0-19.4)
Percentage of all residents covered by Medicaid, median (IQR)
64.3 (50.3-74.7)
Part of a chain
8142 (57.1)
Hospital based
742 (5.2)
Ownership Not for profit
3569 (25.3)
For profit
9919 (70.4)
Government
597 (4.2)
Abbreviation: IQR, interquartile range.
Independent Variables We obtained 5 indicators of SNF performance from Medicare’s Nursing Home Compare website, using data listed there as of the date of hospital discharge. Performance indicators included 3 clinical measures for postacute care residents (the percentage of SNF residents with delirium, with new or worsening pressure ulcers, and reporting moderate to severe pain15,16), a categorical summary rating of staffing intensity that ranged from 1 to 5 stars,17 and a categorical summary rating based on health deficiencies identified through site inspections that also ranged from 1 to 5 stars.17 Information on other SNF facility characteristics came from the Online Survey, Certification, and Reporting survey closest in time to hospital discharge. Facility characteristics included nursing home size (50 beds or fewer; 51-100 beds; 101150 beds; or 151 beds or more),18-20 the percentage of patients 1544
covered by Medicare and Medicaid within each facility,21-24 occupancy rate, 20,25 chain membership, 26-28 location in a hospital,20,29 and ownership (not for profit, for profit, or government owned).28,30-32 We obtained data on patient age, race,33 sex, and the indication for the index hospitalization from Medicare Provider Analysis and Review files. We categorized indications for hospitalization into 5 broad groups based on ICD-9, Clinical Modification diagnosis and procedure codes, using Hospital Readmission Reduction Program algorithms12: surgical and gynecologic conditions, respiratory conditions and heart failure, cardiac and noncardiac vascular conditions, neurologic conditions, and other general medical conditions. We also obtained Hospital Readmission Reduction Program–defined variables on 31 risk factors and 173 admission diagnosis categories, using hospital discharge claims from the index discharge and all hospitalizations occurring in the 12 months before the index.12
Statistical Analyses We used χ2 tests and the Wilcoxon rank-sum test to assess differences in the baseline characteristics of patients according to outcomes at 30 days. We used linear probability models to test the association between SNF factors and risk of readmission or death within 30 days of discharge. Models evaluated the association between risk of readmission or death and the 5 SNF performance measures, SNF facility characteristics, and the combination of available SNF performance measures and facility characteristics. All regression models adjusted for age, sex, and race; the indication for the index hospitalization; and all 204 risk factor and admission diagnosis variables. Because observed differences in rates of readmission or death across SNFs could reflect differences in quality of the discharging hospital, our regression models included hospital fixed effects to account for time-invariant hospital characteristics. In other words, each regression was a “within-hospital” analysis that compared outcomes among patients who were discharged from the same hospital to different SNFs. We handled missing data via listwise deletion (ie, omitting from each model all observations with missing data on a variable included in that model); of our sample of 1 530 824 discharges, 4 (
Original Investigation
Association Between Skilled Nursing Facility Quality Indicators and Hospital Readmissions Mark D. Neuman, MD, MSc; Christopher Wirtalla, BA; Rachel M. Werner, MD, PhD Editorial page 1517 IMPORTANCE Hospital readmissions are common, costly, and potentially preventable. Little is known about the association between available skilled nursing facility (SNF) performance measures and the risk of hospital readmission. OBJECTIVE To measure the association between SNF performance measures and hospital readmissions among Medicare beneficiaries receiving postacute care at SNFs in the United States.
Related article page 1531 CME Quiz at jamanetworkcme.com and CME Questions page 1585
DESIGN AND PARTICIPANTS Using national Medicare data on fee-for-service Medicare beneficiaries discharged to a SNF after an acute care hospitalization between September 1, 2009, and August 31, 2010, we examined the association between SNF performance on publicly available metrics (SNF staffing intensity, health deficiencies identified through site inspections, and the percentages of SNF patients with delirium, moderate to severe pain, and new or worsening pressure ulcers) and the risk of readmission or death 30 days after discharge to a SNF. Adjusted analyses controlled for patient case mix, SNF facility factors, and the discharging hospital. MAIN OUTCOMES AND MEASURES Readmission to an acute care hospital or death within 30 days of the index hospital discharge. RESULTS Of 1 530 824 patients discharged, 357 752 (23.3%; 99% CI, 23.3%-23.5%) were readmitted or died within 30 days; 72 472 died within 30 days (4.7%; 99% CI, 4.7%-4.8%), and 321 709 were readmitted (21.0%; 99% CI, 20.9%-21.1%). The unadjusted risk of readmission or death was lower at SNFs with better staffing ratings and better facility inspection ratings. No. (%) of SNFs
30-d Risk of Readmission or Death, % (99% CI)
Lowest Rating
Highest Rating
Lowest Rating
Highest Rating
Unadjusted risk, by staffing rating
2742 (19.2)
959 (6.7)
25.5 (25.3-25.8)
19.8 (19.5-20.1)
Unadjusted risk, by facility inspection rating
2867 (20.1)
1396 (9.8)
24.9 (24.7-25.1)
21.5 (21.2-21.7)
23.7 (23.7-23.7)
23.0 (23.0-23.1)
CONCLUSIONS AND RELEVANCE Among fee-for-service Medicare beneficiaries discharged to a SNF after an acute care hospitalization, available performance measures were not consistently associated with differences in the adjusted risk of readmission or death.
Author Affiliations: Department of Anesthesiology and Critical Care, Perelman School of Medicine, University of Pennsylvania, Philadelphia (Neuman); Leonard Davis Institute of Health Economics, University of Pennsylvania (Neuman, Werner); Division of General Internal Medicine, Department of Medicine, Perelman School of Medicine, University of Pennsylvania (Wirtalla, Werner); Center for Health Equity Research and Promotion, Philadelphia VA Medical Center, Philadelphia (Werner).
JAMA. 2014;312(15):1542-1551. doi:10.1001/jama.2014.13513
Corresponding Author: Mark D. Neuman, MD, MSc, Department of Anesthesiology and Critical Care, Perelman School of Medicine, University of Pennsylvania, 423 Guardian Dr, 1119A Blockley Hall, Philadelphia, PA 19104 (neumanm @mail.med.upenn.edu).
Adjusted risk, by facility inspection rating
Adjustment for patient factors, SNF facility factors, and the discharging hospital attenuated these associations; we observed small differences in the adjusted risk of readmission or death according to SNF facility inspection ratings. Other measures did not predict clinically meaningful differences in the adjusted risk of readmission or death.
1542
jama.com
Copyright 2014 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ by a Fudan University User on 05/15/2015
Skilled Nursing Facility Quality and Hospital Readmissions
O
ne in 5 Medicare beneficiaries is readmitted to the hospital within 30 days of discharge.1 Under traditional fee-for-service reimbursement, hospitals had few incentives to invest in reducing readmission rates. However, with Medicare’s Hospital Readmission Reduction Program2 and the increasing prevalence of bundled payments and sharedsavings programs since the passage of the Patient Protection and Affordable Care Act,3-6 hospitals have increased incentives to improve postdischarge management. One commonly discussed way to do so is through more effective use of postacute care.7,8 Skilled nursing facilities (SNFs) represent the most common setting for postacute care in the United States. Rates of readmission from SNFs are high. One in 4 patients discharged to a SNF is readmitted within 30 days,9 and two-thirds of these readmissions may be preventable.10 Because readmission rates vary across SNFs,11 preferential discharge of postacute care patients to high-quality SNFs may be one strategy by which hospitals could decrease the likelihood of readmission among these patients. Information about SNF performance on common quality metrics is widely available through Medicare’s Nursing Home Compare website. However, little is known about whether performance on these metrics is associated with differences in performance that could predict the likelihood of readmission. To address this, we examined the association between available indicators of SNF quality and hospital readmission among Medicare beneficiaries receiving postacute care at US SNFs.
Methods Data This study was approved by the Perelman School of Medicine Institutional Review Board, which waived the requirement for participant informed consent. Data sources included the 2008-2010 100% Medicare Provider Analysis and Review files, which include records of inpatient care for all fee-for-service Medicare beneficiaries; the 2009 and 2010 Nursing Home Minimum Data Set, which includes detailed clinical data on all patients in Medicare-certified SNFs; the 2009 and 2010 Medicare Beneficiary Summary files, which record vital status and health maintenance organization enrollment; the 2009 and 2010 Medicare Online Survey, Certification, and Reporting files, which compile data on SNF facility characteristics; and SNF performance data published on the Nursing Home Compare website in 2009 and 2010.
Study Sample We based our inclusion criteria on methods used to calculate risk-adjusted hospital-wide readmission rates by the Hospital Readmission Reduction Program.12 Our starting sample included all Medicare discharges from nonfederal acute care hospitals between September 1, 2009, and August 31, 2010, to Medicare-certified SNFs for postacute care, as indicated by an appropriate Nursing Home Minimum Data Set admission assessment within 7 days of discharge.
Original Investigation Research
Because we obtained patient comorbidity data from claims filed up to 12 months before the index discharge, we excluded beneficiaries who were younger than 66 years at hospital discharge or who were enrolled in a health maintenance organization during the 12 months before the index because their claims were unavailable in our data. We also excluded patients who were enrolled in a health maintenance organization in the 30 days after hospital discharge because we could not identify readmissions among them; those who were discharged against medical advice or discharged to hospice (as recorded in the Medicare Provider Analysis and Review discharge status field); those for whom the primary reason for hospitalization was a psychiatric condition, rehabilitation, or medical cancer treatment, following Hospital Readmission Reduction Program definitions,12 because readmissions after hospitalizations for these indications are likely to occur for different reasons than readmissions after other acute care hospitalizations; and those who received postacute care at SNFs that were excluded from Nursing Home Compare at the index discharge for 1 or more of the 5 performance measures we examined because of low case volumes or an insufficient duration of participation in the Nursing Home Compare program. For consistency with Hospital Readmission Reduction Program methods, if a patient had more than 1 eligible discharge during the study period, all discharges meeting the above criteria were used in our regression analyses. In other words, our analysis was at the discharge level rather than the patient level.12 However, we also conducted a supplementary patientlevel analysis that included only the first eligible discharge for each patient in our sample. The sample definition used International Classification of Diseases, Ninth Revision (ICD-9) diagnosis and procedure codes for each discharge, grouped by Agency for Healthcare Research and Quality Clinical Classification Software.12
Outcomes Our primary outcome was a composite end point of unplanned readmission or death from any cause within 30 days of hospital discharge. To allow a uniform window for outcomes assessment for each discharge, we did not distinguish between patients who were directly readmitted from a SNF and those who were discharged home from one and subsequently readmitted, as long as this readmission occurred within 30 days of the index discharge. Death within 30 days was included in our primary outcome to prevent inappropriate censoring of observations13,14; however, for purposes of comparison, we conducted a secondary analysis using an end point of readmission at 30 days, rather than a combined end point of readmission or death. We considered a readmission to be unplanned if it involved an admission to an acute care hospital that occurred within 30 days of hospital discharge and the reason for admission was not bone marrow or solid organ transplant, maintenance chemotherapy, rehabilitation, or a potentially planned procedure not performed to treat an acute condition or a complication of previous care.12
jama.com
JAMA October 15, 2014 Volume 312, Number 15
Copyright 2014 American Medical Association. All rights reserved.
Downloaded From: http://jama.jamanetwork.com/ by a Fudan University User on 05/15/2015
1543
Research Original Investigation
Skilled Nursing Facility Quality and Hospital Readmissions
Table 1. Skilled Nursing Facilities Included in the Study Sample (n=14 251) No. (%) Skilled Nursing Facility Performance Measures Clinical measures of postacute care quality, median (IQR) Percentage of postacute care residents with delirium
0.7 (0-2.1)
Percentage of postacute care residents with moderate to severe pain
17.3 (9.3-27.2)
Percentage of postacute care residents with new or worsening pressure ulcers)
12.1 (8.2-16.7)
Staffing rating, stars 1 (lowest)
2742 (19.2)
2
2722 (19.1)
3
3018 (21.8)
4
4810 (33.8)
5 (highest)
959 (6.7)
Facility inspection rating, stars 1 (lowest)
2867 (20.1)
2
3331 (23.4)
3
3326 (23.3)
4
3331 (23.4)
5 (highest)
1396 (9.8)
Skilled Nursing Facility Characteristics Total beds ≤50
1533 (10.8)
51-100
5323 (37.4)
101-150
4817 (33.8)
≥151
2578 (18.0)
Percentage of beds occupied, median (IQR)
87.4 (75.8-93.5)
Percentage of all residents covered by Medicare, median (IQR)
12.8 (8.0-19.4)
Percentage of all residents covered by Medicaid, median (IQR)
64.3 (50.3-74.7)
Part of a chain
8142 (57.1)
Hospital based
742 (5.2)
Ownership Not for profit
3569 (25.3)
For profit
9919 (70.4)
Government
597 (4.2)
Abbreviation: IQR, interquartile range.
Independent Variables We obtained 5 indicators of SNF performance from Medicare’s Nursing Home Compare website, using data listed there as of the date of hospital discharge. Performance indicators included 3 clinical measures for postacute care residents (the percentage of SNF residents with delirium, with new or worsening pressure ulcers, and reporting moderate to severe pain15,16), a categorical summary rating of staffing intensity that ranged from 1 to 5 stars,17 and a categorical summary rating based on health deficiencies identified through site inspections that also ranged from 1 to 5 stars.17 Information on other SNF facility characteristics came from the Online Survey, Certification, and Reporting survey closest in time to hospital discharge. Facility characteristics included nursing home size (50 beds or fewer; 51-100 beds; 101150 beds; or 151 beds or more),18-20 the percentage of patients 1544
covered by Medicare and Medicaid within each facility,21-24 occupancy rate, 20,25 chain membership, 26-28 location in a hospital,20,29 and ownership (not for profit, for profit, or government owned).28,30-32 We obtained data on patient age, race,33 sex, and the indication for the index hospitalization from Medicare Provider Analysis and Review files. We categorized indications for hospitalization into 5 broad groups based on ICD-9, Clinical Modification diagnosis and procedure codes, using Hospital Readmission Reduction Program algorithms12: surgical and gynecologic conditions, respiratory conditions and heart failure, cardiac and noncardiac vascular conditions, neurologic conditions, and other general medical conditions. We also obtained Hospital Readmission Reduction Program–defined variables on 31 risk factors and 173 admission diagnosis categories, using hospital discharge claims from the index discharge and all hospitalizations occurring in the 12 months before the index.12
Statistical Analyses We used χ2 tests and the Wilcoxon rank-sum test to assess differences in the baseline characteristics of patients according to outcomes at 30 days. We used linear probability models to test the association between SNF factors and risk of readmission or death within 30 days of discharge. Models evaluated the association between risk of readmission or death and the 5 SNF performance measures, SNF facility characteristics, and the combination of available SNF performance measures and facility characteristics. All regression models adjusted for age, sex, and race; the indication for the index hospitalization; and all 204 risk factor and admission diagnosis variables. Because observed differences in rates of readmission or death across SNFs could reflect differences in quality of the discharging hospital, our regression models included hospital fixed effects to account for time-invariant hospital characteristics. In other words, each regression was a “within-hospital” analysis that compared outcomes among patients who were discharged from the same hospital to different SNFs. We handled missing data via listwise deletion (ie, omitting from each model all observations with missing data on a variable included in that model); of our sample of 1 530 824 discharges, 4 (
