Hospital Readmission Performance and Patterns of Readmission
Hospital Readmission Performance and Patterns of Readmission
We used Medicare standard analytic and denominator files to identify index admissions to acute care hospitals in 2007-09 with a principal discharge diagnosis of heart failure, acute myocardial infarction, or pneumonia. Cohorts were defined with ICD-9 (international classification of diseases, 9th revision, clinical modification) codes used in readmission measures publically reported by the Centers for Medicare and Medicaid Services ( Table A in Appendix 1). We included patients aged 65 and older with a complete claims history for one year before admission. Reasons for exclusion included death in hospital, less than 30 days after discharge with enrollment in Medicare fee for service, transfer to another acute care facility, and discharge against medical advice. To provide more reliable estimates of hospital performance, we did not include patients from hospitals with fewer than 25 index admissions for each condition or no readmissions over the study period. We required that hospitals be represented in 2008 American Hospital Association survey data.
We then used definitions consistent with current Centers for Medicare and Medicaid Services measures ( Table B in Appendix 1) to identify all qualifying readmissions for any cause to any short term acute care hospital within 30 days of the index admission. We specifically excluded planned readmissions for revascularization procedures after admission for acute myocardial infarction. As with the Centers for Medicare and Medicaid Services measures, only the first readmission within 30 days of discharge was considered as a 30 day readmission. Additional readmissions within this 30 day period were not counted as 30 day readmissions or index admissions for the same condition. Subsequent admissions occurring after 30 days from discharge were counted as index admissions if they met inclusion criteria. All study analyses were performed on these readmission cohorts.
To characterize readmission patterns across hospitals, we classified individual diagnoses for readmitted patients using a modified version of the Centers for Medicare and Medicaid Services condition categories, as previously described. Each of the 189 condition categories is structured around a reasonably well specified disease or medical condition. Because nearly 90% of the 189 condition categories each accounted for less than 1% of all readmissions, however, we consolidated related diagnoses in a shorter list of 30 modified condition categories to make data presentation more clinically meaningful ( Table C in Appendix 1). Based on our opinion, these 30 modified condition categories were designed to be clinically internally consistent and capture the most common diagnostic categories associated with readmission. Cardiopulmonary diagnoses were subdivided into a larger number of diagnostic categories given their expected importance after index admission for heart failure, acute myocardial infarction, or pneumonia.
We classified timing of readmission by day (0-30) after hospital discharge.
For each hospital, we calculated separate 30 day readmission rates standardized for risk after index admission for heart failure, acute myocardial infarction, and pneumonia. The National Quality Forum approved these measures and an independent committee of statisticians nominated by the Committee of Presidents of the Statistical Societies endorsed the validity of the methods. Risk standardized readmission rates are publically reported by the Centers for Medicare and Medicaid Services and have been incorporated into incentive programs within the Affordable Care Act. The modeling strategy used for risk standardization accounts for correlation of observed readmission rates within a hospital and reflects the assumption that, after adjustment for sampling variability and patient characteristics including age, sex, and comorbidities, the remaining variation in readmission rates reflects hospital quality.
To compare readmission diagnoses and timing across hospitals of different performance levels, we used the bootstrap algorithm to construct a 95% interval estimate for each 30 day risk standardized readmission rate and divided hospitals into high, average, and low performers for each index condition. High and low performing hospitals had a 95% or greater probability of having an interval estimate respectively less than or greater than the national rate over the three year period of study. All remaining hospitals were considered average. For each hospital, we calculated performance separately for heart failure, acute myocardial infarction, and pneumonia cohorts.
Readmission Diagnoses by Hospital Performance We identified the percentage of readmissions for each of the 10 most common diagnostic categories by modified condition categories among hospitals with high, average, and low performance and compared these percentages across hospital performance groups.
To deal with potential confounding of results by differences in hospital characteristics among performance groups we also examined whether hospital 30 day risk standardized readmission rates were associated with the percentage of readmissions for each of the 10 most common diagnostic categories by modified condition category after adjustment for hospital factors used in performance reports by the Centers for Medicare and Medicaid Services and previous studies examining hospital readmissions.
Readmission Timing by Hospital Performance We calculated the median time to readmission among hospitals with high, average, and low performance and compared this timing across performance categories.
To deal with potential confounding, we also examined whether hospital 30 day risk standardized readmission rates were associated with median time to readmission after adjustment for the additional hospital factors cited above.
Readmission Diagnoses by Hospital Performance We first identified the percentage of observed 30 day readmissions for the 30 most common readmission diagnostic categories by modified condition category for each hospital in the heart failure, acute myocardial infarction, and pneumonia cohorts. We also noted each hospital's percentage of observed 30 day readmissions for cardiovascular diagnoses after admissions for heart failure and acute myocardial infarction, and pulmonary diagnoses after admissions for pneumonia. The modified condition category groups comprising cardiovascular and pulmonary diseases are listed in Table D and Table E in Appendix 1.
We then calculated a weighted average of the percentage of hospitals' readmissions for the 10 most common readmission diagnoses by modified condition category for high, average, and low performing hospitals. Weighting was proportionate to the number of readmissions for each hospital during 2007-09. We calculated summary statistics for each hospital performance level.
To examine the association between hospital performance and readmission diagnoses across the range of hospital 30 day risk standardized readmission rates, we developed weighted regression models of the relation between the rates and each of the 10 most common readmission diagnoses across hospitals. Regression models were weighted by the inverse of the standard error of the hospital level percentage of each of the 10 most common readmission diagnoses. After univariate analysis, we fitted additional weighted regression models that were adjusted for other hospital characteristics besides the 30 day risk standardized readmission rate including teaching status, urban/rural location, ownership status (private, not for profit, public), safety net status, critical access status, proportion of patients receiving Medicaid health insurance, proportion of African-American patients, and hospital volume with regard to the number of index admissions for each condition. Safety net hospitals in the US provide care for a high proportion of uninsured, underinsured, or Medicaid patients; as has been done previously, we defined safety net hospitals as public or private hospitals with an annual Medicaid caseload >1 SD above their respective state's mean private hospital Medicaid caseload.
Readmission Timing by Hospital Performance We calculated a weighted average of hospitals' median time to readmission for hospitals with high, average, and low performance by 30 day risk standardized readmission rate. We calculated summary statistics for each hospital performance level. We used the Kruskal-Wallis test to examine whether median time to readmission differed between hospital performance groups. We also developed regression models of the relation between hospital level risk standardized readmission rates and hospital level median time to readmission. Models were weighted by the number of readmissions for each hospital. After univariate analysis, we fitted additional weighted regression models that were adjusted for the other hospital characteristics described above.
All significance levels for the regression models were two sided with a P value <0.05, and analyses were carried out with SAS 9.2 (SAS Institute, Cary, NC).
Methods
Data Sources and Study Population
We used Medicare standard analytic and denominator files to identify index admissions to acute care hospitals in 2007-09 with a principal discharge diagnosis of heart failure, acute myocardial infarction, or pneumonia. Cohorts were defined with ICD-9 (international classification of diseases, 9th revision, clinical modification) codes used in readmission measures publically reported by the Centers for Medicare and Medicaid Services ( Table A in Appendix 1). We included patients aged 65 and older with a complete claims history for one year before admission. Reasons for exclusion included death in hospital, less than 30 days after discharge with enrollment in Medicare fee for service, transfer to another acute care facility, and discharge against medical advice. To provide more reliable estimates of hospital performance, we did not include patients from hospitals with fewer than 25 index admissions for each condition or no readmissions over the study period. We required that hospitals be represented in 2008 American Hospital Association survey data.
We then used definitions consistent with current Centers for Medicare and Medicaid Services measures ( Table B in Appendix 1) to identify all qualifying readmissions for any cause to any short term acute care hospital within 30 days of the index admission. We specifically excluded planned readmissions for revascularization procedures after admission for acute myocardial infarction. As with the Centers for Medicare and Medicaid Services measures, only the first readmission within 30 days of discharge was considered as a 30 day readmission. Additional readmissions within this 30 day period were not counted as 30 day readmissions or index admissions for the same condition. Subsequent admissions occurring after 30 days from discharge were counted as index admissions if they met inclusion criteria. All study analyses were performed on these readmission cohorts.
Categorization of Readmission Diagnoses and Timing
To characterize readmission patterns across hospitals, we classified individual diagnoses for readmitted patients using a modified version of the Centers for Medicare and Medicaid Services condition categories, as previously described. Each of the 189 condition categories is structured around a reasonably well specified disease or medical condition. Because nearly 90% of the 189 condition categories each accounted for less than 1% of all readmissions, however, we consolidated related diagnoses in a shorter list of 30 modified condition categories to make data presentation more clinically meaningful ( Table C in Appendix 1). Based on our opinion, these 30 modified condition categories were designed to be clinically internally consistent and capture the most common diagnostic categories associated with readmission. Cardiopulmonary diagnoses were subdivided into a larger number of diagnostic categories given their expected importance after index admission for heart failure, acute myocardial infarction, or pneumonia.
We classified timing of readmission by day (0-30) after hospital discharge.
Hospital Performance
For each hospital, we calculated separate 30 day readmission rates standardized for risk after index admission for heart failure, acute myocardial infarction, and pneumonia. The National Quality Forum approved these measures and an independent committee of statisticians nominated by the Committee of Presidents of the Statistical Societies endorsed the validity of the methods. Risk standardized readmission rates are publically reported by the Centers for Medicare and Medicaid Services and have been incorporated into incentive programs within the Affordable Care Act. The modeling strategy used for risk standardization accounts for correlation of observed readmission rates within a hospital and reflects the assumption that, after adjustment for sampling variability and patient characteristics including age, sex, and comorbidities, the remaining variation in readmission rates reflects hospital quality.
To compare readmission diagnoses and timing across hospitals of different performance levels, we used the bootstrap algorithm to construct a 95% interval estimate for each 30 day risk standardized readmission rate and divided hospitals into high, average, and low performers for each index condition. High and low performing hospitals had a 95% or greater probability of having an interval estimate respectively less than or greater than the national rate over the three year period of study. All remaining hospitals were considered average. For each hospital, we calculated performance separately for heart failure, acute myocardial infarction, and pneumonia cohorts.
Outcomes
Readmission Diagnoses by Hospital Performance We identified the percentage of readmissions for each of the 10 most common diagnostic categories by modified condition categories among hospitals with high, average, and low performance and compared these percentages across hospital performance groups.
To deal with potential confounding of results by differences in hospital characteristics among performance groups we also examined whether hospital 30 day risk standardized readmission rates were associated with the percentage of readmissions for each of the 10 most common diagnostic categories by modified condition category after adjustment for hospital factors used in performance reports by the Centers for Medicare and Medicaid Services and previous studies examining hospital readmissions.
Readmission Timing by Hospital Performance We calculated the median time to readmission among hospitals with high, average, and low performance and compared this timing across performance categories.
To deal with potential confounding, we also examined whether hospital 30 day risk standardized readmission rates were associated with median time to readmission after adjustment for the additional hospital factors cited above.
Statistical Analyses
Readmission Diagnoses by Hospital Performance We first identified the percentage of observed 30 day readmissions for the 30 most common readmission diagnostic categories by modified condition category for each hospital in the heart failure, acute myocardial infarction, and pneumonia cohorts. We also noted each hospital's percentage of observed 30 day readmissions for cardiovascular diagnoses after admissions for heart failure and acute myocardial infarction, and pulmonary diagnoses after admissions for pneumonia. The modified condition category groups comprising cardiovascular and pulmonary diseases are listed in Table D and Table E in Appendix 1.
We then calculated a weighted average of the percentage of hospitals' readmissions for the 10 most common readmission diagnoses by modified condition category for high, average, and low performing hospitals. Weighting was proportionate to the number of readmissions for each hospital during 2007-09. We calculated summary statistics for each hospital performance level.
To examine the association between hospital performance and readmission diagnoses across the range of hospital 30 day risk standardized readmission rates, we developed weighted regression models of the relation between the rates and each of the 10 most common readmission diagnoses across hospitals. Regression models were weighted by the inverse of the standard error of the hospital level percentage of each of the 10 most common readmission diagnoses. After univariate analysis, we fitted additional weighted regression models that were adjusted for other hospital characteristics besides the 30 day risk standardized readmission rate including teaching status, urban/rural location, ownership status (private, not for profit, public), safety net status, critical access status, proportion of patients receiving Medicaid health insurance, proportion of African-American patients, and hospital volume with regard to the number of index admissions for each condition. Safety net hospitals in the US provide care for a high proportion of uninsured, underinsured, or Medicaid patients; as has been done previously, we defined safety net hospitals as public or private hospitals with an annual Medicaid caseload >1 SD above their respective state's mean private hospital Medicaid caseload.
Readmission Timing by Hospital Performance We calculated a weighted average of hospitals' median time to readmission for hospitals with high, average, and low performance by 30 day risk standardized readmission rate. We calculated summary statistics for each hospital performance level. We used the Kruskal-Wallis test to examine whether median time to readmission differed between hospital performance groups. We also developed regression models of the relation between hospital level risk standardized readmission rates and hospital level median time to readmission. Models were weighted by the number of readmissions for each hospital. After univariate analysis, we fitted additional weighted regression models that were adjusted for the other hospital characteristics described above.
All significance levels for the regression models were two sided with a P value <0.05, and analyses were carried out with SAS 9.2 (SAS Institute, Cary, NC).