Stroke is the fourth leading cause of death and the second most common cause of hospital admissions among the elderly. Hospital readmission and early mortality are indicators of the quality and efficiency of hospital-level care. CMS will report 30-day risk-standardized mortality rates (RSMRs) and 30-day risk-standardized readmission rates (RSRRs) for hospitalized stroke patients in 2014. CMS reported that the mean unadjusted mortality rate and readmission rate was 15.5% and 14.8%, respectively, among elderly CMS beneficiaries hospitalized for acute ischemic stroke in 2007/2008. After adjusting for patient and clinical comorbidity, the mean RSMR and RSRR were 15.5% and 14.7%, respectively. To date, little data is available that examines variation in risk-adjusted outcomes between hospitals for patients with stroke and even less information is available about the performance of CMS risk-adjusted mortality and readmissions models specifically for VA facilities.
The objective was to examine facility-level variation in RSMRs and RSRRs between VA hospitals for Veterans hospitalized with stroke.
VA administrative data were used to identify index admissions for stroke in CY2007-2011 for Veterans 65 years of age and older using three case-finding algorithms: (1) ICD-9 codes 433.x1 and 434.x1 (CMS definition of stroke beginning in 2013); (2) ICD-9 codes 433.x1, 434.x1 and 436 (CMS definition prior to 2013); and (3) ICD-9 codes 433.x1, 434.00, 434.x1 and 436 (stroke definition used by the 2007 VA Office of Quality and Performance Stroke Special Project). Per CMS methodology, results are reported in 3-year time periods to improve precision of each facility's outcome estimate and to identify more variation in hospital performance.
The VA Vital Status File was used to identify all-cause mortality within 30-days of stroke admission. The mortality outcome was attributed to the VA facility that admitted the patient, even if the patient was subsequently transferred. , VA and CMS data were used to identify unplanned readmissions at acute care facilities within 30-days of hospital discharge. The readmission outcome was attributed to the VA facility that ultimately discharged the patient. Planned readmissions (e.g., admission for carotid endarterectomy) were excluded from the outcome.
Hierarchical logistic regression models were used to calculate RSMRs and RSRRs. The models adjust for case mix differences (age, sex, clinical comorbidity) at the time of admission and include a hospital-specific random effects intercept. The mortality models also adjust for transfer status from an emergency department at an outside facility. The RSMR and RSRR are calculated as the ratio of the number of predicted outcomes to the number of expected outcomes, multiplied by the national unadjusted outcome rate. We assessed whether the facility's 95% confidence interval (estimated using bootstrapping techniques) overlapped with the national crude rate to determine if a VA facility's performance was significantly different from the national rate. Sensitivity analyses were conducted to assess agreement in facility rankings based on the diagnosis codes used to identify stroke admissions and the source of data (VA only, VA + CMS) to identify readmissions. Model performance was assessed using c-statistics.
There were 8230-8671 index stroke admissions at 122-123 facilities per each 3-year time period in CY2007-2011, with a median number of 57-65 admissions per hospital. The number of index stroke admissions increased 3% per year. The mean age was 77 years and almost all were men. Rates of comorbidity were high and consistent each year.
The crude 30-day mortality rate declined 0.7%, from a high of 9.3% in CY2007-2009 to 8.6% in CY2009-2011. At the facility-level, the unadjusted mortality rate ranged from 0% to 31%, with a median of 9.4% in CY2007-2009, 8.9% in CY2008-2010, and 8.4% in CY2009-2011. After adjusting for patient and clinical covariates, the RSMR ranged from 5.7% to 15.6%, with a mean of 9.5% (CI95: 5.9-14.7%) in CY2007-2009, 9.2% (CI95: 5.9-13.9%) in CY2008-2010, and 8.7% (CI95: 5.5-13.2%) in CY2009-2011,
The crude 30-day readmission rate was 15%. At the facility-level, the unadjusted readmission rate ranged from 0% to 44%, with a median of 15.0% in CY2007-2009 and 14.0% in both CY2008-2010 and CY2009-2011. After risk-standardization, there was less facility-level variation in readmission rates. The RSRRs ranged from 12.5% to 17.9% in CY2009-2011. The mean RSRR was 15.2% (CI95: 12.7-18.2%) in CY2007-2009, 15.3% (CI95: 12.3-18.7%) in CY2008-2010, and 14.7% (CI95: 11.3-19.0%) in CY2009-2011,
More than 99% of VA facilities performed as expected, indicating that risk-standardized mortality and readmission rates were no different than the national VA average. There were no facilities that were identified as performing better than expected (i.e. RSMR and RSRRs lower than the national average)..
Sensitivity analyses showed that the choice of a stroke case-finding algorithm did not change the RSMRs and RSRRs and did not impact facility performance rankings. Readmissions are underestimated by 28% if using only VA data (as compared to VA-CMS data) to identify readmissions; however, facility rankings were not affected by the source of data to identify readmissions.
The models had good discrimination for mortality and poor discrimination for readmissions. Model performance using VA data was similar to performance reported by CMS for mortality (c-statistic: 0.73-0.77 in VA vs. 0.73 in CMS) and readmissions (c-statistic: 0.61-0.63 in VA vs. 0.60 in CMS).
The research supports a Stroke QUERI policy goal of understanding facility-level variation in mortality and readmissions and the utility of using the CMS methodology to profile VA facilities. Our findings indicate that although the CMS models perform as well or better in the VA than in CMS, there was little variation in VA facility performance with regard to 30-day mortality and readmissions. It is difficult to distinguish performance of VA facilities because the stroke volume at each VA facility is small and because there was little variation in RSMRs and RSRRs. Findings indicate that RSMRs among Veterans hospitalized for stroke are significantly lower than rates among CMS beneficiaries (8.6% vs. 15.5%); however, RSRRs are similar in the VA and CMS (14.7% vs. 14.8%). Research is needed to understand factors that may contribute to lower mortality in the VA.
External Links for this Project
None at this time.