Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government

Health Services Research & Development

Veterans Crisis Line Badge
Go to the ORD website
Go to the QUERI website

RRP 12-192 – HSR&D Study

New | Current | Completed | DRA | DRE | Portfolios/Projects | Centers | Career Development Projects

RRP 12-192
Understanding factors associated with 30-day Stroke readmission in the VHA
Salomeh Keyhani MD MPH
San Francisco VA Medical Center, San Francisco, CA
San Francisco, CA
Funding Period: December 2012 - March 2014

BACKGROUND/RATIONALE:
Stroke is the fourth leading cause of death and a leading cause of disability among adults in the US and is the second most common cause of hospitalization in the elderly. In the Veterans Health Administration (VA) more than 6,000 Veterans are hospitalized annually for acute ischemic stroke in a VA medical center (VAMC).2 Within the VA system, stroke is both common and costly; understanding factors associated with and predictive of readmission is important to reducing 30-day readmission and improving outcomes.

OBJECTIVE(S):
30-day stroke readmission is currently under review as a potential measure of hospital quality by the Centers for Medicare and Medicaid Services (CMS). This measure could also be used in the VA and potentially reported on Hospital Compare; however, the proposed CMS stroke readmission model, like all other prediction models developed by CMS for Hospital Compare, only includes age, gender and comorbid conditions. Because severity of general medical illness, severity of stroke and social risk measures may be important predictors of stroke readmission, a CMS based model might misclassify a given hospital as either high performing or a low performing. Since the VA shares outcome data with the Hospital Compare program for other conditions, understanding the impact of including the best clinical and social risk factors available on hospital-level comparisons in the VA is informative to policymakers. In this project we compare detailed 30-day readmission models that include the best clinical and non-clinical data available to a CMS based model. In addition to evaluating the effect of including more or less detailed patient level data, we also evaluated the degree to which the statistical method used for risk adjustment influenced hospital level comparisons.

METHODS:
We used data from the 2007 VA Office of Quality and Performance Stroke Special Project to construct 3 different patient-level models examining predictors of 30-day readmission. First we examined a CMS based 30-day readmission model using methods outlined by CMS that includes only age and comorbid conditions. Second, we compared this model to a clinically detailed model using data unavailable in national datasets (e.g., stroke severity) and finally we compared both models to a model that has both detailed clinical characteristics and social risk factors (e.g., homeless, substance abuse). Third, we ranked hospitals by their 30-day risk standardized readmission rates (RSRR) for each model and examined the agreement in rankings between the three models in order to determine if the clinically detailed model classified hospital performance differently than the CMS-based readmission model. Finally, we compared the three different patient-level models that were generated using the CMS hierarchical approach versus a fixed-effect method.

FINDINGS/RESULTS:
The average age in our sample of 3,494 patients was 67.1 years, and the majority were white (66.9%). Overall, the 30-day readmission rate was 13.8%. Multivariate analyses confirmed that older patients [OR 1.015, 95% CI (1.005-1.025] with a higher prevalence of Comorbidity were more likely to be readmitted (Table 1). Patients with metastatic cancer [OR 1.977, 95% CI (1.083, 3.609)] severe hematologic disorders [OR 2.614, 95% CI (1.190, 5.742)] end stage renal disease [OR 2.299, 95%CI (1.010, 5.234)] and skin ulcers [OR 1.952, 95% CI (1.319, 2.889)] had higher odds of 30-day readmission. Social risk factors did not improve model fit. We profiled hospitals based on RSRR generated from the 3 models using both a hierarchical (HGLM) approach and a fixed-effects (FE) approach. Overall, the 30-day readmission rate was 13.8%. The C-Statistic for the three models was 0.6468, 0.6528 and 0.6572 respectively. Using the HGLM approach to modeling, no facilities were designated as outliers and there was complete agreement between the RSRR calculated from the 3 models (kappa=1). The FE approach identified outlier facilities; however all 3 models identified the same 21 hospital as worse than expected.

IMPACT:
This proposed research supports the Stroke QUERI clinical goal related to improving in-hospital management and the policy goal of informing VA policymakers of the utility of using CMS proposed measures to profile VA hospitals. We found that disease severity impacts 30-day readmission but the models used to identify patients at high risk of readmission still need to be improved upon as they have poor discriminative ability. We also found that in the VA, a CMS based model for hospital-level comparisons performs as well as a 30-day stroke readmission model that includes detailed clinical and social risk factors. In the VA, the modeling approach appears to have more impact on hospital profiling than individual variables included in the model. Another important finding was the for low volume hospitals, the CMS approach render all Va hospitals as similar in performance.

PUBLICATIONS:

Journal Articles

  1. Keyhani S, Myers LJ, Cheng E, Hebert P, Williams LS, Bravata DM. Effect of clinical and social risk factors on hospital profiling for stroke readmission: a cohort study. Annals of internal medicine. 2014 Dec 2; 161(11):775-84.


DRA: Cardiovascular Disease
DRE: Prognosis, Treatment - Observational, Prevention
Keywords: QUERI Implementation
MeSH Terms: none

Questions about the HSR&D website? Email the Web Team.

Any health information on this website is strictly for informational purposes and is not intended as medical advice. It should not be used to diagnose or treat any condition.