2006 HSR&D National Meeting Abstract
3068 — Do Diagnoses Contribute to Predictions of Stroke Rehabilitation Outcomes?
Berlowitz DR (Center for Health Quality, Outcomes and Economic Research)
Vogel WB (Rehabilitation Outcomes Research Center)
Duncan PW (Rehabilitation Outcomes Research Center)
Tsilimingras D (Center for Health Quality, Outcomes and Economic Research)
Hoenig H (Durham VAMC)
Cowper DC (Rehabilitation Outcomes Research Center)
Young L (Rehabilitation Outcomes Research Center)
Wing K (Rehabilitation Outcomes Research Center)
Loveland S (Center for Health Quality, Outcomes and Economic Research)
Risk-adjustment models may be useful in monitoring and improving outcomes for stroke patients receiving VA rehabilitation. Diagnoses are often used in risk-adjustment models but the extent to which they predict rehabilitation outcomes is uncertain. We now examine different approaches to using diagnostic data in determining whether they are useful in predicting important rehabilitation outcomes.
Newly diagnosed stroke patients receiving inpatient rehabilitation in 2001, as well as their admission and discharge Functional Independence Measure (FIM) scores, were identified through the VA Integrated Stroke Outcomes Database. Diagnoses, captured from the National Patient Care Database, were grouped based on three different case-mix measurement systems, Adjusted Clinical Groups (ACGs), Diagnosis Cost Groups (DCGs), and the Charlson Index. Different variations of each case-mix system were examined. Outcome measures examined in regression models included change in FIM score from admission to discharge from rehabilitation, 6-month mortality, and 6-month rehospitalization to an acute care service.
2402 stroke patients receiving rehabilitation were identified; 8.0% died and 27.6% were rehospitalized over the 6 months. R-squared values in linear regression models predicting change in FIM increased from 7% to 9-10% when diagnoses were added to age/sex models. Logistic regression models containing diagnostic data had considerably higher c-statistics than age/sex models predicting mortality and rehospitalization. C-statistics were 0.72 and 0.74, respectively, in ACG and DCG-based models predicting mortality; they were 0.63 in models predicting rehospitalization. ACG and DCG-based models performed considerably better than the Charlson Index for all three outcomes.
Diagnoses are important predictors of mortality and rehospitalization in stroke patients receiving VA inpatient rehabilitation but explain little variability in functional outcomes such as change in FIM scores. ACG and DCG-based models performed similarly and were better than models based on the Charlson Index.
These results suggest that risk-adjustment models incorporating diagnostic data may be developed for some outcomes of VA stroke rehabilitation. Use of these models may facilitate efforts to assess and improve the quality of this care. Best performers may be identified and could serve as benchmarks for the entire VA.