OBJECTIVES: This study aimed to develop a risk adjustment model to estimate stroke patients’ severity by incorporating treatment received, medication and comorbidities from claims data.

METHODS: Adult patients (≧20 years) with ischemic stroke were identified in the Taiwan Stroke Registry data file between 1 January 2006 and 31 December 2009. A generalized linear regression model for estimating stroke severity were developed by incorporating treatment received, medication use, comorbidities, and medical utilization as covariates. A final model was selected using a stepwise process, which used the statistical significance (p <.05) as the stopping rule. The predicted probabilities were transferred into score from 0 to 100 for the purpose of readily comprehensible. Comparison of SSE (stroke severity estimation), NIHSS, Elixhauser Score and Charlson comorbidity index were conducted by calculating the concordance index (C-index) which is numerically equivalent to the area under the receiver operating characteristic (ROC) curve.

RESULTS: For a generalized linear regression model predicting the probability of NIHSS of adult ischemic stroke patients, the C-index revealed a satisfactory value of 0.722 and R-square also reaches 45%. The most important predictors of NIHSS were whether received nasogastric tube (estimate=0.9235), urinary catheterization (estimate=0.4782, score=52), or airway suctioning (estimate=0.2991, score=32). After adjusted for age and gender, the model with SSE has higher AUC and AIC than models with NIHSS, Elixhauser Score or Charlson comorbidity index (SSE, NIHSS, Elixhauser Score, Charlson comorbidity index: AUC 0.79> 0.77> 0.68> 0.62, respectively).

CONCLUSIONS: The results provide a risk adjustment model to estimate stroke severity. Also, SSE can be used to predict the treatment results when physicians are choosing treatment strategies for stroke patients.