Predicting functional outcomes in patients suffering from ischaemic stroke using initial admission variables and physiological data: a comparison between tree model and multivariate regression analysis

This study was aimed to compare the application of a tree model and regression approach for developing data-driven models that identified frisk factors related to functional outcomes among ischaemic stroke patients.

Data were derived from 271 hospitalised patients with a first-ever ischaemic stroke. The Barthel Index (BI) and Modified Rankin Scale (MRS) were used to assess the functional outcomes. The stroke severity at admission and related information from 2006 to December 2007 were extracted retrospectively from a chart review.

In the regression approach, including age, the National Institutes of Health Stroke Scale (NIHSS) score and glucose level were the most significant predictors affecting both the BI and MRS. After applying the tree model, different tree structures were found. For the BI score, the NIHSS score interact with glucose, age and systolic blood pressure to form the tree structure. By contrast, the NIHSS score mainly interact with patients' age to form the tree model for MRS.

Both models have their pros and cons. The tree model otherwise provides risk interactions, and can effectively discriminate the risk groups for different functional outcomes. Applying both models to specific situations will provide a different angle for functional assessment and intervention in stroke rehabilitation.