The Provision Prediction of Assistive Technology Devices in Post-Stroke Veterans by Decision Modeling

Feng H, Winkler SL, Kairalla J, Cowper Ripley DC, Hoenig H. The Provision Prediction of Assistive Technology Devices in Post-Stroke Veterans by Decision Modeling. Poster session presented at: AcademyHealth Annual Research Meeting; 2012 Jun 24; Orlando, FL.

Research Objective: Assistive Technology Devices (ATDs) enhance the independence and quality of life of post-stroke patients. Prediction of device provision may improve the conceptualization of rehabilitation intervention and clinical outcome relationships. Previous studies reported associations between ATD provision and patients' clinical and functional characteristics. However, the ATD provision and its predictors have highly complex nonlinear relationships. In this study, we applied a decision tree approach to examine the dynamic interactions of potential impacting factors on the ATD provision in post-stroke Veterans. Study Design: A retrospective cohort included Veterans hospitalized for a stroke at a VA facility during FY2007-08. Data were acquired from three Veterans Affairs databases: the Medical SAS data from the National Patient Care Database, the Functional Status and Outcomes Database, and the National Prosthetic Patient Database. The outcome variables were whether the following devices were provided: (1) power-wheelchair, (2) scooter, (3) activities of daily living, (4) manual-wheelchair, (5) ankle/knee orthotic, (6) walker/cane, (7) patient lifts, and (8) beds. Independent variables were demographics, clinical characteristics, and facility variables. Chi-square automatic interaction detector with likelihood ratio test was used to examine the interactions between exploratory variables and the status of device prescription, to identify the specific factors most strongly associated with device provision, and to predict the likelihood on the provision in eight separate analyses. A 10-fold cross-validation was conducted to assess how well the tree-based model performs to predict device provision. Population Studied: The 13,041 subjects who met the study criteria had a mean age of 68 12 years with 62% Caucasians and 23% African Americans; 59% received at least one ATD. The mean scores of mobility Functional Independence Measure (FIM), cognitive FIM, and the number of Elixhauser co-morbidities were 50.7 26.4, 25.1 10.2, and 5.9 2.8, respectively. Principal Findings: Co-morbidity number (p < 0.001) was the strongest predictor for provision of devices 1-2, while mobility FIM score (p < 0.001) was the best predictor for provision of devices 3-8. Among subjects with a number of comorbidities > 8, the higher cognitive FIM score (p < 0.001) was strongly associated with scooter provision; but the lower mobility FIM score (p < 0.001) was significantly associated with power-wheelchair provision in subjects having a number of co-morbidities < 3. Among subjects with the lowest mobility FIM score, the best predictors (p < 0.001) for provision of devices 3-8 were co-morbidity number (3 and 6), cognitive FIM score (4), age (5), and race (7and 8); but the higher co-morbidity number was significantly associated with the provision of devices 3, 4, 6, and 8 (p < 0.001) in the subjects having higher mobility FIM scores. The misclassification rates for cross-validation were < 5% for devices 1-2 and 7, < 10% for devices 5 and 8, and 20% - 37% for devices 3-4 and 6. Conclusions: Co-morbidity and mobility FIM scores are the most important predictors for ATD provision. Cognitive FIM score, age and race interact with these two important factors to be significantly associated with device provision. Implications for Policy, Delivery or Practice: Decision modeling shows promise for understanding dynamic interactions among ATD provision and the characteristics of post-stroke Veterans. Understanding ATD provision patterns may enable rehabilitation professionals to make better decisions when caring for Veterans with disabilities.