A BAYESIAN LEARNING MODEL OF STROKE PATIENTS' UTILIZATION OF REHABILITATION SERVICE

Purpose: Although evidence suggests that rehabilitation can improve post-stroke functional status, the outpatient rehabilitation service is often underused. The authors propose a dynamic model within a Bayesian learning framework, and apply it to rehabilitation service usage data of a cohort of stroke patients in Singapore to unveil the behavior of stroke patients and inform policies to raise the uptake of rehabilitation service.

Method: We constructed a discrete choice learning model of the individual behaviors of stroke patients in an environment where there is uncertainty about efficacy of rehabilitation service and hence the optimal quantity of service consumption. In our model, the usage experience gives stroke patients imperfect information about the efficacy of each rehabilitation consumption level. Stroke patients initially hold their prior beliefs about the efficacy of rehabilitation service, and their posterior beliefs about the efficacy are further updated by their usage experience through a Bayesian learning process. In addition, stroke patients make a choice about the quantity to use based on their prescribed amount. Stroke patients’ preference are mapped on to an utility function taking perceived efficacy of different quantity usage and corresponding costs of service into consideration. We calibrate the model against the data using maximum likelihood estimation to quantify the uncertainty of stroke patients’ belief regarding rehabilitation service, as well as unveil their sensitivity to the price of the service. Using the estimated coefficients, we performed 2 policy experiments to find how timing of introducing monetary incentives affects the uptake of the rehabilitation service.

Result: The counterfactual experiments show that introducing 1-month free trials of rehabilitation service to stroke patients in the 1^st month post-discharge can increase more utilization of the service than the free trails which are introduced in the 6^th month post-discharge. Based on the model estimates, the results show that introducing 1-month free trials to those who fully adhere to their prescription in the 1^stmonth post-discharge can increase the uptake of the rehabilitation service by 3%.  

Conclusion: To increase the uptake of the rehabilitation service, the model implies that the policy makers could consider designing incentive to facilitate patients learning about the true efficacy of the service. The timing of introducing the incentive is important, and the first few rehabilitation service visits post-discharge are crucial.