STOCHASTIC MODELING AND DEVELOPMENT OF A DECISION SUPPORT TOOL TO FACILITATE TRANSITIONAL CARE DECISION MAKING

The first 30 days after discharge from a hospital is a critical time in a patient’s recovery process and poorly executed transitions can lead to negative patient outcomes, including readmission.  Readmission during this time also imposes a financial burden on providers and insurers, therefore developing interventions, such as care transitions programs aimed at reducing 30-day readmission rates through focused interventions are fundamentally needed.

Purpose: The purpose of this study is to develop a stochastic model, and ultimately a decision support tool, of a care transitions program that all stakeholders can use to better understand the effect of treatment and compliance decisions on patient outcomes, specifically the risk of rehospitalization, after being discharged from a hospital and returning to a personal residence. 

Method: Several stochastic modeling methods, specifically Markov Chains, Colored Timed Stochastic Petri Nets, and Dynamic Bayesian Networks were evaluated for their ability to incorporate the complex nature of care transitions programs, including patient related attributes, program attributes, and temporal effects.

Result: Markov chains are widely used in modeling medical care processes, treatments and technologies and offer an understandable approach to modeling a care transition program.  However, as the amount of complexity and uncertainty included in the model is increased, the model becomes more difficult to manage and analyze.  The Colored Timed Stochastic Petri Net model can more easily handle the complexities of patient attributes however a firing time distribution must be determined making the analysis more involved.   The final modeling approach to be reviewed is the Dynamic Bayesian Network, which provides a compact representation of patient attributes, program attributes, and the effect of time.   However, as the probabilities associated with the care transition program, and therefore the model, are time variant, the number of parameters that must be determined increases in proportion to the number of time slices considered.

Conclusion: Care transitions programs offer targeted intervention steps designed to reduce hospital readmissions.  In order to design the most effective programs and understand the effect of decisions a stochastic model with strong inference capabilities is needed.  The models presented here represent three approaches to creating a predictive model and balancing the tradeoff between realistic representation and analytic capability.