Patients with type 2 diabetes mellitus have high risk for cardiovascular events, and the risk derives from multiple sources including elevated glucose, blood pressure, lipids, and other factors. Prior studies have assessed cardiovascular risk in diabetes patients and several evidence-based clinical goals have been identified that independently reduce risks of future adverse cardiovascular events. Prior studies have estimated projected risks of future events for patients with type 2 diabetes, but these studies do not systematically evaluate strategies, risks, and treatment costs for diabetes patients. However, prior research has not provided usable information needed to compare the relative risks and benefits of multiple treatment policies that are available for diabetes care at successive points in time. Specifically, no research is available that estimates the relative impact on cardiovascular events or on costs of competing clinical policies that differentially emphasize glucose, BP, or lipid control, or the relative merits and drawbacks of a """"""""feedforward"""""""" versus the more typical """"""""feedback"""""""" clinical policy that typically characterizes care of complex patients. The research proposed here addresses these critical gaps in knowledge using modeling and data mining technologies to discover and structure clinical policies that most effectively reduce risk of cardiovascular events in complex patients with diabetes. The work will proceed in two steps: (a) Develop modeling methodology to identify physician treatment strategies (combinations of pharmaceutical agents, timing of clinical interventions, complexity of regimen, risky prescribing events) that minimize cost or risk of major cardiovascular complications in complex patients with diabetes, and (b) Apply computational modeling and data mining techniques to identify the optimal combinations of pharmaceutical agents to minimize pharmaceutical costs while achieving pre-specified degrees of reduction in risk of major cardiovascular complications in complex patients with diabetes. Specific objectives will examine the relative merits of clinical policies that prioritize different clinical domains, and the relative merits of """"""""feedforward"""""""" versus """"""""feedback"""""""" clinical strategies. Results will contribute to the important ongoing debate about comparative effectiveness of alternative clinical policies for complex patients with diabetes, including cost data needed to inform the development of clinical guidelines and public policy for the care of complex patients, whose needs are not well addressed by existing clinical guidelines. Moreover, the methods used in this project will provide a useful prototype for comparative effectiveness research that can be applied to diverse clinical domains and patient populations.
| Vock, David M; Wolfson, Julian; Bandyopadhyay, Sunayan et al. (2016) Adapting machine learning techniques to censored time-to-event health record data: A general-purpose approach using inverse probability of censoring weighting. J Biomed Inform 61:119-31 |
| Wolfson, Julian; Bandyopadhyay, Sunayan; Elidrisi, Mohamed et al. (2015) A Naive Bayes machine learning approach to risk prediction using censored, time-to-event data. Stat Med 34:2941-57 |
