Analysis via Markov decision process to evaluate glycemic control strategies of a large retrospective cohort with type 2 diabetes: the ameliorate study

Abstract

Aims: Our aim was to explore optimal treatment decisions for HbA1c control for type 2 diabetes mellitus patients and assess the impact on potential improvements in quality of life compared with current guidelines. Methods: We analyzed a large dataset of HbA1c levels, diabetes-related key risk factors and medication dispensed to 70,069 patients with type 2 diabetes from polyclinics and a large public hospital in Singapore during January 1, 2008, to December 31, 2015. A Markov decision process (MDP) model was developed to determine the optimal treatment policy concerning medication management for glycemic control over a long-term treatment period. We assessed the model performance by comparing quality-adjusted life years (QALYs) gained by the model with those derived by a conventional Markov model informed by current clinical guidelines. Results: Numerical results showed that optimal treatment strategies derived by the MDP model could increase the total expected QALYs by as much as 0.27�years for patients at higher risk such as old age, high HbA1c levels and smokers. In particular, the improvements in QALYs gained for patients with HbA1c levels of 9% (75�mmol/mol) and above were higher than those with lower HbA1c levels. However, the potential improvements appeared to be marginal for patients at lower risk compared with current guidelines. Conclusions: Use of data-driven prescriptive analytics would help clinicians make evidence-based treatment decisions for HbA1c control for patients with type 2 diabetes, in particular for those at high risk. ? 2020, Springer-Verlag Italia S.r.l., part of Springer Nature.