Using functional data analysis to identify physician decision strategies which lead to better type 2 diabetes patient outcomes

An approach was developed and tested to identify strategies used by physicians treating patients with type 2 diabetes. In a previously published study 19 physicians treated 3 simulated type 2 diabetes patients to a standard blood glucose (A1c) goal. A1c trajectories of physicians treating each patient were analyzed using functional data analysis (FDA). Two characteristic patterns were observed. One pattern was associated with rate of progress to goal as the basis for treatment decisions. The second pattern was associated with treatment decisions based on the difference between current and previous A1c values. Computational models were developed to represent and analyze individual physician treatment decisions. Two decision-making strategies were identified within a process control paradigm: feedback - making treatment decisions based on current patient state, and feedforward - making treatment decisions based on anticipated future patient states. Using similarity in patterns of FDA plots as a metric, each physician's treatment of a simulated patient was associated with the model that best represented their decision strategy. Physicians and their representative models achieved similar A1c outcomes when treating the same simulated patients. FDA is a promising tool for identifying strategies which lead to better outcomes in the treatment of patients with a chronic disease.