We conducted a simulation experiment to identify strategies primary care physicians (PCP) use to treat patients with type 2 diabetes (T2D). In the experiment 19 PCPs were given 3 simulated patients with T2D. The simulated patients (initial A1c 9.5 to 10.5%) represented 3 canonical types of patients: (1) compliant to treatment orders, requiring initiation/titration of oral medications (meds); (2) non-compliant due to undiagnosed depression, needing adjustment of oral and insulin med doses; (3) compliant, needing initiation/titration of insulin. PCPs treated simulated patients over multiple visits using an electronic medical record-like interface. PCPs could order tests, schedule visits and prescribe meds; simulated patients responded in clinically appropriate ways to actions taken. PCP actions and patient responses were recorded. Analysis of data showed that PCPs treatment decisions reflected 1of 2 strategies: feedback strategies (FB) in which decisions were based on current patient state information, and feedforward strategies (FF) in which decisions were based on anticipated future patient state information. FB increased med doses when delayed effects of meds were estimated to be negligible. FF estimated an amount of meds needed to reach goal and increased doses so as to achieve a constant rate of progress (ROP) toward goal. On case 1, PCPs using FF (9) reduced A1c 0.7%/month, while PCPs using FB reduced A1c 0.3%/month. On case 2, PCPs using FF (7) reduced A1c 0.16%/month and PCPs using FB reduced A1c 0.18%/month. On case 3, PCPs using FF (4) reduced A1c 0.6%/month and PCPs using FB reduced A1c 0.3%/month. The pattern of strategy-use varied within and between PCPs across cases. On cases 1 and 3 where the patients responded to treatments in a more predictable manner FF reduced time to A1c goal. On case 2 where the patient responded less predictably PCPs using FB achieved higher ROP toward goal by basing treatments on current conditions rather than inaccurately estimated future states. From this study we concluded that patients benefit when PCPs are trained to use FF with patients who show that they respond in predictable ways to treatments and FB with other types of patients.