Useful Variation in Clinical Practice under Uncertainty: Diversification and Learning


Speaker:Charles Manski, Northwestern University
Date: Friday 11 April 2014
Time: 14.30
Location: Bateman Lecture Theatre, Building One

Further details

Commentary on health care delivery prominently expresses two recommendations regarding variation in clinical practice. One calls for systematically different treatment of patients who differ in their observed covariates. The other calls for uniform treatment of patients who are observationally similar. Analysis of patient care as an optimization problem has provided foundation for both recommendations in settings where knowledge of treatment response is strong enough to determine optimal treatments. This paper assesses the recommendations when treatments are chosen under uncertainty. I find that the first is well-motivated, but the second is not. Indeed, variation in treatment of observationally similar patients can be useful. To develop this conclusion, I first consider decision making by a health planner who wants to optimize treatment for a population of patients. When treatment response is uncertain, two motives—diversification and learning—encourage a planner to randomize the treatment of observationally similar patients. I then consider a decentralized health care system in which clinicians make treatment decisions. Variation in clinical practice may not emulate the random variation that a health planner would choose. Nevertheless, it may still provide useful observational evidence about treatment response.