I am an economics PhD candidate at Yale University. My research interests are in health, public and labor economics. I am especially interested in patient-provider matching and its implementation. Contact me if you are interested in running experiments in this area!

Papers

Comparative Advantages and Patient-Provider Matching  Job Market Paper

If certain healthcare providers are especially proficient at treating certain types of patients, there may be large benefits to allocating patients to providers whose comparative advantages align with the patient's type. In this paper, I model a provider’s comparative advantages in terms of their ability to causally reduce one-year mortality for higher- and lower-risk patients, separately. To start, I derive sufficient statistics to measure the gains from optimal patient-provider matching that minimizes mortality compared to a random assignment under capacity constraints assuming true comparative advantages are known (first-best). 

I then propose a methodology to measure gains from optimal matching using observational comparative advantages in the presence of patient sorting (second-best). I derive a low-dimensional set of parameters that allows us to identify the relationship between observational and true comparative advantages and thus unbiasedly quantify the gains from optimal matching. I use quasi-experiments that arise from provider exits to identify these parameters.

Applying this methodology to the Veterans Health Administration data, I find considerable variation in provider skill in treating lower- and higher-risk patients. A provider who is one s.d. above average for lower-risk patients can reduce one-year mortality by 0.37 p.p. (a 26.8% reduction of the average rate). A provider who is one s.d. above average for higher-risk patients can reduce one-year mortality by 0.68 p.p. (a 14.8% reduction of the average rate). Optimal matching between PCPs and new primary care patients can reduce one-year mortality by 0.3 p.p. (SE: 0.1) on average (a 14.6% reduction) if we assume higher- and lower-risk patients take up the same workload. If we assume that higher-risk patients take up twice the workload compared to lower-risk ones, optimal matching reduces one-year mortality by 0.2 p.p. (S.E.: 0.1) on average compared to random assignment (a 10.9% reduction).

Complementing Public Care with Private: Evidence from Veterans Choice Act (with Hiroki Saruya and Todd Wagner)

This paper studies how to complement public care with private care, a question faced by public health systems. Leveraging a policy at the Veterans Health Administration that generates discontinuity in private care access, we find that eligibility to private care increases private outpatient care by $53 and decreases VA outpatient care by $20 per patient-year. The policy leads to a 0.1 p.p. decrease in one-year mortality (2.8% reduction) as a result of decreased wait times and increased access to care otherwise hard to obtain. The mortality reduction benefit of allowing patients access to private care significantly outweighs the increased costs. Our results suggest that counterfactual policies that expand the eligibility criteria would be highly cost effective.

Fixing Misallocation with Guidelines: Awareness vs. Adherence (with Jason Abaluck, Leila Agha, David Chan and Daniel Singer)

Expert decisions often deviate from evidence-based guidelines. If experts are unaware of guidelines, dissemination may improve outcomes. If experts are aware of guidelines but continue to deviate, promoting stricter adherence has ambiguous effects on outcomes depending on whether experts have information not in guidelines. We study guidelines for anticoagulant use to prevent strokes among atrial fibrillation patients. By text-mining physician notes, we identify when physicians start using guidelines. After mentioning guidelines, physicians become more guideline-concordant, but adherence remains far from perfect. To evaluate whether nonadherence reflects physicians’ superior information, we combine observational data on treatment choices with machine learning estimates of heterogeneous treatment effects from eight randomized trials. Most departures from guidelines are not justified by measurable treatment effect heterogeneity. Promoting stricter adherence to guidelines could prevent 24% more strokes, producing much larger gains than broader guideline awareness.