Dan Zeltzer

I am an assistant professor at Tel Aviv University School of Economics. I received my PhD in Economics from Princeton University. I work in applied microeconomics and health economics. My interests include various aspects of healthcare delivery, including physician referral networks, technology adoption, the use of machine learning for predictive modeling in healthcare, and telemedicine.

Contact Information

Berglas School of Economics, Room 216

Tel Aviv University, Tel Aviv, Israel 6997801

+972 (3) 640 5824




CV (pdf)

Research Papers

Gender Homophily in Referral Networks: Consequences for the Medicare Physician Earnings Gap, AEJ: Applied Economics, April 2020

Featured as AEA Journals Chart of the Week

I assess the extent to which the gender gap in physician earnings may be driven by physicians' preference for referring to specialists of the same gender. Analyzing administrative data on 100 million Medicare patient referrals, I provide robust evidence that doctors refer more to specialists of their own gender. I show that biased referrals are predominantly driven by physicians' decisions rather than by endogenous sorting of physicians or patients. Because most referring doctors are male, the net impact of same-gender bias by both male and female doctors generates lower demand for female relative to male specialists, pointing to a positive externality for increased female participation in medicine.

Prediction Accuracy with Electronic Medical Records versus Administrative Claims Data (joint with Ran Balicer, Tzvi Shir, Natalie Flaks-Manov, Liran Einav, and Efrat Shadmi), Medical Care, July 2019.

We evaluate the incremental predictive power of electronic medical record (EMR) data, relative to the information available in more easily accessible and standardized insurance claims data. Using both EMR and Claims data, we predicted outcomes for 118,510 patients with 144,966 hospitalizations in 8 hospitals, using widely used prediction models. We use cross-validation to prevent overfitting and tested predictive performance on separate data that were not used for model training. We achieve nearly the same prediction accuracy using both EMR and claims data relative to using claims data alone in predicting 30-day readmissions [area under the receiver operating characteristic curve (AUC): 0.698 vs. 0.711; positive predictive value (PPV) at top 10% of predicted risk: 37.2% vs. 35.7%], and 1-year mortality (AUC: 0.902 vs. 0.912; PPV: 64.6% vs. 57.6%). EMR data, especially from the first 2 days of the index admission, substantially improved prediction of length of stay (AUC: 0.786 vs. 0.837; PPV: 58.9% vs. 55.5%) and inpatient mortality (AUC: 0.897 vs. 0.950; PPV: 24.3% vs. 14.0%). Results were similar for sensitivity, specificity, and negative predictive value across alternative cutoffs and for using alternative types of predictive models. Results suggest that EMR data are useful in predicting short-term outcomes. However, their incremental value for predicting longer-term outcomes is smaller. Therefore, for interventions that are based on long-term predictions, using more broadly available claims data is equally effective.

Why is End-of-Life Spending So High? Evidence from Cancer Patients (joint with Liran Einav, Amy Finkelstein, Tzvi Shir, Salomon Stemmer, and Ran Balicer)

Revision requested, Review of Economics and Statistics

We analyze rich data on 160,000 cancer patients to study why healthcare spending is highly concentrated at the end of life. Among patients with similar initial prognoses, monthly spending in the year post diagnosis is over twice as high for those who die within the year than for survivors. This elevated spending is almost entirely driven by higher inpatient spending, particularly low-intensity admissions. However, most low-intensity admissions do not result in death - even among cancer patients with poor prognoses at the time of the admission - making it difficult to target reductions. In addition, among patients with the same cancer type and initial prognosis, end-of-life spending is substantially more concentrated for younger patients compared to older patients, suggesting that preferences play a role in driving end-of-life spending patterns. Taken together, our results cast doubt on the view that end-of-life spending is a clear and remediable source of waste.

Supply-Side Variation in the Use of Emergency Departments (joint with Liran Einav, Avichai Chasid, and Ran Balicer)

Revision requested, Journal of Health Economics

We investigate the role of person-specific and place-specific factors in explaining geographic variation in emergency department (ED) utilization using detailed data on 150,000 patients who moved regions within Israel. We observe a sharp change in the probability of an ED visit following a move that is equal to half of the destination-origin difference in the average ED utilization rate. In contrast, we find no change in the probability of having an unplanned hospital admission (that is, via the ED), implying that the entire change is driven by ED visits that do not lead to hospital admission. Similar results are obtained in our complementary event-study analysis, which uses hospital entry as a source of variation. The results from both approaches suggest that supply-side variation in ED access affects only the less severe cases - for which close substitutes likely exist - and that variation across ED physicians in their propensity to admit patients is not explained by place-specific factors, such as differences in incentives, capacity, or diagnostic quality.

Drug Diffusion through Peer Networks: The Influence of Industry Payments (joint with Lelia Agha)

NBER Working Paper No. 26338


Pharmaceutical companies' marketing efforts primarily target physicians, often through individual detailing that entails monetary or in-kind transfers. We study how peer influence broadens these payments' reach beyond the directly paid physicians. Combining Medicare prescriptions and Open Payments data for anticoagulant drugs, we document that pharmaceutical payments target highly connected physicians. We exploit within-physician variation in payment exposure over time to estimate the payments' influence. Unlike the paid doctor, peer physicians are not directly selected by the pharmaceutical company on the basis of their expertise or enthusiasm for the target drug. Yet, following a large payment, prescriptions for the target drug increase both by the paid physician and the paid physician's peers. These peer effects influence doctors who share patients with the paid physician, even when the two doctors are not affiliated with the same group practice. We find no evidence that payments reduce prescriptions among high-risk patients. Over the period 2014--2016, physician payments associated with anticoagulant marketing increased the drugs' prescription volume by 23 percent, with peer spillovers contributing a quarter of the increase.

Horizon Effects and Adverse Selection in Health Insurance Markets (joint with Olivier Darmouni)

Revision requested, Canadian Journal of Economics

blog post: https://www8.gsb.columbia.edu/articles/ideas-work/there-s-easier-obamacare-fix

This paper highlights the idea that increasing contract length affects adverse selection in health insurance markets. Although health risks may be more predictable over a short planning horizon, private health insurance contracts in the United States have short, one-year terms. Short-term, community-rated contracts allow patients to increase their coverage only after risks materialize, which leads to market unraveling. Longer contracts ameliorate adverse selection because both demand and supply exhibit horizon effects. Intuitively, over a longer horizon risk is mean reverting, thus allowing for better pooling of risk within individuals over time, as opposed to just across individuals, consequently lowering equilibrium premiums. Evidence from administrative claims data suggests that risk is indeed mean reverting. Counterfactual analysis illustrates that a simple reform that implements two-year instead of one-year community-rated contracts could increase equilibrium coverage and yield non-trivial welfare gains.

Can Targeting High-Risk Patients Reduce Readmission Rates? Evidence from Israel (with Efrat Shadmi, Tzvi Shir, Natalie Flaks-Manov, Liran Einav, and Ran Balicer)

Revision requested, Journal of Applied Economics

We study a large intervention to reduce hospital readmission rates by the largest Israeli integrated healthcare system. Since 2012, the intervention flagged patients aged 65 and older with high readmission risk to providers, both upon admission and after discharge. Risk scores were based on patient-specific prior healthcare utilization. Analyzing 171,541 covered admissions during 2009–2016, we find that the intervention reduced 30-day readmission rates by 5.9% among patients aged 65–70 relative to patients aged 60–64, who were not targeted for inclusion in the intervention, and for whom no scores were calculated. The largest reduction, 12.3%, was among high-risk patients. Primary care post-discharge follow-up encounters were significantly expedited. The magnitude of the estimated effect peaked during the first two years, and it declined subsequently, after incentive payments by the Israeli Ministry of Health to organizations that reduce readmission rates were discontinued. Taken together, the evidence demonstrates that informing providers about patient risk in real time can improve care continuity and reduce hospital readmissions, and that maintaining such efforts on an ongoing basis is important to sustain their impact.

Draft available upon request