Dilara Sonmez Lesniak
Data Science · Causal Inference · Optimization
I'm an operations researcher and data scientist completing my Ph.D. in Management Science and Operations Management at the University of Chicago Booth School of Business. My research sits where large-scale optimization meets causal inference. My dissertation reimagines how primary care is delivered: it pairs a study on the Medicare savings that come from continuous, well-timed care with an optimization model that redesigns physician panels and capacity under value-based contracts, and a structural model that recovers the true cost of care from how patients actually seek it. Working with terabyte-scale Medicare claims data, I've partnered with health systems like UChicago Medicine and Geisinger to take these models from theory into practice. I'm now looking for data science and operations management roles where I can turn complex, high-stakes data into decisions that improve real operational and financial outcomes.
Research
Dissertation, “Optimizing Primary Care Visit Patterns and Panel Design”
Advisor: Prof. Daniel Adelman • Healthcare Analytics Lab, University of Chicago Booth. A three-part program linking empirical evidence, optimization, and structural estimation to help accountable care organizations (ACOs) succeed under value-based care
I. Primary-care visit patterns and Medicare savings (empirical)
- Engineered the analytic cohort down from terabyte-scale national Medicare claims (a nationally representative 5% sample), reducing raw fee-for-service records to beneficiaries meeting study criteria; built novel visit-pattern metrics: frequency, temporal regularity (coefficient of variation), and a continuity measure
- Risk-adjusted outcomes with CMS methodology, propensity-score methods, and gamma generalized linear models with fixed effects; proactive (regular, highly continuous) care was associated with 176% greater savings, 16.6% lower risk-adjusted expenditures, 40.5% fewer ED visits, and 53.3% fewer hospitalizations (all P < .001)
- Uncovered concave savings-frequency relationships with risk-dependent optimal visit frequencies (peak savings near ten visits/year for higher-complexity patients)
II. Reconfiguring primary-care resources for value-based care (optimization)
- Built a framework to embed primary-care visit patterns as decision variables in a resource-constrained optimization, jointly optimizing with physician panels, hiring, and capacity to maximize ACO financial success under two-sided MSSP contracts
- Modeled physician availability with a queueing model with abandonment and cast hiring as fixed-charge facility location with nonlinear, coupled assignment benefits (mixed-integer nonlinear program)
- Developed a segment-based aggregation with an LP-integral (totally-unimodular) matching step that cut solve time at a small optimality gap; applied to UChicago Medicine to redesign staffing and panels against continuity targets
III. Recovering the cost of visit frequency (structural estimation)
- Developed the first structurally estimated semi-Markov physician-scheduling model with a continuously-evolving latent health-risk process, formalizing the trade-off between visit burden now and emergency risk later
- Represented latent health as Brownian motion (log-odds scale) with geometric time-to-emergency and Gumbel scheduling shocks; derived transition kernels, conditional choice probabilities, and equilibrium (fixed-point) constraints
- Cross-checked the causal effect with instrumental-variables regression, then estimated structural parameters via maximum likelihood with equilibrium constraints (MPEC); confirmed exact parameter recovery on ~499,000 transitions in simulation; fed recovered cost curves into the panel-optimization model for Geisinger’s Medicare Advantage patients
Publications
Published: Sonmez D, Weyer G, Adelman D. Primary Care Continuity, Frequency, and Regularity Associated With Medicare Savings. JAMA Network Open. 2023;6(8):e2329991. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2808555
Working papers:
- Sonmez D, Weyer G, Adelman D. Reconfiguring Primary Care Resources for Clinical and Financial Success in Value-Based Care
- Sonmez D, Adelman D. Recovering the Cost of Visit Frequency from Dynamic Choice and Health Transition
Experience
Build optimization and machine-learning models on terabyte-scale national CMS/Medicare claims data (SQL, R, Python, Dataiku DSS) in partnership with UChicago Medicine and Geisinger. Led two quarter-long MBA project teams as end-to-end consulting engagements: owned the client relationship through weekly stakeholder meetings to define needs and arrange data access, built the datasets and project blueprint before the quarter, managed timeline and deliverables, contributed the modeling, coached teammates from non-technical backgrounds, and presented results to stakeholders and their C-suite
Customer-experience and website-capability engagements for media and energy clients
Process design and optimization for point-of-sale (POS) delivery to businesses
Time-study analysis of manufacturing processes; job-shop layout design
Teaching
Healthcare Data for Researchers (PhD-level), Healthcare Business Analytics, and the Healthcare Analytics Lab, guided students working with large real-world CMS datasets and collaborative data-science tooling
Pricing and Revenue Management; Production Planning. TA'd the graduate pricing and revenue optimization course for three semesters: price-demand estimation, pricing under constrained supply, price differentiation, customized and dynamic pricing, markdown management, revenue management, network revenue management, and overbooking. Also teaching assistant for Production Planning
Education
Dissertation: Optimizing Primary Care Visit Patterns and Panel Design. Advisor: Prof. Daniel Adelman
Contact
Interested in collaborating or learning more?
Email Me