Research

Publications and Preprints

• Husar, K., Pittman, D. C., Rajala, J., Mostafa, F., & Allen, L. J. (2024). Lyme Disease Models of Tick-Mouse Dynamics with Seasonal Variation in Births, Deaths, and Tick Feeding. Bulletin of Mathematical Biology, 86(3), 1-38, Link.

• Husar, K., Volfovsky, A. (2026). DARTS: Targeting Prognostic Covariates in Budget-Constrained Sequential Experiments. Under review. arXiv:2605.06608.

• Husar, K., Pandey, M., Larsen, I. G., Pliego San Martin, J., Purohit, S., Hamelsky, J., Tang, B., Tackett, M., & Schramm-Sapyta, N. L. The Intersection of Cash Bail Reform, Serious Mental Illness, and Substance Use Disorders in a Southern County Jail. Under review.

• Husar, K., Volfovsky, A. Rerandomization with Missing Data (in preparation).

Research Experience

• Research Science Intern, Amazon, Summer 2026–Present
  Team: Supply Chain Optimization Technologies (SCOT)

  • Analyzed the production impact of design-stage stratification within the SCOT A/B testing platform to evaluate variance reduction and inform baseline experimental strategies.
  • Proposed a novel, easily implementable causal estimator that resolves zero-coverage degradation caused by subgroup unit depletion during concurrent experiments, while maintaining the production baseline under no depletion.
  • Built and launched an internal user-interface application that tracks treatment and control arm convergence after experiment rollouts and rollbacks, surfacing resource conflicts and identifying when units can re-enter the sampling pool; now being integrated into the primary experimentation platform.

• Research Assistant, Duke University, October 2025–Present
  Topic: DARTS: Targeting prognostic covariates in budget-constrained sequential experiments
  Advisor: Alexander Volfovsky

  • Developed a sequential experimental design method for trials where pre-treatment data must be collected under a fixed budget, learning across batches which covariates best predict the outcome and spending the budget on those.
  • Established that adapting data collection to earlier batches preserves randomization validity, so estimates and confidence intervals retain their guarantees, and that the budget is allocated near-optimally.
  • Across 11 simulation settings, reduced MSE by up to 72% (median 34%) over standard randomized designs, outperforming competing adaptive methods and closing much of the gap to the oracle design.

• Bass Connections Project, Duke University, Fall 2025–Spring 2026
  Topic: Cash bail reform, serious mental illness, and substance use disorders in a county jail
  Faculty leads: Nicole L. Schramm-Sapyta, Maria Tackett

  • First author on a study of a county cash bail policy change, linking jail detention records to health system data and applying interrupted time series and matched difference-in-differences analyses.
  • Showed that the reform largely formalized an existing trend toward release on recognizance, and that defendants with co-occurring mental illness and substance use disorders saw significantly higher rebooking rates after release.

• Research Assistant, Duke University, Spring 2024–Present
  Topic: Rerandomization with missing data
  Advisor: Alexander Volfovsky

  • Established theoretical guarantees showing that enhanced randomization strategies (rerandomization) improve the efficiency of treatment-effect estimation even when participant data are partially missing.
  • Ran large-scale simulation studies in R using cluster computing, covering 62+ experimental setups to assess real-world performance and achieved up to a 60% reduction in MSE compared to standard RCT designs.

• MARS Project, Duke University, Spring 2024
  Topic: Causal effect of migraine medication on retinal stroke risk
  Investigators: Jay B. Lusk, Brian Mac Grory, Fan Li, Lauren Wilson, Natalie Smith, Alonso M. Guerrero Castañeda, Kat Husar

  • Collaborated with a 7-member multidisciplinary team (including clinicians) on a study evaluating the effect of abortive migraine medication exposure on retinal stroke risk.
  • Used the Snowflake platform to clean, structure, and run complex SQL queries on 1M+ insurance and clinical records to create analysis-ready datasets.

• REU Program, Texas Tech University, Summer 2021
  Topic: Lyme Disease Models of Tick-Mouse Dynamics with Seasonal Variation in Births, Deaths, and Tick Feeding
  Advisor: Linda J. Allen

• Knots and Graphs Program, The Ohio State University, Summer 2019, Summer 2020
  Topic: Signed posets and a B-symmetric generalization of Stanley’s acyclicity theorem
  Advisor: Sergei Chmutov

Conference Talks and Posters

• ISBA World Meeting, Poster, Summer 2026
  DARTS: Targeting Prognostic Covariates in Budget-Constrained Sequential Experiments.

• Electronic Conference on Teaching Statistics (eCOTS), Breakout Talk, Summer 2026
  Breaking the Syntax Barrier: Empowering Students to Code in Any Language.
  with Marie Neubrander

• American Causal Inference Conference, Poster, Spring 2026
  DARTS: Targeting Prognostic Covariates in Budget-Constrained Sequential Experiments.

• American Causal Inference Conference, Poster, Spring 2025
  Rerandomization with Missing Data in Pre-Treatment Covariates.

• Duke StatSci Research Alumni Symposium, Poster, Fall 2024
  Rerandomization and Regression Adjustment in Studies with Missing Values in Pre-Treatment Covariates.

• Society for Mathematical Biology, Talk, Summer 2023
  Lyme Disease Models of Tick-Mouse Dynamics with Seasonal Variation in Births, Deaths, and Tick Feeding.

• Young Mathematicians Conference, Talk, Summer 2021
  Lyme Disease Models of Tick-Mouse Dynamics with Seasonal Variation in Births, Deaths, and Tick Feeding.

• Young Mathematicians Conference, Talk, Summer 2020
  Signed posets and a B-symmetric generalization of Stanley’s acyclicity theorem.