Research Abstract

I am interested in harnessing existing sources of data to improve the care of patients with gynecologic cancers, by providing credible evidence about the most effective and individualized care.

The treatment that a cancer patient receives is influenced by prediagnosis characteristics, disease factors, and provider attributes that may also affect the outcomes being studied. When enough confounders have been measured, it may be possible to accurately estimate causal associations between treatments and outcomes using a classical epidemiological approach. Unfortunately, it is common that the available data is missing information about important confounders. Unmeasured (and unmeasurable) confounding is a central threat to the validity of causal claims about cancer treatments when such claims are based on analyses of observational data.​

In my observational research, I use analytical approaches that can, sometimes, overcome unmeasured confounding by identifying natural experiments in cancer care delivery. Specifically, I am interested in how focusing on the variability of treatments across time, space, and provider can help to estimates treatment effects, and overcome treatment selection bias.​

I use methods like difference-in-differences, instrumental variables analysis, interrupted time series, and regression discontinuity designs to generate evidence about the causal effects of cancer treatments on survival, treatment-related toxicity, and medical costs. As a surgeon, I am especially interested in studying the effect of surgical care on cancer outcomes. ​

By generating credible clinical evidence from observational cancer data, I seek to help patients and oncologists make better and more informed treatment decisions.