Yi Li
University of Michigan, MI·Mathematics & Statistics
Hospital records and patient registries contain enormous amounts of information about how people fare after cancer treatment or surgery, but they were not collected as experiments. Patients at different hospitals differ, treatment choices vary by site, and the data are recorded in different formats.
- Fields
Mathematics & Statistics
Public Health & Epidemiology
Medicine & Clinical Research
- Topics
causal inference
machine learning
survival analysis
- Grants
- 2 known grants, through Apr 2029
What they work on
- Develop integrative causal machine learning methods that pool multiple high-dimensional observational cancer cohorts, using the Boston Lung Cancer Survival Cohort and the International Lung Cancer Consortium, to compare treatments despite unbalanced patient characteristics across sites.
- Build survival analysis methods for time-to-event cancer outcomes that give unconfounded, interpretable group comparisons across cohorts with heterogeneous data structures and site-varying treatment assignment, producing findings generalizable to broad patient populations.
- Create methods to predict how much opioid an opioid-naive patient will be prescribed after surgery and to quantify the uncertainty of those predictions, using Michigan Surgical Quality Collaborative data from 21,033 patients at 70 hospitals.
- Model individual opioid refill patterns after surgery from postoperative survey and perioperative data, to support data-driven prescription guidelines.
- Develop statistical methods to estimate how lung muscle metrics, muscle-based body-composition measures, influence lung cancer mortality in the Boston Lung Cancer Survival Cohort, and to detect new prognostic biomarkers.
More about the research
That makes it hard to say whether one treatment truly works better than another, or whether a finding at one hospital applies anywhere else. Statistical methods that handle these mismatches, and that can cope with thousands of measured variables per patient, are the focus of this work. Two strands run through the projects: drawing credible cause-and-effect conclusions from pooled observational cancer data, and predicting adverse outcomes for individual patients.
Why it matters
Lung cancer kills roughly 150,000 people a year in the United States, and opioid overdose deaths have tripled over a decade, with most prescription-linked. Methods that compare treatments fairly across hospitals could make cancer evidence apply to wider patient populations, while individualized forecasts of post-surgical opioid need could support prescribing guidelines that limit excess pills. Linking muscle measures to lung cancer survival may point to clinical, nutritional and physical interventions tailored to a patient.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Jennifer Waljee
Indiana University-Purdue University-Indianapolis, IN·Public Health & Epidemiology
Measure how state opioid prescribing limits affect opioid prescribing and adverse events after surgery, and how effects vary by policy feature, patient, procedure and prescriber, using commercial, Medicare and Medicaid claims with quasi-experimental methods.
Peng Liu
Iowa State University, IA·Mathematics & Statistics
Develop functional regression models for imaging outcomes built on a distributed learning framework, so that estimation and inference stay accurate yet scalable for large imaging datasets.
Hong Zhu
University of Virginia, VA·Mathematics & Statistics
Develop a copula-based model, a statistical tool that links the timing of two related events, with time-varying effects and random effects, for clustered semi-competing risks data such as cancer recurrence and death.
The school
University of Michigan: 124 labs on PI Pages, $2.1B a year on research, 963 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.