My research uses smartphone sensing, cohort data and causal methods to study mobility and cognition in aging and stroke.
Using smartphones to capture life-space mobility in older adults
Question. How does outdoor temperature change how far older adults travel from home?
Data. Passive GPS from 82 older adults in the REGARDS cohort, collected over two weeks on their own iPhones.
Methods. A processing pipeline for noisy consumer GPS (gaps, location jitter, irregular sampling); life-space measures including ellipse area and total distance traveled; seasonal temperature analysis.
Result. Life space contracted sharply once daily average temperatures rose above about 29°C (84°F), the first study to show that heat constricts life space.
My role. Designed the GPS processing pipeline and developed the life-space measures.
Outputs. Innovation in Aging, 2026 · JMIR Formative Research, 2026 · Press: San Francisco Chronicle, Stanford Medicine
Cardinal LifeSpace: smartphone digital phenotyping after stroke
Question. Can an iPhone app measure real-world mobility and well-being in stroke survivors reliably enough to use as an outcome?
Data. Cardinal LifeSpace, an iPhone app built on Stanford Spezi, collecting GPS, Apple HealthKit and ecological momentary assessment (EMA) data from stroke survivors.
Methods. Life-space metrics from GPS (standard deviational ellipse, DBSCAN location entropy, time out of home, maximum distance from home); test–retest reliability with ICCs; HealthKit data handled through Firebase and BigQuery.
My role. Designed and lead the study, including the EMA instruments and passive data streams, in collaboration with the Odden and Buckwalter labs and Stanford Biodesign. Funded by an NIH F31 fellowship from the National Institute on Aging.
Code and app. GitHub repository (contributor) · App Store. Open-source iOS app built on Stanford Spezi.
Status. Dissertation Aim 2; manuscript in preparation.
Pre-stroke mobility and cognition after stroke (Cardiovascular Health Study)
Question. Does mobility before a stroke predict how cognition changes after it?
Data. The Cardiovascular Health Study, a cohort with 25+ years of follow-up.
Methods. Exploratory factor analysis; linear mixed models; inverse probability of attrition weighting with SuperLearner.
Status. Dissertation Aim 1; manuscript in preparation.
Related first-author work in CHS. Immune cell subsets and longevity, J Gerontol A, 2025 · Hearing, social networks and depressive symptoms, Age and Ageing, 2022
Also...
Strategic Advisor, Data & Technology, Health in Climate AI hackathons. London Climate Action Week June 2026 at UCL, and "Resilient People. Resilient Places." in New York (September 2026) at Cornell Tech, with the NYC Department of Health's Division of Mental Hygiene and McKesson as partners.
Consultant, Phase IIa trial operations, Halo Biosciences. Built the investigational drug supply process for two clinical trials. Developed analysis pipelines.
Methods & tools
Causal inference. Marginal structural models, IPW for treatment and attrition, propensity scores, mediation, SuperLearner. Used in: CHS post-stroke cognition; guest lecture on MSMs for causal mediation (Stanford EPI 227).
Longitudinal modeling. Linear mixed models, GAMMs, survival analysis, EFA/PCA. Used in: CHS, PRIMA.
Sensor and mobile data. GPS processing pipelines, Apple HealthKit via Firebase/BigQuery, EMA, Google Street View imagery. Used in: heat and life-space study, Cardinal LifeSpace, PRIMA.
Machine learning. XGBoost, random forest, penalized regression, clustering; fairness evaluation of clinical ML. Used in: research and coursework (Stanford BIOMEDIN 223).
Clinical trial operations. Investigational drug supply, IND drafting support, IRB submissions. Used in: Halo Biosciences (SATURN, SOLIS).
Code. R (tidyverse, lme4, mgcv), Python (pandas, scikit-learn), SQL.
Also familiar. EHR data (MIMIC in coursework; STARR, OMOP); PyTorch and self-supervised learning (independent project on WESAD).