Work

Research

My research addresses a fundamental challenge in modern medicine: we know which treatments work, but millions of patients either never receive them or fail to use them as intended.

Browse publications C4HDS research group ↗

My team designs, rigorously tests, and scales solutions to the problem of suboptimal medication use — combining the tools of behavioral science with the power of artificial intelligence and digital technology. Our work spans the full lifecycle of prescribing: from helping patients take medications they need, to safely stopping medications they shouldn't be taking.

01
Primary focus

Medication adherence

Why patients don't take their prescribed medications — and how to fix it at population scale. Despite robust evidence for many treatments, nearly half of patients with chronic conditions do not take their medications as prescribed. The consequences are profound: preventable hospitalizations, strokes, and deaths.

My team has tested a wide range of interventions in partnership with health insurers and delivery systems across the United States — including financial incentives that eliminate cost barriers, behavioral nudges drawing on insights from economics and psychology, pharmacist-led coaching, and reminder devices.

Selected trials
STIC2IT — A cluster-randomized trial (n=4,078) testing a pharmacist-led, behaviorally tailored intervention combining telephone consultations, text messaging, and automated report cards for patients with poorly controlled hyperlipidemia, hypertension, and diabetes. JAMA Internal Medicine, 2018.
REMIND — A randomized trial (n=53,480) evaluating low-cost reminder devices — pill organizers, blister packs, and vibrating reminder caps — as scalable tools to improve adherence to chronic disease medications. JAMA Internal Medicine, 2017.
ENGAGE-DM — A pragmatic randomized trial (n=1,400) of a pharmacist-delivered behavioral intervention — combining brief negotiated interviewing and shared decision-making — to improve medication adherence and glycemic control in adults with poorly controlled type 2 diabetes. PLOS ONE, 2019.
02
Emerging priority

Artificial intelligence & predictive analytics

Behavioral interventions work better when they are precisely targeted. A critical frontier in my work is predicting which patients will become non-adherent — before it happens — and using that prediction to deliver interventions to those most likely to benefit.

My team has developed and validated novel quantitative methods for clustering patients into longitudinal adherence trajectories, demonstrated the capacity to predict trajectory membership with high accuracy using claims data and electronic health records, and explored novel data sources — including retail purchasing information — to improve predictive ability. The application of reinforcement learning represents a particularly promising direction: algorithms that learn, in real time, which message content elicits adherence behavior for a given individual — and adapt automatically.

Selected work
REINFORCE — A randomized trial (n=60) testing a reinforcement learning algorithm that identified each patient's individual responsiveness to text message content and personalized messaging accordingly. Improved diabetes medication adherence by 13.6 percentage points versus control. npj Digital Medicine, 2024.
TARGIT — A 3-arm randomized trial (n=6,000) using predictive analytics to identify patients with type 2 diabetes at highest risk of insulin non-adherence and direct pharmacist-led outreach to those most likely to benefit. JAMA Network Open, 2019.
Data-driven prediction of health care spending — Applied group-based trajectory modeling to Medicare claims to characterize long-term spending patterns and demonstrate that trajectory membership — and future high cost — can be predicted with high accuracy from early data. JAMA Network Open, 2020.
Group-based trajectory models for medication adherence — Introduced group-based trajectory modeling to adherence science, enabling classification of patients into distinct longitudinal adherence patterns and prediction of trajectory membership from claims data. Medical Care, 2013.
03
Growing focus

Prescribing quality & deprescribing

The medication use problem is bidirectional. Just as too many patients fail to take medications they need, too many patients — especially older adults — continue taking medications that are potentially harmful, unnecessary, or no longer appropriate. Deprescribing these high-risk medications is a recognized priority, yet clinician behavior is difficult to change.

More broadly, improving prescribing quality requires addressing not just what patients take, but how and what clinicians prescribe — including reducing harmful or unnecessary prescribing and increasing appropriate preventive care. My work in this area applies behavioral science and electronic health record–based tools to change prescribing behavior at the point of care.

Selected work
NUDGE-EHR — A two-stage adaptive randomized trial (n=1,146) testing 14 behavioral science–informed EHR tools — including pre-commitment strategies and booster reminders — for deprescribing benzodiazepines and sedative hypnotics in adults aged 65 and older. Increased deprescribing by 40% versus usual care. JAMA, 2026.
NUDGE-EHR — design and rationale — Trial protocol describing the two-stage adaptive design, 14 candidate EHR interventions, and implementation strategy for the NUDGE-EHR trial — illustrating how behavioral science principles and adaptive methodology can be embedded within routine clinical workflows. Implementation Science, 2021.
MOTIVATE — A five-arm trial (n=228,000) with the White House Social and Behavioral Science Team demonstrating that a single behaviorally designed mailed letter increased influenza vaccination rates in Medicare beneficiaries — an example of behavioral science driving population-level engagement with evidence-based preventive health interventions. Nature Human Behaviour, 2018.
04
Cross-cutting theme

Health policy & implementation

Effective interventions mean nothing if they don't reach patients. My research examines the policy conditions under which behavioral interventions succeed and fail — including the role of benefit design, drug costs, and insurance structure in determining whether patients receive the medications they need.

This includes work on value-based insurance design (VBID), comparative effectiveness research, cost-effectiveness modeling, and the design of large-scale trials within health insurance systems. Much of this work has been conducted in direct partnership with health systems and insurers positioned to implement findings at scale.

Selected work
MI FREEE — A national randomized trial (n=5,855) demonstrating that eliminating medication copayments after myocardial infarction improved adherence, reduced major vascular events, and lowered patient out-of-pocket spending without increasing insurer costs. Aetna subsequently changed its national benefits policy. New England Journal of Medicine, 2011.
Value-based insurance design features and medication adherence — An observational study across multiple employer VBID programs identifying five structural features of benefit design associated with higher medication adherence rates. Health Affairs, 2014.
Randomized, controlled trials in health insurance systems — A methodological framework establishing how large-scale randomized trials can be embedded within health insurance systems, describing the unique opportunities and design considerations of this approach. New England Journal of Medicine, 2017.

300+ peer-reviewed publications

In NEJM, JAMA, Health Affairs, BMJ, and other leading journals.

Browse publications Google Scholar ↗