Using Reinforcement Learning to Personalize Electronic Health Record Tools to Facilitate Deprescribing (REINFORCE-EHR)

The overall goal of the proposed research is to refine and adapt and perform efficacy testing of a novel reinforcement learning-based approach to personalizing EHR-based tools for PCPs on deprescribing of high-risk medications for older adults. The trial will be conducted at Atrius Health, an integrated delivery network in Massachusetts, and will intervene upon primary care providers. The investigators will conduct a cluster randomized trial using reinforcement learning to adapt electronic health record (EHR) tools for deprescribing high-risk medications versus usual care. 70 PCPs will be randomized (i.e., 35 each to the reinforcement learning intervention and usual care [no EHR tool] in each arm) to the trial and follow them for approximately 30 weeks. The primary outcome will be discontinuation or ordering a dose taper for the high-risk medications for eligible patients by included primary care providers, using EHR data at Atrius. The primary hypothesis is that the personalized intervention using reinforcement learning will improve deprescribing compared with usual care.

Study Overview

• Status: Completed
• Conditions: Aging
• Intervention / Treatment: Behavioral: Reinforcement learning

• Study Type: Interventional
• Enrollment (Actual): 1249
• Phase: Not Applicable

Contacts and Locations

Study Locations

• United States
  • Massachusetts
    • Boston, Massachusetts, United States, 02215
      • Atrius Health

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

The trial will intervene upon primary care providers (including physicians and PCP-designated nurse practitioners and physician assistants) at Atrius Health.

Patients of the PCPs will be included in the intervention and analysis if they are >/=65 years of age and have been prescribed >/= 90 pills of high-risk medications in the prior 180 days based on EHR data.

Exclusion Criteria:

• Not a primary care provider at Atrius Health

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Health Services Research
• Allocation: Randomized
• Interventional Model: Parallel Assignment
• Masking: Double

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: Reinforcement learning intervention; The intervention is a reinforcement learning program that personalizes EHR-based tools for PCPs to promote deprescribing high-risk medications over follow-up. The reinforcement learning intervention selects a tool for each provider based on an algorithm from an inventory of EHR tools and chooses tools that are predicted to motivate action for the individual provider. The effectiveness of each tool will be assessed on a selected interval based on whether a deprescribing action is taken by PCPs for eligible patients. The algorithm is trained to maximize these actions over time.	Behavioral: Reinforcement learning; The intervention is a reinforcement learning program that personalizes EHR-based tools for PCPs to promote deprescribing high-risk medications over follow-up. The reinforcement learning intervention selects a tool for each provider based on an algorithm from an inventory of EHR tools and chooses tools that are predicted to motivate action for the individual provider. The inventory of EHR tools from which the algorithm will choose include the following potential factors: open encounter, order entry, cold-state outreach, simplification, and risk framing.
No Intervention: Usual care; No EHR-based tools provided beyond those used in regular clinical practice.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Discontinuation or taper for high-risk medication	Deprescribing will be assessed using routinely collected data from the EHR system for eligible patients flagged as in need of deprescribing. The deprescribing outcome will be a "reduction" in inappropriate prescribing, defined as either discontinuation of one the the medication classes of interest or ordering a dose taper.	Through trial completion, up to 7 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Discontinuation of high-risk medication	Discontinuation of medication assessed using routinely-collected data from the EHR system for eligible patients flagged as in need of deprescribing.	Through trial completion, up to 7 months

Collaborators and Investigators

• Sponsor: Brigham and Women's Hospital
• Collaborators: National Institute on Aging (NIA); Atrius Health

Study record dates

Study Major Dates

• Study Start (Actual): August 11, 2025
• Primary Completion (Actual): March 27, 2026
• Study Completion (Actual): March 27, 2026

Study Registration Dates

• First Submitted: October 24, 2024
• First Submitted That Met QC Criteria: October 25, 2024
• First Posted (Actual): October 28, 2024

Study Record Updates

• Last Update Posted (Actual): April 2, 2026
• Last Update Submitted That Met QC Criteria: March 30, 2026
• Last Verified: November 1, 2025

More Information

Terms related to this study

• Keywords: Medication optimization; High-risk medications
• Additional Relevant MeSH Terms: Machine Learning; Artificial Intelligence; Algorithms; Mathematical Concepts; Reinforcement Machine Learning
• Other Study ID Numbers: 2024P002700; 2P30AG064199-06 (U.S. NIH Grant/Contract)

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No