Healthcare Provider vs. AI Communication and Chronic Disease Behavior in Rural India

How to cite

Rice, Lauren; Jha, Madhavi. 2026. "Healthcare Provider vs. AI Communication and Chronic Disease Behavior in Rural India." AEA RCT Registry. AEARCTR-0017602. https://doi.org/10.1257/rct.17602-1.3

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Lauren Rice* , Madhavi Jha

* Primary investigator

Registry ID
AEARCTR-0017602
DOI
https://doi.org/10.1257/rct.17602-1.3
First registered
Last updated
Status
In development
Trial dates
2026-01-10 – 2026-07-28
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17602

Abstract

This project investigates whether and how the quality of healthcare interactions affects patient health behaviors and outcomes through a randomized controlled trial with healthcare providers, an AI chatbot, and patients in rural West Bengal, India. We will randomly assign providers to receive communication skills training and cross-randomize patient access to an AI chatbot for blood pressure management advice. This design allows us to examine: i) whether the nature of patient-provider interactions affects patient behavior, ii) how AI substitutes or complements provider communication; and iii) what specific elements of communication lead patients to update their beliefs. We study these questions in the context of hypertension management in India, where 28% of adults have hypertension, yet only 12% achieve controlled blood pressure levels, and there is evidence of substantial gaps in both patient understanding of hypertension and medication adherence behaviors and minimal provider communication efforts.

Intervention

- Soft skill/communication training for healthcare providers
- AI voice-to-voice chatbot for hypertension management advice

Experimental design

Healthcare providers from our partner's network will be randomized into control (hard skill hypertension training only) or treatment (hard skill + soft skill training). Providers will be randomized within strata by geographic block. Providers will then host hypertension clinics, where high-BP patients will be randomized into three chatbot exposure arms: control (no chatbot), pre-provider interaction chatbot exposure, and post-interaction chatbot exposure.

We analyze heterogeneity by:
- Patient beliefs about provider and provider beliefs about patient
- Baseline provider medical and communication skills (by which we stratify)
- Patient characteristics: demographics, hypertension understanding and behavior

Early observation has revealed there is variation in engagement with both trainings and the chatbot. We thus will plan to analyze heterogeneity by treatment engagement and conduct a version of the analysis without un-engaged individuals.

Design details

Randomization method
Provider randomization done by computer, and patient randomization done by SurveyCTO
Randomization unit
Provider randomization is at the individual level, and patient treatment is at the individual level within provider group.
Clusters
200 providers
Observations
200 providers, 2400 patients
Arms
100 providers in control (hard skill training only) 100 providers in soft skill/communication training Within provider, patients randomized into chatbot arms: 800 in control (provider interaction only) 800 in pre-interaction chatbot exposure arm 800 in post-interaction chatbot exposure arm
Primary outcomes
- Healthcare provider behavior and soft skills - Patient-provider and patient-chatbot interaction quality (analyzed through video recordings or transcripts of interactions) - Provider and patient hypertension knowledge and beliefs - Patient hypertension management behavior and status
Secondary outcomes
- Markers of nature of communication (e.g. information disclosure, questions asked) - Patient beliefs about providers and provider choice - Provider beliefs about patients - Patient and provider self-beliefs

Cited by

No citation data available for this registration yet.

Keywords