Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Behavioral AI: Reinforcement Learning to Enhance Pharmacy Services

Ana Fernández del Río, Michael Brennan Leong, Paulo Saraiva, Ivan Nazarov, Aditya Rastogi, Moiz Hassan, Dexian Tang, África Periáñez

TL;DR

Adaptive Behavioral AI applies reinforcement-learning driven nudges to pharmacy operations in LMICs by delivering personalized item-pair recommendations through SwipeRx. The approach uses contextual bandits, specifically a Gauss-Gamma linear bandit with Thompson sampling, to maximize short-term pharmacy expenditure while evaluating via mixed-method impact analyses. Two experiments (XP1 and XP2) show small but positive expenditure effects and potential delayed benefits, validating the feasibility of RL-based adaptive interventions in real-world health apps. This framework promises scalable improvements in inventory management, public health responsiveness, and patient care across health systems that rely on digital tools.

Abstract

Pharmacies are critical in healthcare systems, particularly in low- and middle-income countries. Procuring pharmacists with the right behavioral interventions or nudges can enhance their skills, public health awareness, and pharmacy inventory management, ensuring access to essential medicines that ultimately benefit their patients. We introduce a reinforcement learning operational system to deliver personalized behavioral interventions through mobile health applications. We illustrate its potential by discussing a series of initial experiments run with SwipeRx, an all-in-one app for pharmacists, including B2B e-commerce, in Indonesia. The proposed method has broader applications extending beyond pharmacy operations to optimize healthcare delivery.

Adaptive Behavioral AI: Reinforcement Learning to Enhance Pharmacy Services

TL;DR

Adaptive Behavioral AI applies reinforcement-learning driven nudges to pharmacy operations in LMICs by delivering personalized item-pair recommendations through SwipeRx. The approach uses contextual bandits, specifically a Gauss-Gamma linear bandit with Thompson sampling, to maximize short-term pharmacy expenditure while evaluating via mixed-method impact analyses. Two experiments (XP1 and XP2) show small but positive expenditure effects and potential delayed benefits, validating the feasibility of RL-based adaptive interventions in real-world health apps. This framework promises scalable improvements in inventory management, public health responsiveness, and patient care across health systems that rely on digital tools.

Abstract

Pharmacies are critical in healthcare systems, particularly in low- and middle-income countries. Procuring pharmacists with the right behavioral interventions or nudges can enhance their skills, public health awareness, and pharmacy inventory management, ensuring access to essential medicines that ultimately benefit their patients. We introduce a reinforcement learning operational system to deliver personalized behavioral interventions through mobile health applications. We illustrate its potential by discussing a series of initial experiments run with SwipeRx, an all-in-one app for pharmacists, including B2B e-commerce, in Indonesia. The proposed method has broader applications extending beyond pharmacy operations to optimize healthcare delivery.
Paper Structure (17 sections, 2 figures, 1 table)

This paper contains 17 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Nudges for Pharmacists and Public Health
  3. The AI Platform
  4. The App: SwipeRx
  5. Methodology
  6. General Intervention Setup
  7. Contextual Bandits
  8. Item Pair Recommendation
  9. Impact Analysis
  10. RCT angle
  11. Bandit angle
  12. Recommendation success angle
  13. Experiments
  14. First item pair recommendation experiment (XP1)
  15. Second item pair recommendation experiment (XP2)
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Screenshots of the SwipeRx application.
  • Figure 2: Difference between mean accumulated daily expenditure between users in the adaptive arm vs. control for all users in XP2. Confidence intervals (90 %) are shaded. The black horizontal line is drawn across 0, and vertical lines represent the intervention period's beginning and end.