Improving Health Information Access in the World's Largest Maternal Mobile Health Program via Bandit Algorithms

TL;DR

Kilkari, the world's largest maternal mHealth program, faces dwindling engagement and automatic dropouts under resource constraints. CHAHAK introduces a non-Markovian Time-Series Bandit framework for multi-action intervention planning, leveraging a Time-Series Forecasting model (favoring a lightweight $LSTM$ with history $h=4$ weeks) to predict future listenership and an MA-TARI index $m_{n,i}=u_{n,i}/v_n$ to rank interventions. The approach combines AVM time-slot optimization and non-Markovian multi-action scheduling, comparing ILP, Greedy, Random, and Control policies, with real Odisha data showing improvements in listenership and dropout prevention for ASHA and automated interventions. The results indicate the method scales toward national deployment in India and is applicable to other multi-action, non-Markovian RMAB problems, offering a data-driven path to maximize information access for marginalized communities.

Abstract

Harnessing the wide-spread availability of cell phones, many nonprofits have launched mobile health (mHealth) programs to deliver information via voice or text to beneficiaries in underserved communities, with maternal and infant health being a key area of such mHealth programs. Unfortunately, dwindling listenership is a major challenge, requiring targeted interventions using limited resources. This paper focuses on Kilkari, the world's largest mHealth program for maternal and child care - with over 3 million active subscribers at a time - launched by India's Ministry of Health and Family Welfare (MoHFW) and run by the non-profit ARRMAN. We present a system called CHAHAK that aims to reduce automated dropouts as well as boost engagement with the program through the strategic allocation of interventions to beneficiaries. Past work in a similar domain has focused on a much smaller scale mHealth program and used markovian restless multiarmed bandits to optimize a single limited intervention resource. However this paper demonstrates the challenges in adopting a markovian approach in Kilkari; therefore CHAHAK instead relies on non-markovian time-series restless bandits, and optimizes multiple interventions to improve listenership. We use real Kilkari data from the Odisha state in India to show CHAHAK's effectiveness in harnessing multiple interventions to boost listenership, benefiting marginalized communities. When deployed CHAHAK will assist the largest maternal mHealth program to date.

Figures (5)

• Figure 1: (a) Around 23% beneficiaries are unreached despite 9 attempts (b) Number of beneficiaries (from the set of those enrolled in January 2022) remaining in the program after each month
• Figure 2: Relative (with respect to $h=1$) improvement in log likelihood. The probability of observing the historical trajectories in our dataset increases as we increase the order $h$ of the underlying Markov model. This suggests non-Markovian behaviour for the beneficiaries in Kilkari.
• Figure 3: [Left] Average number of calls performed each week for each beneficiary until a call is successfully picked up. [Right] Fraction of beneficiaries for which the learned pickup rate of the best slot has converged (within a difference of $\leq 0.15$ from the true mean).
• Figure 4: Comparison of different policies for planning multi-action interventions as compared to Control group.
• Figure 5: [Left] Distribution of Multi-Action TAR Indices for beneficiaries that receive interventions. [Right] Prevention of dropouts compared to control.