The widespread availability of cellphones has enabled non-profits to ship crucial well being data to their beneficiaries in a well timed method. Whereas superior functions on smartphones permit for richer multimedia content material and two-way communication between beneficiaries and well being coaches, easier textual content and voice messaging providers could be efficient in disseminating data to giant communities, notably these which might be underserved with restricted entry to data and smartphones. ARMMAN1, one non-profit doing simply this, is predicated in India with the mission of enhancing maternal and baby well being outcomes in underserved communities.
 |
| Overview of ARMMAN |
One of many packages run by them is mMitra, which employs automated voice messaging to ship well timed preventive care data to anticipating and new moms throughout being pregnant and till one yr after start. These messages are tailor-made in keeping with the gestational age of the beneficiary. Common listenership to those messages has been proven to have a excessive correlation with improved behavioral and well being outcomes, akin to a 17% improve in infants with tripled start weight at finish of yr and a 36% improve in girls understanding the significance of taking iron tablets.
Nevertheless, a key problem ARMMAN confronted was that about 40% of ladies regularly stopped partaking with this system. Whereas it’s doable to mitigate this with reside service calls to girls to elucidate the benefit of listening to the messages, it’s infeasible to name all of the low listeners in this system due to restricted assist employees — this highlights the significance of successfully prioritizing who receives such service calls.
In “Area Examine in Deploying Stressed Multi-Armed Bandits: Helping Non-Income in Bettering Maternal and Youngster Well being”, revealed in AAAI 2022, we describe an ML-based answer that makes use of historic information from the NGO to foretell which beneficiaries will profit most from service calls. We deal with the challenges that include a large-scale actual world deployment of such a system and present the usefulness of deploying this mannequin in an actual examine involving over 23,000 members. The mannequin confirmed a rise in listenership of 30% in comparison with the present normal of care group.
Background
We mannequin this useful resource optimization drawback utilizing stressed multi-armed bandits (RMABs), which have been nicely studied for utility to such issues in a myriad of domains, together with healthcare. An RMAB consists of n arms the place every arm (representing a beneficiary) is related to a two-state Markov resolution course of (MDP). Every MDP is modeled as a two-state (good or unhealthy state, the place the nice state corresponds to excessive listenership within the earlier week), two-action (corresponding as to if the beneficiary was chosen to obtain a service name or not) drawback. Additional, every MDP has an related reward operate (i.e., the reward gathered at a given state and motion) and a transition operate indicating the chance of shifting from one state to the following underneath a given motion, underneath the Markov situation that the following state relies upon solely on the earlier state and the motion taken on that arm in that point step. The time period stressed signifies that every one arms can change state no matter the motion.
 |
| State of a beneficiary could transition from good (excessive engagement) to unhealthy (low engagement) with instance passive and energetic transition possibilities proven within the transition matrix. |
Mannequin Improvement
Lastly, the RMAB drawback is modeled such that at any time step, given n complete arms, which ok arms needs to be acted on (i.e., chosen to obtain a service name), to maximise reward (engagement with this system).
The chance of transitioning from one state to a different with (energetic chance) or with out (passive chance) receiving a service name are due to this fact the underlying mannequin parameters which might be crucial to fixing the above optimization. To estimate these parameters, we use the demographic information of the beneficiaries collected at time of enrolment by the NGO, akin to age, earnings, schooling, variety of kids, and many others., in addition to previous listenership information, all in-line with the NGO’s information privateness requirements (extra beneath).
Nevertheless, the restricted quantity of service calls limits the info equivalent to receiving a service name. To mitigate this, we use clustering strategies to be taught from the collective observations of beneficiaries inside a cluster and allow overcoming the problem of restricted samples per particular person beneficiary.
Particularly, we carry out clustering on listenership behaviors, after which compute a mapping from the demographic options to every cluster.
 |
| Clustering on previous listenership information reveals clusters with beneficiaries that behave equally. We then infer a mapping from demographic options to clusters. |
This mapping is helpful as a result of when a brand new beneficiary is enrolled, we solely have entry to their demographic data and haven’t any data of their listenership patterns, since they haven’t had an opportunity to hear but. Utilizing the mapping, we will infer transition possibilities for any new beneficiary that enrolls into the system.
We used a number of qualitative and quantitative metrics to deduce the optimum set of of clusters and explored totally different combos of coaching information (demographic options solely, options plus passive possibilities, options plus all possibilities, passive possibilities solely) to attain essentially the most significant clusters, which might be consultant of the underlying information distribution and have a low variance in particular person cluster sizes.
 |
| Comparability of passive transition possibilities obtained from totally different clustering strategies with variety of clusters s = 20 (pink dots) and 40 (inexperienced dots), utilizing floor fact passive transition possibilities (blue dots). Clustering based mostly on options+passive possibilities (PPF) captures extra distinct beneficiary behaviors throughout the chance house. |
Clustering has the added benefit of lowering computational value for resource-limited NGOs, because the optimization must be solved at a cluster degree moderately than a person degree. Lastly, fixing RMAB’s is understood to be P-space laborious, so we select to unravel the optimization utilizing the favored Whittle index method, which in the end supplies a rating of beneficiaries based mostly on their probably good thing about receiving a service name.
Outcomes
We evaluated the mannequin in an actual world examine consisting of roughly 23,000 beneficiaries who have been divided into three teams: the present normal of care (CSOC) group, the “spherical robin” (RR) group, and the RMAB group. The beneficiaries within the CSOC group observe the unique normal of care, the place there are not any NGO initiated service calls. The RR group represents the state of affairs the place the NGO typically conducts service calls utilizing some systematic set order — the concept right here is to have an simply executable coverage that providers sufficient of a cross-section of beneficiaries and could be scaled up or down per week based mostly on accessible assets (that is the method utilized by the NGO on this specific case, however the method could fluctuate for various NGOs). The RMAB group receives service calls as predicted by the RMAB mannequin. All of the beneficiaries throughout the three teams proceed to obtain the automated voice messages impartial of the service calls.
 |
| Distributions of clusters picked for service calls by RMAB and RR in week 1 (left) and a pair of (proper) are considerably totally different. RMAB could be very strategic in selecting just a few clusters with a promising chance of success (blue is excessive and pink is low), RR shows no such strategic choice. |
On the finish of seven weeks, RMAB-based service calls resulted within the highest (and statistically important) discount in cumulative engagement drops (32%) in comparison with the CSOC group.
 |
| The plot reveals cumulative engagement drops prevented in comparison with the management group. |
| RMAB vs CSOC | RR vs CSOC | RMAB vs RR | |
| % discount in cumulative engagement drops | 32.0% | 5.2% | 28.3% |
| p-value | 0.044 | 0.740 | 0.098 |
Moral Issues
An ethics board on the NGO reviewed the examine. We took important measures to make sure participant consent is known and recorded in a language of the group’s selection at every stage of this system. Information stewardship resides within the fingers of the NGO, and solely the NGO is allowed to share information. The code will quickly be accessible publicly. The pipeline solely makes use of anonymized information and no personally identifiable data (PII) is made accessible to the fashions. Delicate information, akin to caste, faith, and many others., will not be collected by ARMMAN for mMitra. Subsequently, in pursuit of making certain equity of the mannequin, we labored with public well being and subject consultants to make sure different indicators of socioeconomic standing have been measured and adequately evaluated as proven beneath.
 |
| Distribution of highest schooling obtained (prime) and month-to-month household earnings in Indian Rupees (backside) throughout a cohort that obtained service calls in comparison with the entire inhabitants. |
The proportion of beneficiaries that obtained a reside service name inside every earnings bracket fairly matches the proportion within the general inhabitants. Nevertheless, variations are noticed in decrease earnings classes, the place the RMAB mannequin favors beneficiaries with decrease earnings and beneficiaries with no formal schooling. Lastly, area consultants at ARMMAN have been deeply concerned within the improvement and testing of this technique and have offered steady enter and oversight in information interpretation, information consumption, and mannequin design.
Conclusions
After thorough testing, the NGO has at the moment deployed this technique for scheduling of service calls on a weekly foundation. We’re hopeful that this may pave the way in which for extra deployments of ML algorithms for social affect in partnerships with non-profits in service of populations which have up to now benefited much less from ML. This work was additionally featured in Google for India 2021.
Acknowledgements
This work is a part of our AI for Social Good efforts and was led by Google Analysis, India. Due to all our collaborators at ARMMAN, Google Analysis India, Google.org, and College Relations: Aparna Hegde, Neha Madhiwalla, Suresh Chaudhary, Aditya Mate, Lovish Madaan, Shresth Verma, Gargi Singh, Divy Thakkar.
1ARMMAN runs a number of packages to offer preventive care data to girls by way of being pregnant and infancy enabling them to hunt care, in addition to packages to coach and assist well being staff for well timed detection and administration of high-risk situations. ↩
