2 Comments
Carolina Bastidas

In response to Provide evidence that this technology works.

Please expand on how the model accuracy in this prediction would translate into nurse burden, changes in priorities and decreased infant mortality. Thanks

Ruchit Nagar

We are working on predicting a set of health behavior and health outcomes. Our current models look at early infant malnutrition, and we are refining models on health outcomes: low birth weight, delivery complications of the mother and neonate, and maternal hospitalization during pregnancy and health behavior outcomes: complete antenatal care coverage and institutional delivery.

Before predicting health behavior and health outcomes for maternal and newborn health, we have a data quality layer which with over 90% accuracy can classify data of frontline health workers as fraudulent or trustworthy, based on statistical clustering around domain-specific rules, verified by field-visits over a 3-year, 150 health worker cohort data set. This data quality layer acts as a data filter prior to moving forward with subsequent layers of the neural network to predict our maternal and newborn health outcomes.

Practically for the frontline health worker, we plan to use the AI driven models to enhance our heuristically-driven high-risk scores, and accordingly color code and sort the due-lists of beneficiaries, which appear in the Khushi Baby mobile application, which has been deployed by the government for tracking the maternal child health continuum of care in Rajasthan. By prioritizing the due-list, we aim to reduce the burden of the frontline health worker and make her work more impactful, through early intervention.



 
    Back
to Top