Using machine learning model to improve perinatal outcomes in Zanzibar

In February 2020, D-tree International working in partnership with Zanzibar ministry of health and N/lab at the University of Nottingham launched a two-year machine learning initiative which aimed to develop and deploy a machine learning model to identify pregnant women from Zanzibar community health program who were at risk of their child dying during and after birth. In this project, a machine learning model was integrated into the Jamii ni Afya app to guide community health workers (CHWs) to identify high risk pregnant women. For those pregnant women who were identified as high risk, the Jamii ni Afya app would prompt CHWs to deliver additional services like additional CHW pregnancy follow up visits, tailored care and counselling to help decrease the risk.

The machine learning model was deployed in North A and North B districts in Zanzibar, 186 CHWs and 20 supervisors were trained on the model based risk assessment and the additional care to be provided to pregnant mothers. The model worked well offline and it showed that CHWs could be assisted by the technology in the prioritisation and personalization of care to 947 pregnant women involved. When configuring the machine learning model into the Jamii ni Afya app; it was difficult to run the model on a form and include the results on the same form and the model code also had a substantial size for the app framework. To help address some of the technical challenges encountered, Medic and D-tree have collaborated to co-develop the extension libraries function, the functionality makes it possible to run JavaScript code from a form and return values back to the form to help inform a CHT user further actions. Additionally using the extension libraries functionality, it is possible to share logic between the forms and parts of the configurations.

The CHT extension libraries function is now available in CHT v4.2.0, this feature now makes it easier for CHT App Developers to configure machine learning models in CHT to help provide real time decision support to health care workers. Thank you @nKataraia and @jkuester for the great presentation and demo in our July 2023 CHT Round-up call, you can find more details about the Jamii ni Afya machine learning project presentation and the extension libraries demo in the recording here.