AI Predictive Adherence Counseling (South Africa)

Palindrome and Right to Care, South Africa (September 2021)

Countries: South Africa

Intervention Description

Model that can predict whether a patient will miss their next scheduled appointment. Patient gets a tailored intervention based on the results.

Presenter: Tom Compton, Right to Care; Kieran Sharpey-Schafer, Palindrome

DUC Meeting Recording

Intervention Details

Data Elements	Viral load Demographics Clinical data (age, gender, viral load history, visit pattern)
Evidence	Compared to adherence scorecard Model performance: Loss to follow up: Accuracy – 66% Specificity- 94% Sensitivity -61% Graphs were presented on tracking LTFU at different stages of care
Technology Requirements / Interoperability	Machine learning and AI Patient risk score are used to assist workers in nudging towards intervention.
Calculations / Algorithms	Machine Learning Algorithm Total Adherence Score
Factors to Scale
Implementation Considerations	What are the impacts of telling someone they’re high risk?
Governance Considerations