Skip to main content

Research

Design at a glance

  • Use anonymised survey data from adolescents in Kenya and South Africa to build the foundation.
  • Create large “item banks” of mental health questions using multidimensional item response theory (IRT) which is a statistical method that shows which questions are most informative.
  • Run simulations to see how the adaptive test performs in practice.
  • Develop an easy-to-use digital prototype of the AfriCAT tool that can be tested on devices.

Approved Datasets

  • NAMHS–Kenya (national survey)
  • MMAP – Cape Town, South Africa
  • ALIVE – Cape Town, South Africa (baseline)
  • DoBAt – Mpumalanga, South Africa (baseline; no clinical validation sample)

Modelling and algorithmic approach:

Bifactor MIRT calibration of item banks (e.g., DISC-5, K-SADS, PHQ-A, GAD-7 items).

Adaptive selection and stopping rules to reach target precision.

Exploration of MoDN (modular clinical decision support networks) to enhance interpretability and flexibility.

Evaluation metrics:

Diagnostic accuracy: sensitivity, specificity, PPV, NPV, AUROC, Youden’s index, likelihood ratios.

Efficiency: items required; time to complete.

Validity: convergent and divergent.

Validity: convergent and divergent.

Lived-experience and stakeholder engagement:

Adolescents with lived experience, caregivers, educators, PHC workers, NGOs, policymakers.

Mixed methods: workshops, interviews, brief surveys across Nairobi, Cape Town, and Agincourt.

Goal: feasible, acceptable, appropriate implementation pathways.

Ethics and data governance:

Approvals from Wits HREC (Medical) and Aga Khan University.

POPIA, Kenya DPA, and AU Malabo Convention compliant.

Secure storage (MRC/Wits-Agincourt), strict access control, open-science commitment (open-access outputs; code and analytical dataset under OSI-compliant license where permissible).

Behind the technology

AfriCAT combines advanced data science with human insight. The team is building large banks of questions taken from trusted mental health tools and using modern psychometric methods to understand which questions give the most useful information about different levels of depression and anxiety.

Instead of giving every adolescent the same long questionnaire, AfriCAT uses adaptive algorithms that adjust in real time. Each answer guides the system to choose the next most relevant question until it reaches a precise result. This makes the test shorter, faster and more accurate.

The team is also exploring new ways to make digital mental health tools easier to understand and more flexible by using modular decision-support networks. This approach helps explain why certain questions are asked and how specific results are reached.

AfriCAT is evaluated on several criteria. These include how accurately it identifies depression and anxiety, how efficiently it performs by measuring how many questions are needed and how long the test takes, and how closely its results match gold standard clinical measures.

Just as important as the technical work is the contribution of people with lived experience. Adolescents who have experienced depression or anxiety, together with caregivers, teachers, primary healthcare workers, NGOs and policymakers, take part in workshops, interviews and surveys in Nairobi, Cape Town and Agincourt. Their involvement ensures that the tool is scientifically sound and also practical, relatable and usable in real settings.

All data used in the project are handled in accordance with strict ethical and governance processes. These include approvals from the Wits and Aga Khan University ethics committees, full compliance with South African, Kenyan and African Union data protection standards, and a commitment to open and transparent science.