The research agenda behind the platform

The programme treats surveillance, environmental data and decision science as one analytical chain rather than three disciplines. The first requirement is an auditable historical record; the second is a defensible account of what drives transmission; the third is the ability to test an intervention or financing decision before it is made. Each layer is built only on top of a validated version of the one beneath it, which is why the current release ships surveillance alone.

1. 01

   Intervention efficiency and allocative decision-making

   Where does an additional net, IRS campaign or IPTp contact avert the most cases? Coverage series are read against burden trajectories to identify settings where coverage gains have not translated into incidence decline, and to characterise the conditions under which they do.

   → Country typologies · efficiency frontiers
2. 02

   Surveillance quality and reporting integrity

   Confirmation rate, RDT-to-microscopy mix and ACT-to-confirmed-case ratios are analysed jointly as measures of case-management and reporting integrity rather than service volume, including detection of implausible reporting discontinuities.

   → Quality indices · anomaly detection
3. 03

   Financing, resourcing and outcome divergence

   Domestic malaria expenditure indexed against burden identifies divergence between resourcing and results — a starting point for programme diagnostics and financing dialogue, not a verdict in isolation.

   → Financing–outcome analyses
4. 04

   Environmental and climate determinants of transmission

   Rainfall, temperature and vegetation dynamics as lagged covariates of transmission intensity, with attention to the spatial and temporal alignment problems that make naive correlation misleading at national scale.

   → Covariate alignment methods
5. 05

   Explainable AI for risk forecasting

   Short-horizon national risk models with out-of-sample validation, prediction intervals and per-prediction feature attribution, developed under an explicit rule that no forecast is published without its validation record.

   → Validated forecast models
6. 06

   Digital twin simulation for intervention planning

   A calibrated simulation environment for intervention, financing and elimination scenarios, reporting the calibration period, the parameters varied and the sensitivity of every conclusion drawn from it.

   → Scenario simulation engine

• Every figure carries a named source, a documented aggregation rule and a visible count of reporting countries.
• Modelled estimates are labelled as estimates and never presented alongside observations without that distinction.
• Model outputs are published only with out-of-sample validation, stated horizon, error metrics and prediction intervals.
• Limitations are published in the same place as results, not in an appendix or omitted.
• Snapshots are versioned, so any figure cited from a prior release remains reproducible.

The current rules are documented in the methodology and every source is registered on the data provenance page.

FSI-AIBR works with national malaria control programmes, ministries of health, INGOs, multilateral agencies and academic partners. We are specifically interested in licence-compatible datasets that would allow the platform to operate below national level — sub-national case series, facility registers, entomological surveillance and programme campaign records — and in joint validation of forecasting outputs against in-country observed data.

• Sub-national risk stratification at admin-1 and admin-2 level where national programmes share data under agreement.
• Insecticide and antimalarial resistance surveillance as an explicit covariate in risk models.
• Cost-effectiveness modelling linking intervention mix to averted cases under budget constraints.
• Transfer of the platform architecture to tuberculosis, HIV and neglected tropical diseases once the malaria layer is validated end to end.