Data-driven malaria prevalence prediction in large densely populated urban holoendemic sub-Saharan West Africa

Biobele J. Brown, Petru Manescu, Alexander A. Przybylski, Fabio Caccioli, Gbeminiyi Oyinloye, Muna Elmi, Michael Shaw, Vijay Pawar, Rémy Claveau, John Shawe‐Taylor, Mandayam A. Srinivasan, Nathaniel K. Afolabi, Geraint Rees, Adebola E. Orimadegun, Wasiu A. Ajetunmobi, Francis Akinkunmi, Olayinka Kowobari, Kikelomo Osinusi, F O Akinbami, Samuel Omokhodion, Wuraola A. Shokunbi, Ikeoluwa Lagunju, Olugbemiro Sodeinde, Delmiro Fernández-ReyesScientific 2020 · DOI

Abstract

Over 200 million malaria cases globally lead to half-million deaths annually. The development of malaria prevalence prediction systems to support malaria care pathways has been hindered by lack of data, a tendency towards universal "monolithic" models (one-size-fits-all-regions) and a focus on long lead time predictions. Current systems do not provide short-term local predictions at an accuracy suitable for deployment in clinical practice. Here we show a data-driven approach that reliably produces one-month-ahead prevalence prediction within a densely populated all-year-round malaria metropolis of over 3.5 million inhabitants situated in Nigeria which has one of the largest global burdens of P. falciparum malaria. We estimate one-month-ahead prevalence in a unique 22-years prospective regional dataset of > 9 × 10<sup>4</sup> participants attending our healthcare services. Our system agrees with both magnitude and direction of the prediction on validation data achieving MAE ≤ 6 × 10<sup>-2</sup>, MSE ≤ 7 × 10<sup>-3</sup>, PCC (median 0.63, IQR 0.3) and with more than 80% of estimates within a (+ 0.1 to - 0.05) error-tolerance range which is clinically relevant for decision-support in our holoendemic setting. Our data-driven approach could facilitate healthcare systems to harness their own data to support local malaria care pathways.

Introduction: The Need for Localized Surveillance

Malaria remains a global health crisis, particularly in sub-Saharan Africa, where the majority of the half-million annual deaths occur. While the World Health Organization identifies surveillance as a "core pillar" for eradication, current predictive systems are often "monolithic"—broad models designed to cover entire countries or continents. These generalized models often fail at the local level because malaria dynamics are hyper-sensitive to regional factors like urban density, specific rainy season patterns, and local healthcare-seeking behavior.

Furthermore, most models focus on long-term trends, whereas clinical providers need "short-lead" (one-month-ahead) predictions to manage immediate needs like blood bank supplies and staffing. This study addresses this gap by creating a bespoke, data-driven model for Ibadan, a metropolis of 3.5 million people in Nigeria where malaria is "holoendemic" (present year-round).

Geographical context of the study area in Ibadan, Nigeria.

Methods and Datasets

The researchers developed the Region-specific Elastic-Net based Malaria Prediction System (REMPS). The study utilized a "gold-standard" prospective dataset collected over 22 years (1996–2017) at the University College Hospital (UCH) in Ibadan, encompassing over 90,000 participants. The dataset included 15 variables: temporal data (year, month), demographics (median age of cases), clinical metrics (parasite density), and environmental data (rainfall and temperature).

The team evaluated several supervised machine learning approaches, including Generalized Linear Models (LASSO, Ridge Regression, Elastic Net), Ensemble Methods (Random Forests), and Support Vector Machines (SVM).

The machine learning pipeline for model selection and evaluation.

Elastic Net (EN) emerged as the superior choice. By combining the penalties of LASSO (L1) and Ridge Regression (L2), EN is exceptionally robust at handling datasets with highly correlated features—a common trait in epidemiological data where temperature, rainfall, and parasite density often fluctuate together. The model was tuned to predict the following month's prevalence based on input variables from the current and previous months. The data was split into a Training Set (1996–2014) and a Validation Set (2015–2017).

Key Insights: Prioritizing Clinical Utility

A central tenet of this research is that clinical relevance should outweigh mathematical perfection. The authors introduced a "clinically relevant error-tolerance range" of +0.1 to -0.05. In high-burden settings, a slight over-prediction (+0.1) is acceptable for resource preparedness, whereas under-prediction (-0.05) poses a significant risk to patient care.

By focusing strictly on Ibadan, the model captured local nuances, such as the specific transition dynamics between the eight-month rainy season and the dry season. Interestingly, the model also functioned as a "novelty detector." During validation, it significantly over-predicted prevalence for September 2017. Investigation revealed a nationwide healthcare strike that month, suggesting the model can identify anomalies in healthcare delivery.

Results and Performance

The REMPS model demonstrated impressive generalization on the 2015–2017 validation data:

  • Accuracy: Achieved a Mean Absolute Error (MAE) ≤ 0.06 and a Mean Squared Error (MSE) ≤ 0.007.
  • Reliability: More than 80% of predictions fell within the clinically relevant error-tolerance range.
  • Seasonal Robustness: During the long rainy season (April–November), the model was extremely robust, with 95% of predictions falling within the usable range.
  • Directional Agreement: The model successfully predicted both the magnitude and the "up or down" direction of prevalence shifts.

Conclusion and Future Directions

The primary challenge remains the transition periods, specifically March, where the shift from dry to rainy seasons creates complex dynamics that the model occasionally under-predicted. Future work may involve "feature enrichment"—incorporating more granular environmental or socio-economic data—to improve accuracy during these seasonal shifts.

The authors propose a distributed ledger network for future deployment. Instead of a single global database, individual regional health centers could maintain their own REMPS nodes. These nodes could share learned insights via a decentralized network, allowing for localized accuracy while contributing to a broader, more responsive global surveillance system. This data-driven approach facilitates healthcare systems in harnessing their own data to support local malaria care pathways effectively.

Summary updated

Abstract

Over 200 million malaria cases globally lead to half-million deaths annually. The development of malaria prevalence prediction systems to support malaria care pathways has been hindered by lack of data, a tendency towards universal "monolithic" models (one-size-fits-all-regions) and a focus on long lead time predictions. Current systems do not provide short-term local predictions at an accuracy suitable for deployment in clinical practice. Here we show a data-driven approach that reliably produces one-month-ahead prevalence prediction within a densely populated all-year-round malaria metropolis of over 3.5 million inhabitants situated in Nigeria which has one of the largest global burdens of P. falciparum malaria. We estimate one-month-ahead prevalence in a unique 22-years prospective regional dataset of > 9 × 10<sup>4</sup> participants attending our healthcare services. Our system agrees with both magnitude and direction of the prediction on validation data achieving MAE ≤ 6 × 10<sup>-2</sup>, MSE ≤ 7 × 10<sup>-3</sup>, PCC (median 0.63, IQR 0.3) and with more than 80% of estimates within a (+ 0.1 to - 0.05) error-tolerance range which is clinically relevant for decision-support in our holoendemic setting. Our data-driven approach could facilitate healthcare systems to harness their own data to support local malaria care pathways.
BibTeX
@inproceedings{brown2020data,
  author = {Biobele J. Brown and Petru Manescu and Alexander A. Przybylski and Fabio Caccioli and Gbeminiyi Oyinloye and Muna Elmi and Michael Shaw and Vijay Pawar and Rémy Claveau and John Shawe‐Taylor and Mandayam A. Srinivasan and Nathaniel K. Afolabi and Geraint Rees and Adebola E. Orimadegun and Wasiu A. Ajetunmobi and Francis Akinkunmi and Olayinka Kowobari and Kikelomo Osinusi and F O Akinbami and Samuel Omokhodion and Wuraola A. Shokunbi and Ikeoluwa Lagunju and Olugbemiro Sodeinde and Delmiro Fernández-Reyes},
  title = {{Data-driven malaria prevalence prediction in large densely populated urban holoendemic sub-Saharan West Africa}},
  booktitle = {Scientific},
  year = {2020},
  doi = {10.1038/s41598-020-72575-6},
}
Figure