Can machine learning help anticipate sickle cell crises by combining patient information with regional climate conditions?
This study presents a proof-of-concept framework for predicting vaso-occlusive crises among people living with sickle cell anaemia across six Nigerian ecological regions.
The researchers generated a synthetic multimodal dataset containing 12,000 patient-month observations. It combined clinical and demographic variables with environmental factors such as temperature, humidity, rainfall, season, and fine particulate matter.
Five machine-learning models were evaluated: logistic regression, Random Forest, support vector machine, XGBoost, and a neural network.
Logistic regression achieved the strongest overall performance, recording an accuracy of 0.78, an F1-score of 0.74, and a receiver operating characteristic area under the curve of 0.84.
Random Forest followed closely with an ROC-AUC of 0.83, while the support vector machine, XGBoost, and neural network each achieved values of approximately 0.82.
The simulated regional analysis showed a pronounced northโsouth gradient. The North-East recorded the highest modeled crisis probability at 80.1%, followed by the North-West at 72.4% and North-Central at 49.2%.
Lower simulated probabilities were recorded in the South-West, South-East, and South-South, with the South-South showing the lowest value at 10.4%.
Temperature, particulate matter below 2.5 ยตm, humidity, region, and previous hospitalization emerged as influential predictors. Higher temperature and dust exposure were associated with increased modeled risk, while higher humidity showed a protective direction in the simulation.
The results demonstrate that both linear and nonlinear models can recover climate-linked crisis patterns embedded in synthetic data. They also highlight the potential value of integrating environmental monitoring with clinical information.
However, the authors emphasize that the findings are based entirely on synthetic data and should not be interpreted as validated clinical predictions or actual regional prevalence estimates.
Real-world validation using patient-level hospital records and contemporaneous meteorological data is required before the framework can support health surveillance, hospital preparedness, or patient-facing early-warning tools.
๐ Read the full article here:
https://doi.org/10.61298/pnspsc.2026.3.337
Published in: Proceedings of the Nigerian Society of Physical Sciences