Trends and global distribution of malaria burdens
Malaria burden trends (2000-2022): (a) Global malaria cases (millions, left Y-axis) and the number of affected countries (black curve, right Y-axis). (b) Global malaria deaths with (left Y-axis) and the number of affected countries (black curve, right Y-axis). (c) Geographical distribution of malaria cases (millions). (d) Geographical distribution of malaria deaths.
The global trend in malaria burden from 2000 to 2022 highlights fluctuations in malaria cases and deaths in various regions. The number of malaria cases was highest, recorded in 2022 (251.75 million), while malaria-related deaths peaked in 2020 (99,554) (Table S3). Over the years, Nigeria (1,332.99 million) reported the highest malaria cases, followed by the Democratic Republic of the Congo (623.16 million) and India (31983 million). Meanwhile, South Sudan (149,753), Zambia (143,546) and the Central African Republic (124,801) experienced the highest malaria-related mortality rates. Countries such as Burkina Faso (485.6, 194.5), Sierra Leone (404.59, 188.3), and Niger (355.45, 154.6) showed the most average malaria incidence and mortality rates. Additionally, Bolivia (plural state) (41,612) and the Republic of Korea (37,169) together with Guyana (42,021) and Afghanistan (37,492) reported the highest indigenous malaria cases. Falciparum while Malaria Plasmodium vivax Cases were most common in Bolivia (plural state) (51,034) and Eritrea (44,310) (Figures 1 and 2 and Tables S3 – S6).
Spatial patterns of malaria incidence and mortality rates
(a) Moran II test (per 1000 population), (b) Moran mortality (per 100,000 population), (c) Malaria incidence (per 100,000 population) using Getis-ord Gi* (d) Malaria mortality (hot spot detection of malaria mortality per 100,000 population).
Analysis of spatial autocorrelation using local moran I identifies statistically significant spatial clusters of malaria incidence and mortality. High-clusters of malaria incidence (high regions surrounded by other high values) were observed in Burundi (1.562, p-value = 0.029), Benin (2.593, p-value < 0.001), and Burkina Faso (3.751, p-value = 0.001). Similarly, clusters with high malaria mortality were found in Benin (0.713, p-value = 0.001), Central African Republic (0.793, p-value = 0.012), and Liberia (3.055, p-value = 0.001) (Tables S7-S8). These findings highlight areas with high malaria incidence and mortality rates, indicating the areas of interest for the intervention (Figure 3). Gi* statistics further identifies hotspots of malaria incidence in Benin (3.655, <0.001), Burkina Faso (3.205, P-Value = 0.001), Central African Republic (2.534, P-Value = 0.011), Congo (2.676, p-value = 0.007) (2.534, p-value = 0.011), Burkina Faso (3.205, p-value = 0.001). 0.001). Hotspots of malaria mortality were identified in Benin (3.377, p-value = 0.001), Central African Republic (2.525, p-value = 0.012), Democratic Republic of the Congo (2.094, p-value = 0.036), and Ghana (3.448, p-value = 0.001) (Table S6). Other regions either showed insignificant clustering or were identified as spatial outliers (Fig. 3).
Correlation analysis of malaria and its determinants
There is a significant negative correlation between malaria incidence and arable land (-0.64), suggesting that areas with arable land tend to place a lower malaria burden. The average life expectancy and the prevalence of hypertension are positively correlated (r = 0.48), access to basic drinking water services is negatively associated with malaria incidence (r= -0.49). Furthermore, air pollution is positively correlated with malaria incidence, with healthcare infrastructure and economic factors showing mixed effects, highlighting the complex interactions of determinants that shape the global malaria burden (Figures S1 and Table S9).
Select the best model
Among the models evaluated, Xgboost demonstrated the best performance on RMSE (0.63), R² (0.93), Adjusted R² (0.92), and MAE (0.46). This is significantly outperforming other models, such as the Naive Bayesian model (RMSE: 2.36, R²: -0.03, adjusted R²: -0.04, MAE: 2.22) and SVM (RMSE: 1.05). DT (RMSE: 1.31, R²: 0.68, Adjusted R²: 0.65, MAE: 0.94) also performed moderately, while LightGBM (RMSE: 0.68, R²: 0.91, Adjusted R²: 0.91, MAE: 0.5) worked well but not strong. The XGBoost model was integrated with XAI and CAI technologies to enhance interpretability and causal analysis, improving reliability and transparency of the results (Figure S2 and Table S10).
The importance of function in predicting malaria incidence and mortality rates
A critical analysis of the ability to utilize mean shap values identifies the 10 most important variables in predicting malaria incidence and mortality. Key features include the proportion of the population using at least basic hygiene services (1.003), at least basic drinking water services (0.426), and population growth (0.339). Access to electricity (0.318), percentage of farmland (0.132), and number of physicians per 10,000 population (0.159) are also important contributors. More important features include total population (0.102), hospital bed density (0.141), air pollution (0.065), and mortality rate for under-5 years (0.051). These factors contribute significantly to the model's predictive capabilities and highlight the complex interactions between health outcomes, environmental variables, demographic and socioeconomic factors, and health system infrastructures regarding malaria incidence and mortality (Figure 4 and Table S11).
(a) Summary Plot: Most SHAP Prediction Features in the Top 10. (b) ranking of function classes, (c) casual relationships between the top 10 malaria features, and (d) importance of function (using CAI). Y1: Doctors (per 10,000 people); Y2: Population using at least basic drinking water services (%); Y6: GDP growth (per cent of annual); Y7: Hospital bed density (per 10,000 people); Y9: Pharmaceutical personnel density (per 10,000 people); Y12: Mortality rate under 5 years of age per 1000 people born. Y14: Access to electricity. Y15: Land area (square km); Y18: Forest area (%); Y19: Air pollution;
The categories that contribute most to the model's predictive power are health outcomes and environmental factors (57.15%), followed by demographic and socioeconomic factors (20.99%) and health systems and infrastructure (19.52%) (Figure 4 and Table S12). Structural equation modeling (SEM) results further emphasize the importance of these variables. The most notable factors include access to electricity (-1.145), indicating a lower incidence of malaria in areas with higher electricity access. Population growth (0.892) is positively related, suggesting that higher population growth is associated with increased malaria incidence. Mortality rate under the age of 5 (0.906) is also a significant predictor, with higher mortality rates associated with higher incidence of malaria. Other factors such as the number of doctors per 10,000 population (-0.199), hospital bed density (-0.592), and air pollution (0.678) play important roles. These findings highlight the complex interactions of health, environmental, demographic, and infrastructure factors that influence malaria incidence and mortality (Figure 4 and Tables S13, S14).
Accuracy and future trends in predicting malaria incidence and mortality rates
The prediction accuracy for the 2022 model is very high, with R-squared values (0.969) and area under the curve (AUC) (0.834) (Table S15). These metrics confirm the robustness of the model in capturing variability and the ability to distinguish malaria incidence and mortality (Figures 5, S3, and Table S15).
Future forecasts for (a) actual and (b) predicted malaria incidence (per 1000 population) 2022, (c) malaria incidence (per 1000 population) and (D) malaria mortality (per 100,000 population) (2023-2040). The left Y-axis shows the average mortality rate. The right Y-axis indicates the affected countries (black curve).
The forecast for 2023-2040 shows concerns about an upward trend in malaria incidence (per 1,000 population) and mortality (per 100,000 population). The average incidence rate (per 1000 population) is projected to rise from 81.83 in 2023 to 160.22 in 2040, affecting more and more countries. Similarly, the average mortality rate (per 100,000 population) is expected to increase from 19.14 in 2023 to 56.12 in 2040. This escalation underscores the urgent need for an intensified malaria control effort (Figure 5). Countries predicted to have the highest average malaria incidence (per 1000 population) for 2023-2024 are Burundi (828.52) and the Solomon Islands (804.86), while Liberia (252.55) and SAO Tome and Principe (238.62) are expected to experience the same period (100,000 population (100,000 population). These findings provide important insights for policymakers and health authorities to implement targeted interventions to effectively allocate resources and mitigate the impact of malaria in the most affected areas (Figure 5 and Tables S16, S17).
