Spatial and spatiotemporal pattern of stroke relative risk in Ghana using Bayesian modelling approach

Stroke ranks as the second-leading cause of death and third in combined death and disability globally. In Ghana, there is a significant incidence of stroke, yet systematic reviews highlight a lack of comprehensive data on stroke in Sub-Saharan Africa, including Ghana. Additionally, information on the spatial and spatiotemporal distribution of stroke risk across Ghana's 16 regions remains limited. The objective of this study was to study the spatial and spatiotemporal distribution of stroke relative risk, identify regions with high-risk regions and associated predictors.

Stroke risk for each region was estimated using Bayesian spatial and spatiotemporal models, and these risks were mapped to visualize areas with elevated stroke relative risk. The Random Forest and Gradient Boosting models, nonparametric ensemble machine learning algorithms, were used to investigate the effects of potential predictors on stroke risk. The study utilized annual stroke data from 2018 to 2022 obtained from the Ghana Health Service (GHS) via the District Health Information Management System version 2 (DHIMS2) and utilized parameter estimates within the Integrated Nested Laplace Approximation via R software version 4.3.2.

Stroke relative risk decreases significantly over the study period. Some regions exhibited elevated risk, where the spatial model identified Volta, Central, Eastern, Bono, Upper East, Bono East, and Oti as high-risk regions, while the spatiotemporal model pinpointed Eastern, Ahafo, Bono East, Upper West, Savannah, Bono, and Western as high-risk regions. Clustering and variability in stroke risk were observed among regions. The study highlighted that gross national income significantly decreases the risk of stroke occurrence. While temperature and diabetes prevalence showed increased stroke risk, they were not statistically significant.

This study offers valuable insights that can inform resource allocation to regions experiencing elevated stroke risk. Identified high-risk regions can inform targeted screening strategies, referral pathway strengthening, and resource prioritization. Diagnostic capacity (including CT/MRI access), health-facility reporting quality, and surveillance system upgrades needed to reduce measurement bias and improve case ascertainment. Furthermore, we situate the findings within Ghana's existing noncommunicable disease (NCD) policy frameworks and describe how routine updates of the model using new DHIMS2 data can support ongoing decision-making.