Investigators examined routine monthly data from Uganda’s DHIS2 platform spanning January 2016 to December 2025. The number of annual confirmed stroke cases rose from 3,089 in 2016 to 8,449 in 2025. Deaths recorded in facilities increased from 342 to 908 over the same interval. These absolute increases translated to population rates that roughly doubled: incidence per 100,000 increased from 8.08 to 16.65, while mortality per 100,000 rose from 0.90 to 1.79. Short sentences clarify magnitude. The study team observed that these trends were consistent on graphical inspection and supported by formal time-series modeling.
The authors modeled monthly counts using Bayesian negative binomial regression to address over-dispersion in facility-reported counts. Time (in months) was the principal predictor and monthly interpolated population figures were entered as an offset (log-population), allowing coefficients to be interpreted as rate changes. Exponentiated time coefficients were reported as incidence rate ratio (IRR). The reported IRRs indicated a small but statistically significant monthly increase: IRR 1.005 for confirmed cases (95% CI 1.004–1.006) and IRR 1.006 for deaths (95% CI 1.005–1.007), both with p < 0.001. The investigators tested model adequacy with autocorrelation diagnostics and used posterior-based inference to quantify uncertainty.
For forecasting, the authors compared Naive, ARIMA, and exponential smoothing state-space (ETS) approaches. Model selection used RMSE and MAPE metrics, and a parsimonious ETS model was selected for reporting. Training used data through 2024, with 2025 reserved as a test year for out-of-sample performance. Forecast uncertainty was presented as 80% prediction intervals, reflecting an operational focus on plausible short-term planning ranges.
The team applied the Kitagawa decomposition to model-derived annual counts to partition the observed increases into two components: a population effect and an epidemiological effect. According to their estimates, the epidemiological effect—representing rises in incidence and mortality independent of population size—was approximately twice the magnitude of the population growth effect. This indicates that increasing rates of stroke, not population expansion alone, explain most of the burden growth. The decomposition outputs were expressed in both absolute counts and rates per 100,000 to preserve interpretability across stakeholders.
Routine DHIS2 counts showed within-year variability. Suspected stroke reports had recurrent short-term peaks and greater volatility; confirmed cases displayed smoother but upward trajectories. The authors ran STL decomposition to inspect trend and seasonality and used the tso() routine with auto.arima() to identify additive outliers. Detected outliers were replaced with ARIMA-adjusted values before formal analyses. Stationarity was evaluated using unit root testing and autocorrelation structure was assessed with Ljung–Box tests; these diagnostics informed model choice and residual checks. The source did not report disaggregated seasonal drivers (for example, specific months consistently higher after adjustment), only that seasonality was assessed visually and with decomposition plots.
Using the selected ETS framework, investigators projected that, absent effective prevention measures or major reporting changes, monthly confirmed stroke cases could reach approximately 927 by December 2028 (80% PI: 709–1,156). Monthly stroke deaths were forecast to reach about 126 by that date (80% PI: 94–160). The forecasting horizon covered 2026–2028 after model validation in the held-out 2025 data. Short-term prediction intervals were intentionally set at 80% to provide narrower operational bounds for planning. The source did not provide scenario-based forecasts incorporating potential interventions or changes in risk-factor prevalence.
Despite the rising counts of cases and deaths, the case fatality rate remained relatively stable over the study period, fluctuating between 9.9% and 12.4%. The authors noted that this stability suggests proportional increases in both incidence and in-facility deaths rather than a worsening in short-term fatality among admitted patients. From a service perspective, sustained rises in absolute admissions and deaths will increase demand for acute stroke care capacity, rehabilitation services, and long-term support—unless primary prevention or acute care improvements alter the trajectory.
The investigators acknowledged constraints intrinsic to routine DHIS2 data: variable data quality, possible completeness issues, and sensitivity to changes in reporting practices. The source did not provide age- or sex-stratified estimates, nor did it report community-level ascertainment outside facility-based counts. These limitations restrict causal interpretation and the ability to attribute trends to specific risk-factor dynamics. The authors recommend strengthened primary prevention, expanded acute stroke care capacity, and improved surveillance systems to mitigate the projected increases; the report did not include implementation details, intervention effect sizes, or timelines. Remaining questions documented by the source include how much of the epidemiological effect reflects true incidence rises versus improved detection or reporting, and which prevention strategies would most efficiently reduce the projected burden.