Spatio-temporal Distribution and Relative Risk of Dengue Fever in Peninsular Malaysia Authors: Mohamad Afiq Amsyar Hamedin, Kamarul Imran Musa, Mohd Rahim Sulong Published: 2026 ABSTRACT ObjectivesDengue fever remains a major public health concern in Malaysia, with marked spatial heterogeneity and evolving temporal dynamics. Although the incidence of dengue has increased substantially in recent years, district-level risk patterns and their changes over time remain insufficiently characterized. This study aimed to assess the spatio-temporal distribution of dengue risk in Peninsular Malaysia.MethodsA retrospective analysis of dengue surveillance data was conducted using district-level monthly case counts from 2022 to 2024. Dengue risk was modeled with Bayesian spatio-temporal models incorporating spatial effects, temporal effects, and space–time interaction terms. Spatial variation was modeled using the Besag–York–Mollié 2 specification, while temporal trends were modeled using a first-order random walk. Competing models were compared using the deviance information criterion (DIC) and Watanabe–Akaike information criterion (WAIC). Relative risk (RR) and exceedance probability maps were generated from the best-fitting model.ResultsDengue incidence increased markedly after 2022, with incidence rates nearly doubling in 2023 and remaining high in 2024. Incidence maps showed pronounced spatial heterogeneity, with persistent high-incidence districts concentrated in urban and peri-urban regions. Spatio-temporal models incorporating space–time interaction substantially outperformed the additive model, with lower DIC and WAIC values. The best-fitting model identified the highest RRs in districts within Selangor, while several districts in Penang, and Johor had high posterior probabilities of elevated risk [P(RR>1)>0.8].ConclusionsThis study demonstrates substantial spatial heterogeneity and evolving temporal dynamics in dengue risk across Peninsular Malaysia. Bayesian spatio-temporal disease mapping provides robust, localized, time-specific estimates of dengue risk and may support targeted surveillance and intervention in endemic settings. INTRODUCTION Dengue fever remains a major public health challenge in tropical and subtropical regions worldwide. The global burden of dengue has increased substantially in recent decades, with an estimated 390 million infections occurring annually [1]. Malaysia is among the countries where dengue is endemic, and recurrent outbreaks place a considerable burden on public health systems [2]. Dengue risk exhibits pronounced spatial heterogeneity, with substantial variation across relatively small geographic areas [3,4]. This heterogeneity reflects complex interactions among environmental conditions, vector ecology, socio-demographic factors, and human behavior [5,6]. Analyses based on crude case counts or incidence rates may mask underlying spatial patterns, particularly when population sizes differ across administrative units. Therefore, spatial epidemiological methods that account for population structure and spatial dependence are essential for targeted disease control. Bayesian disease mapping approaches have been widely applied to model spatial variation in disease risk, particularly for infectious diseases characterized by unstable small-area counts [7]. The Besag–York–Mollié (BYM) model and its extensions allow spatial autocorrelation and unstructured heterogeneity to be modeled simultaneously, producing smoothed and interpretable estimates of relative risk (RR) [8]. Reparameterisations such as the BYM2 model further improve model identifiability and interpretability while retaining the capacity to capture spatial clustering in disease risk [9]. When extended to include temporal components, Bayesian spatio-temporal models integrate spatial and temporal effects, as well as space –time interactions, to characterise persistent and transient changes in disease risk over time. This feature is particularly relevant to dengue epidemiology, which is characterized by strong seasonal patterns and longer-term temporal trends [3]. Such models are therefore well suited to dengue surveillance because they can identify districts with consistently elevated risk. Unlike previous Malaysian studies that focused on either purely spatial risk estimation or descriptive spatio-temporal patterns without explicit space–time interaction, this study applies a fully Bayesian spatio-temporal framework with formal interaction structures and exceedance probability mapping to quantify district-level uncertainty in dengue risk [10–12]. This approach enables clearer visualization of the spatio-temporal evolution of dengue risk by identifying districts with persistently elevated RR and high exceedance probabilities. These model outputs may support the prioritization of intensified surveillance, targeted vector control, and resource allocation within Malaysia’s district-based dengue prevention framework. METHODS This retrospective secondary data analysis used 3 years of surveillance data collected from 2022 to 2024. District-level dengue data were obtained from eDengue, the Malaysian Ministry of Health’s management and monitoring system for dengue activity. Confirmed dengue cases in Peninsular Malaysia were defined by a positive non-structural protein 1 antigen test or dengue immunoglobulin M test and registration in the eDengue system. Cases were aggregated by administrative district and epidemiological week and then summarised into monthly counts for analysis [13]. Monthly aggregation was used to stabilise small-area counts, align with district-level operational reporting practices, and reduce overdispersion associated with weekly case counts. The study area was Peninsular Malaysia, which comprises 86 administrative districts with substantial heterogeneity in population density, urbanization, and environmental conditions. The region has a tropical climate characterized by relatively stable temperatures and seasonal rainfall patterns, both of which influence dengue transmission [14]. Districts were selected as the spatial unit of analysis to ensure compatibility with routine dengue surveillance and district-level prevention and control activities. Annual district-level population estimates were obtained from official population statistics [15]. Population counts were assumed to remain constant within each calendar year and were applied uniformly across all months of that year. Although short-term population mobility may occur, particularly in highly urbanized districts, annual population estimates are routinely used in district-level dengue surveillance and small-area disease mapping. Given the 3-year study period and the focus on relative rather than absolute risk, this assumption is unlikely to have materially affected district-level RR estimation. These population data were used to calculate expected case counts for risk standardization. Annual district-level population estimates were available separately for 2022, 2023, and 2024. For each district, the population estimate for a given year was assumed to remain constant and was applied to all 12 monthly observations within that year. Expected dengue case counts were calculated at the district-month level using internal standardization. A single pooled monthly reference incidence rate for the entire study period was calculated as: where Yit is the observed dengue case count for district i in month t, and Niy(t) is the district population estimate for the calendar year corresponding to month t. The expected count for district i in month t was then calculated as [16]: Because the year-specific population estimate was repeated across all monthly observations within the corresponding year, r(s) represents a pooled monthly reference rate. The resulting Eit values were used as the offset in the Bayesian spatio-temporal model. The RR of dengue for each district and month was modeled using the approach proposed by Knorr-Held [16], with the temporal index t =1,…,36 representing consecutive months from January 2022 to December 2024. Let Yit denote the observed number of dengue cases in district i, where i =1, …,86, during month t, where t =1, …,36 represents consecutive months from January 2022 to December 2024. The observed district-month counts were assumed to follow a Poisson distribution: Where, Here, Eit is the internally standardised expected dengue count for district i in month t, and RRit is the corresponding RR. Equivalently, where log (Eit) was included as an offset and ηit = log(RRit) was the linear predictor. The log-RR was specified as: where α is the overall intercept; si is the district-level spatial effect; ut is an unstructured temporal effect; vt is a structured temporal effect; and δit is the space–time interaction. The spatial effect si was modeled using the BYM2 parameterization [9]: where θi ~ N(0,1) represents unstructured district-level heterogeneity, ψi represents the scaled spatially structured component defined by the district adjacency graph, τs is the spatial precision parameter, and φ is the proportion of spatial variance attributable to structured spatial dependence. The unstructured temporal effects were assumed to be independent and identically distributed: The structured temporal effect followed a first-order random walk [17]: allowing dengue risk to vary smoothly between consecutive months. Four alternative space–time interaction structures were evaluated according to the Knorr-Held framework [16] to allow dengue risk to vary across districts over time. Four alternative space–time interaction models were considered, representing increasing flexibility in allowing district-specific temporal trajectories. These structures ranged from unstructured interactions to models incorporating structured dependence in space, time, or both, consistent with the heterogeneous and context-specific nature of dengue transmission dynamics. This hierarchical approach allowed evaluation of whether local temporal deviations in dengue risk were independent, spatially structured, temporally structured, or jointly structured. In addition, an additive model including only the main spatial and temporal effects and excluding the space–time interaction term was fitted as a baseline comparator. The specifications of all interaction structures are provided in Supplemental Material 1, and all five models were compared to identify the most appropriate structure for the data. Bayesian inference was conducted using the Integrated Nested Laplace Approximation approach [18]. Penalised complexity priors were assigned to the spatial and temporal random effects to limit overfitting and to favor simpler model structures unless greater complexity was supported by the data [19]. Model performance was assessed using the Deviance Information Criterion (DIC) and the Watanabe–Akaike Information Criterion (WAIC). Differences greater than 10 were considered indicative of meaningful improvement in model fit [20]. The model with the lowest information criterion values was selected as the best-fitting model and used for subsequent risk estimation and mapping. The model therefore estimated a separate posterior RR for each of the 86 districts at each of the 36 monthly time points. Posterior summaries of key hyperparameters from the selected spatio-temporal model were reported using posterior means and standard deviations (SDs). District-level RRs were estimated from the best-fitting model [21]. Monthly RR maps were generated to visualize the spatio-temporal evolution of dengue risk across Peninsular Malaysia. In addition, exceedance probability maps, defined as the posterior probability that the RR exceeded one (P[RR>1]), were produced to identify districts with strong statistical evidence of elevated dengue risk [22]. All analyses were conducted using R version 4.5.1 (R Foundation for Statistical Computing, Vienna, Austria). Ethics Statement The research followed the principles of the Declaration of Helsinki and received ethical clearance from the Human Research Ethics Committee of Universiti Sains Malaysia under approval code USM/JEPeM/KK/25010107. The Malaysia Medical Research and Ethics Committee also granted approval under code NMRR ID-25-00144-OVW. RESULTS Table 1 summarizes the dengue burden from 2022 to 2024. The total number of reported dengue cases increased markedly, from 58 770 cases in 2022 to 115 676 cases in 2023, and remained high in 2024, at 116 229 cases. A similar pattern was observed for mean monthly cases, which nearly doubled from 4897.5 cases per month in 2022 to approximately 9640 cases per month in 2023 and remained similarly high in 2024. Correspondingly, the crude dengue incidence rate increased from 223.8 per 100 000 population in 2022 to 431.1 per 100 000 in 2023, with a slight decrease to 426.0 per 100 000 in 2024. These findings indicate a substantial escalation in dengue transmission after 2022, followed by persistently high incidence in the subsequent years. Monthly dengue case counts from 2022 to 2024 are shown in Supplemental Material 2, illustrating substantial temporal variability and increased dengue incidence after 2022. Figure 1 presents the spatial distribution of crude dengue incidence rates by district for each year from 2022 to 2024. In 2022, higher incidence rates were observed in several urbanized districts, while most districts had relatively lower incidence. In 2023, dengue incidence increased substantially across a wider geographical area, with pronounced clustering of high-incidence districts, particularly in parts of the central and western regions of Peninsular Malaysia. This elevated incidence pattern largely persisted into 2024, although some districts showed reductions compared with the previous year. Overall, the maps reveal marked spatial heterogeneity and highlight the presence of persistent high-incidence areas as well as temporal changes in the geographic distribution of risk. Models incorporating space–time interaction demonstrated substantially better fit than the additive model, with DIC and WAIC values more than 300 points lower. Among the interaction models, the type II space–time interaction model (model 3) had the lowest DIC (18 473.29) and WAIC (18 367.16). This model was therefore selected as the best-fitting model and used for subsequent RR estimation and spatio-temporal mapping. Detailed results of the model comparison are provided in Supplemental Material 3. Table 2 shows posterior estimates of key hyperparameters from the selected spatio-temporal model. The spatial SD (σs=0.884) indicates substantial heterogeneity across districts. However, the relatively low spatial mixing parameter (ϕ=0.101) suggests that spatial heterogeneity in dengue risk was dominated by district-specific effects rather than strong spatial autocorrelation, indicating substantial localized drivers of transmission. Temporal variation was characterized by both unstructured (σu=0.074) and structured components, with the random walk of order 1 temporal SD (σt=0.170) indicating smooth temporal changes in dengue risk over the study period. The space–time interaction SD (σst=0.318) was larger than the temporal main-effect SDs, providing evidence of district-specific temporal deviations and supporting the inclusion of space–time interaction terms in the model. The overall RR map (Figure 2) indicates marked spatial heterogeneity in dengue risk across Peninsular Malaysia. Elevated RR was observed in several districts along the west coast, particularly Petaling, Klang, Gombak, Hulu Langat, and Kuala Lumpur, as well as in parts of southern Peninsular Malaysia, including Johor Bahru. In contrast, many districts in the northern and eastern regions exhibited lower RR. The exceedance probability map (Figure 3) further identifies districts with strong statistical evidence of elevated dengue risk. A threshold of P[RR>1]>0.8 was used to indicate a high posterior probability of excess risk, reflecting substantial statistical support rather than absolute certainty. Monthly RR and exceedance probability maps for 2022–2024 are presented in Supplemental Materials 4 and 5. These maps illustrate temporal fluctuations in dengue risk within districts across months while showing broadly consistent spatial patterns of elevated risk over the study period. DISCUSSION This study provides a comprehensive district-level spatio-temporal assessment of dengue risk in Peninsular Malaysia from 2022 to 2024 using a Bayesian disease mapping framework. By integrating spatial dependence, temporal trends, and space–time interaction, the analysis provides insights into changing dengue transmission dynamics, persistent high-risk areas, and the extent to which spatial heterogeneity shapes dengue risk beyond simple seasonal patterns. Figure 1 demonstrates a marked increase in dengue incidence after 2022, with incidence rates in Peninsular Malaysia nearly doubling in 2023 and remaining high in 2024. Abrupt changes in dengue incidence have also been reported in other endemic settings and have been hypothesized to coincide with shifts in dominant dengue virus serotypes or genotypes, which may alter population immunity and outbreak potential [1,23,24]. In Malaysia, periodic serotype replacement has been documented and associated with major outbreaks, particularly when a previously less prevalent serotype becomes dominant in a population with low herd immunity [25]. Although serotype data were not available for this analysis, the observed temporal pattern is consistent with previous reports suggesting that changes in serotype circulation can substantially influence short-term dengue trends independently of climatic variability [26]. Similar post-pandemic rebounds in dengue incidence have been reported across Southeast Asia, including Thailand, Indonesia, and Vietnam, after reduced transmission during the coronavirus disease 2019 (COVID-19) period [27]. These regional trends suggest that changes in population mobility, disruption of vector control, and viral dynamics may collectively contribute to dengue resurgence across tropical settings. The spatial analyses revealed pronounced heterogeneity in dengue risk across districts, with several areas exhibiting consistently elevated RR over the 3-year period. Persistent high-risk districts were predominantly located in highly urbanized and peri-urban regions, particularly districts within Selangor, Penang, and Johor. These district-level patterns are consistent with previous studies suggesting that higher population density and human mobility are associated with sustained dengue transmission in metropolitan and surrounding areas [28–30]. However, because of the ecological design of this study, these findings reflect area-level associations and should not be interpreted as indicating individual-level risk. Exceedance probability maps identified districts with strong statistical evidence of excess risk, including some peri-urban areas. This finding suggests that elevated dengue risk is not confined to dense urban cores but may extend to transitional zones where rapid land-use change, construction activity, and suboptimal vector control create favorable breeding conditions for Aedes mosquitoes [5,6,31]. Collectively, these results support spatially targeted interventions beyond city centres while acknowledging the ecological scope of inference. Model comparison indicated that spatio-temporal models incorporating interaction terms substantially outperformed the additive model, which assumed uniform temporal trends across districts. The selected type II interaction model allowed district-specific temporal deviations, thereby capturing localized changes in dengue risk beyond overall seasonal patterns. Epidemiologically, this finding suggests that dengue transmission dynamics vary across districts, likely reflecting differences in local vector control intensity, urban development, and population movement [6,17]. Similar approaches have improved risk estimation in other dengue-endemic settings [4,32,33]. Although dengue is often described as a seasonal disease influenced by rainfall and temperature, the present results indicate that dengue risk in Peninsular Malaysia is not entirely seasonally driven. The strong spatial heterogeneity and persistent high-risk districts observed in this study suggest that structural and environmental factors play an important role in shaping dengue transmission. Urban infrastructure, water storage practices, waste management, and socioeconomic conditions have all been shown to influence Aedes breeding and dengue risk independently of seasonal climate variation [6,13,30,32]. This observation is consistent with studies from other tropical regions showing that although climate modulates dengue transmission, spatial determinants often explain a substantial proportion of long-term risk patterns [34]. Consequently, dengue prevention strategies should integrate spatial risk profiling with seasonal preparedness measures. The combined use of RR and exceedance probability maps provides actionable information for dengue surveillance and control. Identifying districts with both elevated risk and high probability enables prioritization of intensified vector control activities, targeted community engagement, enhanced entomological surveillance, and pre-emptive resource deployment. Such spatial risk stratification may also support early warning efforts and more efficient allocation of limited public health resources in endemic settings [35]. Several limitations should be acknowledged. First, this study relied on routine surveillance data, which are subject to underreporting and variation in case detection across districts. Second, the absence of serotype, entomological, and detailed environmental data limited the ability to directly assess drivers of the observed patterns. Third, population estimates were assumed to be static within each year, which may not fully capture short-term population movement. Despite these limitations, the Bayesian spatio-temporal framework helped stabilise small-area risk estimates through spatial and temporal smoothing, reduce instability from sparse counts, and explicitly quantify uncertainty through posterior distributions and exceedance probabilities. During the 2022–2024 study period, dengue risk in Peninsular Malaysia exhibited marked spatial heterogeneity and district-specific temporal variation, highlighting the importance of accounting for localised transmission dynamics when assessing dengue burden. By providing population-adjusted and uncertainty-informed estimates, the combined use of RR and exceedance probability mapping offers a practical framework for identifying districts with elevated risk and strong statistical evidence of excess transmission. This approach may support the prioritisation of targeted vector control, enhanced surveillance, community engagement, and resource allocation. The findings also suggest that dengue risk is shaped not only by seasonal factors but also by persistent spatial determinants related to urbanisation and local environmental conditions. Future studies incorporating serotype, entomological, and environmental data may further clarify the drivers of these patterns and strengthen early warning capabilities.