Local spatial autocorrelation analysis of dengue and dengue hemorrhagic fever in Ho Chi Minh City: Spatial Insights from the 32nd Week

*Corresponding author: Danh-tuyen Vu, Email: Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution License 4.0. Local spatial autocorrelation analysis of dengue and dengue hemorrhagic fever in Ho Chi Minh City: Spatial Insights from the 32nd Week Danh-Tuyen Vu 1, *, Anh-huy Hoang 2 and Tien-thanh Nguyen 1 1 Faculty of Surveying, Mapping and Geographic Information, Hanoi University of Natural Resources and Environment, Hanoi, Vietnam. 2 Faculty of Environment, Hanoi University of Natural Resources and Environment, Hanoi, Vietnam. World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 Publication history: Received on 10 July 2025; revised on 17 August 2025; accepted on 19 August 2025 Article DOI: https://doi.org/10.30574/wjbphs.2025.23.2.0765 Abstract Background : Dengue and dengue hemorrhagic fever (DF/DHF) remain critical public health challenges in Ho Chi Minh City, Vietnam, where recurrent outbreaks highlight the need for fine-scale spatial analysis. Identifying clusters and outliers of disease incidence is essential for guiding effective and targeted control strategies. Methods : Weekly DHF case data for the 32nd epidemiological week of 2025 were aggregated at the commune level. Spatial distribution was first visualized using choropleth mapping and descriptive histograms. A spatial weight matrix was constructed to define commune-level neighborhood structures, and Local Moran’s I (LISA) was applied to detect spatial autocorrelation. The results were interpreted using LISA cluster maps, LISA value distributions, and Moran’s I scatterplots to identify statistically significant hot spots, cold spots, and outliers. Results : The spatial distribution map revealed marked heterogeneity, with higher DHF incidence concentrated in central and southwestern communes, while peripheral areas exhibited lower case numbers. The global Moran’s I was modest but positive (0.192), suggesting localized clustering rather than widespread citywide autocorrelation. LISA results indicated that 14 communes (8.4%) formed high-high clusters, 7 communes (4.2%) formed low-low clusters, and 12 wards (7.2%) were spatial outliers (6 high-low and 6 low-high). Most communes (80.2%) were not significant, indicating that dengue clustering is highly localized. High–high clusters were concentrated in the central-northern urban core, while outliers were located along southern and peripheral districts, reflecting transitional or isolated neighborhood effects. Conclusions : Local Moran’s I analysis demonstrated that dengue transmission in Ho Chi Minh City during the 32nd week of 2025 was characterized by spatially concentrated hot spots and discrete outliers within a largely neutral background. These findings underscore the value of spatial autocorrelation techniques for identifying high-risk neighborhoods, supporting targeted interventions, and enhancing the efficiency of dengue surveillance and vector control programs. Keywords: Dengue Hemorrhagic Fever (DHF); Spatial autocorrelation; Local Moran’s I; Spatial Clustering; Ho Chi Minh City, Vietnam. 1. Introduction Dengue and dengue hemorrhagic fever (DHF) remain major public health concerns in tropical and subtropical regions, particularly in Southeast Asia, where rapid urbanization and climatic variability contribute to recurrent epidemics (1,2). World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 287 Vietnam is among the countries most affected, with Ho Chi Minh City frequently experiencing high transmission intensity due to dense population, complex urban landscapes, and favorable conditions for Aedes mosquito breeding (3,4). Understanding the spatial distribution of dengue incidence is therefore crucial for designing efficient control strategies. Spatial epidemiology has been increasingly applied to infectious disease research, offering robust tools for identifying spatial heterogeneity and clustering (5,6). Spatial autocorrelation statistics, particularly Moran’s I, have proven useful for detecting and quantifying patterns of disease distribution beyond what is visible through raw incidence mapping (7,8). While the Global Moran’s I provides an overall measure of clustering, it may obscure localized variations; hence, Local Indicators of Spatial Association (LISA), including Local Moran’s I, have been widely employed to detect spatial clusters and outliers at finer scales (8,9). Applications of local spatial autocorrelation methods in dengue research have gained momentum in recent years. Studies in Thailand, Indonesia, and the Philippines have demonstrated that dengue incidence often clusters within specific neighborhoods, reflecting socio-environmental and vector-related factors (10– 12). In Vietnam, a study by (13) and (14) highlighted that the spatial distribution of dengue is shaped by urban density, climate variability, and mobility patterns. More recent analyses further emphasize the importance of combining geospatial statistics with epidemiological data to inform localized intervention programs (3,4). The use of Local Moran’s I for dengue mapping has several advantages. First, it enables the identification of “hot spots” (high-high clusters), where interventions should be prioritized, and “cold spots” (low-low clusters), which may represent areas of reduced vulnerability (8,15). Second, it reveals spatial outliers (high-low or low-high), which may indicate emerging outbreaks or protective contexts that warrant further study (9). Such insights are critical for optimizing resource allocation, particularly in urban megacities like Ho Chi Minh City, where both socio-demographic and environmental heterogeneity strongly influence transmission dynamics (16,17). In summary, the literature indicates that spatial clustering analysis, especially through local spatial autocorrelation methods, provides an effective approach to understanding dengue epidemiology in complex urban settings. Building upon this foundation, the present study applies Local Moran’s I to characterize spatial clustering of dengue and DHF in Ho Chi Minh City during the 32nd week of 2025, with the goal of identifying localized patterns of transmission and informing targeted public health interventions. 2. Data used and methods 2.1. Data used Data on confirmed dengue fever (DF) and dengue hemorrhagic fever (DHF) cases were obtained from the Ho Chi Minh City Center for Disease Control (HCDC) for the 32nd epidemiological week of 2025. The dataset included the number of laboratory-confirmed cases aggregated at the commune level across all 168 communes of the city. Communes with zero reported cases were also retained in the dataset to avoid bias in spatial analysis. Case incidence was normalized by commune population size, using the 2025 population statistics from the Ho Chi Minh City Statistical Office, to generate incidence rates per 100,000 population. In addition, administrative boundary shapefiles of Ho Chi Minh City at the district and commune levels were obtained from the Department of Land Administration, Ministry of Agriculture and Environment of Vietnam. These shapefiles were used to construct the spatial framework for analysis and to link epidemiological data to geographic units. 2.2. Methods 2.2.1. Global Moran’s I statistic This study employed the global Moran’s I statistic to assess the presence of spatial clustering of DHF cases at the global scale (7,18). The global Moran’s I statistic is formally defined in equation (1): I= n S0 ∑ ∑ Wij(xi−x )(xj−x ) n j=1 n i=1 ∑ ∑ Wij n j=i n i=1 ∑ (xi−x )2 n i=1 ………… (1) where xi and xj are the number of DHF confirmed cases for each commune i and province/city j; x is the mean of DHF cases and be given by x =∑xi n n i=1 ; n is the total number of communes in the whole study area; and Wij is a (n×n) spatial weight matrix (19). World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 288 The global Moran’s I coefficient ranges between –1 and +1 (19). Positive values indicate positive spatial autocorrelation, whereas negative values signify negative spatial autocorrelation (20). Values approaching zero suggest the absence of spatial autocorrelation, corresponding to a random distribution of DHF cases. 2.2.2. Local Moran’s I statistic The global Moran’s I statistic provides a measure of overall spatial autocorrelation. However, previous studies have indicated that global autocorrelation may obscure local variations; therefore, local indicators of spatial association (LISA) were employed to identify specific local clusters (18). Moreover, even in the absence of global spatial autocorrelation, local spatial autocorrelation analysis can reveal clusters at the local level. Accordingly, the local Moran’s I statistic was applied to quantify the spatial clustering of high and low DHF incidence within each commune (19). The local Moran’s I statistic (Ii) for DHF at commune i is given by the following equation (8): Ii=(xi−x) σ2∑ Wij(xj−x) N j#i,j∈Ji (2) where xi, xj, x, and Wij are defined in equation (1); N is the total number of neighborhood communes (19); Ji denotes the neighborhood set of DHF confirmed cases at province/city i; j#i implies that the sum of all (xj−x) of nearby neighbourhood commune i but not including xj; and σ2 is the variance of x, given in equation (3). Wij defines neighbor connectivity and can be constructed using first order and second of contiguity. σ2=1 N∑(xi−x) N j=1 (3) A positive local Moran’s I value indicates that a feature is surrounded by neighbors with similarly high or low attribute values, suggesting its inclusion in a cluster. Conversely, a negative value indicates neighboring features with dissimilar values, identifying the feature as an outlier. In both cases, statistical significance is determined by the associated p-value. The degree of spatial clustering of DHF in each commune was evaluated using the local Moran’s I statistic. 2.2.3. The Moran’s scatterplot The Moran scatterplot is a graphical tool developed to visualize and interpret spatial autocorrelation, complementing the numerical output of Moran’s I statistic. In the context of dengue hemorrhagic fever (DHF), where case distribution is influenced by complex interactions between environmental, demographic, and vector-related factors, the Moran scatterplot provides an effective means to explore the spatial dependency of disease incidence. In the scatterplot, the standardized values of DHF incidence (z-scores) are plotted on the x-axis, while the spatially lagged values of DHF are represented on the y-axis. The slope of the regression line through these points corresponds to the global Moran’s I value, thereby quantifying the degree of overall spatial autocorrelation. More importantly, the Moran scatterplot allows the classification of spatial relationships into four quadrants: high-high (HH) and low-low (LL) indicating clusters of similar values, and high-low (HL) and low-high (LH) indicating spatial outliers. This visualization is particularly useful in epidemiological studies of DHF because it highlights areas where high incidence clusters (hotspots) or low incidence clusters (coldspots) occur, as well as unusual patterns where a high-value area is surrounded by low values or vice versa. Identifying such patterns is crucial for understanding the spatial dynamics of disease transmission, detecting early warning signals of outbreaks, and guiding targeted interventions in vector control and public health planning. 3. Results and discussion 3.1. Spatial distribution of the DHF cases in the 32th week Data from Figure 1 illustrates both the spatial distribution (left) and the frequency histogram (right) of DHF incidence in Ho Chi Minh City during the 32nd epidemiological week. The spatial distribution map shows marked heterogeneity across districts, with higher incidence rates concentrated in several central and southwestern communes, whereas many peripheral areas exhibited comparatively lower case numbers. This pattern suggests the presence of localized transmission dynamics, possibly influenced by variations in population density, urban infrastructure, and vector breeding environments. Data from the histogram further supports this observation, indicating that the majority of communes reported a relatively low to moderate incidence of DHF, clustered within the range of 12.7-34.1 cases. Only a limited number of communes fell into the highest incidence classes (47.0-77.0 cases), highlighting the emergence of World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 289 distinct hotspots rather than widespread uniform distribution. Conversely, a few communes were characterized by very low case counts (≤ 12.7), aligning with the cold spots detected in peripheral zones of the city. These findings highlight the uneven burden of DHF within the city, where a few localized hotspots contribute disproportionately to overall transmission risk. Such clustering patterns provide an important rationale for applying local spatial autocorrelation techniques (e.g., Local Moran’s I) to identify statistically significant high–high clusters and spatial outliers, which will be further discussed in subsequent sections. Figure 1 Map (left) and histogram (right) of the incidence of DHF in the 32th week in Ho Chi Minh city, Vietnam 3.2. Local spatial autocorrelation analysis of dengue and dengue hemorrhagic fever 3.2.1. Spatial-weights diagnostics and Implications for Local Moran’s I Prior to computing Local Moran’s I, we examined the spatial weights to ensure an appropriate neighborhood structure. The connectivity map (Figure 2, left) shows a well‐connected urban core with dense linkages among central communes, while peripheral areas, particularly the southern and southeast districts, display sparser linkages. The histogram of the number of neighbors (Figure 2, right) is unimodal with its mass between 4-7 neighbors per commune, a thin left tail at 0-1, and a shorter right tail extending to 9-10. This distribution reflects the city’s heterogeneous areal geometry: small, tightly packed communes in the center naturally have more adjacent units, whereas larger peripheral communes (and water‐bounded communes) have fewer. Variation in neighbor counts matters for LISA inference. Units with many neighbors (e.g., inner‐city communes) pool information from a broader local context, generally providing more stable local statistics; units with few neighbors (outer communes) yield higher variance local statistics and reduced power to detect clustering. Any communes with zero neighbors (islands) cannot have a LISA value unless connected (e.g., via nearest‐neighbor or distance band augmentation). Given the observed heterogeneity (0-10 neighbors), row standardization or an equivalent normalization is advisable so that each commune contributes comparably to its neighbors and densely connected cores do not dominate the statistic. Because cluster boundaries can be sensitive to the choice of weights (queen/rook contiguity vs. distance band vs. k–nearest neighbors), results should be interpreted as local patterns conditional on this weights specification. The histogram indicates that our specification avoids pervasive isolation while preserving local structure, which is appropriate for detecting fine‐scale clustering in week 32. Nonetheless, edge communes with ≤2 neighbors warrant cautious interpretation, and reporting their neighbor counts alongside LISA results improves transparency. Overall, the weights matrix provides (i) a strongly connected core conducive to detecting coherent clusters and (ii) a sparser periphery where outliers are more likely. With these diagnostics in place, the ensuing Local Moran’s I results can be read with confidence that observed hot spots, cold spots, and outliers are not artifacts of poor network specification but reflect meaningful local spatial dependence in DHF incidence during the 32nd epidemiological week. 3.2.2. Analysis of LISA distribution and interpretation of positive vs. negative LISA. Data from the map in Figure 3 shows the Local Moran’s I (LISA) values for DHF incidence in week 32 and shows their frequency distribution. The city-wide LISA values range from -1.205 to 3.803. The histogram is right-skewed with a concentration near zero, indicating that most communes exhibit only weak local spatial dependence. Specifically, ~80% (134/168) of communes fall within the middle bands (-0.144 to 0.646), while ~10% (17/168) occupy the upper tail World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 290 (0.646-3.803) and ~10% (17/168) the lower tail (-1.205 to -0.144). Positive LISA indicates local similarity, communes whose DHF incidence resembles their neighbors (potential high-high or low–low areas). The upper tail (top 10–1%) therefore flags strongly clustered neighborhoods where local conditions are spatially reinforcing. Negative LISA indicates local dissimilarity - communes that differ from their neighbors (potential high–low or low-high outliers). The small group in the bottom 10–1% suggests discrete outliers, often found at transitions between higherand lowerincidence areas or where commune geometry/connectivity is atypical. The map reveals a central–northern belt of high positive LISA, consistent with a cohesive cluster structure in the urban core. In contrast, several peripheral and riverbounded communes display lower or negative LISA, consistent with outlier behavior at the edge of clusters or in sparsely connected areas. This pattern aligns with the connectivity diagnostics: densely connected inner-city communes tend to yield more stable and often larger positive LISA values, whereas sparsely connected fringe communes are more prone to weak or negative local association. Figure 2 Connectivity map (left) and histogram (right) of neighbor number in Ho Chi Minh city, Vietnam Figure 3 Map (left) and histogram (right) of local Moran’s I obtained DHF datasets collected in the 32nd week in 2025 in Ho Chi Minh city, Vietnam The presence of a limited number of strongly positive LISA neighborhoods points to high-priority zones where intensified vector control, case finding, and community engagement are likely to be most efficient. The negative LISA Hotspots and coldspots World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 291 outliers warrant situational review, either as genuine micro-hotspots embedded in lower-risk surroundings (high-low) or as resilient low-incidence pockets adjacent to higher-risk areas (low-high), to tailor interventions at the commune boundary. In the next subsection, we use the LISA quadrant and significance maps to distinguish hot spots, cold spots, and spatial outliers and to translate these patterns into actionable public-health guidance. 3.2.3. Analysis of global pattern and local clustering and outliers The Moran’s I scatterplot plots standardized DHF incidence (x-axis) against the spatially lagged incidence (y-axis), essentially comparing each commune’s value with the mean of its neighbors. The Moran’s I scatterplot (Figure 4, right) shows a positive slope of 0.192, indicating modest global spatial autocorrelation in DHF incidence during week 32. This implies the citywide pattern is not uniformly clustered, but contains localized pockets where neighboring communes display similar values. (Isolates in the weight matrix were removed from this calculation, so the estimate reflects the connected system only.). The slope of the regression line corresponds to Moran’s I = 0.192, indicating a modest but positive global spatial autocorrelation. The scatterplot supports the interpretation that dengue transmission in Ho Chi Minh City during week 32 is not randomly distributed, but is characterized by a few meaningful clusters and spatial outliers within a largely neutral background. This pattern reflects localized drivers of transmission (e.g., breeding sites, density, microclimate) rather than uniform citywide risk. The modest Moran’s I (0.192) emphasizes that local analysis (LISA) is more informative than relying solely on global measures. The LISA cluster map (Figure 4, left) identifies 14 high-high (HH) communes (~8.4%), 7 low-low (LL) communes (~4.2%), and 12 spatial outliers split between high-low (HL, 6; ~3.6%) and low-high (LH, 6; ~3.6%). The remaining 134 communes (~80.2%) are non-significant at the chosen threshold. HH communes form cohesive nuclei in the central–northern urban core, consistent with dense connectivity and previously observed higher incidence; LL communes are peripheral and river-bounded, suggesting contiguous low-risk neighborhoods. Outliers align with cluster margins, notably along southern/southeastern fringes, where commune geometry and fewer neighbors can yield abrupt transitions in local association. The balance of mostly non-significant LISA values with a small set of strong clusters matches the right-skewed LISA distribution reported earlier. Given heterogeneous neighbor counts across communes, the interpretation of outliers in sparsely connected edges should be cautious. Reported results are conditional on the chosen weights; sensitivity checks (e.g., queen vs. rook contiguity or distance-band/k-NN) and permutation-based significance help ensure robustness. When used for action, cluster boundaries should be paired with field validation and population-standardized rates to avoid artifacts from small denominators. Week-specific clustering suggests targeted, place-based responses rather than blanket citywide measures. Priority actions include: (i) intensifying source reduction and larval control in HH communes and immediately adjacent LH communes; (ii) active case finding and community engagement within identified clusters; and (iii) cross-boundary coordination across contiguous HH communes so interventions are synchronized across administrative borders. Continuous LISA monitoring can track cluster persistence or drift in subsequent weeks to guide resource allocation. Figure 4 Map of spatial clustering of DHF (left) and Moran’s I scatterplot (right) 4. Conclusion This study applied Local Moran’s I to examine local spatial autocorrelation of dengue and dengue hemorrhagic fever cases in Ho Chi Minh City during the 32nd epidemiological week of 2025. The results demonstrate that DHF incidence World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 292 is not randomly distributed across the city but instead exhibits distinct spatial clustering patterns. Specifically, high– high clusters were concentrated in central and northern districts, while low–low clusters were mainly located in peripheral communes. In addition, several high–low and low–high outliers were identified, reflecting transitional zones or localized anomalies in disease transmission. The global Moran’s I statistic (0.192) confirmed the presence of modest but significant spatial autocorrelation, indicating that local rather than citywide clustering is the dominant feature of dengue spread in this context. These findings highlight the importance of using local spatial autocorrelation measures, such as LISA, to detect fine-scale variations in disease risk that are obscured by global analyses. From a public health perspective, the identification of high–high clusters provides evidence for prioritizing targeted interventions in specific neighborhoods where transmission is spatially reinforcing. Similarly, the detection of spatial outliers can inform strategies to prevent the expansion of hotspots and to strengthen protective factors in resilient low-incidence communes. By integrating spatial statistical analysis with dengue surveillance, municipal health authorities can optimize resource allocation, enhance early warning systems, and improve the overall effectiveness of vector control programs. Compliance with ethical standards Acknowledgments The authors would like to thank Center for Disease Control of Ho Chi Minh City for providing the data and the anonymous reviewers for their careful reading of our manuscript and their many insightful comments and suggestions Disclosure of conflict of interest The authors declare no conflict of interest. Statement of ethical approval Ethical approval was obtained from respective ethical committee. Statement of informed consent Informed consent was obtained from all individual participants included in the study. References [1] Narvaez F, Gutierrez G, Pérez MA, Elizondo D, Nuñez A, Balmaseda A, et al. Evaluation of the traditional and revised WHO classifications of dengue disease severity. PLoS Negl Trop Dis. 2011;5(11):e1397. [2] Bhatt S, Gething PW, Brady OJ, Messina JP, Farlow AW, Moyes CL, et al. The global distribution and burden of dengue. Nature. 2013;496(7446):504–7. [3] Cuong HQ, Vu NT, Cazelles B, Boni MF, Thai KTD, Rabaa MA, et al. Spatiotemporal dynamics of dengue epidemics, southern Vietnam. Emerg Infect Dis. 2013;19(6):945. [4] Gibb R, Colón-González FJ, Lan PT, Huong PT, Nam VS, Duoc VT, et al. Interactions between climate change, urban infrastructure and mobility are driving dengue emergence in Vietnam. Nat Commun. 2023;14(1):8179. [5] Waller LA, Gotway CA. Applied spatial statistics for public health data. John Wiley & Sons; 2004. [6] Lawson AB. Bayesian disease mapping: hierarchical modeling in spatial epidemiology. Chapman and Hall/CRC; 2018. [7] Cliff AD, Ord JK. Spatial processes: models & applications. (No Title). 1981; [8] Anselin L. Local indicators of spatial association—LISA. Geogr Anal. 1995;27(2):93–115. [9] Tiefelsdorf M, Boots B. The exact distribution of Moran’s I. Environ Plan A. 1995;27(6):985–99. [10] Arcari P, Tapper N, Pfueller S. Regional variability in relationships between climate and dengue/DHF in Indonesia. Singap J Trop Geogr. 2007;28(3):251–72. [11] Khormi HM, Kumar L. Modeling dengue fever risk based on socioeconomic parameters, nationality and age groups: GIS and remote sensing based case study. Sci Total Environ. 2011;409(22):4713–9. World Journal of Biology Pharmacy and Health Sciences, 2025, 23(02), 286-293 293 [12] Salje H, Lessler J, Maljkovic Berry I, Melendrez MC, Endy T, Kalayanarooj S, et al. Dengue diversity across spatial and temporal scales: Local structure and the effect of host population size. Science (80- ). 2017;355(6331):1302– 6. [13] Warren MB, Bullard SA. First elucidation of a blood fluke (Electrovermis zappum n. gen., n. sp.) life cycle including a chondrichthyan or bivalve. Int J Parasitol Parasites Wildl. 2019;10:170–83. [14] Wang D, Zhang F, Yang S, Xia N, Ariken M. Exploring the spatial-temporal characteristics of the aerosol optical depth (AOD) in Central Asia based on the moderate resolution imaging spectroradiometer (MODIS). Environ Monit Assess. 2020;192:1–15. [15] Getis A, Ord JK. The analysis of spatial association by use of distance statistics. Geogr Anal. 1992;24(3):189–206. [16] Louis VR, Phalkey R, Horstick O, Ratanawong P, Wilder-Smith A, Tozan Y, et al. Modeling tools for dengue risk mapping-a systematic review. Int J Health Geogr. 2014;13(1):50. [17] Cromley EK, McLafferty SL. GIS and public health. Guilford Press; 2011. [18] Getis A, Ord JK. Local spatial statistics: An overview. Spatial analysis: Modeling in a GIS environment. Longley, P., and M. Batty. Wiley, New York; 1996. [19] Vu D-T, Nguyen T-T, Hoang A-H. Spatial clustering analysis of the COVID-19 pandemic: A case study of the fourth wave in Vietnam. Geogr Environ Sustain. 2021;14(4). [20] Nguyen TT, Vu TD. Identification of multivariate geochemical anomalies using spatial autocorrelation analysis and robust statistics. Ore Geol Rev. 2019;111.