Study design

We conducted a retrospective observational study of measles transmission dynamics in Mexico using national surveillance data from the General Directorate of Epidemiology (DGE) [20]. The dataset was retrieved on Feb 16, 2026, from the official open‑data repository and included all confirmed cases with a diagnosis date up to Feb 10, 2026. Available variables comprised demographic characteristics (age and sex), geographic identifiers (municipality and state of residence), final case classification, and diagnostic criteria (clinical, epidemiological, or laboratory-based). Diagnosis date was used as a proxy for symptom onset due to inconsistent reporting of the latter in the public database.

A disruption in data availability occurred between late 2025 and early 2026. Although the dataset released on Dec 17, 2025, remained accessible, a subsequent update dated Dec 26, 2025, could not be retrieved due to persistent server errors (HTTP 404). This resulted in a reporting gap from Dec 16, 2025, to Jan 4, 2026. The discontinuity was explicitly addressed during temporal preprocessing to minimize bias in the spatiotemporal analysis.

Setting

Mexico comprises 32 federal entities (31 states and Mexico City), in turn subdivided into 2,471 municipalities, the smallest administrative units for which population and health surveillance data are systematically collected. The 2025 national population was estimated at 129 million inhabitants, with substantial heterogeneity in population density, urbanization, and access to health services across federal entities. This study covered the entire national territory, with the municipality as the geographic unit of analysis for spatiotemporal scan statistics. All confirmed measles cases and population denominators were aggregated at the municipal level. A map illustrating the geographical division of Mexico’s federal entities is provided as S1 Fig, and a reference map of the same boundaries is also available from the National Institute of Statistics and Geography (INEGI) [21].

Case definition and population data

Measles case definitions followed Mexico’s official epidemiological surveillance guidelines for vaccine-preventable diseases [22]. A probable case was defined as any person presenting with fever, maculopapular rash, and one or more of the following: cough, coryza, conjunctivitis, or lymphadenopathy (retroauricular, occipital, or cervical). A confirmed case was defined as any probable case with laboratory-confirmed measles virus infection through IgM antibodies or RT-PCR or epidemiologically linked to a laboratory-confirmed case. A discarded case was a probable case with negative laboratory results or ruled out by expert committee review. The “1” diagnostic code in the surveillance database corresponds to confirmed cases.

The initial dataset included 21,104 suspected cases (14,430 in 2025; 6,674 in 2026), of which 8,205 were confirmed (5,738 in 2025; 2,467 in 2026). Laboratory confirmation was available for 94.4% of confirmed cases (7,745/8,205), with the remainder classified through clinical and epidemiological association.

Population denominators were obtained from the National Population Council (CONAPO) 2025–26 municipal projections, calculated as the sum of male and female estimates [23]. Incidence rates were expressed per 100,000 inhabitants and classified into discrete categories (“No cases”, 1–10, 11–50, 51–150, 151–400, and > 400 per 100,000) to improve interpretability given the highly skewed distribution of municipal rates.

Data preparation and geospatial analysis

Geographic identifiers were standardized into five-digit municipality codes by concatenating state (01–32) and municipality (three-digit) identifiers, consistent with the INEGI framework. A vector layer was constructed using the 2025 Geostatistical Framework [24], and municipal centroids were calculated to serve as the coordinate base for spatial analysis.

Data were aggregated into a Case-Aggregation Structure (CAS) by municipality and epidemiological week. Cases were assigned to epidemiological weeks based on diagnosis date. The study period spanned 50 epidemiological weeks from February 19, 2025 (Epidemiological Week 8) to February 10, 2026. Weekly case counts at the municipal level served as the input for spatiotemporal scan statistics. Weekly incidence rates were calculated for each municipality as the number of confirmed cases per 100,000 inhabitants per week, using CONAPO’s 2025 mid-year population estimates as fixed denominators. Cumulative incidence for the entire study period was calculated as a summary measure (total cases per 100,000 inhabitants) for descriptive purposes only and was not used in spatiotemporal analysis.

To maintain the continuous temporal sequence required for scan statistics despite the mid-December reporting gap, a temporal realignment was performed. Week 1 was defined as the week of the first confirmed case (Feb 19, 2025; Epidemiological Week 8), and subsequent weeks were numbered sequentially; this renumbering was purely cosmetic for intuitive presentation and had no impact on the spatiotemporal analysis, as SaTScan operates on actual dates rather than numerical week labels.

The period with missing data (previously Weeks 45–46) was bypassed by recoding the next active week (Week 47) as the subsequent chronological step (Week 45), preserving temporal contiguity without introducing artificial zero-case intervals. Three alternatives were considered for handling this gap: i) inserting artificial zero-case weeks, which would violate Poisson model assumptions and introduce spurious temporal patterns; ii) truncating the analysis before the gap, discarding valuable late-outbreak-phase information; or iii) removing the gap period while preserving temporal continuity. We selected third option as the most conservative approach, maintaining correct temporal sequence for scan statistics while acknowledging potential underestimation during that interval.

Spatial analysis and mapping

Spatial analyses and mapping were conducted in R 4.4.1 (R Core Team) using the ggplot2 package. The municipal shapefile used as the base map was developed by the research team from publicly accessible geographic coordinates of municipal boundaries provided by INEGI’s 2024 Geostatistical Framework. No proprietary basemaps, tiles, or satellite imagery were used. All spatial layers were projected to the official national grid (EPSG: 6362) for consistency across analyses.

Spatiotemporal statistical analysis

Spatiotemporal clusters of measles incidence were identified using Kulldorff’s space–time scan statistic, implemented in SaTScan version 10.3.3 (Martin Kulldorff and Information Management Services Inc.; Boston, MA, USA). A discrete Poisson model was used, if the number of cases in each municipality followed a Poisson distribution proportional to the population at risk. The scanning window was defined as a cylinder with a circular geographic base and a temporal height. The maximum spatial window size was set to 50% of the population at risk, and the maximum temporal window to 50% of the study period. These thresholds represent standard recommendations in scan statistic methodology, balancing sensitivity to detect large-scale regional transmission with specificity to avoid overly broad clusters lacking actionable geographic precision. Purely spatial and purely temporal clusters were not evaluated, as our research objective required identifying where and when transmission was most intense simultaneously to inform both geographic targeting and response timing. Statistical significance was assessed using 999 Monte Carlo replications, and clusters were reported under the no-geographical-overlap criterion.

Single-municipality clusters were retained if they met the statistical significance threshold (p < 0.001) and the minimum case requirement (≥2 cases). The potential for false-positive detection was addressed through the 999 Monte Carlo replications (correcting for multiple testing), the stringent significance threshold, and the discrete Poisson model which accounts for expected random variation based on population size and temporal distribution. Single-municipality clusters with markedly elevated relative risk represent epidemiologically plausible microoutbreaks, consistent with measles transmission dynamics following introduction into highly susceptible populations.

All data preprocessing and visualization were conducted in R version 4.5.2 (R Foundation for Statistical Computing; Vienna, Austria) using the tidyverse [25], sf [26], lubridate [27], and data.table [28] packages. The full SaTScan parameter file, along with the case (CAS), population (POP), and geographic coordinates (GEO) input files, are provided as open data to ensure full reproducibility.

Ethical considerations

This study used de-identified, aggregated public health data from open-access repositories. As the data were anonymized and available in the public domain, the study was exempt from review by an Institutional Review Board, and informed consent was not required.

Overview of national measles activity

A total of 8,205 confirmed measles cases were included in the spatiotemporal analysis, distributed across 337 of Mexico’s 2,471 municipalities (13.7%) during the 50‑week study period. Among all confirmed cases, 50.7% were female (n = 4,164) and 49.3% male (n = 4,044). The median age was 17 years (interquartile range 5–29), with a broad age distribution: 9.6% were infants < 1 year, 14.6% aged 1–4 years, 21.1% aged 5–14 years, 29.7% aged 15–29 years, 22.5% aged 30–49 years, and 2.5% aged ≥ 50 years.

The mean weekly incidence rate was 50.0 cases per 100,000 population (SD 142.7), with a median of 5.4 (IQR 1.5–23.6). The highest municipal rates occurred in four municipalities in Chihuahua (1,122.1 [week 32], 1,079.2 [week 19], 790.0 [week 4], and 746.6 [week 17]) and in one municipality in Guerrero (745.6, week 44).

Fig 1 shows cumulative measles incidence rates per 100,000 inhabitants using a discrete, categorized color scale designed to improve interpretability given the highly skewed distribution of municipal rates. Municipalities were grouped into six incidence intervals (“No cases”, 1–10, 11–50, 51–150, 151–400, and > 400 per 100,000), allowing clearer visual comparison across regions. This representation highlights a highly focal and heterogeneous spatial pattern: the highest incidence was concentrated in selected municipalities of Chihuahua, where small populations and clustered cases produced rates in the 750–1,100 per 100,000 range. Additional high‑incidence pockets were observed in parts of central and western Mexico, whereas most municipalities fell within the lowest categories or reported no cases, indicating geographically circumscribed outbreaks against a backdrop of widespread apparent absence of transmission.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Cumulative measles incidence rate, Mexico, 2025–26.

Notes: Municipalities shown in grey reported no confirmed measles cases during the 50‑week study period. The map uses the Lambert Conformal Conic coordinate system (EPSG:6362).

https://doi.org/10.1371/journal.pone.0353018.g001

Spatial heterogeneity and cluster detection

Building on this spatial heterogeneity, the space–time scan statistic identified 23 clusters of elevated risk, of which 19 were statistically significant, confirming that the observed hotspots represent true excess transmission rather than random variation. These clusters were highly concentrated in a limited number of municipalities, consistent with the localized patterns shown in Fig 2.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Independent SaTScan-detected space–time clusters of elevated measles risk and municipal relative risk (RR) in Mexico, 2025–26.

Notes: Only the seven principal clusters are shown. The map uses the Lambert Conformal Conic coordinate system (EPSG:6362).

https://doi.org/10.1371/journal.pone.0353018.g002

Primary cluster

The most likely cluster was in northern Mexico, with a centroid at 28.72°N, 106.95°W and a spatial radius of 295 km, encompassing 89 municipalities. SaTScan identified an optimal temporal window spanning epidemiological weeks 6–30. Within this cluster, 4,012 cases were reported compared with 75.0 expected, yielding an observed-to-expected ratio of 53.5 and a relative risk (RR) of 103.7. The aggregated incidence was 2,403.6 per 100,000 population. The cluster had a log-likelihood ratio of 13,187.8 and a p value < 10 ⁻ ¹⁷, indicating a highly significant concentration of transmission in space and time.

Fig 3 places this cluster within the national epidemic curve by overlaying its temporal window onto the weekly number of confirmed cases. The rise in incidence during weeks 6–27 overlaps with the active period of the primary cluster. The figure also shows that subsequent fluctuations in weekly case counts occur during the time windows of several secondary clusters.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Weekly number of confirmed measles cases during the 50‑week study period, with the temporal extent of all detected space–time clusters, Mexico, 2025–26.

Notes: C labels denote cluster identifiers. A disruption in data availability between mid‑December 2025 and early January 2026 created a reporting gap, which was corrected through temporal realignment to maintain a continuous weekly sequence for the spatiotemporal analysis.

https://doi.org/10.1371/journal.pone.0353018.g003

The spatial extent of this high-intensity aggregation is shown in Fig 4, which maps the 89 municipalities encompassed by the primary cluster across northern Mexico. The figure illustrates the broad, contiguous zone of high risk spanning central and western Chihuahua and extending into adjacent areas of Sonora, Durango, and Sinaloa. The large radius of 295 km reflects the wide geographic reach of sustained transmission during weeks 6–30, consistent with the high observed-to-expected ratio and relative risk.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Spatial distribution and case burden of municipalities within the primary measles cluster, Mexico, 2025–26.

Notes: Map of northern Mexico showing the 89 municipalities included in the primary space–time cluster detected by SaTScan (centroid 28.72°N, 106.95°W; radius 295 km) during epidemiological weeks 6–30. Relative risk is shown as a continuous background surface, while bubble size is proportional to the number of confirmed cases in each municipality. The cluster spans central and western Chihuahua and extends into adjacent areas of Sonora, Durango, and Sinaloa.

https://doi.org/10.1371/journal.pone.0353018.g004

Secondary clusters

Secondary clusters were detected in multiple regions of the country. The largest (cluster 2) occurred in western Mexico, centred at 21.57°N, 105.49°W with a radius of 272 km, spanning weeks 45–49 and comprising 113 municipalities. This cluster contained 1,493 observed cases versus 57.5 expected (RR 31.5), corresponding to an aggregated incidence of 1 166·8 per 100 000 population.

A third high-intensity cluster (cluster 3) was detected in the Bajío region, centred at 20.49°N, 102.34°W with a radius of 31 km, spanning weeks 32–41. It included 295 cases versus 4.7 expected (RR 64.8), reflecting a high observed‑to‑expected ratio within a geographically limited area.

Several smaller clusters showed extreme localized risk. A single-municipality cluster (cluster 4) in southern Mexico persisted from weeks 25–49 and reported 134 cases versus 0.58 expected (RR 235.8). Additional single-location clusters showed similarly elevated risks, including cluster 9 (RR 54.2), cluster 15 (RR 24.5), and cluster 16 (RR 92.1).

Other clusters were more diffuse but still significant. Cluster 5 in southeastern Mexico spanned weeks 46–50, comprising 22 municipalities and reporting 193 cases versus 9·3 expected (RR 21.2). A mid-outbreak cluster in north-central Mexico (cluster 6) spanned weeks 32–39, with 108 cases versus 7.2 expected (RR 15.1). Several additional clusters (clusters 8–23) showed moderate but significant elevations in risk, often restricted to one or a few municipalities and lasting only a single week.

Short‑duration and point clusters

Short‑duration and point clusters were also detected. An example occurred in northern Mexico during week 4 (cluster 7), where six neighboring municipalities reported 31 cases despite an expectation of 0.30 (RR 104.8). A second short‑duration cluster emerged in central Mexico (cluster 11) during weeks 47–49, comprising 80 cases versus 21.2 expected (RR 3.8). Several single‑municipality point clusters (clusters 9, 15, 16, and 18) reported 5–21 cases over brief intervals, with RR values exceeding 50.

Low‑intensity widespread cluster

Cluster 10 was a large, diffuse aggregation spanning 244 municipalities across central and northern Mexico during weeks 42–43. It contained 157 observed cases versus 60.3 expected (RR 2.64).

Clusters with borderline significance

Two clusters met reporting criteria despite higher p‑values. Cluster 20 (RR 32.2) comprised six cases during weeks 36–37 (p = 0.114), while cluster 21 (RR 74.8) involved four cases in week 10 (p = 0.410). These clusters were retained under the no‑geographical‑overlap rule. Table 1 summarizes the significant space–time clusters of measles detected using a discrete Poisson model.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Space–time clusters of measles detected using a discrete Poisson model, Mexico, 2025–26.

https://doi.org/10.1371/journal.pone.0353018.t001

Strengths and limitations

A major strength of this study is the use of high‑resolution municipal data and a robust, widely validated statistical framework to detect and quantify spatiotemporal clusters. The analysis captures both sustained regional transmission and short‑duration microoutbreaks, offering a granular view of epidemic dynamics. The use of open‑access surveillance data and reproducible SaTScan inputs enhances transparency and replicability.

However, several limitations warrant consideration. First, reliance on diagnosis date rather than symptom onset may introduce temporal imprecision, although this limitation is inherent to the public dataset. This imprecision may affect the exact timing of cluster initiation and termination, particularly for short‑duration events. Second, a reporting gap in late 2025 required temporal realignment to maintain sequence continuity for scan statistics. While this approach preserved the analytical framework, it introduces several potential biases. Any cases occurring during the unreported period (approximately mid-December 2025) would not be captured in our analysis, potentially underestimating the true magnitude of transmission during the Decline-to-Late phase transition. Clusters with temporal windows spanning this period may have artificially truncated durations or underestimated case counts, and if transmission intensity varied significantly during the gap, our characterization of the late-phase diffuse cluster (cluster 10, weeks 42–43) and the temporal boundary between outbreak phases may be affected. These uncertainties may influence both the apparent size of late‑phase clusters and the precision with which phase transitions are delineated. Additionally, state-level heterogeneity in reporting completeness during this period could have differentially impacted the detection of secondary clusters outside Chihuahua, where surveillance infrastructure may have been more strained. As a result, regional comparisons of cluster magnitude or persistence should be interpreted with caution. These factors should be considered when interpreting cluster duration, magnitude, and the precise timing of epidemic phase transitions. Third, under‑ascertainment of cases, particularly in municipalities with limited surveillance capacity, may have led to underestimation of cluster magnitude or failure to detect smaller clusters in remote areas. This could bias the spatial distribution of detected clusters toward better‑resourced municipalities. Fourth, the scan statistic assumes Poisson distribution proportional to population size, which may not fully capture heterogeneities in contact patterns or health‑seeking behavior. Moreover, SaTScan results are sensitive to parameter selection (e.g., maximum spatial and temporal window sizes), and different specifications could yield variations in cluster detection.

Fifth, no formal assessment of global spatial autocorrelation was performed; the spatial heterogeneity described in this study is based on the scan statistic and visual patterns rather than metrics such as Moran’s I. It is important to note that spatial heterogeneity was not formally assessed using global measures; instead, the patterns described here reflect localized excess risk identified through the scan statistic and visual inspection of municipal‑level incidence. The use of municipal centroids as proxies for population distribution and the absence of adjustments for temporal trends or global spatial dependence may influence the detection, size, and duration of clusters.

Accordingly, findings reflect localized clustering of excess risk rather than a characterization of broader spatial dependence across the municipal distribution of incidence. Sixth, the analysis does not incorporate vaccination coverage or mobility data at the individual level, which could further elucidate the drivers of cluster formation and transmission dynamics. The absence of these data limits the ability to contextualize clusters in terms of underlying susceptibility or movement patterns, and therefore the findings should be interpreted as descriptive rather than explanatory.

Implications for policy and practice

The marked contrast between sustained regional transmission and localized surges underscores the need for a dual‑track public health strategy. First, targeted “mop‑up” immunization campaigns should prioritize the high‑risk northern and western regions identified in this study, where sustained transmission was most pronounced. Strengthening routine immunization in these areas is essential to prevent reestablishment of endemic circulation. Second, the detection of high‑intensity microclusters highlights the need for rapid‑response capacity at the municipal level. Surveillance systems must be capable of detecting small increases in incidence and deploying interventions within days to prevent localized introductions from evolving into regional outbreaks. Third, the late‑phase diffuse cluster suggests that national‑level efforts to restore routine vaccination, rebuild trust in immunization programs, and improve cross‑jurisdictional coordination remain critical. Given the descriptive nature of this analysis, these considerations should be interpreted as contextual insights rather than prescriptive recommendations. Integrating spatiotemporal analytics into routine surveillance could support more agile, geographically tailored responses. Such applications should be viewed as potential areas for operational strengthening rather than direct policy directives derived from causal inference.