Ameloblastoma at a Tertiary Referral Hospital: Spatial Distribution and Area-Level Sociodemographics in South Sulawesi Authors: Fenty Napitupulu, Yossy Yoanita Ariestiana, Abul Fauzi, Andi Tajrin, Mohammad Ghazali, Kei Tomihara, Kumiko Fujiwara, Ristya Widi Endah Yani, Nisrina Ekayani Nasrun, Hasanuddin, Muhammad Ruslin, Muhammad Assegaf Valianra, Muhammad Irfan Rasul Published: 2026 ABSTRACT BackgroundAmeloblastoma is a locally aggressive odontogenic tumour whose reported occurrence varies geographically. In South Sulawesi, Indonesia, epidemiological data remain scarce and district-level disparities are undocumented. This study mapped the residential distribution of patients treated at a tertiary referral hospital and examined area-level correlates of that distribution.MethodsAn analytical ecological study used the 24 districts/cities of South Sulawesi as spatial units. All histopathologically confirmed cases (n = 160, 2022-2024) were attributed to district of residence and ecologically associated with district-level health system capacity and socioeconomic indicators. Spatial distribution was assessed by choropleth mapping with Jenks classification, and population-adjusted rates per 100,000 were calculated. Differences across urban, peri-urban, and rural groupings were evaluated using Kruskal–Wallis and pairwise Poisson tests, and ecological associations using Pearson’s correlation and exploratory geographically weighted regression.ResultsCases concentrated in the southwestern corridor, mainly in urban areas. Parepare recorded the highest reported case rate (15.54 per 100,000), whereas Makassar ranked sixth (1.84 per 100,000). Urban (rate ratios = 3.27) and peri-urban (rate ratios = 3.59) districts showed higher rates than rural districts (both P < .001), although the Kruskal–Wallis omnibus test was not significant (P = .076), plausibly reflecting limited power with unequal group sizes. Reported rates correlated strongly with health system capacity indicators (r ≈ 0.83-0.89; P < .0001), negatively with poverty (r = −0.493) and positively with expenditure, schooling and unemployment (r = 0.617-0.698; P < .01), reflecting ecological patterns of case detection rather than individual risk. Exploratory geographically weighted regression indicated spatial nonstationarity, with 22 districts showing at least one locally significant socioeconomic variable.ConclusionReported ameloblastoma rates concentrate in urban referral centres and track health system capacity, indicating patterns of case detection and diagnostic access rather than true disease occurrence. These hypothesis-generating findings support district-specific strategies to improve diagnostic equity rather than uniform interventions. Introduction Ameloblastoma is an aggressive odontogenic tumour characterized by invasive growth and high recurrence. It develops from enamel-forming odontogenic epithelium.1,2 The tumour arises from dysregulated odontogenesis, involving transformation of the odontogenic cyst epithelium or dental lamina remnants.3 The aetiology of ameloblastoma remains unclear; while trauma, inflammation, and nutritional factors have been suggested, no definitive causative factors have been established.4 Ameloblastomas exhibit geographical variation worldwide. A Nigerian study across 10 hospitals documented 1246 cases, highlighting regional differences.5 A previous review has indicated a higher prevalence in Africa and Asia, particularly in tropical regions,6 whereas Europe and North America report lower incidence rates and later diagnosis ages.7,8 Case detection and reporting were strongly influenced by health system capacity; regions with better such as southwestern Nigeria, recorded higher incidence rate due to both environmental factors and healthcare resources, whereas areas with limited facilities often underreport, shaping observed disease pattern.6,9,10 Sociodemographic factors contribute to disparities in ameloblastoma detection and management. Evidence from Eastern Nigeria indicates that patients originate from low-income rural populations with limited education.1 These conditions lead to delayed healthcare and advanced disease due to financial constraints and limited awareness. Individuals with higher socioeconomic status typically have improved access to diagnostic facilities, enabling earlier detection. Similar patterns exist in high-income countries. Diez Roux and Mair9 in New York indicated that low-income populations face higher morbidity and delayed diagnoses due to limited healthcare access. While these studies examine individual-level characteristics, the present study adopts an ecological approach to assess whether district-level socioeconomic indicators are associated with spatial variation in reported case detection, recognizing that area-level associations do not imply individual-level risk. The geographic context, healthcare capacity, and sociodemographic conditions emphasize the requirement for geospatial approaches to understand ameloblastoma distribution. Geographic information systems provide tools for visualizing disease patterns and examining the relationships between the environment and healthcare access.8,11 Despite their value, geospatial methods have remained underutilized in odontogenic tumour studies. In eastern Indonesia, epidemiological data on ameloblastoma are limited to hospital referrals. Ruslin et al12 conducted a 6-year study of 84 patients, showing patterns linked to healthcare capacity. However, spatial variations or district-level healthcare and socioeconomic factors remain unexamined. Despite geographic and resource disparities, there remains a notable lack of comprehensive geospatial epidemiological data on ameloblastoma in South Sulawesi Province. Existing Indonesian data primarily stem from tertiary hospitals and often lack a spatial context, leaving interdistrict healthcare and socioeconomic disparities unexamined in relation to ameloblastoma distribution.13,14 The present study addresses this gap using an ecological design, examining district-level associations rather than individual patient risk factors. This gap underscores the critical requirement for a spatial epidemiological approach to understand the influence of contextual factors on ameloblastoma patterns in South Sulawesi. Therefore, the present study aimed to1: map the spatial distribution of histopathologically confirmed ameloblastoma cases across South Sulawesi districts2; examine ecological associations between reported case frequencies and district-level health system capacity indicators; and3 explore spatial heterogeneity in associations between socioeconomic factors and reported case frequencies using geographically weighted regression (GWR). This study does not claim to measure true disease incidence or healthcare accessibility directly; rather, it maps the observable distribution of patients presenting to a tertiary referral centre and its ecological correlates to generate hypotheses and inform future population-based surveillance. Material and methods Study design and setting The present study used a quantitative observational ecological design to examine the distribution of confirmed ameloblastoma cases across districts in South Sulawesi Province, Indonesia. A retrospective analysis of secondary data from January 2022 to December 2024 was conducted, with each of the 24 administrative districts represented as the spatial unit of analysis, reflecting geographic, demographic, and healthcare access heterogeneity. An ecological design was selected to assess district-level variations in reported ameloblastoma case rates in relation to health system capacity and socioeconomic context, as this approach was well suited for identifying structural disparities, informing policy-and service-planning interventions, and generating hypotheses for future individual-level studies, despite its inherent limitations in individual risk inference. All histopathologically confirmed ameloblastoma cases meeting the inclusion criteria during the study period were included through total sampling (n = 160 patients). The unit of analysis for all ecological and spatial inferential analyses was the district/city (N = 24). Sample size and statistical power The sample size for spatial analysis was fixed at N = 24 districts/cities, determined by the administrative geography of South Sulawesi Province. A sensitivity analysis of the minimum detectable effect size (G*Power 3.1.9.7) indicated that with N = 24 and α = 0.05 (two-tailed), Pearson correlation analysis achieves 80% power to detect |r| ≥ 0.50 (large effect). Correlations of moderate magnitude (|r| = 0.30-0.40) are detectable with 45% to 65% power, indicating elevated Type II error risk. For GWR, simulation studies recommend N ≥ 50 to 100 spatial units for stable local coefficient estimation; the present study falls well below this threshold. All inferential findings are therefore exploratory and hypothesis-generating rather than confirmatory. Ethical approval The present study utilized secondary anonymized data and received ethical exemption from the Health Research Ethics Committee of the Faculty of Dentistry, Hasanuddin University (Ethical Exemption No.253/KEPK FKG-RSGMP UH/EE/VIII/2025; Protocol No.: 246/FKGUH/PPDGS_BM-006/31.VII/2025). The study was conducted in accordance with the principles of the Declaration of Helsinki. Owing to the retrospective design and the use of deidentified data, the requirement for individual informed consent was formally waived by the ethics committee. Eligibility criteria Inclusion criteria were1: diagnosis of ameloblastoma confirmed by histopathological examination demonstrating characteristic odontogenic epithelial islands with peripheral palisading, reverse nuclear polarization, and stellate reticulum-like areas, consistent with WHO 2022 classification criteria2; ICD-10 coding as D16.4 or D16.53; case registered at a participating hospital in South Sulawesi during January 2022 to December 2024; and4 patient domicile recorded as within South Sulawesi Province. Exclusion criteria were1: absence of histopathological confirmation (clinical or radiological diagnosis only)2; patient domicile outside South Sulawesi Province3; duplicate records identified through cross-database verification; and4 ICD-10 code present but histopathological report unavailable or from a noncertified laboratory. Ameloblastoma case data source The study population included all patients diagnosed with ameloblastoma in South Sulawesi referred to participating hospital electronic medical records (EMRs) during the study period. The data were extracted by FN using a standardized protocol to ensure consistency. All eligible cases that met the inclusion criteria were identified and included through total sampling of the EMRs. Patient domicile district, as recorded in the EMR at the time of registration, was used as the geographic unit of attribution for all spatial analyses. This approach reflects the patient’s district of residence rather than the treating hospital’s district, thereby minimizing geographic misclassification arising from interdistrict referral. Of 196 initially identified EMR records, 36 cases (18.4%) were excluded from all spatial and statistical analyses for the following reasons: patient domicile outside Sulawesi (other Indonesian islands; n = 27), absence of clear histopathological confirmation of ameloblastoma (n = 6), and duplicate records identified through cross-database verification (n = 3). The final analytic sample comprised 160 histopathologically confirmed ameloblastoma cases with documented domicile in South Sulawesi Province. The case selection flow is summarized in Supplementary Figure 1. Population and district-level explanatory data District-level population data were obtained from the Central Bureau of Statistics (Badan Pusat Statistik) in South Sulawesi Province. Health system capacity and sociodemographic indicators for the study period (2022-2024) were retrieved from official government databases, including those of the Ministry of Health and the Central Bureau of Statistics, to ensure standardized and comparable measurements across districts. Outcome variable The primary outcome was district-level ameloblastoma case rate, defined as histopathologically confirmed and registered cases recorded in each district during the study period. This outcome reflects reported cases, which are influenced by healthcare access and diagnostic capacity rather than true biological incidence. Individual-level descriptive variable Individual-level variables were extracted from hospital records solely for descriptive purposes to characterize the patient cohort and included patient age, which was categorized into four groups (0-11, 12-25, 26-45, and 46-80 years), biological sex (male or female), highest educational attainment, and occupation. These individual-level characteristics were used to summarize the demographic profiles of the patients and were not incorporated into inferential ecological or spatial regression analyses, which were conducted exclusively using district-level variables. Sociodemographic-related outcome variables The sociodemographic variables included individual- and district-level indicators. Individual variables (age, sex, education, and occupation) were obtained from hospital records. Age was categorized according to the Indonesian Ministry of Health classification (0-11, 12-25, 26-45, and 46-80 years), while sex was recorded as male or female. Education level and occupation status were documented from administrative records. District-level indicators from the Indonesian Central Bureau of Statistics included the poverty rate (defined by the World Bank’s threshold of USD 234.28 [IDR 3921088.79] monthly for South Sulawesi), per capita expenditure (average monthly household expenditure), mean years of schooling (average number of years of formal education for adults aged 15+), unemployment rate (percentage of the labour force without work), Gini ratio (measure of income inequality) and life expectancy (average number of years a person is expected to live). Spatial visualization Geographic boundaries at the district/city level were used to generate choropleth maps illustrating the spatial distribution of reported ameloblastoma cases. Reported case counts were classified using the Jenks natural breaks method, which minimizes within-class variance and maximizes between-class variance, thereby highlighting meaningful spatial patterns. Statistical validation of urban–rural differences To formally evaluate whether differences in reported ameloblastoma case rates across district classification groups were statistically significant, a Kruskal–Wallis test was performed with districts categorized into three groups: urban centres (Makassar, Parepare); peri-urban transitional districts (Gowa, Sidenreng Rappang, Pinrang); and rural districts (the remaining 19 districts). Pairwise rate comparisons between groups were performed using Poisson exact rate ratios (RR) with 95% confidence intervals and a Bonferroni-corrected significance threshold (α = 0.05). Spatial diagnostic tests and correlation analysis Global Moran’s I statistics evaluated spatial autocorrelation of reported ameloblastoma case rates across districts. Spatial heterogeneity and heteroscedasticity were examined using the Breusch–Pagan test and the spdep package.15 Correlation analysis identified sociodemographic variables relevant to reported ameloblastoma case rates using the ggplot2 and correlation packages.16,17 Pearson’s correlation analysis evaluated the associations between reported ameloblastoma case rates and district-level explanatory variables. Only variables indicating statistically significant relationships were retained, while nonsignificant variables were excluded as they did not provide strong contributions to spatial variation in reported ameloblastoma case rates.18 Geographically weighted regression Given the evidence of spatial heterogeneity, GWR was applied as a hypothesis-generating exploratory tool to examine potentially heterogeneous local associations between sociodemographic factors and reported ameloblastoma case frequencies using the GWmodel package.19 However, with N = 24 spatial units – substantially below the conventionally recommended minimum of 50 to 100 for stable GWR estimation – local coefficient estimates, particularly in rural districts with few reported cases, should be interpreted with considerable caution as exploratory rather than confirmatory findings. An adaptive bisquare kernel was selected because the 24 administrative districts of South Sulawesi are heterogeneous in geographic extent (urban districts such as Makassar and Parepare are compact, while eastern districts such as Luwu Timur span substantially larger territories); a fixed-distance kernel would produce insufficient neighbours for geographically extensive rural districts, whereas an adaptive kernel adjusts the radius to include a consistent number of neighbouring units at each regression point. The optimal bandwidth was selected by minimizing the corrected Akaike Information Criterion (AICc) as implemented in the GWmodel package. AICc-based selection was preferred over R² maximization to penalize model complexity and reduce overfitting risk, a particular concern given N = 24 spatial units. The AICc-optimized bandwidth yielded 22 nearest neighbours. Prior to GWR estimation, Variance Inflation Factors (VIF) were calculated for all candidate predictors; only predictors with statistically significant Pearson correlation with reported case rates and VIF < 5 were retained. All predictors were standardized prior to model fitting. A Poisson GWR sensitivity analysis was conducted in parallel using ggwr.basic with a log link function, given the highly skewed case-count distribution; results are presented in Supplementary Table S1. Local regression coefficients were estimated for each district and mapped to visualize spatial variations in relationships. Given the small spatial-unit sample, results are not intended for confirmatory inference; local coefficient estimates in districts with 1 to 3 reported cases are presented with appropriate caution. The interpretation of GWR results incorporated urban–rural contextual differences, particularly in relation to healthcare accessibility and referral patterns. Bias The main biases in the present study were referral and detection biases from histopathologically confirmed ameloblastoma cases at urban centres. To mitigate these, only cases with definitive histopathological diagnoses were included to minimize misclassification and ensure accuracy. Records were extracted using predefined criteria. Cases were included if diagnosed with ameloblastoma based on ICD-10 codes (D16.4-D16.5) and confirmed through histopathological examination demonstrating characteristic odontogenic epithelial islands with peripheral palisading, reverse nuclear polarization, and stellate reticulum-like areas according to the World Health Organization classification criteria. The data were cross-checked across databases to identify and reduce duplication. Any discrepancy was resolved by discussion with clinical experts (AT, YYA). District-level aggregation standardized spatial analysis units and prevented facility-level reported case rate inflation. Ecological correlations were interpreted within a health systems framework, recognizing that recorded case rates may reflect diagnostic capacity rather than true disease occurrence. GWR reduced spatial nonstationarity bias by allowing local variations in socioeconomic relationships, while urban-rural stratification contextualized case concentration and nonsignificant spatial autocorrelation cautioned against clustering overinterpretation. Natural breaks of reported ameloblastoma case rates The reported case rates of ameloblastoma were grouped into classes based on the natural distribution of the data using the natural breaks method. This method was selected due to its ability to identify class boundaries that statistically separate the data values into natural clusters. The main purpose of using natural breaks is to enable the identification of areas with similar levels of vulnerability, thereby supporting a more accurate spatial interpretation.20 The results of the classification and spatial analyses were visualized in the form of thematic maps using the tmap and sf packages. Statistical and spatial analysis Descriptive statistics were used to summarize patient characteristics and district-level variables. Pearson’s correlation analysis was performed to evaluate the global associations between reported ameloblastoma case rates and district-level explanatory variables. The correlation coefficients (r) and corresponding P values were reported. All statistical analyses were conducted using R software version 4.4.0 and RStudio version 2025.05.0+496 with the assistance of various relevant statistical and spatial packages. Results Patient characteristics A descriptive analysis of patients from South Sulawesi revealed an approximately equal distribution of sexes, with a slight predominance of males (Table; 51% male, 49% female). The majority of patients were within the productive age range, specifically 26 to 45 years, followed by those aged 12 to 25 years. Educational attainment was primarily at the secondary school level, although a significant proportion of patients had not completed formal education. Employment status indicated a high proportion of individuals who were either unemployed or engaged in informal work. The demographic profile of the patients predominantly comprised working-age adults with moderate to low levels of educational attainment and socioeconomic status. Regarding histological subtype, all 160 histopathologically confirmed cases were documented solely as ‘Ameloblastoma’ without further subtype specification, reflecting the diagnostic reporting capacity of participating regional hospitals in South Sulawesi. All cases met inclusion criteria based on ICD-10 codes (D16.4-D16.5) and histopathological confirmation of characteristic odontogenic epithelial features consistent with WHO 2022 classification criteria. Subtype-stratified analysis was therefore not feasible.TableSociodemographic characteristics of patients with ameloblastoma in South Sulawesi (2022-2024).Table dummy alt textVariableCategoryn%SexMale8251.25Female7848.75Age group (y)0-11127.5012-255131.8826-455936.8846-803823.75Highest educational attainmentNo formal education2918.13Primary school2012.50Junior secondary school1811.25Senior secondary school7245.00Diploma/bachelor’s degree1911.88Master’s degree21.25OccupationUnemployed3924.38Daily labourer85.00Private employee127.50Housewives2817.50Student2616.25Retirement21.25Farmer/fisherman95.63Civil servant106.25Self-employed2616.25 TableSociodemographic characteristics of patients with ameloblastoma in South Sulawesi (2022-2024).Table dummy alt textVariableCategoryn%SexMale8251.25Female7848.75Age group (y)0-11127.5012-255131.8826-455936.8846-803823.75Highest educational attainmentNo formal education2918.13Primary school2012.50Junior secondary school1811.25Senior secondary school7245.00Diploma/bachelor’s degree1911.88Master’s degree21.25OccupationUnemployed3924.38Daily labourer85.00Private employee127.50Housewives2817.50Student2616.25Retirement21.25Farmer/fisherman95.63Civil servant106.25Self-employed2616.25 Sociodemographic characteristics of patients with ameloblastoma in South Sulawesi (2022-2024). Spatial classification of reported ameloblastoma case rates Utilizing the Jenks natural breaks method,21 district-level classification revealed reported case rate variation across 24 districts/cities in South Sulawesi (Figure 1). During 2022 to 2024, 15 districts (62.5%) had the lowest reported case rates (1-3 cases), mainly in rural areas with limited diagnostic services. Four districts (Barru, Bone, Jeneponto, and Luwu Utara) reported 3 to 9 cases, while Gowa and Sidenreng Rappang had 9 to 17 cases. These districts lie along urban peripheries and referral corridors. The highest reported case rates (23-27 cases) occurred in Makassar and Parepare, the main urban centres with concentrated hospitals and diagnostic facilities. This pattern reveals an urban–rural gradient in reported ameloblastoma case rates and service availability (Figure 1).Fig. 1Natural breaks classification of reported ameloblastoma case rates across districts/cities in South Sulawesi Province during 2022 to 2024. The numbers in parentheses indicate the number of districts/cities in each reported case rate class. Reported case rates were classified using the Jenks natural breaks method to identify meaningful spatial groupings based on the distribution of the reported cases. This figure was generated using R software version 4.4.0 (https://cran.r-project.org/) and RStudio Desktop version 2025.05.0+496 for the Windows operating system. Several R packages were utilized to produce this figure, including sf (https://r-spatial.github.io/sf/), dplyr (https://github.com/tidyverse/dplyr), tmap (https://github.com/r-tmap/tmap), and leaflet (https://rstudio.github.io/leaflet/). The spatial classification revealed five reported case rate categories as follows: (1-3 cases): Luwu Timur, Toraja Utara, Tana Toraja, Palopo City, Luwu, Wajo, Soppeng, Pangkajene and Islands, Maros, Sinjai, Bantaeng, Bulukumba, Selayar Islands, Enrekang, and Takalar (15 districts/cities). (3-9 cases): Barru, Bone, Jeneponto, and Luwu Utara (four districts/cities). (9-17 cases): Gowa and Sidenreng Rappang (two districts/cities). (17-23 cases): Pinrang (one district). (23-27 cases): Makassar and Parepare (two cities).Fig 1 dummy alt text Natural breaks classification of reported ameloblastoma case rates across districts/cities in South Sulawesi Province during 2022 to 2024. The numbers in parentheses indicate the number of districts/cities in each reported case rate class. Reported case rates were classified using the Jenks natural breaks method to identify meaningful spatial groupings based on the distribution of the reported cases. This figure was generated using R software version 4.4.0 (https://cran.r-project.org/) and RStudio Desktop version 2025.05.0+496 for the Windows operating system. Several R packages were utilized to produce this figure, including sf (https://r-spatial.github.io/sf/), dplyr (https://github.com/tidyverse/dplyr), tmap (https://github.com/r-tmap/tmap), and leaflet (https://rstudio.github.io/leaflet/). The spatial classification revealed five reported case rate categories as follows: (1-3 cases): Luwu Timur, Toraja Utara, Tana Toraja, Palopo City, Luwu, Wajo, Soppeng, Pangkajene and Islands, Maros, Sinjai, Bantaeng, Bulukumba, Selayar Islands, Enrekang, and Takalar (15 districts/cities). (3-9 cases): Barru, Bone, Jeneponto, and Luwu Utara (four districts/cities). (9-17 cases): Gowa and Sidenreng Rappang (two districts/cities). (17-23 cases): Pinrang (one district). (23-27 cases): Makassar and Parepare (two cities). Population-adjusted reported case rate After adjustment for district-level population, the spatial ranking of reported ameloblastoma case rates per 100,000 population differed substantially from the ranking based on absolute case counts (Figure 2). Parepare recorded the highest population-adjusted rate (15.54 per 100,000), followed by Pinrang (4.95), Sidenreng Rappang (4.23), and Barru (3.70). Despite recording the highest absolute case count (n = 27), Makassar ranked sixth (1.84 per 100,000), reflecting its substantially larger population base. The lowest rates were observed in Tana Toraja (0.34) and Toraja Utara (0.36), both rural highland districts with limited referral infrastructure. These findings demonstrate that reliance on absolute counts substantially overstates the burden in the most populous urban centre and underestimates the relative burden in smaller cities and peri-urban districts along the referral corridor. Statistical validation of urban–rural differences Reported case rates were 3.25 per 100,000 in the peri-urban group, 2.97 in the urban group, and 0.91 in the rural group. The Kruskal–Wallis test indicated no statistically significant overall difference among groups (P = .076); however, the P value approaching .05 suggests a trend towards higher rates in urban and peri-urban districts. Given the small sample size (N = 24) and unequal group sizes (urban n = 2, peri-urban n = 3, rural n = 19), the test may have been underpowered to detect moderate omnibus differences. Pairwise Poisson exact rate comparisons, which test directional hypotheses, demonstrated significant differences (see below). Pairwise Poisson exact rate comparison showed no significant difference between urban and peri-urban groups (RR = 0.91; 95% CI: 0.61-1.37; P = .696), but both the urban group (RR = 3.27; 95% CI: 2.21-4.85; P < .001) and the peri-urban group (RR = 3.59; 95% CI: 2.40-5.35; P < .001) showed significantly higher rates than rural districts. These findings confirm that both urban and peri-urban regions exhibit substantially higher ameloblastoma reported case rates compared with rural regions. Association between health system capacity and reported ameloblastoma case rates Pearson’s correlation analysis indicated strong positive associations between reported ameloblastoma case rates and health system capacity indicators at the district level (Figure 3). The number of referral facilities was strongly correlated with reported case rates (r = 0.826; P < .0001). The number of anatomical pathologists indicated the highest correlation (r = 0.885; P < .0001), followed by oral and maxillofacial surgeons/oncologists (r = 0.876; P < .0001).Fig. 3The correlation matrix shows the indicators of health personnel and facilities for reported ameloblastoma case rates.Fig 3 dummy alt text The correlation matrix shows the indicators of health personnel and facilities for reported ameloblastoma case rates. Association between socioeconomic indicators and reported ameloblastoma case rates in South Sulawesi (2022-2024) Districts with extensive healthcare resources recorded more confirmed ameloblastomas, suggesting a link between diagnostic capacity and case ascertainment. Districts with limited facilities reported fewer cases, whereas urban centres contributed more, indicating that reported case rate variations align with health service distribution (Figure 4). Poverty rate indicated a negative correlation (r = −0.493; P < .01), whereas per capita expenditure (r = 0.698; P < .001), years of schooling (r = 0.617; P < .01) and unemployment rate (r = 0.639; P < .01) indicated positive correlations. Population growth rate (r = −0.158; P ≥ .05), Gini ratio (r = 0.268; P ≥ .05) and life expectancy (r = 0.315; P ≥ .05) did not exhibit significant differences. Higher reported case rates of ameloblastoma were observed in economically active and educated districts, whereas high-poverty districts reported fewer cases, reflecting healthcare access variations rather than biological effects.Fig. 4The correlation matrix shows the socioeconomic indicators of reported ameloblastoma case rates in South Sulawesi (2022-2024).Fig 4 dummy alt text The correlation matrix shows the socioeconomic indicators of reported ameloblastoma case rates in South Sulawesi (2022-2024). Spatial modelling of district-level sociodemographic variation and case detection Preliminary spatial diagnostics indicated lack of global spatial autocorrelation in reported ameloblastoma case rates (Global Moran’s I, P = .264). The Breusch–Pagan test revealed spatial heterogeneity (P = .021), indicating nonconstant variance between socioeconomic variables and reported case rates. These findings support the use of GWR. GWR analysis indicated local variations in socioeconomic determinants (Figure 5). Of the 24 districts/cities, 22 demonstrated at least one significant socioeconomic variable (α = 0.05), whereas Luwu Utara and Luwu Timur exhibited none. Significant differences were noted in the poverty rate in 16 districts, in the mean years of schooling in 12 districts, and in the unemployment rate in 11 districts. Per capita expenditure demonstrated significant differences in four districts.Fig. 5Influence of sociodemographics on reported ameloblastoma case rates with GWR in South Sulawesi (2022-2024). Results are exploratory and should be interpreted with caution given N = 24 spatial units.Fig 5 dummy alt textFig. 2Geographic distribution and population-adjusted reported ameloblastoma case rates across districts/cities in South Sulawesi Province, Indonesia (2022-2024). The upper panel presents a choropleth inset map of South Sulawesi overlaid on the Indonesia archipelago, illustrating the spatial distribution of absolute reported case counts, with warmer colours (orange to red) indicating higher case frequencies. The lower panel presents a horizontal bar chart of reported case rates per 100,000 population for all 24 districts/cities, ranked in descending order. Parepare recorded the highest population-adjusted rate (15.54 per 100,000), substantially exceeding all other districts; Makassar, despite the highest absolute count, ranked sixth (1.84 per 100,000) after population adjustment. The figure was generated using R version 4.4.0 and RStudio Desktop 2025.05.0+496 (sf, tmap, ggplot2 packages); population data: Central Bureau of Statistics (BPS) South Sulawesi Province (2024).Fig 2 dummy alt text Influence of sociodemographics on reported ameloblastoma case rates with GWR in South Sulawesi (2022-2024). Results are exploratory and should be interpreted with caution given N = 24 spatial units. Geographic distribution and population-adjusted reported ameloblastoma case rates across districts/cities in South Sulawesi Province, Indonesia (2022-2024). The upper panel presents a choropleth inset map of South Sulawesi overlaid on the Indonesia archipelago, illustrating the spatial distribution of absolute reported case counts, with warmer colours (orange to red) indicating higher case frequencies. The lower panel presents a horizontal bar chart of reported case rates per 100,000 population for all 24 districts/cities, ranked in descending order. Parepare recorded the highest population-adjusted rate (15.54 per 100,000), substantially exceeding all other districts; Makassar, despite the highest absolute count, ranked sixth (1.84 per 100,000) after population adjustment. The figure was generated using R version 4.4.0 and RStudio Desktop 2025.05.0+496 (sf, tmap, ggplot2 packages); population data: Central Bureau of Statistics (BPS) South Sulawesi Province (2024). Full GWR model diagnostics are presented in Supplementary Table S2. The unadjusted GWR R² was 0.7766, and the adjusted GWR R² was 0.3144, with an effective number of parameters of 15.51 and effective degrees of freedom of 8.49. The discrepancy between unadjusted and adjusted R² reflects high model parameterization relative to N = 24 spatial units, consistent with the exploratory rather than confirmatory purpose of the analysis. Local R² values ranged from 0.2343 to 0.8518 (median 0.7905; mean 0.7520), with lower local fit in rural districts with fewer reported cases. The AICc-optimized bandwidth of 22 of 24 nearest neighbours indicates that each local regression incorporates the majority of available observations, producing spatially smooth rather than sharply localized estimates. Pre-GWR VIF diagnostics confirmed no severe multicollinearity (all VIF < 5; range 1.288-3.956; Supplementary Table S3). Discussion The present study mapped ameloblastoma cases across South Sulawesi districts, examined associations between reported case rates and health system/socioeconomic factors, and used GWR to explore spatial variations in socioeconomic determinant effects, considering urban-rural differences. Spatial mapping revealed that reported ameloblastoma case concentrations were notably higher in urban centres, such as Makassar and Parepare, with sporadic cases in peripheral districts and an absence of significant global clustering. The spatial concentration in Makassar and Parepare is most parsimoniously explained by the concentration of diagnostic infrastructure, referral hospitals, anatomical pathologists, and oral and maxillofacial surgeons in these urban centres, rather than by a true elevation in disease occurrence. This interpretation is directly supported by the strong correlations between health system capacity indicators and reported case frequency (r = 0.826-0.885; P < .0001), which substantially exceed associations with socioeconomic variables. Population adjustment further revealed that Parepare, rather than Makassar, carried the highest relative reported case rate (15.54 vs 1.84 per 100,000), a distinction obscured by absolute counts alone. Accordingly, the primary finding is not that ameloblastoma is more common in urban South Sulawesi, but that it is more frequently detected and confirmed there owing to superior diagnostic access. Interventions should therefore prioritize expanding diagnostic capacity and specialist availability in rural and peripheral districts to reduce the detection gap, rather than targeting disease prevention in urban areas. To contextualize whether the urban concentration of ameloblastoma cases reflects a disease-specific geographic pattern or a general feature of oncological detection in South Sulawesi, provincial cancer prevalence data from the 2023 Indonesian National Health Survey (Survei Kesehatan Indonesia/SKI 2023)22 were examined. The provincial diagnosed cancer prevalence in South Sulawesi was 0.8 per mil (95% CI: 0.6-1.2), below the national average of 1.2 per mil.22 These findings carry direct implications for the geographic organization of diagnostic and oral healthcare services. In a comparable ecological, geographic information systems–based study of dental accessibility across Japanese municipalities, Huang et al23 recommended that local governments increase the number of dental clinics in areas of disadvantaged accessibility and optimise their geographic distribution, arguing that dental facilities should assume an active role in oral primary health care by promoting preventive care, raising oral health awareness, and encouraging regular examinations. Although their work addressed the prevention of dental caries rather than the detection of odontogenic tumours, the underlying spatial logic is transferable to the present context: where diagnostic infrastructure is unevenly distributed, geographic proximity to competent services becomes a determinant of whether disease is identified and confirmed at all. In the case of ameloblastoma, however, the priority is not the density of general dental clinics but the strategic placement of histopathology and specialist maxillofacial capacity, since it is the scarcity of these confirmatory services rather than of primary care per se that most plausibly underlies the pronounced urban concentration of confirmed cases observed here. Nationally, the SKI 2023 data revealed a pronounced urban–rural gradient in diagnosed cancer prevalence (1.5 per mil urban vs 0.8 per mil rural), as well as positive associations with educational attainment (0.6 per mil for no formal education vs 2.5 per mil for tertiary-educated individuals) and socioeconomic status (1.0 per mil in the lowest economic quintile vs 1.8 per mil in the highest). These nationally representative patterns are strikingly parallel to the socioeconomic and urban–rural correlates of reported ameloblastoma case rates observed in South Sulawesi, suggesting that the observed distribution largely mirrors broader patterns of diagnostic access and healthcare utilization rather than representing a disease-specific geographic phenomenon. Notably, the SKI 2023 data are population-based rather than hospital-based, offering an important complement to the hospital-registry data informing the present study. Ecological analysis revealed strong associations between reported case rates and health system capacity indicators, moderate associations with socioeconomic indicators, and a lack of significant relationships with demographic variables.10,14 GWR analysis indicated spatial variations in socioeconomic determinant effects across districts, with distinct urban–rural patterns, highlighting the requirement for contextual interpretations rather than uniform explanations.11,24 Geographic inequalities in oral health outcomes, as observed in the present study, often reflect disparities in access to dental care and diagnostic services rather than true biological variation in disease occurrence.14,25 This interpretation is consistent with evidence from health systems and cancer care disparities research in low- and middle-income countries, indicating that reported disease occurrence frequently reflects healthcare service availability rather than true population burden.10 Consequently, areas with greater healthcare infrastructure and specialist availability tend to report higher disease detection rates owing to improved diagnostic capacity and case ascertainment14; the spatial patterns of reported case rates observed in this study are most parsimoniously explained by such disparities in diagnostic access rather than by true variation in ameloblastoma occurrence across districts, as evidenced by the substantially higher correlations between health system capacity indicators (r = 0.826-0.885) and reported case frequencies compared with socioeconomic predictors. Similar patterns have been reported in Nigeria, where referral hospitals recorded higher numbers of ameloblastoma cases owing to more advanced diagnostic infrastructure.6 The present study revealed a strong association between health system capacity and reported ameloblastoma case rates. Advanced facilities, pathologists, and maxillofacial surgeons indicated high correlation coefficients annually. Since histopathological confirmation is the diagnostic standard for ameloblastoma,26 districts lacking these services are disadvantaged in terms of case detection. Similar patterns have been observed in hospital studies from South Asia and Africa, where case numbers reflect referral bias.27, 28, 29 Beyond health system capacity, socioeconomic factors also contribute to variations in reported ameloblastoma case rates across districts. Specifically, the present analysis revealed that poverty rates were negatively correlated with reported case rates, while higher per capita expenditure and education indicated positive associations, likely reflecting greater health literacy and improved access to diagnostic care in these areas.9,30 These are ecological associations and should not be interpreted as individual-level determinants. Higher unemployment rates may indicate the referral of advanced cases from vulnerable populations, supporting referral bias in low-resource systems.1,31 The counterintuitive direction of the socioeconomic associations observed here echoes findings from the wider oral health services literature, in which economic disadvantage does not translate simply into greater service use. In a 28-year time-series analysis of public oral healthcare utilization among Malaysian adults, Tan et al32 reported that higher relative poverty incidence and higher unemployment were each significantly and negatively associated with utilization, contrary to the expectation that greater need would drive greater uptake. They attributed this pattern to indirect costs, opportunity costs, transportation barriers, and the perception of dental care as nonessential during periods of financial instability, alongside institutional advantages enjoyed by the formally employed. This interpretation reinforces the argument advanced in the present study: the positive correlations between reported case rates and expenditure, schooling, and urban-industrial context, together with the negative correlation with poverty, are more consistent with differential diagnostic access and case ascertainment than with a true biological gradient in disease occurrence. Notably, Tan et al32 also identified internet access as a positive enabling factor, underscoring the role of information and digital connectivity in linking populations to services – an observation of particular relevance to the geographically dispersed and unevenly connected districts of South Sulawesi, where awareness of, and referral to, tertiary diagnostic centres may itself be spatially patterned. A significant methodological contribution of the present study is the application of GWR, which allowed the expansion of the analysis beyond global correlations and revealed spatial nonstationarity in socioeconomic determinants. However, with N = 24 districts below the conventionally recommended minimum for robust GWR estimation, local coefficient estimates, particularly in rural districts with very few cases, should be interpreted with considerable caution. These findings are best understood as generating hypotheses for future, better-powered spatial epidemiological studies with larger spatial units or multiprovince datasets. The observed absence of global spatial autocorrelation with significant heterogeneity underscore that a single factor cannot explain reported ameloblastoma case rates across South Sulawesi. Instead, poverty, education, unemployment, and service capacity effects vary by district, supporting place-based health perspectives.24 Similar GWR analyses have shown that socioeconomic effects on health are spatially contingent at the local level.11 Urban centres indicated consistent locally significant determinants, particularly poverty and healthcare facility availability, reflecting concentrated diagnostic services. Previous studies of urban health inequality report similar patterns, in which socioeconomic-health relationships are more stable owing to dense populations.33,34 Rural districts indicated higher heterogeneity in service-related and socioeconomic factors, which was consistent with the findings that rural health outcomes are shaped by fragmented services and context-specific constraints.35,36 This spatial heterogeneity indicates that the importance of socioeconomic factors varies by district, reflecting interactions between conditions and healthcare accessibility. Several limitations affect the findings of the present study. First, because the analyses were conducted using district-level aggregated data, the observed associations represent ecological relationships and should not be interpreted as reflecting individual-level risk, thereby introducing the possibility of ecological fallacy. Second, the study was based on histopathologically confirmed ameloblastoma patients treated at a tertiary referral hospital rather than a population-based registry. Consequently, the mapped spatial patterns represent the residential distribution of patients presenting for specialist care rather than the true geographic incidence or burden of ameloblastoma. Referral pathways, healthcare-seeking behaviour, accessibility to tertiary services, and regional differences in diagnostic capacity may have influenced the observed distribution, potentially leading to underrepresentation of patients from remote or underserved districts. Third, the study relied on secondary EMR data, which are subject to variability in data completeness and quality. Fourth, the study was constrained by a small number of spatial units (N = 24 districts/cities), determined by the fixed administrative geography of South Sulawesi Province. This sample size limits statistical power for all inferential analyses. This sensitivity analysis indicates that Pearson correlation achieves 80% power only for large effects (|r| ≥ 0.50); moderate associations (|r| = 0.30-0.40) are detectable with 45% to 65% power, increasing the risk of Type II error. The Kruskal–Wallis test, with unequal group sizes (urban n = 2, peri-urban n = 3, rural n = 19), may have been underpowered to detect moderate omnibus differences, though pairwise Poisson comparisons were adequately powered for directional hypotheses. For GWR, the sample size is well below the recommended minimum of 50 to 100 spatial units for stable local coefficient estimation; accordingly, all GWR results are exploratory and hypothesis-generating. Future studies should incorporate multiprovince data or aggregate to larger spatial units (eg, regional clusters) to achieve adequate power for confirmatory spatial regression. This limitation is inherent to the administrative structure of the study area and cannot be mitigated through analytical adjustments. Accordingly, all findings should be interpreted as descriptive and hypothesis-generating rather than confirmatory. Future studies incorporating population-based registries, multicentre data from multiple referral hospitals, or multiprovince surveillance systems are warranted to validate these spatial patterns and estimate the true geographic burden of ameloblastoma. The lower reported case frequency in rural districts may, in principle, reflect a genuinely lower true prevalence of ameloblastoma due to different demographic profiles or unmeasured biological factors. However, the pattern of socioeconomic correlations observed in the present study renders this purely biological interpretation less plausible. Specifically, reported case rates were positively associated with per capita expenditure (r = +0.698; P < .001), mean years of schooling (r = +0.617; P < .01), and unemployment rate (r = +0.639; P < .01), and negatively associated with poverty rate (r = −0.493; P < .01) a configuration in which reported case frequency increases with economic activity, educational attainment, and urban-industrial context, and decreases with poverty. This pattern is more consistent with differential diagnostic access and case ascertainment across districts than with a true biological gradient, since most established biological risk factors for neoplastic disease are typically more prevalent in disadvantaged populations and would be expected to produce the opposite correlation profile. Definitively distinguishing residual true biological variation from ascertainment-driven variation would nonetheless require population-based surveillance incorporating both confirmed and clinically suspected but unbiopsied cases. The reliance on EMRs also introduces a structural ascertainment bias: districts with paper-based or incomplete digital health record systems are systematically underrepresented in the dataset, independent of true disease burden. This limitation cannot be mitigated through analytical adjustments and represents a fundamental constraint of secondary data-based spatial epidemiology in low-resource settings. The study period (2022-2024) partially overlaps with the postacute phase of the COVID-19 pandemic in Indonesia, which may have influenced case ascertainment. Residual disruptions to hospital-based services, patient reluctance to seek care, and reduced diagnostic activity, particularly in rural districts, during 2022 may have led to under-reporting in that year. Additionally, government referral regulations implemented during this period may have altered referral pathways, potentially affecting the geographic distribution of recorded cases. Conversely, a rebound in healthcare utilization in 2023 to 2024 may have captured previously delayed diagnoses. These dynamics could have contributed to apparent temporal and geographic variation in reported case frequencies and should be considered when interpreting the spatial patterns observed. Histological subtype classification was unavailable for all 160 cases, as participating regional hospitals recorded diagnoses uniformly as ‘Ameloblastoma’ without further subtype specification. This reflects an institutional reporting limitation rather than a data extraction failure and precluded subtype-stratified spatial or demographic analysis. The inability to distinguish conventional (solid/multicystic), unicystic, peripheral, and desmoplastic subtypes represents a meaningful constraint, as these subtypes differ in biological behaviour, treatment approach, and recurrence risk. Standardization of histopathological reporting according to WHO 2022 classification criteria is strongly recommended for future surveillance and research systems. In summary, the present study provides the first geospatial analysis of ameloblastoma in South Sulawesi, integrating district-level mapping and socioeconomic determinants to demonstrate the way by which spatial context and health systems shape disease patterns. It was found that ameloblastoma cases exhibited geographic variation, primarily concentrated in urban referral centres without global clustering, which strongly suggested service-driven patterns. Strong ecological associations were identified between reported case rates, healthcare capacity indicators, and socioeconomic factors affecting access to services. Furthermore, GWR revealed spatial nonstationarity and urban-rural differences in socioeconomic effects. These findings collectively indicate that the observed reported case rate patterns largely reflect differential case detection across districts, underscoring the necessity for district-specific strategies to enhance diagnostic equity and surveillance. Importantly, these are ecological patterns; future studies with individual-level data are required to examine whether patient sociodemographic characteristics independently predict ameloblastoma risk or treatment outcomes. Conclusion Reported ameloblastoma case rates in South Sulawesi (2022-2024) showed spatial concentration in Makassar and Parepare as urban referral centres. After population adjustment, Parepare carried the highest district-level reported case rate (15.54 per 100,000), while Makassar ranked sixth (1.84 per 100,000), demonstrating that absolute case counts overestimate the relative burden of the largest cities. District-reported case rates correlated with health system capacity, including advanced facilities and specialist availability. Socioeconomic factors poverty, expenditure, education, and unemployment were associated with reported case rates, whereas demographic growth and income inequality exhibited no correlation. Geospatial analysis revealed local heterogeneity without global clustering. These findings reflect the distribution of histopathologically confirmed cases within the participating hospital network and should not be interpreted as estimates of true population-based ameloblastoma incidence or as direct measures of healthcare accessibility. A foundational recommendation of this study is the establishment of a standardized, province-wide electronic health record system for oral pathology and odontogenic tumours in South Sulawesi. Without such infrastructure, population-based spatial epidemiology of rare diseases such as ameloblastoma will remain constrained by differential digital capacity across districts. This study represents a baseline analysis intended to motivate investment in surveillance infrastructure and district-specific strategies to enhance diagnostic equity. Author contributions All authors contributed substantially to the conception, design, data acquisition, analysis, and interpretation of data. All authors participated in drafting or critically revising the manuscript and approved the final version. Funding This research received no external funding. Ethics statement This study was granted ethical exemption by the Health Research Ethics Committee of the Faculty of Dentistry, Hasanuddin University (No. 253/KEPK FKG-RSGMP UH/EE/VIII/2025). Data availability The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Conflict of interest The authors declare that they have no conflicts of interest related to this study.