Spatiotemporal co-occurrence patterns and related risk factors of esophageal and lip and oral cavity cancers among older adults across 203 countries and territories: a machine learning-driven systematic analysis Authors: Minsheng Zheng, Chi Zhang, Boyu Yang, Chenxinzi Lin, Chongjie Zhu, Yawen Chen, Qilong Wan Published: 2026 ABSTRACT AbstractBackgroundThe country-level co-burden of esophageal cancer and lip and oral cavity cancer (LOC) among older adults remains poorly characterised. We examined global spatiotemporal patterns and associated population-level exposures.Materials and methodsWe analyzed Global Burden of Disease 2021 data for adults ≥ 55 years across 203 countries and territories. Incidence rates were classified into quartile-based co-burden categories, comparing 1990–2021 trends with Socio-demographic Index (SDI) trajectories. After screening 58 candidate exposures, random forests and SHapley Additive exPlanations ranked 36. Following variance inflation factor screening, the top 25 entered multivariable negative binomial models with population offsets.ResultsCountries and territories were classified as consistent (n=73), esophageal cancer-dominant (n=61), or LOC-dominant (n=69). Increasing SDI generally coincided with declining esophageal cancer incidence, whereas LOC incidence increased in several high-SDI settings. Both cancers were associated with high alcohol use (esophageal RR=1.32, 95% CI: 1.13–1.53; LOC RR=1.27, 95% CI: 1.13–1.43) and diets low in nuts and seeds (esophageal RR=1.16, 95% CI: 1.05–1.27; LOC RR=1.08, 95% CI: 1.01–1.16). Esophageal cancer was additionally associated with iron deficiency (RR=1.31, 95% CI: 1.14–1.52), occupational exposure to particulate matter, gases, and fumes(RR=1.29, 95% CI: 1.13–1.49), and high red meat diets (RR=1.16, 95% CI: 1.02–1.33). LOC was associated with chewing tobacco (RR=1.32, 95% CI: 1.22–1.44) and high-sodium diets (RR=1.10, 95% CI: 1.01–1.20).ConclusionsThese ecological associations indicate geographic heterogeneity in cancer co-burden and exposure profiles, informing regionally tailored prevention priorities without establishing individual-level causality.Keywords: Esophageal cancer, Global Burden of Disease, lip and oral cavity cancer, risk factors, spatiotemporal analysis Background Esophageal cancer and lip and oral cavity cancer (LOC) constitute a substantial component of the global burden of gastrointestinal and head and neck malignancies [1]. According to the International Agency for Research on Cancer, approximately 511,000 individuals are newly diagnosed with esophageal cancer annually worldwide, and esophageal cancer remains one of the leading causes of cancer-related mortality [2]. LOC is also among the most common head and neck malignancies, with a global age-standardized incidence rate of 4.74 per 100,000 population in 2021 [1]. Despite advances in clinical treatment, both malignancies remain associated with poor survival and severe impairment of swallowing, speech, oral function, and facial integrity [3,4]. The long-term treatment and rehabilitation required for these cancers impose substantial economic and social costs. In China, the direct medical expenditure attributable to esophageal cancer was estimated to reach 33.4 billion USD in 2013, whereas the total annual societal cost of LOC in Australia was projected to exceed 2.1 billion AUD when productivity losses were included [5,6]. Importantly, these two malignancies share upper aerodigestive tract susceptibility and several modifiable carcinogenic exposures, which may increase the probability of co-occurrence and further complicate clinical management [7,8]. Beyond these aggregate estimates, both cancers show pronounced geographic heterogeneity. Esophageal cancer incidence is disproportionately concentrated in parts of East Asia, Central Asia, and eastern and southern Africa, whereas LOC burden is prominent in South Asia and parts of Europe [1,2,9]. Understanding their combined burden and shared determinants is therefore important for developing efficient prevention and screening strategies, particularly for older adults and resource-limited settings [9–11]. Available literature indicates that dominant risk profiles vary substantially across regions. Tobacco and alcohol are established risk factors for upper aerodigestive tract cancers, and their joint association is particularly well documented for esophageal squamous cell carcinoma [7,12]. In South Asia, smokeless or chewing tobacco and betel-quid use are strongly associated with oral cancer, while smokeless tobacco products from Bangladesh contain high concentrations of carcinogenic tobacco-specific nitrosamines [13,14]. In some Western populations, smoking and heavy alcohol consumption have historically been prominent risk factors for esophageal squamous cell carcinoma, but these exposures do not fully explain the high incidence observed in other regions [15]. In the high-incidence esophageal squamous cell carcinoma regions of Asia and eastern and southern Africa [9,15], additional regionally variable exposures may include dietary factors, high-temperature beverages, poor oral health, nitrosamines, and polycyclic aromatic hydrocarbons [15,16]. These geographic differences provide a rationale for evaluating shared and disease-specific risk factors within a unified global framework. Biologically and epidemiologically, a shared burden of esophageal cancer and LOC is plausible because both arise within the upper aerodigestive tract and may be influenced by field cancerization, chronic mucosal injury, alcohol and tobacco exposure, dietary imbalance, nutritional deficiencies, and occupational carcinogens [7,8,17–20]. However, most epidemiological investigations have examined these malignancies separately or within selected regional populations. Consequently, it remains unclear whether they show coordinated population-level burden patterns across countries and territories and which behavioral, dietary, occupational, and socioeconomic factors are associated with their geographic heterogeneity. Clinically, it is also uncertain whether regional co-burden patterns can help identify older high-risk populations who may benefit from coordinated risk assessment and screening [10,11]. To address this gap, we conducted a country/territory-level ecological analysis using GBD 2021 data to characterize the spatial co-occurrence patterns of esophageal cancer and LOC among adults aged 55 years and older. We aimed to determine whether these malignancies cluster into shared or cancer-dominant burden patterns, identify independent shared and disease-specific risk factors, and generate a composite risk index to support region-specific public health prioritization. The specific aims of the study were: 1) to measure and compare the 30-year spatiotemporal evolution and co-burden of these dual malignancies across 203 countries and territories using Global Burden of Disease (GBD) 2021 data and the socio-demographic index (SDI); 2) to systematically screen 58 candidate variables to identify and estimate the impact of independent shared and disease-specific risk factors; and 3) to construct a composite risk index to map the regional accumulation and exposure intensity of these core hazards. Methods Overview and data sources To systematically define the geographical co-burden landscape and isolate the driving determinants of esophageal and LOC cancers, we designed a global ecological study utilizing the GBD 2021 framework. Our primary analysis targeted populations aged 55 years and older, capturing cross-sectional epidemiological data across 203 countries and territories for the year 2021. The GBD taxonomy classifies both esophageal cancer and LOC as Level 3 causes of health loss. Specifically, disease endpoints were mapped using the International Classification of Diseases (ICD) systems: esophageal cancer was defined by ICD-10 code C15 and ICD-9 codes 150-150.9, whereas LOC corresponded to ICD-10 C00–C08 and ICD-9 140-145.9. Alongside the 2021 incidence metrics, we also extracted longitudinal incidence rates spanning from 1990 to 2021 to trace the temporal evolution of these dual malignancies. To investigate upstream drivers, we retrieved the 2021 exposure estimates for a comprehensive panel of 58 environmental, occupational, and behavioral candidate risk factors. Concurrently, country-specific SDI values for 1990 and 2021 were collected to serve as a macroeconomic covariate. Each of the 203 regions functioned as an independent analytical unit, allowing us to spatially align disease incidence with localized risk exposures and SDI transitions. Map data were derived from the ArcGIS World Vector Map Set, which includes information on the political and economic levels of countries and territories worldwide. Each country or territory was defined as the basic analytical unit, and spatial matching was performed between map data and disease incidence as well as exposure data. Of the 204 countries and territories initially available from GBD 2021, Tokelau, a three-island group in the South Pacific, was excluded because it could not be matched to the ArcGIS map data, yielding a final study area of 203 countries and territories. Analytically, we diverged from relying solely on traditional regressions by coupling SHapley Additive exPlanations (SHAP) machine learning with negative binomial regression models [21]. This integrated approach was utilized to systematically screen the 58 candidate variables, isolate independent macro-level hazards, and quantify their relative impact on the combined disease burden. Finally, a composite risk index was calculated and visualized through spatial mapping to illustrate the regional intensity of cumulative risk exposures worldwide. Geographical mapping of disease co-occurrence To evaluate the spatial overlap and distinct regional burdens of these two malignancies, national incidence rates were first stratified into four ordinal tiers based on quartile thresholds: low (<25th percentile), lower-middle (25th–50th percentiles), upper-middle (50th–75th percentiles), and high (>75th percentile). Subsequently, we cross-referenced the respective tier assignments for esophageal cancer and LOC within each of the 203 analytical units. A country or territory was designated as a ‘consistent region’ if the incidence rates for both malignancies fell into the exact same quartile. Alternatively, areas demonstrating a higher quartile ranking for esophageal cancer relative to LOC were classified as ‘esophageal cancer-dominant’. Conversely, territories where the LOC incidence quartile strictly surpassed that of esophageal cancer were defined as ‘LOC-dominant’ [21]. This tripartite classification framework enabled the systematic mapping of divergent and convergent geographical patterns across the global disease burden. Spatiotemporal trends and SDI analysis To trace the long-term epidemiological evolution of the two malignancies, we calculated the Average Annual Percent Change (AAPC) in incidence rates from 1990 to 2021 using log-linear regression models [22]. Furthermore, to explore the macroscopic socioeconomic drivers behind these temporal shifts, we plotted the country-specific SDI trajectories (SDI 2021 versus SDI 1990). A diagonal reference line (y = x) was established to indicate the magnitude of socioeconomic development over the 30-year period. By mapping the AAPC direction and magnitude against these SDI trajectories across the three distinct co-occurrence regions, we systematically visualized how regional economic transitions influenced the dynamic disease burden of esophageal cancer and LOC. Independent risk factor screening protocol Rather than defaulting to the pre-established risk attributions within the GBD framework, we implemented an independent, country-level screening protocol to evaluate 58 comprehensive risk variables. This de novo approach was designed to validate macroscopic epidemiological drivers without relying on prior individual-level assumptions [23]. The feature filtering methodology was executed in progressive stages. Initially, leveraging domain expertise alongside preliminary bivariate analyses, we eliminated candidate variables lacking plausible biological or statistical relevance to upper aerodigestive malignancies. This step removed 22 covariates—including neonatal conditions, unrelated occupational hazards, and generalized climatic or sociodemographic proxies—leaving 36 viable exposures for further evaluation. In the subsequent phase, these remaining 36 parameters were integrated into a Random Forest machine learning architecture. Model predictive performance was internally verified utilizing the out-of-bag (OOB) error metric. To accurately quantify the relative predictive weight of each exposure, we calculated SHapley Additive exPlanations (SHAP) values across all geographic units. A global feature importance hierarchy was then constructed by aggregating the mean absolute SHAP scores. Finally, to optimize the balance between model interpretability and comprehensive risk coverage, we established a predefined threshold to retain only the top 25 highest-ranking features for downstream regression. This constraint successfully discarded redundant variables that offered minimal additive value. The deployment of SHAP is particularly advantageous in this context; as a robust model-agnostic technique, it adeptly untangles the complex, non-linear interactions and collinearities inherent in multifactorial population health data far more effectively than conventional linear methods [23]. Quantification of disease risk attribution To evaluate the epidemiological impact of the prioritized features, negative binomial regression frameworks were applied to compute the relative risks (RRs) alongside their corresponding 95% confidence intervals (CIs) [23]. These statistical models were intentionally constructed as unadjusted analyses, incorporating solely the selected risk-factor exposures without controlling for further covariates to isolate their direct associations. For full methodological transparency, a comprehensive catalog detailing all 58 initial candidate exposures, the step-by-step filtering outcomes, and the specific biological or statistical justifications for every exclusion is documented in Additional File 1, Table S1. Cumulative risk indexing and geospatial exposure mapping To systematically evaluate the cumulative burden of risk exposures on a national scale, a composite hazard index was formulated. The continuous exposure metrics for the seven finalized risk determinants were transformed into discrete ordinal variables based on their global quartile distributions. Each quartile was assigned an integer value ranging from 0 (lowest) to 3 (highest). A summative index was subsequently generated for every individual territory by aggregating these assigned scores, providing a macroscopic metric of combined vulnerability. Furthermore, to visually map the spatial concentration of these primary hazards, an exposure intensity matrix was developed by computing the Mean Summary Exposure Value (Mean SEV) for each core factor across the three predefined disease co-occurrence zones. The application of SEV was a deliberate methodological choice; unlike the Population Attributable Fraction (PAF), which is intrinsically linked to disease outcomes, SEV isolates and quantifies the absolute magnitude of upstream behavioral and environmental risks. This independent quantification ensures that any subsequent geographic correlations with disease incidence clusters are robust and free from tautological bias. All statistical computations and modeling were executed within the R programming environment (version 4.4.2). Statistical significance was defined at an alpha level of 0.05 utilizing two-sided testing. Results Global spatial co-occurrence patterns of esophageal cancer and LOC An epidemiological assessment of populations aged 55 and older across 203 geographic regions demonstrated that esophageal cancer poses a substantially greater global burden than LOC (31.44 versus 18.91 cases per 100,000 individuals, respectively). Analysis of quartile-stratified incidence data revealed distinct but overlapping geographical footprints. ‘Consistent regions’—where the incidence tiers for both malignancies matched—were widespread, covering Russia, Oceania, and the majority of South America. However, pronounced regional disparities emerged elsewhere: esophageal cancer was the dominant threat across East Asia and Africa, whereas LOC incidence was disproportionately elevated in North America and select European nations, generating substantial intersection on a global scale (Figure 1). Further details on these spatial dynamics are provided in Figure 1 and Additional File 1, Table S2. Global spatial distribution and co-occurrence patterns of esophageal cancer and lip and oral cavity cancer (LOC) incidence. (A, B) Global maps illustrating incidence risk levels of LOC (A) and esophageal cancer (B), categorized into quartiles. (C) Spatial distribution of co-occurrence characteristics, classifying countries into three regions. (D) Matrix showing country counts by joint incidence quartiles. The 203 analyzed territories were distributed relatively evenly across our predefined categories: 73 nations (35.96%) fell into the consistent area, 61 (30.05%) were esophageal cancer-dominant, and 69 (33.99%) were LOC-dominant. Geographically, the most severe joint burdens were concentrated in North America (e.g. Greenland: LOC 35.72, esophageal cancer 74.97 per 100,000), East Asia (e.g. Taiwan: LOC 57.25, esophageal cancer 37.83), and South Asia (e.g. Bangladesh: LOC 44.01, esophageal cancer 40.28). At the opposite end of the spectrum, the lowest dual incidence rates were recorded in the Middle East (e.g. United Arab Emirates: LOC 1.45, esophageal cancer 1.49) and West Africa (e.g. Gambia: LOC 3.59, esophageal cancer 4.84). Territories designated as LOC-dominant were heavily clustered in Europe and North America (e.g. United States, Spain, Italy, Poland), while esophageal cancer-dominant nations were primarily situated in East Asia and Africa (e.g. China, Japan, Kenya, South Africa). Spatiotemporal evolution and socioeconomic drivers To contextualize these spatial patterns within socioeconomic development, we analyzed the spatiotemporal evolution of incidence trends alongside SDI trajectories from 1990 to 2021 (Figure 2). Overall, the vast majority of countries exhibited substantial SDI improvements across the three decades. However, distinct epidemiological trajectories emerged across the predefined regions. In the esophageal cancer-dominant area, advancing SDI was largely accompanied by pronounced decreases in esophageal cancer incidence. Conversely, within the LOC-dominant area, which predominantly consists of high-SDI nations, LOC incidence rates frequently displayed alarming upward trends, indicating that enhanced socioeconomic status did not naturally eradicate LOC burdens but rather exposed vulnerabilities to persistent modern lifestyle risks. Spatiotemporal evolution of incidence trends and Socio-demographic Index (SDI) from 1990 to 2021. Scatter plots demonstrating the relationship between SDI shifts and the Average Annual Percent Change (AAPC) for esophageal cancer and lip and oral cavity cancer across the three co-occurrence regions. Identification and quantification of shared and specific risk factors To evaluate the potential impact of the 36 selected candidate variables on disease incidence, SHAP analysis was applied (Figure 3). Because these variables are ubiquitous environmental and behavioral exposures rather than disease-specific markers, our investigation was intentionally restricted to features demonstrating positive associations with cancer incidence. The distribution of SHAP values revealed a much wider variance for esophageal cancer drivers compared to LOC. For LOC specifically, the predictive influence of features ranked below the top 15 began to plateau. To maintain a standardized analytical approach and guarantee sufficient explanatory power for both malignancies, we advanced the top 25 positively associated variables into the negative binomial regression phase. SHAP summary plots of candidate risk factors for lip and oral cavity cancer (A) and esophageal cancer (B). Features are ranked top-to-bottom by overall importance. Colors indicate the original feature value, and the horizontal axis shows the SHAP value. Initial screening isolated twelve prominent predictors for esophageal cancer, encompassing high alcohol use, high red and processed meat diets, iron deficiency, low fruit intake, occupational asbestos and PMGF exposure, diets low in nuts/seeds, chewing tobacco, low seafood omega-3, diesel exhaust exposure, and low dietary fiber. Conversely, thirteen variables emerged as primary contributors to LOC incidence, including chewing tobacco, high alcohol intake, diets high in processed meat and sodium, smoking, elevated Low-density lipoprotein (LDL) cholesterol, high red meat consumption, occupational arsenic, asbestos, and sulfuric acid exposure, alongside diets low in fiber, whole grains, and nuts/seeds. Following stringent multivariable adjustments—which included the removal of highly collinear variables (Variance Inflation Factor > 10) and standardizing all exposure metrics to optimize model stability—seven variables remained as statistically significant, independent risk determinants (Figure 4). We categorized these into two shared hazards and five disease-specific drivers (three for esophageal cancer and two for LOC). Forest plots displaying relative risks (RR) and 95% confidence intervals for disease-specific and common risk factors associated with lip and oral cavity cancer (A) and esophageal cancer (B). Notably, heavy alcohol consumption and insufficient intake of nuts and seeds functioned as universal risk factors across both cancers. High alcohol exposure exerted a strong driving effect on both esophageal cancer (RR = 1.32, 95% CI: 1.13–1.53) and LOC (RR = 1.27, 95% CI: 1.13–1.43). A diet low in nuts and seeds similarly increased the incidence risk for both malignancies (esophageal cancer RR = 1.16, 95% CI: 1.05–1.27; LOC RR = 1.08, 95% CI: 1.01–1.16). Distinctly, esophageal cancer incidence was uniquely exacerbated by iron deficiency (RR = 1.31, 95% CI: 1.14–1.52), occupational PMGF exposure (RR = 1.29, 95% CI: 1.13–1.49), and high red meat consumption (RR = 1.16, 95% CI: 1.02–1.33). In contrast, the unique independent drivers for LOC were limited to chewing tobacco (RR = 1.32, 95% CI: 1.22–1.44) and a high-sodium diet (RR = 1.10, 95% CI: 1.01–1.20). Composite risk index and exposure intensity mapping The composite risk index, established based on quartiles of the seven risk factors, ranged from 2 to 17. A higher value indicated a more severe combined exposure and an increased risk of esophageal cancer and LOC. Overall, regions with the highest risk index were relatively concentrated, mainly in Southeast Asia (e.g. Thailand, Lao People’s Democratic Republic, and Viet Nam) (Figure 5A; Additional file 1, Table S3). Conversely, regions with the lowest risk index were mainly in Western Asia and the Middle East (e.g. Iran [Islamic Republic of], Syrian Arab Republic, and Lebanon). Among the 203 countries and territories, the overall frequency of the risk index approached a normal distribution, with the most common value being 10 (Figure 5B). Global spatial distribution of the composite risk index and exposure intensity matrix. (A) Spatial distribution of the calculated composite risk index based on seven independent risk factors. (B) Stacked histogram displaying the frequency distribution of the composite risk index among countries. (C) Heatmap matrix displaying the mean summary exposure value of the seven identified risk factors across the three distinct co-occurrence regions. To further decode the underlying determinants of the regional co-occurrence patterns, an exposure intensity matrix was evaluated (Figure 5C). The heatmap revealed that the formation of the LOC-dominant area was closely aligned with its disproportionately high regional exposures to a diet high in sodium (mean SEV = 54.23) and high alcohol use (mean SEV = 6.66). In stark contrast, the esophageal cancer-dominant area was profoundly shaped by severe regional vulnerabilities to iron deficiency (mean SEV = 7.86) and extensive occupational PMGF exposure (mean SEV = 17.21). This precise spatial alignment between the aggregated exposure intensities and the initial epidemiological clusters thoroughly validates that the global spatial heterogeneity of these dual malignancies is profoundly driven by synergistic, localized risk exposures. Discussion The present study provides three main findings. First, esophageal cancer and LOC formed three interpretable global co-occurrence patterns: consistent, esophageal cancer-dominant, and LOC-dominant regions. Second, SDI-related temporal patterns differed between the two malignancies, suggesting that socioeconomic development alone may not eliminate the burden of LOC in highly developed settings. Third, two shared risk factors and five disease-specific risk factors were identified, supporting the need for region-specific prevention strategies rather than uniform cancer-control programs. Our study reported a roughly equal number of countries categorized into the consistent, esophageal cancer-dominant, and LOC-dominant areas. However, the 73 countries within the consistent high-burden area demand urgent public health attention, particularly those with the highest incidence rates for both malignancies, including Greenland in North America, Taiwan in East Asia, and Bangladesh in South Asia. Previous epidemiological investigations have likewise identified a disproportionate burden of upper aerodigestive tract cancers in these regions [14,24,25]. Conversely, the lowest incidence rates for both diseases were mainly observed in the Middle East, such as the United Arab Emirates, and in West Africa, such as Gambia. This geographical contrast creates marked intra-continental heterogeneity, particularly across Asia: the Middle East generally shows low incidence of both cancers, East Asia contains many esophageal cancer-dominant countries, and South Asia includes several consistent high-burden areas. The exposure intensity matrix further showed regional differences in hazards such as iron deficiency and occupational PMGF exposure [26,27], alongside differences in healthcare infrastructure and cancer registry coverage [14,24,25,28,29]. These patterns highlight the importance of region-specific prevention and resource allocation. At the global level, country-level co-exposures to the identified risk factors showed distinct regional clustering that reflected local socioeconomic, dietary, and occupational conditions [30]. The composite risk index was particularly high in Southeast Asia, including Thailand, the Lao People’s Democratic Republic, and Viet Nam. Its geographical distribution closely aligned with the spatial distribution of incidence rates for both malignancies, supporting a potential role for the combined accumulation of localized risk exposures. Previous ecological studies have similarly associated regional economic transitions and cultural practices with co-occurrence patterns of upper aerodigestive tract tumors [31]. Our spatiotemporal analysis further revealed divergent patterns between the two cancers: increasing SDI was generally accompanied by decreasing esophageal cancer incidence in transitioning countries, whereas LOC incidence increased in several highly developed LOC-dominant regions. This contrast suggests that socioeconomic development alone may not eliminate LOC burden when lifestyle-related exposures persist [32]. Together, these findings indicate that local socio-environmental and behavioral conditions remain important to the global distribution of both cancers. Identifying regions with the highest composite risk may facilitate the targeted allocation of oncological resources [33]. In regions with high shared exposure to alcohol use and low intake of nuts and seeds, integrated noncommunicable disease prevention platforms could combine alcohol taxation or restriction policies, community-based alcohol harm-reduction campaigns, and nutrition programs that improve access to protective plant-based foods. In esophageal cancer-dominant regions with high occupational PMGF exposure and iron deficiency, priorities could include workplace air-quality regulation, occupational protection, and nutrition-oriented surveillance. In LOC-dominant regions, interventions could focus on chewing or smokeless tobacco control, cessation support, and sodium-reduction strategies tailored to local dietary practices. In high co-burden regions, health systems could incorporate oral examination and standardized alcohol, tobacco, dietary, and occupational risk assessment into routine care for adults aged 55 years and older, with risk-based referral for upper gastrointestinal evaluation in accordance with local clinical guidelines and healthcare resources [10,11]. Continued surveillance is also important in regions currently showing lower composite risk, such as Western Asia and the Middle East, because globalization, lifestyle transitions, and population aging may reshape future cancer burdens [12,34]. For esophageal cancer, the disease-specific factors identified in our analysis were occupational PMGF exposure, iron deficiency, and a diet high in red meat. Chronic occupational PMGF exposure may induce mucosal inflammation and DNA damage through the swallowing of inhaled carcinogenic particles [35]. Iron deficiency can lead to epithelial atrophy and may increase mucosal susceptibility to exogenous carcinogens [36], whereas high red meat intake may contribute through the endogenous formation of N-nitroso compounds and heme-induced oxidative stress [37,38]. For LOC, chewing tobacco and a diet high in sodium emerged as the predominant disease-specific factors. Prolonged chewing tobacco use causes sustained contact between the oral mucosa and carcinogens, promoting field cancerization and malignant transformation [39]. Excessive sodium intake may also exacerbate chronic mucosal irritation and systemic inflammation [40]. The exposure intensity matrix indicated that PMGF exposure and iron deficiency were prevalent across East, South, and Southeast Asia and sub-Saharan Africa, whereas chewing tobacco use and high-sodium diets were concentrated in parts of Asia, Africa, and Eastern Europe. Accordingly, prevention in these regions should prioritize occupational safety and workplace PMGF control, nutrition-oriented interventions in populations with a high burden of iron deficiency, and strengthened regulation and cessation programs for smokeless or chewing tobacco use. High alcohol use and a diet low in nuts and seeds were common factors for both esophageal cancer and LOC. Reducing these modifiable exposures may therefore benefit both upper aerodigestive tract malignancies. Chronic alcohol consumption causes mucosal damage and oxidative stress, while acetaldehyde acts as a carcinogen and enhances mucosal permeability to other carcinogens [41]. Because alcohol exposure was more prominent in countries across North America, Europe, and Oceania, relevant strategies include regulation, taxation, and community-based alcohol harm-reduction programs [42]. A diet low in nuts and seeds highlights the potential protective role of specific nutritional components. Nuts and seeds contain dietary fiber, polyunsaturated fatty acids, and antioxidants with anti-inflammatory properties that may help maintain mucosal integrity and neutralize reactive oxygen species [43]. This suboptimal dietary pattern was common across Russia, Central and Eastern Europe, Western and Central Asia, Southern Africa, and Southern Latin America. Population-level responses could include updating nutritional guidelines and improving access to healthy plant-based foods. Integrating alcohol harm reduction and dietary improvement into existing noncommunicable disease prevention platforms may provide a practical shared-risk strategy. A major strength of this study is its comprehensive ecological design, which integrates 30 years of global data with machine learning to screen a broad range of potential risk factors and characterize disease co-burden. Several limitations should nevertheless be considered. First, increases in incidence in some transitioning countries may partly reflect improved cancer registration and surveillance rather than a purely biological increase, potentially leading to overestimation [44]. Second, the composite risk index uses an unweighted sum and therefore does not account for differences in the estimated magnitudes of individual risk factors. Third, the cross-sectional comparison of contemporary mean SEV exposures with current incidence rates cannot fully account for the latency between exposure and cancer occurrence. Finally, because this is a country-level ecological analysis, the observed spatial alignments are hypothesis-generating and cannot establish individual-level causality. The findings should therefore be applied cautiously when considering individual clinical practice. Conclusions In conclusion, by integrating 30-year spatiotemporal trajectories with machine learning-driven risk attribution, this study comprehensively decodes the global co-occurrence patterns of esophageal cancer and LOC. We demonstrate that the striking geographical heterogeneity of these dual malignancies is profoundly driven by the synergistic, localized accumulation of specific behavioral, dietary, and occupational exposures. Crucially, advancing socioeconomic development alone is insufficient to organically eradicate this co-burden, as evidenced by the persistent vulnerability to LOC in highly developed regions driven by entrenched modern lifestyle risks. Mitigating this global health challenge requires moving beyond generic cancer control. Policymakers should use regional exposure profiles to guide alcohol harm reduction, dietary improvement, workplace PMGF control, nutrition-oriented management of iron deficiency, chewing or smokeless tobacco cessation, sodium-reduction programs, and coordinated oral and upper aerodigestive risk assessment for older adults in high-burden settings. Supplementary Material Additional file 1.docx Acknowledgements The authors highly appreciate the Global Burden of Disease (GBD) collaborators for providing the publicly available data used in this research. Funding Statement This study was supported by grants from School and Hospital of Stomatology, Wuhan University [No. LYZX202002]. Ethical approval and consent to participate The study protocol was reviewed and approved by the Ethics Committee of School and Hospital of Stomatology, Wuhan University (Approval No. [WDKQ2026] B38). The requirement for informed consent was waived by the ethics committee because the study solely utilized publicly accessible, de-identified data. Availability of data and materials All data were obtained from publicly available databases, and therefore informed consent was not required. The datasets generated during and/or analyzed during the current study are available in the GBD repository [http://ghdx.healthdata.org/gbd-results-tool]. Disclosure statement The authors declare that they have no competing interests. References Associated Data Supplementary Materials Additional file 1.docx Data Availability Statement All data were obtained from publicly available databases, and therefore informed consent was not required. The datasets generated during and/or analyzed during the current study are available in the GBD repository [http://ghdx.healthdata.org/gbd-results-tool].