Data acquisition and sources

This is a secondary analysis of the Global Burden of Disease Study 2021 (GBD 2021) data. We used the GBD 2021 dataset to assess smoking-attributable digestive system cancers among populations aged ≥60 years. Data were obtained via the Global Health Data Exchange (GHDx) Results Tool, which provides open-access global and regional health metrics. GBD 2021 covers 371 diseases and injuries with estimates of incidence, prevalence, mortality, and disability-adjusted life years (DALYs), as well as 88 risk factors across behavioral, environmental, occupational, and metabolic domains¹. Our analysis focused on smoking-attributable mortality and DALYs for five major digestive system cancers: EC, SC, LC, PC, and CRC in adults aged ≥60 years worldwide. In this study, smoking-attributable estimates were based on the GBD level 3 risk factor ‘smoking’, which captures active tobacco smoking (current and former smokers of any smoked tobacco products). Secondhand smoke exposure is modeled as a separate risk factor in GBD and was not included in our definition of smoking exposure. Population attributable fractions (PAFs) and smoking-attributable deaths and DALYs for adults aged ≥60 years were taken directly from the GBD 2021 comparative risk assessment results for the level 3 risk factor ‘smoking’, rather than being recalculated from primary exposure and relative risk data. We relied on the standard GBD counterfactual risk assessment framework as described previously¹. DALYs, defined as the sum of years lived with disability (YLDs) and years of life lost (YLLs) due to premature mortality, were used to capture total disease burden^1,11. The SDI is a composite measure ranging from 0 to 1, derived from a region’s total fertility rate for those aged ≤25 years, mean years of education among individuals aged ≥15 years, and lag-distributed per capita income. Based on this metric, locations are categorized into five SDI quintiles: low, low-middle, middle, high-middle, and high^13,14. We analyzed data for 204 countries and territories from 1990 to 2021 to provide temporal trends and a comprehensive assessment of the global burden of smoking-attributable digestive system cancers in older adults. We did not identify any material missingness in the extracted GBD 2021 estimates for the included strata, so no missing-data imputation was performed. Ethical approval and informed consent were not required because only publicly available, de-identified data were used. Reporting followed the GATHER guidelines¹⁵.

Population analysis and global burden analysis

We quantified age-standardized mortality rates (ASMRs), age-standardized DALY rates (ASDRs), and their 95% uncertainty intervals (UIs) for smoking-attributable digestive system cancers among adults aged ≥60 years using GBD 2021 estimates across 204 countries and territories, 22 GBD regions, and five SDI quintiles¹. Outcomes were stratified by gender (male and female) and by eight elderly age subgroups to capture granular age-specific patterns within older populations. This stratification encompassed the following age brackets: 60–64, 65–69, 70–74, 75–79, 80–84, 85–89, 90–94, and ≥95 years. To characterize geographical heterogeneity, we produced high-resolution choropleth maps depicting the global burden of smoking-attributable digestive cancers and disparities across sociodemographic and regional contexts, highlighting priority areas for targeted interventions. Temporal trends in ASMRs and ASDRs from 1990 to 2021 were assessed using estimated annual percentage changes (EAPCs), derived from log-linear regression of ln (age-standardized rate) on calendar year (Supplementary file Equation S1). EAPCs were calculated using log-linear regression models and are presented with 95% confidence intervals (CIs), while GBD-derived estimates are reported with 95% uncertainty intervals (UIs) and Bayesian projections with 95% credible intervals (CrIs). Joinpoint regression was not applied; temporal trends were evaluated solely using EAPCs derived from log-linear regression models. To delineate distinct epidemiological patterns and burden trajectories, we performed cancer-specific analyses focusing on five digestive cancers: esophageal, stomach, liver, pancreatic, and colorectal.

Decomposition analysis

We conducted a decomposition analysis using the Das Gupta stepwise replacement method to quantify the contributions of population growth, population aging, and epidemiological change to changes in the absolute numbers of smoking-attributable deaths and DALYs from 1990 to 2021. Decomposition was performed on counts rather than age-standardized rates; trends in age-standardized rates were assessed separately using EAPCs. Following established GBD analytic conventions, the total change in burden was partitioned into three components: 1) population growth, reflecting increases due to larger absolute population size; 2) population aging, reflecting shifts in age structure toward older demographics; and 3) epidemiological change, reflecting alterations in age-specific rates independent of demographic shifts^11,16. The decomposition of changes in the number of smoking-attributable deaths and DALYs followed the identity shown in Supplementary file Equation S2.

We implemented the decomposition separately by cancer type, sex, and SDI quintile to delineate subgroup-specific patterns. This approach clarifies whether increases in burden are predominantly attributable to demographic expansion or to adverse shifts in age-specific risk, thereby informing priorities for prevention and control¹⁶.

SDI correlation analysis

We assessed the association between socioeconomic development and smoking-attributable digestive cancer burdens through SDI-based correlation analyses at both country and regional levels. At the country level, we computed Spearman rank correlation coefficients (ρ) between 2021 SDI values and age-standardized mortality and DALY rates for each cancer type across all 204 countries and territories. Because SDI and burden metrics showed skewed distributions and potentially monotonic but non-linear relationships, we used Spearman’s rank correlation rather than Pearson’s product–moment correlation to assess associations between SDI and ASMRs/ASDRs and between SDI and estimated annual percentage changes (EAPCs). Statistical significance was evaluated using two-tailed tests with α=0.05, and the Bonferroni correction was applied for multiple comparisons across the five cancer types. For regional analyses, we aggregated country-level estimates within each of the 22 GBD regions using population-weighted averages to ensure representativeness. We visualized SDI–burden relationships with scatter plots and fitted lines to screen for nonlinearity. To flexibly capture complex or non-monotonic patterns potentially indicative of differing stages of epidemiological transition, we applied locally weighted scatterplot smoothing (LOWESS) to the regional data^13,14.

Prediction analysis

To inform policy and resource planning, we stratified analyses by sex (female, male) and applied a Bayesian age-period-cohort (BAPC) modeling framework from 2022 to 2036. The model integrates age-specific patterns and temporal (period) variation to project future burdens with uncertainty, offering a comprehensive outlook for planning. We implemented BAPC using integrated nested Laplace approximation (INLA) to obtain accurate approximations to marginal posterior distributions while avoiding the mixing and convergence challenges common with Markov chain Monte Carlo (MCMC) methods. Priors for smoothness were specified via second-order random-walk (RW2) structures on age, period, and cohort effects, with hyperparameters estimated from the data. Model performance was evaluated through out-of-sample cross-validation and posterior predictive checks. Forecasts are presented as posterior medians with 95% credible intervals^17,18.

In greater detail, for each cancer type and sex, we modeled age-specific smoking-attributable deaths and DALYs using 5-year GBD age groups among adults aged ≥60 years (60–64, 65–69, 70–74, 75–79, 80–84, 85–89, 90–94, and ≥95 years). Calendar years from 1990 to 2021 were treated as periods and used as training data, and projections were generated for 2022–2036, consistent with the primary analyses. We modeled age- and period-specific smoking-attributable counts using a Poisson log-linear age-period-cohort model, with log (population) as an offset. We report posterior predictive medians and 95% credible intervals, and projected ASMRs/ASDRs were obtained via direct standardization to the GBD standard population. Model fit and predictive performance were assessed using information criteria and a hold-out validation, comparing observed versus fitted values in the most recent years.

Statistical software

The study utilized a suite of statistical and geospatial tools for data analysis. For statistical modeling, we employed R (version 4.3.3; R Foundation for Statistical Computing, Vienna, Austria)¹⁹ and Stata (version 18; StataCorp LLC, College Station, TX, USA)²⁰, developing custom scripts for decomposition, correlation, and sensitivity analyses. Spatial patterns were analyzed and visualized using ArcGIS Pro (version 3.0; Esri, Redlands, CA, USA)²¹ and QGIS (version 3.28; QGIS Development Team)²², which enabled the creation of high-resolution choropleth maps to display global burden and disparities. Finally, data visualization was performed in R to generate forest plots, scatter plots, and multi-panel figures. Time series analyses and EAPC calculations were also conducted in R²³.

Statistical significance

Statistical significance was assessed using two-sided p-values with p<0.05 considered significant. For sets of related tests (e.g. correlations across the five cancer types), we accounted for multiple comparisons but did not apply a single, strict Bonferroni-adjusted threshold across all subgroup- and region-specific analyses. P-values are therefore treated as descriptive and interpreted in relation to effect sizes and the overlap or non-overlap of 95% uncertainty intervals. Given the large number of subgroups and location-specific comparisons, we did not perform exhaustive pairwise hypothesis testing across all strata (sex, SDI quintile, age group, and region). Instead, we focused on the magnitude and direction of differences in ASMRs, ASDRs, and EAPCs, evaluated against their 95% uncertainty intervals. Credibility intervals (CrIs) for model-based estimates and UIs for GBD-derived quantities were calculated at the 95% level, with UIs from GBD data reflecting both parameter and model uncertainty in the underlying estimation process^15,24.

Global burden of smoking-attributable digestive system cancers among adults aged _≥60 years (1990–2021)

In 2021, total smoking-attributable DALYs among older adults were: EC, 3268824 (95% UI: 2663716–3802166); liver cancer, 708190 (95% UI: 240659–1196883); PC, 1102207 (95% UI: 943669–1273484); gastric cancer, 1674301 (95% UI: 1282517–2170914); and CRC, 727595 (95% UI: 451515–1021401). The corresponding ASDRs per 100000 were 298.69 (95% UI: 240.45–360.12) for esophageal, 64.33 (95% UI: 21.83–108.81) for liver, 100.44 (95% UI: 85.88–116.20) for pancreatic, 153.26 (95% UI: 117.34–198.66) for gastric, and 66.51 (95% UI: 41.24–93.47) for CRC. Compared with 1990, DALYs rose by 57.5% (esophageal), 83.1% (liver), 94.0% (pancreatic), 2.8% (gastric), and 58.0% (colorectal). EAPCs in ASDRs confirmed downward trends: -1.18 (95% CI: -1.25 – -1.10) for esophageal, -0.69 (95% CI: -0.82 – -0.56) for liver, -0.37 (95% CI: -0.42 – -0.32) for pancreatic, -2.48 (95% CI: -2.53 – -2.44) for gastric, and -1.19 (95% CI: -1.23 – -1.16) for CRC (Supplementary file Tables S1–S10).

Global burden of disease and trends in digestive system cancers attributable to smoking by sex and age

Regarding sex, deaths, DALYs, and their age-standardized rates (ASRs) for smoking-attributable EC, LC, PC, SC, and CRCs have remained consistently higher in men than in women over the long-term, reflecting historically greater smoking prevalence and exposure among males. In 2021, among adults aged ≥60 years, male-to-female mortality ratios were consistently elevated across major gastrointestinal cancers. These ratios showed a pronounced male predominance: approximately 6.3:1 for EC (24892 vs 3940 deaths), 2.6:1 for LC (22202 vs 8588), 3.4:1 for PC (38020 vs 11174), 4.9:1 for SC (68540 vs 14116), and 2.9:1 for CRC (23546 vs 8070). Corresponding DALY ratios were similarly elevated at 6.4:1 for EC (577145 vs 90274), 2.7:1 for LC (735625 vs 276956), 3.4:1 for PC (774545 vs 226654), 4.9:1 for SC (1344196 vs 274739), and 2.9:1 for CRC (491730 vs 167801). Although these disparities have persisted from 1990 to 2021, the gender gap in ASRs has narrowed slightly, particularly in high-income regions where male smoking has declined. (Supplementary file Figure S1).

From 1990 to 2021, absolute deaths and DALYs from all five cancers rose among adults aged ≥60 years, paralleling global population aging and patterns of cumulative long-term smoking exposure described in prior studies. The steepest rises occurred in those aged 75–89 years; for instance, SC deaths in this band increased from about 25000 in 1990 to over 45000 in 2021, accounting for nearly half of the total increase across ages. Similar patterns were observed across cancers, with PC and CRCs showing the highest proportional growth in ages 80–84 and 85–89 years (about 120–150% increases), likely compounded by comorbidities and delayed diagnosis. Despite rising counts, ASMRs and ASDRs generally declined relative to 1990 across age bands, with the most pronounced improvements in the younger elderly (60–69 years), where EAPCs ranged roughly from -2.5 to -1.0 for most cancers, a pattern consistent with concurrent advances in prevention and treatment access. However, small late-life reversals were observed: in CRC, ASMRs rose slightly from 4.5 to 4.8 per 100000 and ASDRs from 95 to 102 per 100000 among those aged >80 years in 2021 versus 1990, potentially linked to rising obesity and dietary risks interacting with lifelong smoking; LC ASDRs also edged up among those aged ≥90 years (EAPC +0.2) (Supplementary file Figure S2).

Global burden of disease and trends in digestive system cancers attributable to smoking in different SDI levels

From 1990 to 2021, pronounced disparities were evident in the smoking-attributable burden of digestive system cancers across sociodemographic index (SDI) strata. High SDI regions consistently showed the lowest ASMRs and ASDRs for all major digestive cancers, whereas low and low-middle SDI groups bore the highest burdens throughout. In 2021, for example, the male ASMR for smoking-attributable EC was about 3.7 per 100000 in high SDI regions versus over 8.0 per 100000 in low SDI regions. The patterns observed for DALYs closely mirrored those for mortality. High-SDI regions, which began with lower ASDRs in 1990, subsequently experienced the most substantial declines. This was evidenced by a reduction of >50% in SC DALY rates in these regions. In contrast, many low and low-middle SDI settings demonstrated only modest improvements or a trend toward a plateau. Despite improvements in age-standardized metrics, absolute deaths and DALYs increased in most SDI groups, especially in low and middle-SDI regions, reflecting population growth and aging; for instance, low-middle SDI regions experienced substantial rises in liver and PC deaths, whereas high SDI regions maintained stable or slightly declining counts. Across all SDI categories and time points, males experienced substantially higher burdens than females, with male-to-female ratios for both death and DALY rates typically exceeding 2:1 and remaining relatively stable over three decades (Supplementary file Figure S3).

Global burden of disease and trends in digestive system cancers attributable to smoking in different regions

From 1990 to 2021, East Asia and South Asia consistently contributed the largest absolute numbers of smoking-attributable deaths and DALYs from digestive system cancers, driven by large and aging populations alongside persistently high smoking prevalence. In contrast, high-income regions such as high-income North America and Western Europe achieved the steepest declines in ASMR and DALY rates (ASDRs) owing to robust tobacco control, earlier detection, and better treatment access. In 1990, historically high ASMRs and ASDRs for esophageal and liver cancers were observed in regions with entrenched tobacco epidemics, notably East Asia and Central Europe. Driven by decades of prior smoking prevalence, the rates in these areas frequently surpassed 15–25 per 100000. Over the subsequent three decades, comprehensive public health measures in affluent areas substantially reversed these trends, while many low- and middle-income regions, including parts of Sub-Saharan Africa and the Caribbean, experienced slower declines or plateaus by 2021; for example, PC rates in Eastern Sub-Saharan Africa hovered around 8–12 per 100000 versus below 4 per 100000 in Australasia. Despite widespread improvements in age-standardized metrics, the absolute numbers of deaths and DALYs rose in most regions (Supplementary file Figure S4).

Regional heterogeneity remains pronounced. East Asia continues to bear the heaviest absolute burden due to its sizable elderly population and smoking behaviors. By 2021, Western Europe registered consistently among the lowest ASMRs and ASDRs across digestive cancers, a trend attributable to comprehensive smoke-free policies and mature screening programs. These interventions facilitated sharp declines, as seen in CRC, where the ASDRs fell from roughly 150 to about 70 per 100000, corresponding to EAPCs between -2.5 and -1.8. While in South Asia, absolute DALYs saw notable surges; for instance, LC rose from approximately 120000 in 1990 to 210000 in 2021, highlighting the persistent middle-income challenges in the context of rapid industrialization. From 1990 to 2021, a rising burden of cancer deaths and DALYs was observed across many low- and middle resource settings. Southeast Asia and Central Asia maintained the highest SC ASMRs through 2021, at approximately 10–15 per 100000. Meanwhile, Sub-Saharan Africa, specifically Eastern and Southern subregions, saw the slowest progress against EC, with ASDRs declining only marginally from about 180 to 160 per 100000 (EAPC near -0.5), a challenge compounded by resource constraints and overlapping HIV epidemics (Supplementary file Tables S1 and S2). This period also saw some of the sharpest increases in the Asia-Pacific, exemplified by a more than 90% rise in PC deaths in South-East Asia, from roughly 10000 to 19000. While a broad global decline in ASRs was evident, its magnitude varied: high-income North America achieved substantial reductions (e.g. LC ASMR from about 5.5 to 3.2 per 100000), whereas Andean Latin America and Oceania showed minor rises in certain ASDRs (e.g. EAPC +0.2 for SC), reflecting vulnerabilities in geographically isolated or economically transitional contexts (Supplementary file Figure S5).

Global burden of disease and trends in digestive system cancers attributable to smoking in 204 countries and territories

From 1990 to 2021, the largest declines in ASMRs and ASDRs for smoking-attributable digestive cancers were achieved by countries that implemented comprehensive tobacco control and robust screening programs. Leading examples include Australia, Canada, the United States, Germany, and the United Kingdom, where reductions often exceeded 50–70% across most cancer sites. In contrast, progress was slower or reversed in parts of Sub-Saharan Africa, the Caribbean, and Central Asia, where weak enforcement, socioeconomic constraints, and rising initiation yielded EAPCs near zero or positive. Declines also varied by cancer within countries: Japan saw steep drops in SC (from ~15.2 to ~5.1 per 100000; EAPC -3.5) but more modest decreases for EC, influenced by alcohol and cultural factors. China experienced rapid declines in LC and CRC rates. This progress was second only to leading nations such as Australia and Sweden for liver cancer (EAPC -2.8), and Brazil for CRC, where rates fell from approximately 9.5 to 4.5 per 100000. In contrast, PC incidence rose from about 3.5 to 5.2 per 100000 (EAPC +0.9). Mongolia and Thailand exhibit some of the highest esophageal and SC rates (e.g. esophageal in Mongolia >20 per 100000 in 2021). India, Russia, and South Africa remain among the highest for liver and PCs (e.g. liver in India ~11–14 per 100000). For CRC, Hungary, Cuba, and Bolivia rank highest. Marked gender disparities in cancer burden persist. For instance, men in Russia and India face ASMRs for EC and LC that are three to five times higher than those of women. Geographically, deaths and DALYs are heavily concentrated in a few populous nations, namely China, India, the United States, and Indonesia. These four countries collectively account for a dominant share of global digestive cancer deaths, including 52.1% of liver, 60.4% of esophageal, 68.9% of stomach, 55.7% of pancreatic, and 62.3% of colorectal cancer deaths, with closely corresponding shares for DALYs (Supplementary file Tables S1–S10).

Decomposition analysis of smoking-attributable burden of digestive system cancers among populations aged _≥60 years

Decomposition analyses revealed marked heterogeneity in the drivers of smoking-attributable digestive system cancer burden among adults aged ≥60 years. From 1990 to 2021, population growth was the dominant contributor, accounting for approximately 41.27% of deaths and 43.15% of DALYs across EC, SC, LC, PC, and CRCs, followed by epidemiological transitions (29.86% of deaths; 33.47% of DALYs). Aging had a comparatively smaller overall impact, typically 5–15% across strata. These patterns, derived from aggregated SDI-level data, underscore how demographic shifts in older cohorts amplify the effects of cumulative smoking exposure and age-related vulnerabilities. Drivers varied by SDI level. In high-SDI regions (e.g. high-income North America, Western Europe), population dynamics remained the leading driver of deaths (about 35–50%), alongside reduced epidemiological influence (15–25%) and a stronger aging effect (up to 20%), yielding stable or declining net burdens due to mature tobacco control and geriatric care. As SDI decreased, the contribution of epidemiological transitions grew (reaching 40–50% in low-SDI settings such as Sub-Saharan Africa), while population factors continued to dominate (50–70%), aligning with rising age-specific death rates amid limited access to cessation services and screening. For DALYs, the population contribution declined with higher SDI (from roughly 60% in low-SDI to about 30% in high-SDI), whereas aging became more influential (15–25% in high-SDI), reflecting longer survival accompanied by greater morbidity in older populations. Epidemiological transitions peaked in middle- and middle-high-SDI regions (35–45%): transitional economies, such as those in East Asia, achieved rapid rate reductions for liver and ECs, but these gains were partially offset by rapid aging and population expansion (Figure 1).

Figure 1

Decomposition of changes in smoking-attributable deaths and DALYs from five digestive cancers in adults aged ≥60 years, 1990–2021, by SDI quintile and GBD region. Panels show (A) esophageal cancer, (B) stomach cancer, (C) colon and rectum cancer, (D) liver cancer, and (E) pancreatic cancer. For each cancer, the left panel shows the contributions of population growth, population aging, and epidemiological change to deaths, and the right panel shows the corresponding contributions to DALYs

SDI correlation analysis

Supplementary file Figure S6 illustrates the relationship between SDI and smoking-attributable deaths and DALYs from digestive system cancers at national and regional scales. At the country level, a broadly positive correlation was observed: age-standardized death and DALY rates tended to rise with increasing SDI, peaking when SDI ranged from 0.70 to 0.85, particularly for colorectal and pancreatic cancer. This indicates that more developed countries often bear higher absolute burdens of smoking-attributable digestive cancers. The correlation coefficients (ρ) between SDI and deaths were 0.28 (EC), 0.27 (SC), 0.15 (LC; p=0.032), 0.79 (PC), and 0.65 (CRC), all other p<0.001, indicating moderate associations. Corresponding ρ values for DALYs were 0.25, 0.23, 0.15 (p=0.029), 0.78, and 0.66 (all other p<0.001). Across the SDI spectrum, the smoothed curves showed an overall ‘M-shaped’ relationship: deaths and DALYs rose sharply with SDI when SDI was below 0.45 and again between 0.65 and 0.80, but declined when SDI exceeded 0.80 and in the intermediate range of 0.45–0.65. This complexity likely reflects heterogeneity in age structures, healthcare access, effectiveness of tobacco control, and lagged effects of historical smoking.

Predicted trends in 2036

Using an age–period–cohort projection framework analogous to BAPC, our forecasts indicate that by 2036, the global absolute deaths and DALYs from smoking-attributable digestive cancers in adults aged ≥60 years will continue to increase, driven by population growth and aging, whereas ASMRs are expected to decline across major sites (esophagus, stomach, liver, pancreas, and colorectum) over the subsequent 15 years (to 2036) reflecting ongoing risk reduction and treatment improvements. Specifically, site-wise ASMRs trajectories suggest gradual declines from current levels: EC from approximately the upper 0.7–0.8 per 100000 range toward lower values, SC from around 0.5–0.6 toward 0.3–0.4, LC from roughly 0.3–0.4 toward 0.2–0.3, PC from near 2.0–2.5 toward 1.5–2.0, and CRC from about 1.5–2.0 toward 1.0–1.5 per 100000, with variation by region and sex. Consistent with these trends, the aggregate ASMR for smoking-related digestive cancers in older adults is projected to decline modestly but steadily through 2036, while the corresponding death counts and DALYs rise due to demographic momentum (Supplementary file Figure S7).

Limitations

This analysis has several limitations. First, as a secondary analysis of aggregated, modelled GBD estimates, our findings are subject to measurement and modeling uncertainty (especially in data-scarce settings) and should be interpreted as ecological associations rather than individual-level causal effects. Accordingly, causal inferences cannot be established, and there is potential residual confounding from incompletely captured determinants (e.g. air pollution, occupational exposures, diet/metabolic risks, healthcare access, screening, diagnostic and treatment changes, and broader socioeconomic shifts). In addition, smoking-attributable burden may reflect co-exposures and interactions with other behavioral and environmental factors (and omission of protective factors such as diet quality, physical activity, and aspirin), which we could not explicitly model, potentially confounding SDI relationships. Our BAPC projections assume a relatively smooth continuation of past age–period–cohort patterns and therefore may not capture future shocks or structural breaks (e.g. pandemics, policy reversals, economic disruptions, conflicts, or rapid diffusion of heated tobacco and e-cigarettes)^35,36, with uncertainty increasing over longer horizons. Moreover, projections focused on adults aged ≥60 years may not be generalizable to younger adults, and EAPC summarizes trends under a log-linear assumption that may mask non-linear temporal changes. Nonetheless, convergent patterns across decomposition, SDI correlations, and projections suggest a coherent picture in aging societies: rising absolute burdens occur alongside falling age-specific risks. These descriptive findings may help inform sustained, equitable tobacco control and geriatric-appropriate prevention, detection, and treatment strategies.