Association between the Human Development Index and the burden of major gastrointestinal cancers: results from Global Burden of Disease 2021 Authors: Mehdi Azizmohammad Looha, Ali Saberi Shahrbabaki, Sina Yoosefi, Sara Javadi, Mohamad Amin Pourhoseingholi, Amir Moradi Published: 2025 ABSTRACT AbstractAim:To assess the longitudinal association between the Human Development Index (HDI) and the global burden of major gastrointestinal (GI) cancers from 1990 to 2021.Background:GI cancers contribute substantially to global incidence, mortality, and disability-adjusted life years (DALYs), with wide variation across HDI levels. Methods:Using Global Burden of Disease 2021 data, we analyzed age-standardized incidence (ASIR), mortality (ASMR), disability-adjusted life-year (ASDR), and tobacco-attributable mortality for colorectal, liver, pancreatic, and stomach cancers (1990–2021). Countries were classified as more developed (HDI ≥0.700) or less developed (HDI <0.700), and associations were assessed using longitudinal multilevel mixed-effects models and Spearman correlations.Results:In 2021, HDI was positively correlated with colorectal and pancreatic cancer burden, while liver cancer showed inverse associations only in less developed countries and stomach cancer only in more developed settings. For tobacco-attributable ASMR, associations were positive for colorectal and pancreatic cancers, non-significant for liver, and opposite for stomach (positive in less vs. inverse in more developed countries). Longitudinally (1990-2021), colorectal cancer showed a rising ASIR in both groups, but significantly divergent slopes for ASMR and ASDR, with declining rates in more developed countries and increasing rates in less developed countries. Liver cancer had significantly unfavorable slopes, with ASIR, ASMR, and ASDR increasing in more developed but declining in less developed settings. Pancreatic cancer exhibited increases in all measures, with significantly slower growth in more developed countries. Stomach cancer showed consistent declines across all measures, with considerably steeper reductions in more developed regions. Conclusion:Human development strongly shapes GI cancer burden, with divergent high- and low-HDI trends evident only in longitudinal analyses.Key Words: Gastrointestinal neoplasms, Human development, Incidence, Mortality, Disability-adjusted life years, Longitudinal studies, Socioeconomic factors, Linear models Introduction Gastrointestinal (GI) cancers represent a leading cause of global morbidity and mortality. Among them, colorectal, liver, pancreatic, and stomach cancers are of particular importance, together accounting for millions of new cases and deaths annually worldwide (1). The patterns and trends of GI cancers differ markedly across regions, with less developed areas generally experiencing a disproportionately higher burden than more developed countries (2). These cancers also share several major risk factors, including tobacco use, obesity and metabolic syndrome, unhealthy dietary patterns, chronic infections such as hepatitis viruses and Helicobacter pylori, and hereditary predisposition (3). One factor that may underlie these global differences is the level of human development, often measured by the Human Development Index (HDI) (4-8). The HDI is a composite indicator that reflects a country’s level of development by integrating life expectancy, education, and gross national income per capita (9). Unlike static biological or behavioral risk factors, HDI is dynamic and changes annually within each country, capturing shifts in socioeconomic conditions that may directly or indirectly influence cancer incidence, mortality, and outcomes (10). However, to date, very few studies have systematically examined how temporal changes in HDI relate to trends in age-standardized incidence rates (ASIR), age-standardized mortality rates (ASMR), and age-standardized disability-adjusted life-year (DALY) rates (ASDR) for these cancers (11). Accordingly, we conducted a global, regional, and national analysis to investigate how changes in HDI from 1990 to 2021 are associated with trends in incidence, mortality, and DALYs of the four primary GI cancers. We employed multilevel longitudinal models to examine these associations while controlling for variation across countries, regions, and super-regions. This approach provides a clearer understanding of how development status shapes cancer patterns over time and offers evidence to guide prevention and policy efforts. Methods Study design and setting An ecological, longitudinal study was performed using country-level data from the Global Burden of Disease (GBD) 2021 project. This study was a secondary analysis of data from the GBD 2021 (12, 13). We examined ASIR, ASMR, and ASDR for four central GI cancers: colorectal, liver, pancreatic, and stomach. The analyses were performed at different geographic levels, including 204 countries and territories, 21 regions (Andean Latin America, Australasia, Caribbean, Central Asia, Central Europe, Central Latin America, East Asia, Eastern Sub-Saharan Africa, High-income Asia Pacific, High-income North America, North Africa and Middle East, Oceania, South Asia, Southeast Asia, Southern Latin America, Southern Sub-Saharan Africa, Tropical Latin America, Western Europe, Western Sub-Saharan Africa, Central Sub-Saharan Africa, and North America), as well as seven super-regions (viz. Central and Eastern Europe and Central Asia, High-income countries, Latin America and the Caribbean, North Africa and the Middle East, South Asia, Southeast and East Asia and Oceania, and Sub-Saharan Africa) (14). Since the dataset consisted entirely of publicly available and de-identified data, ethical approval was not required. Study population The study population consisted of all reported cases of colorectal, liver, pancreatic, and stomach cancers from the GBD 2021 dataset, which incorporates information from national surveys, vital registration systems, cancer registries, and surveillance systems (14). Variables Primary outcomes were ASIR, ASMR, and ASDR for colorectal, liver, pancreatic, and stomach cancers, with secondary analyses of tobacco-attributable ASMR. Explanatory variables included year (1990–2021) and country-level HDI, with all 204 countries were classified into 21 regions and seven super-regions (14). Data sources Data were extracted from the GBD 2021 estimates via the Global Health Data Exchange (GHDx) Results Tool (http://ghdx.healthdata.org/gbd-results-tool). HDI values were obtained from the United Nations Development Programme, which defines development in terms of health, education, and standard of living. For analysis, countries were classified into two groups: more developed (high HDI: 0.700–0.799 and very high HDI: 0.800–1.000) and less developed (medium HDI: 0.550–0.699 and low HDI: <0.550) (9, 15). Statistical analysis Cross-sectional analyses (correlation and ranking) We conducted cross-sectional analyses to evaluate associations between HDI and cancer burden indicators in 2021. First, correlation analysis was performed using country-level Spearman rank correlation coefficients (rₛ) with 95% confidence intervals, computed via a Fisher z–transformation approach applied to Spearman’s ρ (normal approximation), stratified into two groups: less developed (HDI < 0.700) and more developed (HDI ≥ 0.700). Correlations were calculated annually from 1990 to 2021, and scatter plots were generated for 2021 to visualize differences in development level, overall patterns, and potential outliers. Second, HDI-based ranking and mean-rank aggregation were carried out by classifying countries into three HDI categories: low (<0.550), medium (0.550–0.699), and high/very high (≥0.700). For each of the 12 measures (ASIR, ASMR, ASDR for four cancers), countries were ranked within their HDI category, and a mean-rank was calculated as the average of these ranks. Countries with the highest mean-rank were highlighted in all categories, and those with the lowest mean-rank were additionally noted in the high/very high group. The same method was applied to tobacco-attributable ASMR. Longitudinal multilevel analysis of HDI and GI cancer burden (1990–2021) To evaluate the longitudinal effect of HDI on cancer burden indicators (ASIR, ASMR, ASDR, and tobacco-attributable ASMR) while accounting for variations across countries, regions, and super-regions from 1990 to 2021, we applied a longitudinal multilevel generalized linear mixed-effects model with a Gamma distribution and log link. The model accounted for hierarchical clustering at the country (k), region (j), and super-region (i) levels across years (t). Let Yijkt denote the outcome (ASIR, ASMR, ASDR, or tobacco-attributable ASMR) for country k in region j and super-region i at year t. Time was treated as a continuous covariate (centered so that t=0 indicates the baseline year), and development status was encoded as a binary indicator Dk (More developed if HDI ≥0.700, Less developed if HDI <0.700, based on UNDP thresholds). This cutoff follows the UNDP classification (high/very high vs. low/medium), and a binary grouping was chosen to maintain sufficient sample size and stable modeling (15, 16). An interaction term between Time and development status (HDI group) was explicitly included to assess whether temporal trajectories differed between more developed and less developed countries. The outcome was modeled as Yijkt​|​μijkt~Gamma,E(Yijkt)=μijkt,Var(Yijkt​|​μijkt)=ϕμijkt2, with the log link ηijkt=logμijkt=β0+β1t+β2Dk+B3(t×Dk)+uk∣ji+vj∣i​+wi​, Where random intercepts were specified as uk∣ji~N(0,τcountry2),​​​​​vj∣i~N(0,τregion2),wi~N(0,τsuper2), Because a log link was used, fixed-effect coefficients were exponentiated and interpreted as multiplicative effects on the mean. In this parameterization, exp(β₀) is the baseline mean outcome at t = 0 for the reference group (Less developed); exp(β₁) is the per-year multiplicative change for the reference group; exp(β₂) is the baseline rate ratio comparing More to Less developed countries at t = 0; and exp(β₃) is the ratio of yearly multipliers for More versus Less developed. Consequently, the per-year multiplicative change for the More-developed group is exp(β₁+β₃) = exp(β₁) × exp(β₃); this multiplicative interpretation is preserved when back-transforming from the log link. Population-level (marginal) mean functions were μijkt(Dk=0)=exp(β0+β1t),μijkt(Dk=1)=exp(β0+β2+(β0+β3)t) Estimation was performed using maximum likelihood (Laplace approximation) in R (version 4.1.2) with the glmmTMB package (17). Inference for fixed effects was based on Wald statistics and confidence intervals computed on the link scale, as required for the Gamma–log specification. Random-effect variance components (τcountry2,τregion2,τsuper2) were summarized by first obtaining confidence intervals on the corresponding standard-deviation scale (using likelihood profiling whenever available, with a validated Wald fallback), then squaring the bounds to report variance-scale intervals. When the software parameterized random effects on an internal scale, the intervals were back-transformed monotonically to the standard-deviation scale before being squared. If an interval could not be estimated (e.g., due to boundary fits), it was reported as 'not available'. The contribution of each random level (country, region, super-region) was assessed with likelihood-ratio tests that compare the full model to reduced models omitting one level at a time. The dispersion parameter ϕ was reported, along with a 95% confidence interval obtained from the intervals for its square-root scale, and then squared. All tabulated estimates and interval bounds were rounded to three decimals (18). Intraclass correlation coefficients (variance-partition coefficients) were computed on the latent (log) scale by adding an observation-level term appropriate for the Gamma–log model. With dispersion ϕ (shape =1/ϕ), the observation-level variance on the log scale was approximated by the trigamma function, σobs,log211ϕ​ and level-specific ICCs were defined as ICCl=τl​2τcountry2+τregion2+τsuper2+σobs,log2 (19). Results We first describe cross-sectional associations between HDI and cancer burden in 2021, then identify country-level outliers, and finally present longitudinal trend analyses. Correlation between HDI and GI cancer burden (ASIR, ASMR, ASDR, and tobacco-attributable ASMR) in 2021 Table 1 showed that in 2021, HDI was strongly associated with a higher incidence of colorectal (ASIR: rₛ = 0.82; p < 0.001) and pancreatic cancer (ASIR: rₛ = 0.80; p < 0.001), with weaker but still significant associations for ASMR and ASDR. Liver cancer showed inverse correlations with HDI across all measures (ASIR: rₛ = –0.19; p = 0.009; ASMR: rₛ = –0.27; p < 0.001; ASDR: rₛ = –0.35; p < 0.001), whereas stomach cancer showed negative and borderline correlations (ASIR: rₛ = –0.14; p = 0.053; ASMR: rₛ = –0.31; p < 0.001; ASDR: rₛ = –0.35; p < 0.001). The inverse association for liver cancer was evident only in less developed countries, while for stomach cancer it appeared only in more developed settings. Consistent with these findings, Table 2 showed that in 2021, tobacco-attributable ASMR was strongly correlated with HDI for colorectal (rₛ = 0.63, p < 0.001) and pancreatic cancer (rₛ = 0.77, p < 0.001). Similar patterns were present in both less and more developed countries, although the magnitudes were smaller within strata (rₛ = 0.47 vs. 0.27 for colorectal; rₛ = 0.59 vs. 0.47 for pancreatic cancer). For stomach cancer, the correlation was positive in less developed countries (rₛ = 0.33, p = 0.003) but inverse in more developed countries (rₛ = –0.24, p = 0.010), giving an overall positive association (rₛ = 0.26, p < 0.001). By contrast, correlations for liver cancer were not statistically significant in either stratum. Table 1Country-level Spearman correlations between HDI and ASIR/ASMR/ASDR for major gastrointestinal cancers in 2021, stratified by development status (HDI ≥ 0.7 = more developed; HDI < 0.7 = less developed).CancerHDI GroupASIRASMRASDRrₛ (95% CI)P-valuerₛ (95% CI)P-valuerₛ (95% CI)P-valueColorectalLess Developed0.43 (0.22, 0.62)<0.0010.25 (0.01, 0.45)0.0300.26 (0.04, 0.46)0.020More Developed0.67 (0.55, 0.76)<0.0010.34 (0.19, 0.48)<0.0010.23 (0.08, 0.37)0.013Total0.82 (0.77, 0.86)<0.0010.61 (0.52, 0.67)<0.0010.53 (0.44, 0.62)<0.001LiverLess Developed-0.37 (-0.56, -0.15)<0.001-0.38 (-0.57, -0.16)<0.001-0.39 (-0.58, -0.17)<0.001More Developed0.14 (-0.04, 0.33)0.1320.05 (-0.15, 0.24)0.634-0.02 (-0.22, 0.18)0.804Total-0.19 (-0.32, -0.05)0.009-0.27 (-0.41, -0.14)<0.001-0.35 (-0.47, -0.22)<0.001PancreaticLess Developed0.47 (0.26, 0.66)<0.0010.47 (0.25, 0.65)<0.0010.47 (0.25, 0.66)<0.001More Developed0.67 (0.58, 0.74)<0.0010.61 (0.50, 0.70)<0.0010.53 (0.40, 0.64)<0.001Total0.80 (0.74, 0.84)<0.0010.77 (0.71, 0.82)<0.0010.75 (0.68, 0.80)<0.001StomachLess Developed-0.01 (-0.25, 0.22)0.940-0.03 (-0.27, 0.20)0.777-0.03 (-0.27, 0.20)0.814More Developed-0.27 (-0.45, -0.09)0.004-0.42 (-0.58, -0.25)<0.001-0.47 (-0.62, -0.31)<0.001Total-0.14 (-0.28, 0.01)0.053-0.31 (-0.44, -0.16)<0.001-0.35 (-0.48, -0.21)<0.001Note: Values are Spearman’s rank correlation coefficients (rₛ) with 95% confidence intervals and two-sided p-values. Correlations were computed across countries within each HDI stratum and for the total sample; countries with missing HDI or outcome values were excluded. ASIR, ASMR, and ASDR are age-standardized incidence, mortality, and DALY rates, respectively (per 100,000 population). P-values <0.001 are reported as “<0.001.” Country-level Spearman correlations between HDI and ASIR/ASMR/ASDR for major gastrointestinal cancers in 2021, stratified by development status (HDI ≥ 0.7 = more developed; HDI < 0.7 = less developed). Note: Values are Spearman’s rank correlation coefficients (rₛ) with 95% confidence intervals and two-sided p-values. Correlations were computed across countries within each HDI stratum and for the total sample; countries with missing HDI or outcome values were excluded. ASIR, ASMR, and ASDR are age-standardized incidence, mortality, and DALY rates, respectively (per 100,000 population). P-values <0.001 are reported as “<0.001.” Table 2Country-level Spearman correlations between HDI and tobacco-attributable ASMR in 2021 by cancer site, stratified by development status (HDI ≥ 0.7 = more developed; HDI < 0.7 = less developed).GroupColorectal CancerLiver CancerPancreatic CancerStomach Cancerrs (95% CI)P-valuers (95% CI)P-valuers (95% CI)P-valuers (95% CI)P-valueLess Developed0.47 (0.29, 0.62)<0.001-0.09 (-0.32, 0.15)0.4280.59 (0.40, 0.72)<0.0010.33 (0.12, 0.50)0.003More Developed0.27 (0.12, 0.42)0.0040.10 (-0.09, 0.28)0.3140.47 (0.35, 0.58)<0.001-0.24 (-0.40, -0.07)0.010Total0.63 (0.54, 0.70)<0.0010.07 (-0.07, 0.20)0.3710.77 (0.70, 0.82)<0.0010.26 (0.10, 0.40)<0.001Note: Values are Spearman’s rank correlation coefficients (rₛ) with 95% confidence intervals and two-sided p-values. Correlations were computed across countries within each HDI stratum and for the total sample; countries with missing HDI or outcome values were excluded. Tobacco-attributable ASMR is the age-standardized mortality rate attributable to tobacco (per 100,000 population). P-values <0.001 are reported as “<0.001.” Country-level Spearman correlations between HDI and tobacco-attributable ASMR in 2021 by cancer site, stratified by development status (HDI ≥ 0.7 = more developed; HDI < 0.7 = less developed). Note: Values are Spearman’s rank correlation coefficients (rₛ) with 95% confidence intervals and two-sided p-values. Correlations were computed across countries within each HDI stratum and for the total sample; countries with missing HDI or outcome values were excluded. Tobacco-attributable ASMR is the age-standardized mortality rate attributable to tobacco (per 100,000 population). P-values <0.001 are reported as “<0.001.” HDI-stratified country-level outliers in GI cancer burden (ASIR, ASMR, ASDR, and tobacco-attributable ASMR) in 2021 Figure 1 (Tables S1–S12 in the Supplementary File) highlighted clear outliers in GI cancer indicators in 2021 across different HDI strata. For colorectal cancer (panels A–C, Table S1-S3), high values were observed in Ethiopia (ET; ASIR 16.4, ASMR 15.9, ASDR 343.0) among low-HDI countries, the Netherlands (NL; ASIR 69.8) and Uruguay (UY; ASMR 27.5) among high-HDI countries, and Hungary (HU; ASDR 615.0) as an extreme outlier. Liver cancer (panels D–F, Table S4-S6) showed elevated rates in Gambia (GM; ASIR 32.3, ASMR 33.9, ASDR 977.0) within the low-HDI group and Mongolia (MN; 74.6, 80.9, 1993.3) among high-HDI settings. For pancreatic cancer (panels G–I, Table S7-S9), Uruguay (UY; ASIR 12.6, ASMR 13.3, ASDR 297.1) stood out among high-HDI countries, while Zimbabwe (ZW; 7.3, 7.8, 191.0) was the main outlier in low-HDI settings. Finally, stomach cancer (panels J–L, Table S10-S12) revealed Afghanistan (AF; ASIR 32.9, ASMR 34.6, ASDR 910.2) as the primary outlier in low-HDI countries, and Mongolia (MN; 36.8, 37.4, 930.5) among high-HDI countries. Association between the HDI and GI cancer burden in 2021. Scatter plots show HDI (x-axis) versus (A) ASIR, (B) ASMR, and (C) ASDR for colorectal cancer; (D) ASIR, (E) ASMR, and (F) ASDR for liver cancer; (G) ASIR, (H) ASMR, and (I) ASDR for pancreatic cancer; and (J) ASIR, (K) ASMR, and (L) ASDR for stomach cancer. Countries are distinguished by development group (HDI ≥ 0.7 = more developed; HDI < 0.7 = less developed). Figure 2 (Tables S13–S16 in the Supplementary File) highlighted outliers in tobacco-attributable ASMR across different HDI strata in 2021. For colorectal cancer (panel A, Table S13), the highest ASMRs were observed in low-HDI countries such as Lesotho (LS, 0.50) and Zimbabwe (ZW, 0.49), and in high/very-high-HDI countries where Croatia (HR, 1.39), Uruguay (UY, 1.26), and Bulgaria (BG, 1.25) stood out. For liver cancer (panel B, Table S14), the highest ASMRs appeared in low-HDI countries such as Gambia (GM, 2.07) and Lesotho (LS, 1.66), and in high/very-high-HDI countries including Mongolia (MN, 7.67), Tonga (TO, 2.86), and Egypt (EG, 2.20). For pancreatic cancer (panel C, Table S15), elevated ASMRs were found in low-HDI countries such as Zimbabwe (ZW, 0.83) and Lesotho (LS, 0.80), and in high/very-high-HDI countries led by Greece (GR, 2.01), Montenegro (ME, 1.93), and Armenia (AM, 1.90). For stomach cancer (panel D, Table S16), elevated ASMRs were observed in low-HDI countries, such as Yemen (YE, 2.22) and Afghanistan (AF, 1.24), as well as in high- and very-high-HDI countries, where Mongolia (MN, 3.42) and China (CN, 3.10) were notable. Association between HDI and tobacco-attributable ASMR in 2021. Scatter plots show HDI (x-axis) versus tobacco-attributable ASMR for (A) colorectal cancer, (B) liver cancer, (C) pancreatic cancer, and (D) stomach cancer. Countries are distinguished by development group (HDI ≥ 0.7 = more developed; HDI < 0.7 = less developed). Regional contrasts in GI cancer burden and tobacco-attributable mortality by HDI, 2021 Using the within-category mean-rank defined in Methods, we observed clear contrasts in 2021 (Table S17, Supplementary File). In the low-HDI group, the highest values occurred mainly in Sub-Saharan Africa (Zimbabwe, Lesotho, Guinea-Bissau, Mali) and also in South Asia (Afghanistan) and the Middle East (Yemen). Among medium-HDI countries, the upper range was found in the Pacific Islands (Nauru, Fiji), Sub-Saharan Africa (Eswatini, Cabo Verde, Gabon), and Latin America (Bolivia). In the high/very-high HDI group, the lowest values appeared in South Asia (Sri Lanka), North Africa and the Middle East (Algeria, Jordan, Tunisia), and the Gulf region (Kuwait, Oman, Saudi Arabia). In contrast, the highest were clustered in East Asia (Japan, Republic of Korea, China), Eastern and Southern Europe (North Macedonia, Bulgaria, Slovakia, Andorra), the Pacific Islands (Palau), and Latin America (Uruguay). Using the within-category mean-rank (Table S18, Supplementary File), we observed distinct regional contrasts in tobacco-attributable burden in 2021. Among low-HDI countries, the upper range was concentrated in Sub-Saharan Africa (Zimbabwe, Lesotho, Guinea-Bissau, Mali), as well as in South Asia (Afghanistan, Pakistan) and the Middle East (Yemen). Within the medium-HDI group, higher values were observed in the Pacific Islands (Nauru, Solomon Islands, Kiribati), Sub-Saharan Africa (Eswatini, Cabo Verde, Gabon), and parts of Latin America (Bolivia, Nicaragua). In the high/very-high HDI group, the lowest values were found in South Asia and the Middle East (Sri Lanka, Oman, Saudi Arabia, Kuwait), while the highest clustered in East Asia (China, Republic of Korea, Japan, Mongolia), Eastern and Southeastern Europe (Bulgaria, Croatia, Bosnia and Herzegovina, North Macedonia), and Latin America (Uruguay, Argentina). Longitudinal multilevel impact of HDI on GI cancer burden (ASIR, ASMR, ASDR), 1990–2021 Year-by-year correlations between HDI and GI cancer burden, as well as tobacco-attributable mortality, were calculated for both less developed and more developed countries (Tables S19–S23, Supplementary File). However, no consistent temporal trend was observed. To capture longitudinal dynamics, we applied multilevel Gamma regression models with a log link. In these models, the time × development status interaction quantified the ratio of the yearly multiplicative change in more developed countries relative to less developed countries, as detailed in Table 3 (Tables S24 and S25 in the Supplementary File). Values below 1.00 indicate a slower increase or faster decline in more developed settings; values above 1.00 indicate the opposite. Across sites, interaction estimates clustered near 1.00, implying essentially parallel trajectories with small but statistically robust divergences. For colorectal cancer, as shown in Figure 3, panels A–C, ASIR rose at comparable annual rates in both strata, while ASMR and ASDR declined about 1.0% faster per year in more developed countries. For liver cancer, illustrated in Figure 3, panels D–F, ASIR, ASMR, and ASDR showed mild upward trends in more developed countries, whereas less developed countries experienced slight declines. The interaction greater than 1.00 reflects this divergence, indicating less favorable trajectories in more developed settings. Table 3Multilevel Gamma GLMM estimates for ASIR, ASMR, ASDR, and tobacco-attributable ASMR in major gastrointestinal cancers, 1990–2021MetricPredictorsEstimates (95% CI)P-valueASIR, colorectalTime------1.008 (1.008, 1.009)<0.001Development statusMore developed vs. less developed1.134 (1.117, 1.151)<0.001Time × Development statusTime × (More developed vs. less developed)0.997 (0.997, 0.998)<0.001ASMR, colorectalTime------1.004 (1.004, 1.005)<0.001Development statusMore developed vs. less developed1.242 (1.224, 1.261)<0.001Time × Development statusTime × (More developed vs. less developed)0.990 (0.990, 0.991)<0.001ASDR, colorectalTime------1.003 (1.003, 1.004)<0.001Development statusMore developed vs. less developed1.247 (1.228, 1.266)<0.001Time × Development statusTime × (More developed vs. less developed)0.990 (0.990, 0.991)<0.001ASIR, pancreaticTime------1.013 (1.013, 1.014)<0.001Development statusMore developed vs. less developed1.161 (1.142, 1.180)<0.001Time × Development statusTime × (More developed vs. less developed)0.993 (0.992, 0.993)<0.001ASMR, pancreaticTime------1.013 (1.013, 1.013)<0.001Development statusMore developed vs. less developed1.182 (1.163, 1.202)<0.001Time × Development statusTime × (More developed vs. less developed)0.992 (0.991, 0.992)<0.001ASDR, pancreaticTime------1.012 (1.012, 1.013)<0.001Development statusMore developed vs. less developed1.198 (1.178, 1.217)<0.001Time × Development statusTime × (More developed vs. less developed)0.991 (0.991, 0.992)<0.001ASIR, liverTime------0.994 (0.993, 0.995)<0.001Development statusMore developed vs. less developed0.727 (0.709, 0.746)<0.001Time × Development statusTime × (More developed vs. less developed)1.015 (1.013, 1.016)<0.001ASMR, liverTime------0.994 (0.993, 0.995)<0.001Development statusMore developed vs. less developed0.750 (0.730, 0.770)<0.001Time × Development statusTime × (More developed vs. less developed)1.013 (1.012, 1.014)<0.001ASDR, liverTime------0.992 (0.991, 0.993)<0.001Development statusMore developed vs. less developed0.757 (0.737, 0.777)<0.001Time × Development statusTime × (More developed vs. less developed)1.012 (1.011, 1.014)<0.001ASIR, stomachTime------0.990 (0.990, 0.991)<0.001Development statusMore developed vs. less developed1.202 (1.183, 1.221)<0.001Time × Development statusTime × (More developed vs. less developed)0.988 (0.987, 0.988)<0.001ASMR, stomachTime------0.990 (0.989, 0.990)<0.001Development statusMore developed vs. less developed1.264 (1.243, 1.286)<0.001Time × Development statusTime × (More developed vs. less developed)0.984 (0.984, 0.985)<0.001ASDR, stomachTime------0.988 (0.988, 0.989)<0.001Development statusMore developed vs. less developed1.262 (1.241, 1.284)<0.001Time × Development statusTime × (More developed vs. less developed)0.985 (0.984, 0.985)<0.001Tobacco-attributable ASMR, colorectalTime------0.999 (0.999, 1.000)<0.001Development statusMore developed vs. less developed1.288 (1.264, 1.312)<0.001Time × Development statusTime × (More developed vs. less developed)0.989 (0.989, 0.990)<0.001Tobacco-attributable ASMR, liverTime------0.988 (0.987, 0.989)<0.001Development statusMore developed vs. less developed0.752 (0.731, 0.775)<0.001Time × Development statusTime × (More developed vs. less developed)1.014 (1.013, 1.015)<0.001Tobacco-attributable ASMR, pancreaticTime------1.006 (1.005, 1.006)<0.001Development statusMore developed vs. less developed1.214 (1.190, 1.239)<0.001Time × Development statusTime × (More developed vs. less developed)0.992 (0.991, 0.992)<0.001Tobacco-attributable ASMR, stomachTime------0.983 (0.983, 0.984)<0.001Development statusMore developed vs. less developed1.332 (1.305, 1.360)<0.001Time × Development statusTime × (More developed vs. less developed)0.982 (0.981, 0.983)<0.001Note: Outcomes were modeled with Gamma regression and a log link. Predictors included Time (years since baseline), Development status (More developed vs. Less developed; HDI ≥ 0.7 vs. < 0.7), and their interaction (Time × Development status). Fixed-effect estimates are presented on the response scale (exponentiated) and interpreted as multiplicative rate ratios (per +1 year for Time; More vs. Less developed for Development status; a ratio of yearly changes for the interaction), with p-values from Wald tests. Rows labeled tobacco-attributable ASMR (TA-ASMR) refer to mortality attributable to tobacco, estimated within the same modeling framework. Abbreviations: ASIR = age-standardized incidence rate; ASMR = age-standardized mortality rate; ASDR = age-standardized disability-adjusted life-year rate; HDI = Human Development Index. Multilevel Gamma GLMM estimates for ASIR, ASMR, ASDR, and tobacco-attributable ASMR in major gastrointestinal cancers, 1990–2021 Note: Outcomes were modeled with Gamma regression and a log link. Predictors included Time (years since baseline), Development status (More developed vs. Less developed; HDI ≥ 0.7 vs. < 0.7), and their interaction (Time × Development status). Fixed-effect estimates are presented on the response scale (exponentiated) and interpreted as multiplicative rate ratios (per +1 year for Time; More vs. Less developed for Development status; a ratio of yearly changes for the interaction), with p-values from Wald tests. Rows labeled tobacco-attributable ASMR (TA-ASMR) refer to mortality attributable to tobacco, estimated within the same modeling framework. Abbreviations: ASIR = age-standardized incidence rate; ASMR = age-standardized mortality rate; ASDR = age-standardized disability-adjusted life-year rate; HDI = Human Development Index. Modeled trends in GI cancer rates by development status, 1990–2021. Panels show annual age-standardized rates for more developed countries (HDI ≥ 0.7, orange line) versus less developed countries (HDI < 0.7, green line), based on multilevel models. (A) colorectal cancer ASIR, (B) colorectal cancer ASMR, (C) colorectal cancer ASDR; (D) liver cancer ASIR, (E) liver cancer ASMR, (F) liver cancer ASDR; (G) pancreatic cancer ASIR, (H) pancreatic cancer ASMR, (I) pancreatic cancer ASDR; (J) stomach cancer ASIR, (K) stomach cancer ASMR, (L) stomach cancer ASDR. Shaded areas indicate 95% confidence intervals around model estimates. For pancreatic cancer, as presented in Figure 3, panels G–I, ASIR, ASMR, and ASDR increased in both strata; however, the yearly rise was approximately 0.8-0.9% slower in more developed countries. For stomach cancer, depicted in Figure 3, panels J–L, ASIR, ASMR, and ASDR decreased more rapidly in more developed countries, with approximately 1.2–1.6% faster annual declines compared to less developed settings. For tobacco-attributable mortality, displayed in Figure 4, panels A–D, ASMR for colorectal, pancreatic, and stomach cancers declined faster in more developed countries. In contrast, liver cancer mortality decreased more strongly in less developed countries. Modeled trends in tobacco-attributable ASMR of GI cancers by development status, 1990–2021. Panels show annual age-standardized rates for more developed countries (HDI ≥ 0.7, orange line) versus less developed countries (HDI < 0.7, green line), based on multilevel models. (A) colorectal cancer, (B) liver cancer, (C) pancreatic cancer, and (D) stomach cancer. Shaded areas indicate 95% confidence intervals around model estimates. Discussion Our study provides the first longitudinal multilevel assessment of how HDI influences trends in GI cancer burden from 1990 to 2021. We found that trajectories of ASIR, ASMR, and ASDR were broadly parallel across development strata, but with notable divergences: more developed countries experienced faster declines in stomach and colorectal mortality and DALYs, slower increases in pancreatic cancer, and comparatively less favorable trends for liver cancer. Secondary analyses demonstrated strong cross-sectional associations between HDI and cancer burden in 2021, with positive correlations for colorectal and pancreatic incidence, inverse associations for liver cancer concentrated in less developed countries, and inverse associations for stomach cancer largely confined to more developed settings. Importantly, our longitudinal analyses of tobacco-attributable mortality showed that ASMR for colorectal, pancreatic, and stomach cancers declined more rapidly in more developed countries, whereas liver cancer mortality decreased more strongly in less developed settings. We also identified marked country-level outliers and distinct regional clustering of both overall and tobacco-attributable burdens. Together, these findings highlight the complex interplay between development status, risk exposures, and cancer burden across time and geography. Previous studies have reported associations between HDI and incidence, mortality, and DALYs for colorectal cancer, but these were generally restricted to single-year estimates, in line with our secondary findings (20-24). In our analysis, we stratified countries into less developed (HDI < 0.700) and more developed (HDI ≥ 0.700) groups. Although overall patterns were similar across strata, the associations were stronger in more developed countries, except for ASDR. Crucially, our study adds novel evidence by assessing the longitudinal impact of HDI on colorectal cancer burden from 1990 to 2021. We found that each unit annual increase in HDI was associated with an approximate 1% change in the slope of burden measures. More developed countries, which initially had higher incidence, mortality, and DALY rates, experienced slower increases in ASIR and steeper declines in ASMR and ASDR compared with less developed countries, such that by the end of the study period their levels were lower. Nevertheless, in 2021 certain high-HDI countries including the Netherlands, Slovakia, New Zealand, Uruguay, Hungary, and Bulgaria still showed elevated values. One possible explanation, as suggested by Zhang et al., is low whole grain intake combined with high red meat consumption (25). However, the relative decline in colorectal cancer burden compared with earlier decades has also been attributed to improved control of risk factors such as tobacco use (26), which aligns with the findings of the present study. These findings highlight the need to strengthen prevention and screening strategies, while future research should clarify the role of lifestyle, environmental, and genetic factors in explaining persistent outlier patterns. Similar to colorectal cancer, pancreatic cancer has previously been linked to HDI in several studies, though mostly through cross-sectional analyses restricted to single years (27-33). Consistent with prior evidence, our findings confirm that higher HDI is associated with a greater burden of pancreatic cancer, with the correlation in 2021 being slightly stronger in more developed countries. Similar patterns were observed for tobacco-attributable mortality. A key novel contribution of our study is the longitudinal perspective. Between 1990 and 2021, each unit increase in HDI was associated with an approximately 1% steeper annual rise in pancreatic cancer incidence, mortality, and DALYs in less developed countries, ultimately resulting in higher burdens in these settings by the end of the study period. One possible explanation is that more developed countries may already have reached a plateau in the prevalence of key risk factors, while advances in screening and treatment have further moderated the growth of pancreatic cancer burden (34, 35). However, in 2021, a number of high and very high-HDI countries, particularly in the Americas (Uruguay), East Asia (Japan), the Middle East (United Arab Emirates), and Europe (Czech Republic), still showed comparatively high burdens of pancreatic cancer, likely reflecting persistent lifestyle-related exposures such as high fasting plasma glucose and body-mass index, tobacco, alcohol, diet, and aging populations (36-42). Unlike colorectal and pancreatic cancers, liver cancer displayed a distinct pattern. Previous studies using cross-sectional correlation have suggested that higher HDI is associated with lower incidence, mortality, and DALYs for liver cancer (43-47), likely reflecting the greater burden of viral hepatitis, hepatocellular carcinoma, and aflatoxin exposure in developing countries (48-50). While our findings support this inverse relationship, we observed that it was confined to less developed countries with HDI < 0.700, whereas no significant association was found in more developed settings. This suggests that beyond a certain threshold, the link between HDI and liver cancer burden may diminish. We also found no significant association between HDI and tobacco-attributable liver cancer mortality in either less or more developed countries. An additional contribution of our study was the longitudinal perspective. From 1990 to 2021, each unit increase in HDI was associated with an approximate 1% annual divergence in the slopes of ASIR, ASMR, and ASDR, with upward trends in more developed countries and downward trends in less developed ones. By the end of the study period, less developed countries had lower levels of these measures. A similar pattern was observed for tobacco-attributable ASMR. These results are consistent with those reported by Amini et al., who analyzed liver cancer trends from 1990 to 2019, although their HDI categorization was not standardized (11). This might be due to increasing metabolic risk factors such as obesity (51-53) in high HDI countries. Another important observation was the exceptionally high liver cancer burden in Mongolia, where Yao et al. highlighted alcohol use and viral hepatitis as major contributors (54) and Batsaikhan et al. documented a rapid increase among women since 1990 (55). Persistently elevated levels were also evident in low-HDI Sub-Saharan African countries such as Gambia, Mali, and Mozambique, which might be explained by widespread HBV infection, aflatoxin exposure, and limited access to vaccination and treatment (56-58). Together, these patterns suggest that liver cancer reflects a dual burden: traditional risk factors continuing to dominate in low-HDI regions, while metabolic and lifestyle-related drivers are increasingly shaping trends in more developed settings. Stomach cancer, in contrast to other GI cancers, exhibited a contradictory pattern in cross-sectional analyses. Previous studies have reported both positive (59-62) and negative (63) associations between HDI and stomach cancer, raising uncertainty about the true direction of this relationship. In our study, by 2021 the association was negative but limited to more developed countries, consistent with the findings of Bray et al. (15), while no significant correlation was observed in less developed settings, as also noted by Mousavi et al. (64). A similar inconsistency was noted for tobacco-attributable mortality, where HDI showed a positive correlation with ASMR in less developed countries but a negative correlation in more developed ones. These findings emphasize the importance of considering smoking status as well as the developmental context when interpreting associations between HDI and stomach cancer burden. The clearest insight emerged from our longitudinal analysis. From 1990 to 2021, both stomach cancer burden indicators and tobacco-attributable ASMR consistently declined, with steeper decreases in more developed countries. Each unit increase in HDI was linked to approximately a 1% greater annual reduction in these measures in more developed compared with less developed settings. Thus, the apparent inconsistency in HDI–stomach cancer associations is likely attributable to whether the analysis is cross-sectional or longitudinal, and to differences between more and less developed regions. Given that longitudinal analyses provide stronger evidence than cross-sectional snapshots, our findings indicate an overall negative association between HDI and stomach cancer burden. Our analysis also identified clear outliers in stomach cancer burden, with exceptionally high ASIR, ASMR, and ASDR values observed in Mongolia (high/very high HDI), Bolivia (medium HDI), and Afghanistan (low HDI). In Mongolia, the high burden has been linked to widespread Helicobacter pylori infection, dietary practices such as high salt consumption and frequent use of salty tea, smoking, and familial predisposition (65). In Afghanistan, Shayan et al. reported that stomach cancer is the leading cancer among men and the third among women (66). Its burden has been associated with H. pylori infection, tobacco use, high-salt and low-fruit diets, and preserved or red meat (67-69), with local studies in Kabul further demonstrating that pro-inflammatory and high-insulin diets significantly increase risk (70, 71). These findings highlight the need for targeted prevention strategies, including improved H. pylori control, dietary interventions, and tobacco reduction, particularly in high-burden countries. A key strength of our study is the application of a longitudinal multilevel generalized linear mixed-effects model, for the first time, to the four major GI cancers. This framework enabled us to assess the long-term impact of HDI on ASIR, ASMR, and ASDR while accounting for heterogeneity at the country, regional, and super-regional levels. By moving beyond single-year analyses, our approach provides a more comprehensive view of how cancer burden trajectories diverge between more and less developed settings. This methodology has previously been applied in other published studies on different health outcomes (11, 72, 73). Another strength of our study is the evaluation of HDI–cancer burden correlations across the full study period, stratified by more developed, less developed, and total groups, which highlighted differences from prior studies and offered a clearer understanding of the HDI–cancer relationship. Finally, outlier countries in each HDI category (low, medium, and high/very high) were also identified, as these may exert statistical influence on linear trends and should be considered in interpretation. The study has several limitations. It relies on GBD 2021 data, which may be affected by reporting bias, incomplete cancer registry coverage, and misclassification, especially in low- and middle-income countries. HDI was used as a proxy for development but does not reflect within-country disparities, health system quality, or access to care. Ecological analyses at the country level risk ecological fallacy and cannot capture individual-level risk factors. Although tobacco-attributable mortality was included, other key risk factors such as alcohol, obesity, diet, H. pylori, and viral hepatitis were not directly modeled. Finally, the observational design precludes causal inference between HDI and cancer burden trends. Despite these limitations, our findings provide valuable insights for cancer control strategies. They suggest that as countries develop, the profile of major GI cancers shifts, calling for adaptable health policies that focus on infection control and basic healthcare in low-HDI settings and on lifestyle modification and early detection in high-HDI settings. Conclusion By applying a longitudinal multilevel approach, this study shows that changes in human development over time are strongly associated with the burden of gastrointestinal cancers. Unlike earlier research that only examined single years, we captured how shifts in HDI shape cancer trends differently in more and less developed countries. Colorectal and pancreatic cancers tend to rise with higher development, while liver and stomach cancers show opposite or mixed patterns. These findings suggest that cancer prevention cannot be uniform across all settings. In lower-HDI countries, expanding vaccination, infection control, and access to care is critical, while in higher-HDI countries efforts should focus on reducing obesity, tobacco use, alcohol consumption, and unhealthy diets. Identifying countries with unusually high burdens also highlights where tailored strategies are most urgent, underscoring the importance of incorporating socio-development indices like HDI into cancer burden projections and policy planning, and pointing to the need for future research on the drivers of divergent trends and evaluation of targeted interventions. Acknowledgment We would like to thank the GBD 2021 collaborators for providing open access to the database. We also used ChatGPT-5 to assist with paraphrasing and improving the clarity of the English language in this manuscript; no content was generated by the tool. Ethics approval All data used in this study were obtained from the publicly available GBD 2021 database. The database contains fully de-identified data at the country/region level and does not include any individual-level or personally identifiable information. As this study used publicly accessible, anonymized aggregate data, no ethical approval or informed consent was required. Availability of data and materials Data can be obtained from the following website: http://ghdx.healthdata.org/gbd-results-tool. Conflict of interests There is no conflict of interest for authors of this article. References Associated Data Data Availability Statement Data can be obtained from the following website: http://ghdx.healthdata.org/gbd-results-tool.