Driving Mechanisms and Future Trends of Migraine and Tension-Type Headache Burden in G20 Countries: A Comparative Analysis Utilizing Decomposition Analysis and Explainable Machine Learning

Abstract

Introduction

Migraine and tension-type headache (TTH) are leading causes of neurological disability globally. While their macro-trends are documented, the nonlinear drivers and the regulatory role of socio-economic development across G20 countries remain under-explored.

Methods

Based on Global Burden of Disease (GBD) 2023 data, the headache burden across G20 members was quantified. Developmental trajectories were identified by integrating the Socio-demographic Index (SDI). Crucially, a comparative analysis was conducted between a decomposition analysis model and an explainable machine learning framework, Shapley Additive Explanations-Extreme Gradient Boosting (SHAP-XGBoost), to quantify nonlinear marginal contributions and complex feature interactions. Finally, future trends from 2024 to 2035 were projected utilizing Autoregressive Integrated Moving Average (ARIMA) models.

Results

In 2023, the number of prevalent cases for migraine and TTH across G20 countries was estimated at 705 million and 1.27 billion, respectively. A significant nonlinear correlation between headache burden and the SDI was identified, with the burden for both disorders peaking at an SDI of approximately 0.8 before plateauing or declining. While population growth was identified as the primary macro-driver in both decomposition and SHAP analyses, a “saturation effect” within the headache models was revealed by SHAP analysis. Specifically, the marginal contribution of population size was found to diminish once specific thresholds were exceeded. Furthermore, according to ARIMA model projections, the age-standardized incidence rate (ASIR) of migraine is expected to stabilize in the coming years, whereas the ASIR of TTH is forecast to maintain a continuous upward trajectory through 2035.

Conclusions

This study provides a comprehensive re-evaluation of migraine and TTH burden across G20 countries, shifting the focus from descriptive status quo to mechanistic explanation. Our findings confirm that while population growth remains a key contributor, variations in burden appear to be associated with differences in development level, age structure, and population scale.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s40122-026-00859-2.

Key Summary Points

Why carry out this study?

• Migraine and tension-type headache (TTH) represent leading causes of neurological disability worldwide, imposing a disproportionately heavy burden on the social productivity and economies of G20 countries, which account for approximately 85% of the global gross domestic product (GDP)

• Traditional epidemiological studies predominantly rely on linear assumptions, which often fail to characterize the complex, nonlinear dynamics and potential threshold effects between demographic or socio-economic factors and the resulting headache burden

• This study aimed to quantify the nonlinear drivers of migraine and TTH burden across G20 countries by integrating traditional decomposition analysis with explainable machine learning Shapley Additive Explanations-Extreme Gradient Boosting (SHAP-XGBoost) and to project their burden trends through 2035

What was learned from the study?

• Our findings reveal a significant nonlinear correlation between headache burden and Socio-demographic Index (SDI), with disease burden peaking at an SDI of approximately 0.8 before plateauing or declining

• While both decomposition and SHAP analyses identified population growth as a primary macro-driver, SHAP attribution uniquely uncovers a “saturation effect” in the models, where the marginal contribution of population size diminishes once specific thresholds are surpassed

• Projections indicate that while the migraine burden is expected to stabilize or even decrease by 2035, the burden of TTH will continue to rise, indicating different future patterns between the two disorders

Introduction

Primary headache disorders—encompassing migraine and tension-type headache (TTH)—are among the most prevalent and burdensome neurological conditions globally, collectively driving substantial disability, healthcare expenditure, and societal productivity loss [1, 2]. Migraine, characterized by recurrent moderate-to-severe unilateral headache with photophobia, phonophobia, or nausea, ranks as the second leading cause of disability worldwide and the top cause of disability in women aged 15–49 years [3, 4]. In contrast, TTH presents as mild-to-moderate bilateral “band-like” pain, with a far higher global prevalence, but lower individual disability impact, yet its sheer caseload translates to considerable population-level burden [5, 6]. Together, these two disorders account for over 50% of primary headache cases, making them a critical focus for public health action [7].

The global burden of migraine and TTH is well-documented, but their impact is disproportionately concentrated in the G20—a bloc of 19 countries and the European Union representing over 80% of global gross domestic product (GDP) and two-thirds of the world’s population [8]. Over the past three decades, these nations have undergone a profound transformation in their disease landscapes, driven by the complex interplay of multifaceted factors.

According to our knowledge, research on comparative headache burden across G20 countries remains limited [9]. In addition, several important knowledge gaps persist in the existing literature. First, many existing studies rely on traditional linear decomposition analysis to identify drivers of disease burden. While effective at a macro level, these methods have important limitations in capturing the complex interaction effects between variables [10, 11]. For instance, the impact of population aging on headache burden may not be uniformly proportional across different population scales, and the coupling between gender, age, and socioeconomic status has received limited quantification [12]. Furthermore, the nonlinear relationships between these drivers and headache burden often remain underexplored in traditional epidemiological approaches. The escalation of disease burden frequently exhibits “threshold effects” or “saturation phenomena” that traditional parametric models may not adequately capture [13], which could lead to policy projections that potentially oversimplify future health needs.

To address these gaps, this study leverages the latest Global Burden of Disease Study 2023 (GBD 2023) data to provide a comprehensive, comparative analysis of migraine and TTH burden across G20 countries. Our objectives are to: (1) quantify the 2023 burden using absolute counts and age-standardized metrics to establish a baseline for both disorders; (2) evaluate the correlation between headache burden and the Socio-demographic Index (SDI) to identify nonlinear developmental trajectories; (3) delineate the drivers of burden change via decomposition analysis, partitioning total variation into population growth, aging, and epidemiological shifts at a macro level; (4) implement an explainable machine learning framework, Shapley Additive Explanations-Extreme Gradient Boosting (SHAP-XGBoost), to contrast with traditional decomposition, specifically quantifying the nonlinear marginal contributions of age, sex, and population size to the disease burden; (5) project 2024–2035 trends using Autoregressive Integrated Moving Average (ARIMA) models to provide evidence-based forecasting for future public health planning.

Methods

Overview

This study used data from the GBD 2023, published by the Institute for Health Metrics and Evaluation (IHME), University of Washington. The GBD database is widely regarded as the global benchmark for disease burden estimation. It integrates 292 causes of death, 375 diseases and injuries, and 88 risk factors across 204 countries and territories, drawing on multiple sources such as vital registration systems, population censuses, epidemiological surveys, and hospital records [14]. All data are harmonized through standardized modeling frameworks to provide consistent long-term estimates across diseases and regions.

Data Sources

Our analysis focused on the G20 members, including Argentina, Australia, Brazil, Canada, China, France, Germany, India, Indonesia, Italy, Japan, Republic of Korea, Mexico, Russian Federation, Saudi Arabia, South Africa, Turkey, the UK, the USA, and the European Union (EU). In this study, the G20 countries were defined according to the GBD 2023 geographic hierarchy (Location Set ID: 1437). Although the African Union was officially admitted as a permanent member of the G20 in September 2023, this analysis adheres to the established G20 composition used within the GBD 2023 data architecture. This study used publicly available data from the GBD 2023 study provided by the IHME. All data were deidentified and aggregated; therefore, ethical approval and informed consent were not required. The GBD database is publicly accessible through the Global Health Data Exchange (GHDx) query tool (https://ghdx.healthdata.org/gbd-results-tool). Our analysis is governed by the GBD 2023 Publication Policy and Collaborative Agreement [15].

Burden Description

This study extracted data on incidence, prevalence, and disability-adjusted life years (DALYs)—both crude and age-standardized—for migraine and TTH from all G20 countries. All metrics, comprising the age-standardized incidence rate (ASIR), age-standardized prevalence rate (ASPR), and age-standardized DALY rate (ASDR), are expressed per 100,000 population and were standardized to the World Health Organization (WHO) world standard population to facilitate comparative analysis. Descriptive analyses and visualizations were conducted using R software (version 4.5.1) to depict the distribution of ASDR, ASIR, and ASPR for migraine and TTH across G20 countries in 2023. The Average Annual Percentage Change (AAPC) was calculated using the Joinpoint Regression Program (version 5.1.0.0). A positive AAPC value indicates an increasing trend in headache burden, whereas a negative value reflects a decreasing trend. Cases of migraine and TTH were defined per the International Classification of Headache Disorders, 3rd edition (ICHD-3) and coded as 8A80 and 8A81, respectively, according to the International Classification of Diseases, 11th Revision (ICD-11).

Socio-Demographic Index

The Socio-demographic Index, developed by IHME in 2015, is a summary measure of a geography’s socio-economic development. Ranging from 0 to 1, the SDI captures the correlation between developmental progress and health outcomes; higher values denote superior levels of regional development [16].

Decomposition Analysis

To disentangle the driving forces behind changes in disease burden, we conducted a decomposition analysis partitioning total variation into three components: population aging, population growth, and epidemiological change. This allowed us to assess how demographic transitions versus disease-specific risk factors contributed to the observed burden over time.

SHAP Analysis

To further scrutinize the drivers of migraine and TTH across G20 countries, we employed the XGBoost algorithm in conjunction with SHAP for systematic model decomposition. Compared with traditional decomposition methods—which often rely on rigid linear assumptions and may fail to capture synergistic or high-order interactions—SHAP provides a robust framework by assigning each feature an additive contribution to the model’s output. This approach provides both local accuracy and consistency, grounded in cooperative game theory [13, 17]. By decomposing the model’s predictions into the marginal contributions of specific variables, we elucidated the granular impacts of biological sex, age-group distributions, and demographic shifts on the evolving burden of headache disorders.

To address the skewed distribution of prevalence data and enhance model stability, we applied a log-transformation to the age-specific DALY rates, prevalence, and incidence $(Log\_Rate=ln(1+Rate))$. The feature set incorporated sex, age group (excluding individuals < 5 years due to data unavailability), normalized calendar year (Year/1000), and the log-transformed total population ($Log\_Population$). Categorical variables, including sex and age, were transformed via one-hot encoding, while continuous variables underwent normalization prior to modeling.

To capture the intrinsic nonlinear relationships between predictors and headache burden, an optimized XGBoost algorithm was implemented, with its predictive performance evaluated using the coefficient of determination (R²). The XGBoost model was optimized using a systematic grid search integrated with a fivefold cross-validation strategy. Optimal configurations for key hyperparameters were identified, including a maximum tree depth (max depth = 3) and a learning rate (eta = 0.05). The optimal number of boosting iterations was dynamically determined through an early stopping mechanism (10-round threshold) to prevent redundant training. Furthermore, regularization techniques and stochastic feature/sample selection were implemented, with both the subsample and colsample_bytree ratios set to 0.8. The final model reached convergence at 4338 iterations, achieving an average validation root mean square error (RMSE) of 0.0094, which indicates robust generalization performance and the effective prevention of overfitting.

Model interpretability was facilitated through the SHAP framework; feature importance was quantified and ranked by mean absolute SHAP values, and SHAP dependence plots were constructed to elucidate the nonlinear trajectories of specific drivers in relation to the disease burden.

Autoregressive Integrated Moving Average Model

To forecast the future burden of migraine and TTH among G20 countries, we applied the ARIMA model, a classical time-series forecasting method [18]. The ARIMA framework captures both trend and seasonality by integrating past values (autoregressive terms), differencing operations to ensure stationarity, and moving average components that account for short-term fluctuations. We fitted separate ARIMA models for ASIR, ASPR, and ASDR for each country. Data from 1990 to 2023 served as the training sample for fitting the ARIMA model, with the optimal model determined using the auto.arima() function based on the Akaike Information Criterion and Bayesian Information Criterion values. Using the fitted models, we generated projections for the period 2024–2035, with 95% confidence intervals (CI) reported to represent forecast uncertainty.

All estimates were accompanied by 95% uncertainty intervals (UI) to account for sampling and model-based error. All the analyses and visualizations in this study were performed in the R language (version 4.5.1), and the differences were considered statistically significant at two-sided P < 0.05.

Results

Overview of G20 Countries

The burden of migraine and TTH across G20 countries is substantial. For migraine, the number of prevalent cases was 704,681,170 (95% UI: 610,455,397–801,370,230), with 51,918,764 (95% UI: 46,201,137–58,828,018) incident cases. The number of DALYs attributable to migraine was 24,873,985 (95% UI: 16,489,401–34,719,363). Regarding TTH, there were 1,269,387,250 prevalent cases (95% UI: 1,126,527,856–1,414,688,569) and 27,448,460 incident cases (95% UI: 24,099,890–30,923,293). The associated DALYs for TTH were 2,854,591 (95% UI: 1,916,150–4,050,260).

In G20 countries during 2023, the age-standardized rates for migraine were as follows: ASDR was 483.8 per 100,000 (95% UI: 320.05–676.05), ASPR was 14,008.3 per 100,000 (95% UI: 12,076.92–15,965.36), and ASIR was 1149.6 per 100,000 (95% UI: 1,011.48–1,307.01). The AAPC from 1990 to 2023 indicated slow but significant increasing trends: ASDR increased by 0.0799% (95% UI: 0.0562–0.1036), ASPR by 0.1007% (95% UI: 0.0835–0.1178), and ASIR by 0.1020% (95% UI: 0.0936–0.1104). For TTH, ASDR was 53.41 per 100,000 (95% UI: 36.03–75.85), ASPR was 25,005.85 per 100,000 (95% UI: 22,224.44–28,018.26), and ASIR was 8926.84 per 100,000 (95% UI: 7838.26–10,084.07). The AAPCs were notably lower. The rates were ASDR: 0.0450% (95% UI: 0.0399–0.0501), ASPR: 0.0450% (95% UI: 0.0399–0.0501), and ASIR: 0.0591% (95% UI: 0.0550–0.0633) (Tables 1, 2).

Table 1.

The numbers (2023)—age-standardized rates (2023) and average annual percent change rate (1990–2023) of DALYs—prevalence and incidence of migraine in G20 country

Migraine location	DALYs (95% UI)			Prevalence (95% UI)			Incidence (95% UI)
location_name	Number	ASDR	AAPC%	Number	ASPR	AAPC%	Number	ASIR	AAPC%
G20	24,873,985 (16,489,401–34,719,363)	483.75 (320.05–676.05)	0.0799 (0.0562–0.1036)	704,681,170 (610,455,397–801,370,230)	14,008.4 (12,076.92–15,965.36)	0.1007 (0.0835–0.1178)	51,918,764 (46,201,137–58,828,018)	1149.57 (1011.48–1307.01)	0.1020 (0.0936–0.1104)
Argentina	200,418 (131,856–278,201)	391.91 (256.48–546.63)	0.0916 (0.0608–0.1224)	5,698,406 (4,880,011–6,571,382)	11,278.31 (9611.58–13,066.56)	0.1158 (0.1097–0.1219)	440,974 (379,523–519,199)	939.53 (814.32–1,110.92)	0.0803 (0.0745–0.0862)
Australia	129,173 (85,334–177,060)	447.48 (293–618.01)	−0.0226 (−0.0449–0.0003)	3,716,479 (3,148,906–4,314,895)	13,365.41 (11,258.88–15,729.05)	0.0049 (0.0046–0.0053)	257,732 (223,325–302,030)	1,090.96 (938.02–1,304.22)	0.0014 (0.0011–0.0017)
Brazil	1,352,948 (916,744–1,838,380)	613.08 (415.76–834.65)	0.0547 (0.0287–0.0808)	39,432,744 (33,853,669–44,684,822)	18,150.64 (15,471.77–20,733.43)	0.0774 (0.0515–0.1033)	2,797,095 (2,466,151–3,173,961)	1481.29 (1298.14–1696.02)	0.0879 (0.0598–0.1160)
Canada	232,619 (151,390–320,391)	565.76 (374.01–783.11)	−0.0932 (−0.1115–0.0748)	6,351,099 (5,375,167–7,294,914)	16,103.57 (13,571.68–18,735.57)	−0.0736 (−0.0790–0.0681)	389,840 (338,531–454,890)	1228.01 (1057.21–1452.03)	−0.0208 (−0.0263–0.0154)
China	6,562,055 (4,404,640–9,001,993)	400.02 (265.94–554.75)	0.2230 (0.1753–0.2708)	179,397,306 (155,129,546–204,557,530)	11,419.64 (9840.95–12,927.13)	0.2075 (0.1686–0.2465)	12,689,685 (11,204,545–14,303,841)	952.39 (840.44–1,090.96)	0.1719 (0.1345–0.2093)
European Union	2,855,791 (1,880,325–3,938,709)	587.57 (388.73–811.99)	0.0110 (−0.0217–0.0437)	77,678,613 (65,751,122–88,590,879)	16,862.38 (14,476.19–19,137.21)	0.0261 (−0.0248–0.0770)	4,692,523 (4,141,439–5,400,589)	1301.84 (1136.7–1524.27)	0.0419 (0.0259–0.0579)
Germany	562,658 (373,030–767,147)	636.55 (418.71–871.72)	0.0108 (−0.0483–0.0700)	15,436,086 (13,164,032–17,699,885)	18,454.05 (15,783.82–21,232.47)	0.0300 (−0.0342–0.0943)	873,331 (761,244–1,006,279)	1379.98 (1197.69–1617.9)	0.0112 (−0.0417–0.0642)
France	434,948 (286,405–598,776)	600 (394–828.03)	−0.0430 (−0.0612–0.0249)	11,699,938 (9,889,560–13,269,091)	16,923.54 (14,462.22–19,476.54)	−0.0287 (−0.0327–0.0248)	758,918 (666,043–883,528)	1320.44 (1139.06–1558.82)	−0.0214 (−0.0236–0.0193)
Indonesia	1,739,938 (1,152,190–2,426,663)	557.78 (369.78–771.43)	−0.0115 (−0.0190–0.0039)	49,188,828 (42,167,651–55,771,623)	15,817.25 (13,582.44–17,910.16)	−0.0168 (−0.0171–0.0164)	3,833,824 (3,426,052–4,366,388)	1308.29 (1160.35–1495.2)	−0.0149 (−0.0151–0.0147)
Japan	487,724 (321,209–670,310)	364.74 (240.19–503.06)	0.0525 (0.0361–0.0688)	12,951,120 (11,165,415–14,660,379)	10,498.16 (9156.44–12,123.64)	0.0761 (0.0653–0.0870)	784,101 (695,128–888,350)	854.57 (745.01–992.92)	0.0734 (0.0681–0.0787)
Mexico	673,438 (457,269–933,440)	478.56 (325.55–663.05)	0.0324 (0.0037–0.0610)	19,135,576 (16,567,368–21,597,408)	13,658.09 (11,775.31–15,426.72)	0.0374 (0.0311–0.0438)	1,488,105 (1,318,811–1,698,392)	1119.78 (984.63–1276.42)	0.0272 (0.0220–0.0324)
India	7,480,365 (4,941,161–10,420,988)	490.19 (325.18–684.4)	0.0217 (0.0065–0.0369)	224,720,137 (193,335,814–256,552,225)	14,619.2 (12,625.34–16,702.6)	−0.0004 (−0.0047–0.0039)	18,443,129 (16,263,763–20,967,851)	1216.55 (1066.03–1379.55)	−0.0111 (−0.0149–0.0072)
Italy	414,160 (275,689–577,195)	663.3 (445.55–920.5)	0.0088 (−0.0804–0.0980)	11,038,572 (9,524,728–12,558,644)	18,870.57 (16,363.78–21,296.99)	0.0149 (−0.0813–0.1113)	628,554 (563,091–705,763)	1457.21 (1295.75–1662.47)	0.0111 (−0.0187–0.0409)
South Korea	235,277 (157,802–322,662)	391.81 (259.99–544.18)	−0.1322 (−0.2446–0.0197)	6,478,631 (5,507,736–7,471,640)	11,538.65 (9692.41–13,665.39)	−0.1552 (−0.2576–0.0527)	381,368 (322,732–441,882)	919.19 (792.81–1090.22)	−0.1047 (−0.1552–0.0541)
Turkey	517,201 (342,578–716,584)	557.93 (369.06–769.01)	0.0139 (−0.0129–0.0408)	13,658,327 (11,587,988–15,657,913)	14,928.48 (12,690.82–17,067.45)	0.0123 (0.0012–0.0234)	1,007,520 (875,472–1,175,150)	1204.31 (1043.9–1414.34)	0.0119 (0.0027–0.0211)
Russian	925,357 (614,483–1,256,093)	532.32 (349.81–730.38)	−0.0100 (−0.0196–0.0004)	23,170,347 (20,085,421–26,289,888)	14,002.59 (12,116.05–15,908.38)	−0.0084 (−0.0156–0.0013)	1,440,818 (1,279,049–1,629,704)	1074.83 (951.26–1232.56)	−0.0020 (−0.0039–0.0000)
Saudi Arabia	173,920 (115,764–241,413)	483.05 (325.02–666.69)	−0.0901 (−0.1169–0.0633)	5,034,537 (4,255,072–5,865,590)	13,683.9 (11,598.55–15,945.94)	−0.0535 (−0.0720–0.0350)	406,340 (347,230–476,447)	1137.55 (982.14–1343.8)	−0.0438 (−0.0551–0.0325)
South Africa	316,794 (213,038–439,525)	467.39 (315.77–649.1)	−0.0466 (− 0.0606–0.0326)	9,310,729 (8,072,618–10,674,312)	13,575.58 (11,820.05–15,529.08)	−0.0067 (−0.0077–0.0058)	747,248 (656,361–858,979)	1110.45 (983–1280.08)	−0.0466 (− 0.0606–0.0326)
United Kingdom of Great Britain and Northern Ireland	426,198 (282,292–592,359)	584.91 (386.9–809.07)	−0.0181 (−0.0313–0.0050)	11,841,806 (10,218,436–13,291,794)	16,945.26 (14,631.71–19,195.49)	0.0221 (0.0186–0.0256)	780,196 (690,962–888,941)	1345.58 (1183.42–1552.22)	0.0144 (0.0108–0.0179)
USA	2,008,793 (1,341,892–2,794,361)	563.61 (379.1–783.48)	−0.1031 (−0.1280–0.0783)	56,420,502 (48,402,973–65,080,175)	16,360.07 (14,103.31–19,044.26)	−0.0824 (−0.0956–0.0691)	3,769,988 (3,367,379–4,268,192)	1295.61 (1152.41–1485.44)	−0.0593 (−0.0670–0.0517)

DALYs disability-adjusted life years, UI uncertainty interval, AAPC average annual percent change, ASDR age-standardized DALYs rate, ASIR age-standardized incidence rate, ASPR age-standardized prevalence rate

Table 2.

The numbers (2023)—age-standardized rates (2023) and average annual percent change rate (1990–2023) of DALYs—prevalence and incidence of TTH in G20 country

TTH location	DALYs (95% UI)			Prevalence (95% UI)			Incidence (95% UI)
location_name	Number	ASDR	AAPC%	Number	ASPR	AAPC%	Number	ASIR	AAPC%
G20	2,854,591 (1,916,150–4,050,260)	53.41 (36.03–75.85)	0.0450 (0.0399–0.0501)	1,269,387,250 (1,126,527,856–1,414,688,569)	25,005.85 (22,224.44–28,018.26)	0.0450 (0.0399–0.0501)	27,448,460 (24,099,890–30,923,293)	8926.84 (7838.26–10,084.07)	0.0591 (0.0550–0.0633)
Argentina	30,450 (20,328–43,616)	58.41 (38.89–83.67)	0.0147 (0.0014–0.0281)	13,149,553 (11,338,611–15,171,530)	25,925.36 (22,154.36–29,977.98)	0.0003 (−0.0001–0.0008)	21,891,308 (19,147,732–25,008,494)	9159.82 (7956.62–10,344.94)	0.0002 (−0.0004–0.0008)
Australia	18,009 (12,006–25,947)	58 (38.58–83.13)	−0.0018 (−0.0059–0.0023)	7,670,882 (6,632,051–8,772,097)	26,450.26 (22,929–30,470.64)	−0.0002 (−0.0005–0.0000)	17,602,177 (15,435,602–19,897,287)	9314.34 (8175.63–10,483.11)	−0.0005 (−0.0007–0.0003)
Brazil	129,487 (88,015–183,477)	55.01 (37.62–78.01)	0.0258 (−0.0137–0.0653)	65,185,607 (57,980,249–72,947,805)	29,171.43 (25,909.39–32,859.98)	0.0114 (−0.0090–0.0318)	13,680,008 (12,101,950–15,334,785)	10,089.5 (8821.97–11,429.19)	0.0632 (0.0585–0.0679)
Canada	29,576 (19,784–42,313)	65.81 (43.92–93.9)	0.0029 (−0.0121–0.0178)	13,666,998 (11,976,750–15,551,758)	32,922.2 (28,632.16–37,680.19)	−0.0014 (−0.0016–0.0012)	12,337,764 (10,762,048–13,903,123)	10,719.09 (9386.93–12,150.4)	−0.0010 (−0.0013–0.0006)
China	709,842 (469,707–1,033,209)	41.43 (27.75–60.23)	0.1381 (0.1293–0.1469)	291,018,721 (256,741,805–327,287,356)	18,564.14 (16,310.82–20,860.9)	0.2330 (0.2190–0.2470)	9,223,500 (8,086,426–10,364,987)	6802.61 (5869.84–7690.11)	0.1733 (0.1624–0.1843)
European Union	389,293 (264,743–560,664)	71.94 (48.77–102.65)	0.0090 (−0.0077–0.0257)	157,472,320 (138,826,697–177,152,099)	32,065.64 (28,170.21–36,532.33)	0.0026 (−0.0042–0.0094)	7,174,936 (6,332,377–8,156,679)	10,747.04 (9422.82–12,056.26)	−0.0048 (−0.0076–0.0021)
Germany	70,349 (47,364–101,641)	69.88 (47.11–101.08)	0.0074 (−0.0299–0.0448)	29,619,581 (26,014,750–33,415,430)	32,591.15 (28,403.41–37,705.64)	0.0189 (−0.0126–0.0505)	40,507,121 (35,677,927–45,920,743)	10,626.14 (9263.59–11,956.43)	0.0009 (−0.0066–0.0083)
France	59,900 (40,015–87,351)	74.49 (49.77–106.81)	0.0048 (−0.0317–0.0413)	23,213,633 (20,111,580–26,565,679)	31,360.19 (27,194.49–35,902.33)	−0.0041 (−0.0183–0.0101)	140,625,619 (122,581,612–158,060,311)	10,554.88 (9173.85–11,887.86)	−0.0009 (−0.0048–0.0030)
Indonesia	156,005 (106,283–219,530)	50.62 (34.56–71.56)	0.0291 (0.0191–0.0390)	78,650,324 (69,088,063–88,244,861)	25,970.86 (22,942.58–29,093.82)	−0.0023 (−0.0027–0.0019)	1,657,054 (1,446,218–1,910,626)	9213.48 (8102.94–10,468.39)	−0.0022 (− 0.0024–0.0019)
Japan	95,388 (64,986–139,306)	62.66 (42.84–89.6)	−0.0106 (−0.0260–0.0047)	40,408,380 (36,428,561–44,811,720)	29,801.9 (26,500.38–33,333.04)	−0.0105 (−0.0136–0.0074)	8,158,989 (7,181,682–9,219,891)	10,580.06 (9253.44–11,944.67)	0.0046 (0.0019–0.0073)
Mexico	74,832 (50,546–106,469)	52.95 (35.84–75.13)	0.0294 (0.0194–0.0394)	34,791,271 (30,630,244–39,155,344)	24,974.02 (21,987.25–28,212.48)	−0.0006 (−0.0009–0.0004)	7,570,622 (6,697,017–8,553,560)	8976.55 (7825.77–10,158.76)	−0.0005 (−0.0007–0.0003)
India	770,341 (520,117–1,086,516)	51.6 (34.87–72.71)	0.0585 (0.0461–0.0709)	389,571,414 (342,984,252–439,787,403)	25,811.73 (22,898.59–28,918.2)	−0.0034 (−0.0132–0.0063)	2,616,630 (2,299,818–2,943,903)	9388.93 (8221.77–10,559.44)	−0.0030 (−0.0099–0.0039)
Italy	51,280 (34,889–74,031)	71.78 (49.32–101.68)	0.0146 (−0.0045–0.0336)	21,495,373 (19,281,875–23,896,497)	33,418.68 (29,861.65–37,077.19)	0.0100 (−0.0065–0.0266)	2,801,426 (2,372,541–3,208,426)	11,510.53 (10,065.17–12,995.84)	0.0036 (−0.0147–0.0219)
South Korea	39,621 (26,376–56,915)	60.15 (40.98–85.41)	0.0921 (0.0588–0.1254)	17,540,108 (15,188,553–20,304,507)	29,949.6 (25,928.18–34,673.26)	0.1024 (0.0300–0.1749)	5,839,826 (5,064,949–6,605,925)	10,028.8 (8830.07–11,338.41)	0.0220 (0.0151–0.0288)
Turkey	62,117 (38,880–92,141)	65.56 (41.47–97.31)	0.0246 (−0.0154–0.0646)	19,394,928 (16,723,676–22,500,691)	21,140.38 (18,120.24–24,716.75)	−0.0157 (−0.0520–0.0207)	104,124,451 (91,607,896–117,235,639)	7952.68 (6979.09–9003.57)	−0.0110 (−0.0329–0.0110)
Russia	177,938 (112,628–260,731)	97.59 (64.23–140.95)	0.0073 (− 0.0041–0.0187)	52,221,542 (46,671,059–58,305,935)	32,056.26 (28,531.98–35,860.02)	0.0185 (0.0125–0.0245)	443,021,448 (389,923,366–498,772,540)	11,250.57 (9810.79–12,708.04)	0.0073 (0.0031–0.0115)
Saudi Arabia	21,371 (13,678–31,315)	61.44 (40.38–89.49)	−0.0053 (−0.0471–0.0364)	7,651,169 (6,365,804–8,974,663)	22,008.81 (18,874.91–25,509.33)	0.0150 (−0.0111–0.0410)	4,563,036 (3,984,299–5,131,642)	8136.23 (7070.11–9241.91)	0.0069 (−0.0080–0.0219)
South Africa	37,395 (25,437–53,753)	56.34 (38.27–80.87)	−0.0121 (−0.0309–0.0067)	16,458,263 (14,408,752–18,625,008)	24,471.71 (21,519.32–27,420.79)	0.0003 (0.0000–0.0005)	4,264,108 (3,732,090–4,820,923)	8753.92 (7653.52–9909.94)	0.0004 (0.0001–0.0006)
United Kingdom of Great Britain and Northern Ireland	57,619 (39,140–82,635)	72.22 (49.62–102.23)	0.0001 (−0.0049–0.0052)	24,853,053 (22,291,181–27,851,160)	33,422.05 (29,731.19–37,826.44)	0.0009 (0.0006–0.0013)	5,545,443 (4,863,491–6,327,298)	11,374.25 (9890.92–12,785.25)	0.0031 (0.0027–0.0035)
USA	263,070 (180,801–375,160)	68.5 (47.2–97.21)	−0.0586 (−0.0646–0.0526)	122,826,448 (109,613,829–137,176,076)	34,104.11 (30,349.05–37,830.18)	−0.0938 (−0.0973–0.0904)	50,549,098 (44,700,219–57,222,048)	11,655.69 (10,181.13–13,146.96)	−0.0288 (−0.0300–0.0275)

DALYs disability-adjusted life years, TTH tension-type headache, UI uncertainty interval, AAPC average annual percent change, ASDR age-standardized DALYs rate, ASIR age-standardized incidence rate, ASPR age-standardized prevalence rate

Burden of Migraine and TTH in G20 Countries

The distribution of migraine and TTH burden across G20 countries in 2023 revealed substantial heterogeneity across all age-standardized metrics.

For migraine, ASDR ranged from 364.74 (95% UI: 240.19–503.06) to 663.3 per 100,000 (95% UI: 445.55–920.5), ASPR from 10,498.16 (95% UI: 9156.44–12,123.64) to 18,870.57 per 100,000 (95% UI: 16,363.78–21,296.99), and ASIR from 854.57 (95% UI: 745.01–992.92) to 1481.29 per 100,000 (95% UI: 1298.14–1696.02) across G20 members. This disparity persisted consistently across all three metrics simultaneously. Countries with similar SDI levels exhibited markedly divergent burden trajectories over the 1990–2023 period. Among temporal trends, China recorded the largest increasing AAPC across all three migraine metrics (ASDR: 0.2230%, 95% UI: 0.1753–0.2708; ASPR: 0.2075%, 95% UI: 0.1686–0.2465; ASIR: 0.1719%, 95% UI: 0.1345–0.2093), while South Korea showed the largest decreasing AAPC (ASDR: −0.1322%, 95% UI: −0.2446 to −0.0197; ASPR: −0.1552%, 95% UI: −0.2576 to −0.0527; ASIR: −0.1047%, 95% UI: −0.1552 to −0.0541) (Fig. 1a–c, Table 1).

Fig. 1.

Geographic distribution of migraine burden in G20 countries in 2023. a–c show ASDR, ASPR, and ASIR, respectively. Countries are shaded according to burden intensity

For TTH, inter-country variation was similarly pronounced, with ASDR ranging from 41.43 (95% UI: 27.75–60.23) to 97.59 per 100,000 (95% UI: 64.23–140.95), ASPR ranging from 18,564.14 (95% UI: 16,310.82–20,860.90) to 34,104.11 per 100,000 (95% UI: 30,349.05–37,830.18), and ASIR ranging from 6802.61 (95% UI: 5869.84–7690.11) to 11,655.69 per 100,000 (95% UI: 10,181.13–13,146.96). Temporal trends highlighted China as the nation with the most rapidly increasing burden across all three TTH metrics: ASDR AAPC of 0.1381% (95% UI: 0.1293–0.1469), ASPR AAPC of 0.2330% (95% UI: 0.2190–0.2470), and ASIR AAPC of 0.1733% (95% UI: 0.1624–0.1843). The USA demonstrated the largest decreasing trends across all three metrics: ASDR AAPC of −0.0586% (95% UI: −0.0646 to −0.0526), ASPR AAPC of −0.0938% (95% UI: −0.0973 to −0.0904), and ASIR AAPC of −0.0288% (95% UI: −0.0300 to −0.0275). Several high-income nations showed stable or declining age-standardized rates over the study period (Fig. 2a–c, Table 2).

Fig. 2.

Geographic distribution of TTH burden in G20 countries in 2023. a–c show ASDR, ASPR, and ASIR, respectively. Countries are shaded according to burden intensity

Age and Sex

Heatmap analysis revealed consistent life-cycle patterns for migraine and TTH burdens across G20 countries in 2023. Across all disease metrics, the burden remained exceptionally low in the 5–9 years age group before escalating sharply starting from age 15–19. The burden peaked within the 20–54 years working-age cohort and subsequently declined in older age groups, highlighting the substantial impact of headache disorders on the labor force.

Regarding gender-specific distributions, migraine exhibited a pronounced female predominance. In most countries and across nearly all age groups, female DALY rates were 2.0–3.0 times higher than those of males, with incidence and prevalence rates also systematically exceeding male figures (Fig. 3a–c). In contrast, TTH demonstrated a more balanced gender distribution, with female-to-male ratios largely hovering between 1.0 and 1.2 (Fig. 4a–c).

Fig. 3.

Age- and sex-specific burden of migraine in G20 countries in 2023. a–c show prevalence, incidence, and DALYs, respectively, along with female-to-male ratios

Fig. 4.

Age- and sex-specific burden of TTH in G20 countries in 2023. a–c show prevalence, incidence, and DALYs, respectively, along with female-to-male ratios

Regional analysis further identified several geographic outliers and demographic extremes. In the Americas, Canada exhibited a significantly higher migraine burden for females across almost all age groups, while Mexico showed a prominent burden peak among females aged 20–49. European entities, including the EU, Germany, France, Italy, and the UK, reported markedly elevated DALY rates for females aged 35–49. Additionally, China displayed a unique profile where TTH incidence and prevalence among females aged 60–74 were significantly higher than those of their male counterparts. Detailed age- and sex-specific burden data are provided in Supplementary Material (Supplementary Tables of Age and Sex).

Relationship Between Headache Burden and SDI

From 1990 to 2023, the burden of migraine and TTH in G20 countries showed a nonlinear correlation with the SDI. The burden of both disorders generally scales upward with socio-economic development, peaking at an SDI of approximately 0.8 before exhibiting a subsequent decline. Geographical heterogeneity was observed: Italy’s migraine indicators far exceeding the average level, while Russia emerged as a prominent outlier with high ASDR for TTH. Conversely, trajectories for East Asian countries—such as China, Japan, and South Korea—typically remained below the average level (Fig. 5–6).

Fig. 5.

Correlation between SDI and migraine burden in G20 countries, 1990–2023. a–c show correlations with ASDR, ASPR, and ASIR, respectively. Points represent country-year observations; curves indicate expected trends

Fig. 6.

Correlation between SDI and TTH burden in G20 countries, 1990–2023. a–c show correlations with ASDR, ASPR, and ASIR, respectively. Points represent country-year observations; curves indicate expected trends

Decomposition Analysis of Migraine and TTH Between 1990 and 2023

From 1990 to 2023, the absolute numbers of DALYs, prevalent cases, and incident cases for both migraine and TTH exhibited a consistent upward trend across G20 countries. Decomposition analysis identified population growth as the primary driver of this increase (Figs. 7, 8). For migraine, population growth accounted for an increase of 55.77%, 62.83%, and 61.43% in DALYs, prevalent cases, and incident cases, respectively (Supplementary Table S1–S3). This driving effect was even more pronounced for TTH, contributing to 79.52%, 89.14%, and 82.98% of the growth across the corresponding metrics (Supplementary Tables S4–S6). Population aging was identified as the secondary driver, while changes in specific epidemiological indicators contributed the least to the overall increase in burden.

Fig. 7.

Decomposition of changes in migraine burden in G20 countries, 1990–2023. a–c show contributions of population growth, aging, and epidemiological change to DALYs, prevalence, and incidence, respectively

Fig. 8.

Decomposition of changes in TTH burden in G20 countries, 1990–2023. a–c show contributions of population growth, aging, and epidemiological change to DALYs, prevalence, and incidence, respectively

Regional analysis revealed several instances of extreme driving effects. The contribution of population growth to the increase in migraine incident cases in China reached 541.87% (Supplementary Table S3). Germany exhibited a unique pattern where migraine DALYs were driven by aging at an extraordinary rate of 12,987.61%, despite a powerful counter-effect of −16,903.54% from population decline (Supplementary Table S1). Additionally, the TTH burden in Russia and Italy showed significant sensitivity to demographic shifts, with aging driving the increase in incident and prevalent cases by 213.95% and 256.97%, respectively (Supplementary Tables S5–S6).

SHAP Analysis

The study employed the XGBoost algorithm to model the disease burden of migraine and TTH across G20 countries. All models demonstrated exceptional predictive performance, with R² > 0.99. SHAP attribution analysis was subsequently applied to quantify and rank the contributions of core features.

The SHAP summary plots identify log-transformed population size (Log_Population) as the dominant predictor for migraine models. SHAP dependence plots further elucidate a nonlinear positive correlation, where the marginal contribution of population size to migraine burden accelerates as the scale increases (Fig. 9a–c). In contrast, the population dependence analysis for TTH reveals a nonlinear trajectory. Specifically, once the Log_Population exceeds a threshold of approximately 19, the SHAP values exhibit a downward trend and transition into negative territory. This phenomenon suggests a “saturation effect” in hyper-populated contexts, where the marginal contribution of the population base to TTH burden diminishes. In these instances, the model applies a negative correction to account for the overflow impact of rapid population expansion on predicted TTH values, marking a significant divergence from the sustained positive pull observed in migraine models (Fig. 10a–c).

Fig. 9.

SHAP analysis of migraine burden in G20 countries. a–c correspond to DALYs, prevalence, and incidence models, respectively. SHAP summary plots indicate feature importance, and dependence plots show nonlinear effects of key predictors

Fig. 10.

SHAP analysis of TTH burden in G20 countries. a–c correspond to DALYs, prevalence, and incidence models, respectively. SHAP summary plots indicate feature importance, and dependence plots show non-linear effects of key predictors

The contribution of different age groups varies across headache types. Dependence plots identify the 5–9 years age group as a “low-risk window” for the global headache burden. Conversely, the high-risk window is primarily concentrated within the working-age population (20–54 years). Furthermore, the high importance of the 95+ years age group in specific morbidity indicators—particularly migraine and TTH incidence (Figs. 9, 10).

Gender plays a pivotal role in differentiating the burden profiles of these two headache types. For migraine-related models, both SHAP summary and dependence plots identify female gender as a potent positive driver of predicted burden. In contrast, TTH dependence plots show while the risk for females remains higher, the gender bias is significantly less pronounced than that observed in migraine (Fig. 9).

Predictive Analysis for 2035

This study utilized ARIMA models to analyze historical data (1990–2023) on the ASDR, ASPR, and ASIR for migraine and TTH, stratified by sex, and subsequently projected the future trends of these indicators from 2024 to 2035. The G20 countries’ prediction results are presented in Figs. 11 and 12, while the predicted values for individual countries are displayed in the Supplementary Material (Fig. S21–S26 and Supplementary Tables of the ARIMA).

Fig. 11.

ARIMA-based projections of migraine burden in G20 countries, 2024–2035. a–c show projected ASDR, ASPR, and ASIR, respectively. Solid lines represent observed values, and dashed lines indicate forecasts with 95% confidence intervals

Fig. 12.

ARIMA-based projections of TTH burden in G20 countries, 2024–2035. a–c show projected ASDR, ASPR, and ASIR, respectively. Solid lines represent observed values, and dashed lines indicate forecasts with 95% confidence intervals

For migraine, the forecasts for G20 countries from 2024 to 2035 indicate that the ASDR is projected to see a minimal decrease from 482.89 to 481.66 per 100,000. In contrast, the ASPR is expected to decline from 13,965.37 to 13,894.89 per 100,000, and the ASIR from 1146.2 to 1133.14 per 100,000. For TTH, the ASDR is predicted to remain nearly stable, with a slight decrease from 53.37 to 52.89 per 100,000. The ASPR is projected to rise from 25,009.67 to 25,177.84 per 100,000, and the ASIR from 8931.82 to 8937.67 per 100,000.

In summary, the ASDR for both headache disorders is forecast to remain largely stable overall. However, the ASPR and ASIR for migraine are projected to experience an overall decline, while those for TTH are projected to see an overall increase (Figs. 11, 12).

Discussion

Our study shows that the increase in migraine and TTH burden across G20 countries is primarily driven by population growth and aging. While demographic shifts remain the dominant macro-level drivers, our SHAP-XGBoost model successfully identifies critical population “saturation effects” and nonlinear marginal contributions of sex, age, and population size. Furthermore, we identify a burden plateau at mid-to-high SDI levels, with projections suggesting the migraine burden will stabilize, whereas TTH prevalence rates and incidence rates will likely to continue to rise by 2035. These results provide a reference for G20 countries in the formulation and optimization of public health strategies to mitigate the burden of migraine and TTH.

A Comparative Analysis of Headache Burden Drivers in G20 Countries

By comparing two analytical methods, we explored the drivers of the headache burden in G20 countries. Both methods converge on the macro trend that population size is the main driver of the increasing burden. However, the analysis also revealed several divergent conclusions. First, decomposition analysis is essentially a linear description, assuming a constant impetus for each unit of population growth. Conversely, the SHAP dependence plot unmasks a profound nonlinear “saturation effect.” Particularly in the TTH model, the contribution rate drops significantly once the population size crosses a specific threshold. This divergence theoretically hints at a potential “resilience” within the healthcare infrastructures of larger G20 countries. Specifically, the coverage efficiency of public health interventions or the adaptability of social groups may, to some extent, offset the load brought by pure physical population growth [19]. However, it may partially reflect modeling artifacts or the mathematical properties of the log-transformation applied to population data, which tends to compress larger values. This finding should be viewed as a predictive association rather than a confirmed causal mechanism, and further empirical research is required to validate these potential drivers.

Second, traditional decomposition methods are based on a linear assumption that treats aging, epidemiological changes, and population growth as mutually independent dimensions. A potential limitation of this linearity is that it may not fully explain differences in burden trends among countries with similar demographic profiles. Because decomposition analysis assigns a fixed contribution value to each incremental change in a single variable, it may mask the complex, nonlinear interactions between variables [10, 11]. In contrast, the SHAP analysis, through a game-theoretic framework, isolates the interactive contributions of variables [13]. The results demonstrate that sex and age do not act in isolation but exhibit a powerful synergistic amplification effect; the marginal contribution of female status during the peak productivity years (ages 20–54) far exceeds the simple sum of the two. This coupling reflects the complexity of headache pathogenesis. While decomposition analysis masks these overlapping risks due to its inability to handle “multicollinearity” between variables, SHAP restores this authentic socio-biological driving logic.

It is essential to clarify that these two methods operate at fundamentally different analytical scales. Decomposition analysis operates on the absolute population scale, partitioning the total change in absolute numbers of cases between 1990 and 2023 into contributions from population growth, aging, and epidemiological changes [20]. It directly answers: “How many additional patients will healthcare systems need to accommodate?” In contrast, SHAP-XGBoost operates on the log-transformed rate scale. This transformation has important consequences: (1) the log transformation compresses larger values, altering relationships compared with the original rate scale [21]; (2) SHAP values represent multiplicative rather than additive effects, where a one-unit log-scale change represents approximately 10% change on the original scale; (3) nonlinear patterns on the log scale may not reflect true nonlinearity on the original rate scale [22]. It directly answers: “Which demographic features are most predictive of distinguishing high-burden from low-burden nations?” These scale differences are complementary, not contradictory. Both methods are necessary: decomposition reveals absolute burden growth requiring capacity expansion, while SHAP reveals which subpopulations drive rate heterogeneity [17]. This transition from “describing the status quo” to “explaining the mechanism” constitutes the core value of the dual-model framework in this study. It serves as a reminder to policymakers: in populous nations, blindly expanding general medical resources may yield diminishing returns; instead, precision management should be implemented for the high-marginal-contribution groups identified by SHAP. Therefore, we provide individualized SHAP analysis results for each G20 country to serve as an evidence-based reference for national policymakers in tailoring headache management strategies (Supplementary Fig. S1–S20).

Sex and Age Disparities: A Complex Interplay of Biology and Psychosocial Factors

Our analysis revealed a pronounced female preponderance in the prevalence and burden of migraine and TTH across G20 countries, with notable variations across the lifespan. This highlights the critical role of sex and age as key biological variables influencing the distribution of headache disorders. The neurobiological underpinning for this sex disparity is partly attributed to the role of estrogen. Fluctuations in estrogen levels are known to modulate the excitability of the trigeminovascular system, leading to its dysregulation. This can result in increased release of calcitonin gene-related peptide (CGRP) and substance P, promoting neurogenic inflammation and contributing to the peripheral and central sensitization processes that underlie migraine attacks [23, 24]. While TTH does not exhibit as pronounced a sex disparity as migraine, female patients commonly report a higher prevalence of psychosocial stress and comorbid mood disorders. These factors are believed to exacerbate the burden of TTH, potentially through mechanisms involving the central dysregulation of pain processing, leading to heightened pain sensitivity and reduced inhibitory control [25]. The refined age-specific patterns we identified further support the dynamic interplay between biology and life stage. The peak incidence in adolescence coincides with pubertal hormonal changes, while the peak DALYs in middle age may reflect the cumulative burden of a chronic condition during peak productive years, as well as perimenopausal transitions in women [26, 27].

Demographic Composition as the Primary Moderator of Inter-Country Burden Heterogeneity

The differential pace and composition of these demographic transitions across G20 members may partly account for the inter-country divergence observed in cross-sectional comparisons. As our decomposition analysis demonstrates, population growth and aging together account for the dominant share of burden variation over time [20, 28].

Nations exhibiting rapidly increasing burden trajectories—most notably China—share key demographic characteristics: large populations undergoing accelerating aging, with young cohorts shifting into the peak-risk window for headache (ages 20–54) [29, 30]. The finding that population growth accounted for 541.87% of migraine incident case increases in China reflects a demographic amplification effect: as large cohorts move through high-risk age windows simultaneously, even stable age-specific rates generate substantial absolute burden increases [31]. This pattern has potential policy relevance: it suggests that burden growth in such nations may continue even in the absence of any change in underlying disease risk, and that interventions targeting age-specific incidence rates could have outsized population-level impact precisely because of the large cohort sizes passing through high-risk ages.

Conversely, nations demonstrating stable or declining age-standardized rates despite overall population growth illustrate a potentially moderating effect of healthcare system maturation. The SHAP analysis revealed a saturation pattern in which the marginal contribution of population size to TTH burden diminishes beyond a threshold of approximately log-population of 19, potentially consistent with the hypothesis that larger, more developed healthcare systems may partially absorb burden through improved case management and preventive care, though this association warrants caution as it reflects a modeling pattern rather than a confirmed causal mechanism.

The age-specific burden patterns identified in our heatmap analyses further reinforce the potential centrality of demographic structure. The sharp escalation of burden from age 15–19 onward and the concentration of peak DALYs within the 20–54 working-age cohort suggest that nations with large and growing working-age populations may face compounding productivity losses in addition to direct healthcare costs [32, 33]. This age-structured burden profile may inform how G20 countries prioritize interventions: countries with younger demographic profiles and rising burden trajectories may be better positioned for early, preventive investment, whereas those with older profiles might benefit from focusing on managing chronic burden in established patient populations.

The Nonlinear Nexus Between SDI and Headache Burden

The association analysis between age-standardized indicators and the SDI reveals that the headache burden exhibits a distinct nonlinear evolutionary pattern alongside social development. In lower SDI stages, prevalence and incidence show an upward trend driven by “diagnostic expansion” resulting from improved medical resources [34]. However, when the SDI exceeds a threshold of approximately 0.8, several nations enter a descending phase. This suggests that advanced medical interventions and preventative strategies in highly developed societies may partially alleviate the disease burden [19].

These findings have potential policy implications for G20 healthcare planning. Countries with SDI below 0.8 should prioritize diagnostic capacity building and primary care training to identify cases within existing infrastructure. Nations at or approaching the SDI threshold of 0.8 face a critical intervention window, requiring simultaneous investment in capacity expansion and innovative therapeutics to effectively manage the diagnostic expansion phase [35, 36]. High-SDI countries should shift from case-finding to treatment optimization and precision prevention, with particular emphasis on reproductive-age women [37].

Divergent Future Trends in Migraine and TTH

The model forecasts for 2035 provide an exploratory projection based on historical trends [21]. The projected decline in migraine ASPR and ASIR is optimistic and likely reflects the continued positive impact of improved treatments and growing disease awareness [38]. In contrast, the rising trajectory for TTH highlights a persistent and potentially neglected challenge. The lack of specific and effective preventive medications for TTH, coupled with its strong association with ubiquitous modern stressors (e.g., psychological stress, poor posture), positions it as an increasingly significant public health issue [28, 39]. The under-recognition and trivialization of TTH as a “normal” headache further impede effective management and contribute to its chronicity.

Limitations

This study presents several limitations that warrant consideration. First, this analysis relies exclusively on GBD 2023, a systematized, modeled dataset that may not reflect differences among countries in diagnostic thresholds, healthcare-seeking behavior, or case ascertainment methods. GBD 2023 was selected as the primary data source because it is the most recently updated global disease burden database covering all G20 member states; however, this was not the only option. Other data sources—including national disease registries, WHO regional health estimates, and population-based surveys—provide higher granularity in specific regions but lack uniform coverage across all G20 countries and employ heterogeneous methodologies, which limit direct cross-national comparison. To comprehensively address the research questions, an ideal database would combine GBD’s methodological standardization with population-based surveys conducted simultaneously across all G20 countries using unified diagnostic criteria—a resource that currently does not exist. Second, GBD does not comprehensively capture environmental and lifestyle factors critical to headache burden epidemiology, which may underestimate modifiable risk factors [40]. Third, ARIMA projections are extrapolations of historical trends and do not explicitly incorporate emerging therapies approved during the study period, and should therefore be interpreted as exploratory rather than definitive predictions [21]. Fourth, the findings are specific to G20 countries and cannot be directly extrapolated to non-G20 countries; policymakers elsewhere should treat mechanistic insights as testable hypotheses applicable to their own contexts rather than direct policy prescriptions. Finally, GBD modelling uncertainty is particularly relevant for lower-SDI G20 members with sparse headache survey data, affecting the precision of decomposition percentages and SHAP feature importance values, though the relative ranking of countries and broad temporal patterns remain robust [41].

Conclusions

This study provides a comprehensive reevaluation of migraine and TTH burden across G20 countries, moving beyond descriptive epidemiology to mechanistic explanation. Our findings confirm that while population growth remains the foundational driver of the absolute burden increase, the “internal logic” of this escalation is governed by complex nonlinearities and synergistic interactions. As we look toward 2035, the divergent trajectories of migraine and TTH underscore the need for a shift in public health strategy. We advocate for a move away from uniform, population-wide resource scaling toward risk-stratified, precision intervention strategies. By identifying nonlinear inflection points and specific high-risk clusters, G20 countries can optimize healthcare delivery, ensuring that interventions are both evidence-based and economically efficient in the face of evolving demographic landscapes.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Zhuochen Chen and Guifen Fan contributed equally to this work. Zhuochen Chen, Guifen Fan, and Kai Dai were responsible for study design, data analysis, and manuscript drafting. Xueting Liu and Xu Cheng contributed to data interpretation and critical revision of the manuscript. Jianhong Ye and Jun Peng supervised the study and provided overall guidance. All the authors assisted with the interpretation of the findings, thoroughly reviewed the content of the manuscript, and approved the final version submitted for publication.

Funding

No funding or sponsorship was received for this study. The Rapid Service Fee was funded by the authors.

Data Availability

The datasets analyzed during the current study are available in the Global Health Data Exchange (GHDx) portal (https://ghdx.healthdata.org/gbd-results-tool).

Declarations

Conflict of Interest

Zhuochen Chen, Guifen Fan, Kai Dai, Xueting Liu, Xu Cheng, Jianhong Ye, Jun Peng have nothing to disclose.

Ethical Approval

This study used publicly available data from the GBD 2023 study provided by the IHME. All data were de-identified and aggregated; therefore, ethical approval and informed consent were not required. The GBD database is publicly accessible through the Global Health Data Exchange (GHDx) query tool (https://ghdx.healthdata.org/gbd-results-tool). Our analysis is governed by the GBD 2023 Publication Policy and Collaborative Agreement [15].