All data used in this study were extracted from the GBD 2021 database, which was updated on May 16, 2024, by the Institute for Health Metrics and Evaluation (IHME) at the University of Washington, Seattle, USA. The GBD 2021 delivers comprehensive assessments of incidence, prevalence, deaths, DALYs, and 88 risk factors for 371 diseases and injuries across 204 countries and territories, including 811 subnational locations. Since the GBD 2021 data are publicly available and intended solely for academic research, with no commercial use involved, ethical approval for data access and use was not required. The URL information of the data sources: https://vizhub.healthdata.org/gbd-results/. In this tool, we selected “Cause of death or injury” as the “GBD Estimate”; “Liver cancer”, “Liver cancer due to hepatitis B”, “Liver cancer due to hepatitis C”, “Liver cancer due to alcohol use”, “Liver cancer due to other causes”, and “Liver cancer due to NASH” as the “Cause”; “Incidence”, “Prevalence”, “Deaths”, and “DALYs” as the “Measure”; “Number”, “Rate” and “Percent” as the “Metric”; “Global”, “China”, and “G20” as the “Location”; “All ages”, “Age-standardized”, “ < 5 years”, “5–9 years”, “10–14 years”, “15–19 years”, “20–24 years”, “25–29 years”, “30–34 years”, “35–39 years”, “40–44 years”, “45–49 years”, “50–54 years”, “55–59 years”, “60–64 years”, “65–69 years”, “70–74 years”, “75–79 years”, “80–84 years”, “85–89 years”, “90–94 years”, and “95+ years” as the “Age”; “Both”, “Male”, and “Female” as the “Sex”; and select all years from 1990 to 2021 as the “Year”. The G20 aggregate follows IHME's standardized location hierarchy where the EU is treated as a distinct entity. When reporting G20 totals, individual EU member states (e.g., France, Germany) contribute only through the EU aggregate to prevent double-counting. The complete list of G20 constituents includes 19 sovereign nations plus the EU. All estimates were extracted directly from GBD's pre-calculated location-specific results using the GBD results tool.

We selected the incidence, prevalence, deaths, and DALYs of liver cancer in China and G20 countries from the GBD 2021 study, along with the corresponding age-standardized incidence rate (ASIR), age-standardized prevalence rates (ASPR), age-standardized mortality rate (ASMR), and age-standardized DALYs rate (ASDR). The 95% confidence intervals (CIs) were demonstrated for each estimated quantity. All age-standardized rates (ASRs) presented in this study were calculated using the GBD world standard population. This standard population structure is defined and maintained by the Global Burden of Disease Collaborative Network specifically for consistent comparability within GBD studies. The methods for age standardization were consistent across all analyses, including those for China, the G20, and globally. Age and sex-specific data for China and G20 countries were downloaded for the period 1990 to 2021 to assess the liver cancer burden. To validate subgroup trends, we tested sex × year interactions using linear regression. The GBD collaborators classify the etiologies for liver cancer into five groups: hepatitis B (HBV), hepatitis C (HCV), alcohol use, non-alcoholic steatohepatitis (NASH), and other causes. “Cryptogenic” or “idiopathic” liver cancer, liver cancer caused by hemochromatosis, autoimmune hepatitis, and Wilson's disease were included in the “other causes” category. The etiology decomposition analysis illustrated how the population, age, and epidemiological changes contributed to the varying liver cancer burden between 1990 and 2021 (15).

To assess trends in liver cancer burden, the Joinpoint regression analysis was conducted using Joinpoint software (version 5.2.0; Statistical Research and Applications Branch, National Cancer Institute, National Institutes of Health, Bethesda, MD, USA). The Joinpoint regression analysis was employed to identify significant temporal trends in the ASRs. The log-transformed ASRs were modeled as ln(y) = α + βx + ε, where y represents the ASR and x denotes the calendar year. The maximum number of Joinpoints (inflection points) was set to six to balance model flexibility against overfitting, given the length of the time series. The optimal model was selected using a sequential Monte Carlo permutation test (4,500 repetitions, α = 0.05) rather than information criteria, as this method is established for identifying statistically significant trend changes while favoring more parsimonious models. The model assumed uncorrelated errors, an appropriate choice for annual population-level vital statistics data where serial autocorrelation is typically minimal. To account for the uncertainty in the ASR, a variance weighting scheme based on Poisson distribution was applied. The annual percent change (APC) with its 95% CI was calculated for each identified trend segment (16, 17). The AAPC over the entire study period, a summary measure of the trend, was derived as 100 × [exp(β) – 1], with its 95% CI obtained from the model (18). An increasing trend was defined as an AAPC with a lower 95% CI boundary > 0, while a decreasing trend was indicated by an AAPC with an upper 95% CI boundary < 0. It is important to note that our Joinpoint regression analysis involved testing for trend changes across multiple time segments, indicators (ASIR, ASPR, ASMR, ASDR), and geographical strata (China and G20 countries). This multiplicity of comparisons increases the family-wise error rate, meaning that the probability of falsely identifying a significant trend change (Type I error) across the entire analysis is higher than the significance level (α = 0.05) used for any single test. While P-values for each APC segment are presented as per conventional reporting, the interpretation of results prioritizes the point estimates (APCs) and their 95% CIs over isolated statistical significance to provide a more robust assessment of trends.

Then, we used the autoregressive integrated moving average (ARIMA) model to predict the incidence, prevalence, deaths, and DALYs of liver cancer in China and G20 countries from 2022 to 2036. The ARIMA model integrates the autoregressive (AR) and moving average (MA) approaches. Its underlying assumption posits that time series represent time-dependent random variables. ARIMA characterizes their autocorrelation and facilitates predictions of future values based on previously observed values (19). Initially, stationarity testing was performed on all series. For non-stationary series, differencing was applied until stationarity was achieved (P < 0.05), with the order of differencing denoted as d. Subsequently, based on the differenced stationary series, autocorrelation function (ACF) plot was examined to preliminarily identify the potential range for the autoregressive order p and moving average order q. The optimal model orders (p, d, q) were ultimately determined using the auto.arima() function from the R forecast package, which performs an automated search aimed at minimizing the akaike information criterion (AIC). Model parameters were estimated via maximum likelihood estimation (MLE). Following model establishment, residual diagnostics were conducted to ensure its adequacy: the residual ACF plot was inspected to confirm the absence of autocorrelation structures, and the Ljung-Box test [with the number of lags set to (10)] was performed to verify that the residuals constituted a white noise series (P > 0.05). To evaluate the applicability of different time-series forecasting methods to the data in this study, we fitted and compared three commonly used models: (1) the ARIMA model, which excels at capturing the autocorrelation and moving average structure of a series; (2) the ETS (Error, Trend, Seasonality) model, based on exponential smoothing, which is suitable for identifying and forecasting underlying trends, seasonality, and error components in a time series; and (3) the Naive Drift model, which serves as a simple benchmark by considering only the average change between observations. Comparing these models is crucial for identifying the forecasting method that best captures the dynamic characteristics of the specific health indicator. Subsequently, model predictive accuracy on the test set was assessed using metrics such as the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), while the AIC and BIC were employed to evaluate the balance between model goodness-of-fit and complexity.

Data analysis and visualization were performed using R statistical software (version 4.4.1) and Joinpoint software (version 5.2.0). A P-value of < 0.05 was considered statistically significant.

In China, the number of liver cancer cases increased from 96,434 (95% CI: 80,971–113,769) in 1990 to 196,637 (95% CI: 158,273–243,558) in 2021. Despite this increase in absolute numbers, the ASIRs experienced a decline from 10.58 per 100,000 people (95% CI: 8.94–12.43) in 1990 to 9.52 per 100,000 people (95% CI: 7.72–11.78) in 2021, resulting in an AAPC of −0.31% (95% CI: −0.39 – −0.23). In G20 countries, liver cancer cases rose from 185,549 (95% CI: 169,557–202,922) to 407,347(95% CI: 365,029–460,802) during the same period, with a slight increase in ASIR from 5.89 per 100,000 people (95% CI: 5.39–6.43) to 6.33 per 100,000 people (95% CI: 5.67–7.16), reflected by an AAPC of 0.20% (95% CI: 0.15–0.24) (Table 1).

In terms of prevalence, the number of liver cancer cases in China more than doubled, increasing from 132,779 (95% CI: 108,924–155,564) in 1990 to 265,539 (95% CI: 212,435–331,149) in 2021. However, the ASPR remained relatively stable, with a slight decrease from 13.51 per 100,000 people (95% CI: 11.20–15.77) to 13.29 per 100,000 people (95% CI: 10.75–16.41), reflected by an AAPC of 0.02% (95% CI: −0.11–0.15). In G20 countries, the ASPR increased from 7.88 per 100,000 people (95% CI: 7.11–8.62) in 1990 to 9.25 per 100,000 people (95% CI: 8.30–10.38) in 2021, reflected by an AAPC of 0.50% (95% CI: 0.46–0.54) (Table 1).

The number of liver cancer deaths in China rose from 94,937 (95% CI: 79,884–111,527) in 1990 to 172,068 (95% CI: 139,621–212,496) in 2021. The ASMR decreased from 10.75 per 100,000 people (95% CI: 9.12–12.61) to 8.35 per 100,000 people (95% CI: 6.80–10.29), with an AAPC of −0.68% (95% CI: −1.25 – −0.10). Meanwhile, the liver cancer deaths in G20 countries increased from 178,992 (95% CI: 163,158–195,991) in 1990 to 361,134 (95% CI: 323,331–408,535) in 2021, while the ASMR remained relatively stable, with a slightly decline from 5.75 per 100,000 people (95% CI: 5.26–6.28) to 5.59 per 100,000 people (95% CI: 5.00–6.32), reflected by an AAPC of −0.04% (95% CI: −0.24–0.15) (Table 1).

The DALY attributed to liver cancer in China escalated from 3,294,864 (95% CI: 2,763,029–3,879,589) in 1990 to 4,890,023 (95% CI: 3,905,089–6,124,599) in 2021. Nonetheless, the ASDR significantly decreased from 334.52 per 100,000 people (95% CI: 281.08–393.14) to 239.91 per 100,000 people (95% CI: 191.98–299.37), resulting in an AAPC of −0.96% (95% CI: −1.36 – −0.56). In G20 countries, DALYs increased from 5,609,295 (95% CI: 5,052,422–6,184,552) in 1990 to 9,250,469 (95% CI: 8,204,183–10,595,225) in 2021, while the ASDR showed a decline from 171.28 per 100,000 people (95% CI: 154.62–188.76) to 146.24 per 100,000 people (95% CI: 129.45–167.60), reflected by an AAPC of −0.46% (95% CI: −0.72 – −0.19) (Table 1).

In China, the number of liver cancer cases increased from 96,434 in 1990 to 196,637 in 2021, representing a 103.91% increase. Among G20 countries, cases rose from 185,549 to 407,347 over the same period, a 119.54% increase. The cumulative number of deaths in China increased by 81.24% between 1990 and 2021, while G20 countries experienced a 191.71% rise. ASIR, ASPR, ASMR and ASDR for liver cancer all declined in China during this period. Conversely, throughout the 1990–2021 period, the G20 countries overall experienced a slight increasing trend in ASIR and ASPR, though ASMR and ASDR decreased. Notably, ASIR, ASPR, ASMR, and ASDR in G20 countries were markedly lower than those in China in both 1990 and 2021.

Joinpoint regression analysis of liver cancer ASIR, ASPR, ASMR, and ASDR trends for China and G20 countries (1990–2021) is presented in Figure 1. The segment-specific APCs are reported in Supplementary 3. In China, ASMR and ASDR exhibited significant increases from 1990 to 2001 (ASMR: APC = 0.53, 95% CI: 0.24 −0.82; ASDR: APC = 0.39, 95% CI: 0.18–0.59). Concurrently during 2000–2005, ASMR and ASDR also demonstrated significant reductions (P < 0.05). From 1995 to 2000, ASIR and ASPR for liver cancer in China showed a significant upward trend (ASIR: APC = 1.58, 95% CI: 1.35–1.86; ASPR: APC = 1.74, 95% CI: 1.33–2.16). However, ASIR and ASPR declined significantly between 2000 and 2005 (ASIR: APC = −3.33, 95% CI: −3.60 – −3.19; ASPR: APC = −3.94, 95% CI: −4.26 – −3.62). From 2005 to 2015, ASIR and ASPR displayed renewed increases (P < 0.05), whereas ASMR and ASDR trends were non-significant. Finally, all four metrics showed significant declines during 2015–2021 (ASIR: APC = −1.36, 95% CI: −1.99 – −1.40; ASPR: APC = −0.65, 95% CI: −0.98 – −0.33; ASMR: APC = −2.16, 95% CI: −3.12 – −1.20; ASDR: APC = −1.92, 95% CI: −2.63 – −1.20).

In contrast, G20 countries experienced significant increases in all four indicators from 1990 to 2000. Throughout 2000–2021, ASIR and ASPR fluctuated but maintained an overall upward trajectory, while ASMR and ASDR displayed overall declines, with ASDR showing a particularly pronounced downward trend.

From 1990 to 2021, ASDR of liver cancer in China and G20 countries showed an overall declining trend. Specifically, in China, the ASDR experienced a slight increase from 1990 to 2000, followed by a significant decrease from 2000 to 2005, and then a gradual decline after 2005 (Figure 2A). In contrast, G20 countries exhibited a steady decline in ASDR. Additionally, from 1990 to 2021, both ASIR, ASPR and ASMR of liver cancer in China and G20 countries showed a slight fluctuation trend (Figure 2B).

Figures 2C, D illustrates temporal trends in ASMR and ASDR for liver cancer across all ages and genders in China and other G20 countries (1990–2021). Notably, Japan and South Korea exhibited among the highest ASMR and ASDR rates within the G20. Similarly, China demonstrated persistently elevated rates, with both indicators showing a consistent upward trajectory over the study period.

Figures 3, 4 display the incidence, prevalence, deaths, and DALYs numbers for liver cancer across different age groups in China and G20 countries from 1990 to 2021 (Figure 3) and the ASRs (Figure 4). Liver cancer is more prevalent in individuals aged over 35 in both China and G20 countries, showing substantial increases between ages 35–85. Age-specific analyses reveal these temporal patterns: For ASIR, peak incidence in 1990 occurred at 85–89 years for males (China, G20) and 90–94 years for Chinese females vs. 85–89 years for G20 females; by 2021, Chinese females peaked earlier at 85–89 years (previously 90–94 in 1990) while Chinese males peaked later at 90–94 years, with G20 males and females maintaining peaks at 85–89 years. For ASPR, 1990 peaks were at 65–69 years for males (China, G20) and 85–89 years for females, shifting to 85–89 years for all groups by 2021. Regarding ASMR, China maintained peaks at 90–94 years for both sexes (1990/2021), while G20 shifted from male (85–89 years) and female (90–94 years) peaks in 1990 to delayed 2021 peaks at 90–94 years (males) and >95 years (females). For ASDR, Chinese males shifted from 65–69 years (1990) to 90–94 years (2021), while females peaked earlier (80–84 years in 2021 vs. 85–89 years in 1990). G20 males shifted from 65–69 years to 90–94 years, with females peaking later (>95 years in 2021 vs. 70–74 years in 1990). Notably, in both 1990 and 2021, the incidence, prevalence, deaths, and DALYs rates for males in China and G20 countries were higher than those for females. It is also noteworthy that the peak incidence and DALYs rate of liver cancer in Chinese women are both advanced.

Figure 5 illustrates the all-age case counts and ASR (incidence, prevalence, mortality, and DALYs) by gender for liver cancer in China and G20 countries from 1990 to 2021. Gender-specific and ASR of incidence, prevalence, mortality, and DALYs exhibited year-to-year fluctuations. Overall, China experienced declining trends in ASMR and ASDR, while ASIR and ASPR showed stable fluctuations. Conversely, all-age case counts of incidence, prevalence, mortality, and DALYs increased consistently. Among G20 countries, ASIR, ASPR, ASMR, and ASDR demonstrated broadly fluctuating patterns, yet all-age case counts trended upward. Notably, males consistently displayed higher ASIR, ASPR, ASMR, ASDR, and all-age case counts than females across all observations.

Based on the data from the sex × year interaction analysis in China and G20 countries (Supplementary 4), the results demonstrate highly significant overall models for all age-standardized metrics (ASIR, ASPR, ASMR, ASDR) across both groups, evidenced by exceptional model fits (adjusted R² > 0.99) and robust overall significances (all F-statistics P < 0.001). Specifically, in China, significant sex-specific temporal interactions (P < 0.05) were detected for ASPR, ASMR, and ASDR, revealing diverging gender trends: ASPR exhibited a female decline (−0.04/year) contrasted with a male increase (+0.03/year), while ASMR and ASDR showed reductions for both sexes, albeit steeper for males (ASMR male: −0.09/year vs. female: −0.05/year; ASDR male: −4.44/year vs. female: −2.20/year). For G20 countries, significant interactions emerged for ASIR, ASMR, and ASDR, with both genders displaying slight rises in ASIR and ASPR (ASIR: male: +0.01/year; female: +0.00/year; ASPR male: +0.07/year; female: +0.01/year) but declines in ASMR and ASDR, where males experienced larger decreases (ASMR: male: −0.02/year vs. female: −0.00/year; ASDR male: −1.70/year vs. female: −0.53/year). Non-significant interactions for ASIR in both groups (P > 0.05) suggested similar, albeit mild, trends by gender, such as gradual female and male reductions in China vs. slight increases in G20. These findings underscore distinct dynamics of ASIR, ASPR, ASMR, ASDR, with China's patterns indicating greater gender divergence in certain indicators compared to G20′s more uniform trends, highlighting contextual influences on sex-specific changes.

Tables 2, 3 summarize the ASRs and case numbers for five etiologies of liver cancer in China and G20 countries for 1990 and 2021, with temporal trends from 1990 to 2021 illustrated in Figure 6. During this period, the case number of incidence, prevalence, deaths, and DALYs attributable to these five etiologies increased in China and G20 countries. In China, NASH exhibited the most substantial growth in incident cases (+178.36%), prevalence (+203.36%), deaths (+152.16%), and DALYs (+104.72%). This pattern of NASH being the most rapidly increasing cause of liver cancer was also consistent in G20 countries and worldwide. Between 1990 and 2021, China experienced a decline in the ASIR for liver cancer related to HBV and HCV, while the ASIRs for NASH and alcohol-related liver cancer rose. The etiological profile of liver cancer in China was generally similar to that in G20 countries. At the global level in 2021, HBV infection was the predominant cause, accounting for 38.99% of incident liver cancer cases, followed by HCV (29.81%), alcohol use (18.75%), NASH (8.09%), and other causes (4.36%). A comparable distribution was observed for mortality, with HBV responsible for 37.25% of deaths, HCV for 31.00%, alcohol use for 18.93%, NASH for 8.55%, and other causes for 4.28% (Figure 6). HBV was the leading etiology in both China and G20 countries, contributing to 60.70% of incident cases and 58.09% of deaths in China, and 41.02% of cases and 39.23% of deaths in the G20 group. HCV was the second leading contributor to the liver cancer burden in these populations.

We fitted the ASIR, ASPR, ASMR, and ASDR of liver cancer stratified by gender from 1990 to 2021 using the ARIMA model and predicted the ASIR, ASPR, ASMR, and ASDR for 2022–2036, selecting the optimal model parameters along with the corresponding AIC and BIC (Supplementary 1). Supplementary 1 detail the performance metrics (RMSE, MAE, and MAPE) of the ARIMA, ETS, and Naive Drift models in forecasting ASIR, ASPR, ASMR, and ASDR for both male and female in China and the G20. The ACF test diagram of ARIMA model and the prediction trend diagram of ETS and naive drift model are summarized in Supplementary 2. The results demonstrate that the ARIMA model achieved the best predictive accuracy (lowest RMSE, MAE, and MAPE) for the majority of indicators, particularly for ASIR and ASPR, which further validates its selection for this analysis.

Between 2022 and 2036, Chinese male ASIR is projected to increase from 14.21 to 14.91 per 100,000 people, representing a 4.93% rise. The female ASIR will climb from 4.82 to 4.93 per 100,000 people, marking a 2.28% increase. In contrast, G20 countries will see their male ASIR rise from 9.48 to 9.68 per 100,000 people (2.11% increase), while the female ASIR will grow from 3.47 to 3.87 per 100,000 people, corresponding to a 11.53% increase. ASPR for males in China is forecasted to increase slightly from 19.71 to 19.95 per 100,000 people (1.22% increase), with the female ASPR rising minimally from 6.49 to 6.52 per 100,000 people (0.46% increase). Conversely, G20 countries are expected to experience a male ASPR decline from 13.94 to 12.92 per 100,000 people (7.32% decrease), while the female ASPR will decrease marginally from 8.20 to 8.17 per 100,000 people (0.36% decrease). The male ASMR in China is predicted to decline slightly from 12.22 to 12.16 per 100,000 people (0.49% decrease), whereas the female ASMR will decrease substantially from 4.48 to 3.64 per 100,000 people (18.75% decline). In G20 countries, male ASMR will decrease from 8.20 to 8.17 per 100,000 people (0.37% decrease), but female ASMR will increase from 3.24 to 3.36 per 100,000 people (3.70% rise). Regarding ASDR, Chinese males will see a marginal reduction from 363.10 to 362.11 per 100,000 people (0.27% decrease), while males in G20 countries will experience a slight increase from 218.74 to 219.35 per 100,000 people (0.28% increase). Chinese females will demonstrate a significant ASDR reduction from 109.70 to 80.15 per 100,000 people (26.94% decline), compared to a minor decrease among females in G20 countries from 75.98 to 75.64 per 100,000 people (0.45% decrease). Overall, model projections indicate that between 2022 and 2036, both China and G20 countries will experience predominantly modest changes in ASRs (Table 4 and Figure 7). For China, the majority of projected changes fall within a narrow range: ASIR increases by 4.93% (male) and 2.28% (female), ASPR increases by 1.22% (male) and 0.46% (female), ASMR decreases by 0.49% (male) and increases by 3.70% (female), and ASDR decreases by 0.27% (male). Notably, Chinese females show a substantial ASMR decrease (18.75%) and ASDR decrease (26.94%). In G20 countries, changes are similarly contained for most metrics: ASIR increases by 2.11% (male) and 11.53% (female), ASPR decreases by 7.32% (male) and 0.36% (female), ASMR decreases by 0.37% (male) and increases by 3.70% (female), and ASDR increases by 0.28% (male) and decreases by 0.45% (female). The most significant deviations from this pattern of limited change are the larger increase in female ASIR in G20 countries and the pronounced decreases in female ASMR and ASDR in China.