The research cohort data were derived from the SEER Research Data (2000–2018), 18 Registries, and November 2020 Sub database. A total of 9,957 patients’ information and follow-up data were collected. The criteria for patient inclusion were the availability of data based on the International Classification of Diseases for Oncology, third edition (ICD-O-3), and disease-specific survival (DSS) for individuals diagnosed with GISTs. Screening requirements for potentially eligible patients with GIST included a histologic type of ICD-O-38936 from one of the specified sites, including the Appendix, Colon, Esophagus, Peritoneum, Rectum, Small intestine, and stomach. As the SEER is a publicly available database with anonymized data, this study was exempted by the Ethics Committee of China Medical University.

Patients with GIST obtained from the SEER database were excluded based on the following criteria: (a) patients who did not undergo surgical intervention; (b) patients under 18 years of age; (c) patients with a survival duration of less than one month post-surgery; and (d) patients with incomplete, duplicate, or inaccurate pathological data. A total of 5318 cases were included in the statistical analysis. The cohort selection process is illustrated in Figure 1.

We excluded patients with missing, duplicate, or inconsistent pathological data. The variables with missing values included tumor size, mitotic index, AJCC stage and tumor site. Other variables such as age, sex, and marital status were complete. Given the lack of a robust multiple imputation strategy for the SEER database in this context and to maintain consistency in the analysis, we opted for complete-case analysis. We acknowledge that this approach may introduce selection bias, particularly if the missing data are not random. However, sensitivity analyses excluding patients with survival <1 month were conducted to assess the robustness of our findings.

Comparisons of baseline characteristics between the sexes were conducted utilizing t-tests and chi-square tests. Survival rates were ascertained using the Kaplan-Meier method, and survival curves were graphically represented. Univariate analysis was performed using the log-rank test, whereas multivariate analysis was performed utilizing the Cox test, with all baseline factors incorporated as adjustment variables. To account for mortality as a competing event, we employed the Fine and Gray subdistribution hazard model in sensitivity analyses to assess the relationship between baseline characteristics and GIST-specific mortality.

Given the potential disparities in variables between the sexes at baseline, propensity score matching (PSM) was employed to address the imbalance. Nearest-neighbor matching with a ratio of 1:1 and a caliper of 0.1 was implemented between different sex groups. The adjusted factors included age, tumor size, marital status, race, origin, tumor site, T stage, N stage, M stage, AJCC stage, radiation, and mitotic index. A standardized mean difference (SMD) of less than 0.1 was considered indicative of balance between variables.

A multiplicative interaction model was employed to analyze the influence of age and sex on the outcomes of patients with GIST. We performed a stratified analysis based on tumor location (gastric vs. non-gastric) to further investigate potential interaction effects between age and sex. To ensure the robustness of the results, a sensitivity analysis was conducted by excluding patients with a survival time of less than one month. To assess the potential non-linear relationship between age and DSS and to visualize how the HR changes continuously with age for sex, we employed restricted cubic spline (RCS) regression. This method is advantageous as it does not impose a pre-specified linear or parametric form on the relationship, allowing the data to reveal the true functional form. The number and positions of the knots, which define the smoothness and flexibility of the spline curve, were chosen based on established statistical recommendations to balance model flexibility and prevent overfitting. The relationships between age, sex, and DSS were investigated utilizing an RCS model. The patients were stratified into three age groups (< 50 years, 50–60 years, and > 60 years), and the association between sex and DSS was examined. To evaluate whether the mediating effect could elucidate the survival differences between females and males, Causal mediation analysis was performed using the ‘mediation’ package in R to quantify the direct and indirect effects of sex on GIST-specific survival. This method is grounded within the counterfactual framework for causal mediation analysis. The mediator variables included AJCC stage, tumor site, T stage, N stage, M stage, radiation, tumor size, and mitotic index. Among these, tumor size was treated as a continuous variable, while all others were included as categorical variables. R software (version 4.3.1) was utilized for all statistical analyses, with the primary packages including survival, tidycmprsk, Matchit, RCS, and mediation. Statistical significance was set at p < 0.05.

The study encompassed 9,957 patients with GIST from the SEER database, spanning the period from 2000 to 2018. Following the exclusion of non-qualifying cases, 5,318 patients were ultimately included in the analysis, comprising 2,647 females and 2,671 males. Table 1 summarizes the clinical and pathological characteristics of the overall cohort and sex subgroups. The demographic factors of men and women differed with respect to their race and marital status. Regarding clinical and pathological features, females exhibited smaller tumor sizes compared to males (66.96 mm vs. 77.52 mm, p < 0.001) and a higher proportion of gastric site involvement (66.2% vs. 60.0%, p < 0.001). Additionally, females demonstrated lower T stage and lower rates of distant metastasis compared to males (8.5% vs. 11%, p = 0.003), lower american joint committee on cancer (AJCC) stage, and lower nuclear mitotic rates (76.18% vs. 80.63%, p < 0.001).

In the entire cohort, females had a median follow-up time of 44 months, whereas males had a median follow-up time of 43 months. In the female GIST cohort, 190 patients experienced disease progression (7.18%), whereas in the male cohort, 262 patients experienced disease progression (9.81%). Univariate analysis revealed that males exhibited a higher hazard ratio (HR) for DSS than females (HR = 1.40, 95%CI: 1.16-1.68, p<0.001) (Figure 2). Multivariate analysis was adjusted for the following variables: race, origin, tumor site, T stage, N stage, M stage, radiation, age, tumor size, marital status, AJCC stage, and mitotic rate. The results demonstrated that the HR for DSS remained significantly higher in males compared to females (HR = 1.38, 95%CI: 1.10-1.74, p=0.006). The associations between sex and DSS remained robust in the competing risk models (Supplementary Table S1).

Following PSM for sex, all variables between the two subgroups exhibited adequate balance (Supplementary Figure S1, Supplementary Table S2). Prognostic analysis of the PSM-adjusted cohort revealed statistically significant differences in DSS for HR between sexes (HR = 1.36, 95% CI: 1.06-1.73, p=0.014) (Table 2). Consequently, both multivariable analysis and PSM models indicated that male sex remained an independent risk factor for GIST.

A statistically significant interaction was observed between age and sex in predicting the prognosis of GIST (p = 0.025). In the gastric tumor cohort, no significant interaction was observed between age and sex (P = 0.23). In contrast, a statistically significant interaction between age and sex was identified in the non-gastric tumor cohort (P = 0.027). Sensitivity analysis after excluding patients with a survival time of less than one month confirmed that the interaction between age and sex remained statistically significant in predicting GIST-specific survival (P = 0.029). RCS analysis revealed that the HR for DSS in males gradually increased with age, demonstrating a significant elevation after 65 years (p < 0.001 for non-linearity). Conversely, the HR for DSS in females displayed a gradual increase between the ages of 20 and 55, maintained relative stability between 55 and 65 years, and subsequently increased rapidly after 65 years (p = 0.010 for non-linearity). Moreover, the HR for DSS in males consistently exceeded that in females (Figure 3).

To investigate the differences in prognosis between sexes across different age subgroups, we categorized age into three subgroups (≤ 50 years, 51–60 years, and > 60 years) and conducted an analysis, as described in Table 3. Through subgroup analysis, we identified variations in the pathological parameters between males and females across different age groups. To validate these potential differences, we performed PSM to adjust for baseline characteristics between the different sexes within each age subgroup. After PSM, the intergroup variables within each age subgroup were well-balanced (Supplementary Figure S2). The results of the subgroup analysis indicated that the differences in DSS between sexes diminished with increasing age, demonstrating statistical significance only in the ≤ 50 years age subgroup. No significant differences were observed between the other age groups (Figure 4). In these three subgroups, survival analysis after PSM revealed that the prognostic differences between males and females were present only in the ≤ 50 years subgroup, where the HR for DSS in males was higher than that in females, with no such differences observed in the other two subgroups (Figure 5). This suggests that after adjusting for various factors, younger and middle-aged female patients with GIST exhibit a favorable prognosis; however, this advantage diminishes with increasing age.

To further investigate the impact of sex on DSS mediated by other factors, we conducted a multiple mediation analysis (Figure 6). Based on the contribution of sex to the mediation of DSS, the five most influential variables, ranked in descending order, were AJCC stage (36.8%), M stage (11.6%), T stage (11.2%), tumor size (6.8%), and N stage (0.9%). These findings suggest that sex may indirectly influence the prognosis of GIST by affecting its biological behavior.