This study aimed to retrospectively analyse data from the Surveillance, Epidemiology, and End Results (SEER) program database. We used SEER Stat software (SEER*Stat 8.4.1, Incidence-SEER 19 Regs Custom Data with additional treatment fields, Nov 2019 Sub [1973–2019 varying]) to retrieve the data for all patients diagnosed with GISTs (M8936, C16.0–C21.1) between 2010 and 2019. Patients with no cancer-directed surgery (n=1,563), M1 disease indicated as “M1 or Mx” (n=596; Mx cases were excluded because an indeterminate metastatic status might have compromised staging accuracy and introduce systematic bias if included), systemic therapy before surgery or before and after surgery (n=494), and missing tumour size or mitotic count data (n=2,019; handled using listwise deletion), were excluded. Patients with survival times listed as “0” (n=42) or with a SEER cause-specific death classification of “missing/unknown cause of death (COD)” (n=9) were also excluded. Ultimately, a total of 3,034 patients were included, of which 2,106 had gastric GISTs and 928 had small intestine and colorectal GISTs. ( Figure 1 ). Because of the AJCC TNM system, gastric GISTs and small intestinal GISTs were classified differently, and colorectal GISTs were classified into the same cohort in this study. Cancer-specific survival (CSS) was chosen over overall or progression-free survival as the primary endpoint because the SEER database provides cause-specific death data but lacks progression information, and overall survival may be confounded by non-cancer-related deaths in older patients. CSS was calculated from the date of GIST diagnosis to either cancer-related death or the most recent follow-up. As the SEER database provides de-identified, publicly available data, which does not involve any risk to personal privacy, institutional review board approval and written consent from participants were not required.

The 8th edition (12) of the AJCC TNM staging system (2017) was applied to classify patients with GISTs. It incorporates tumour size (T), lymph node involvement (N), distant metastasis (M), and mitotic activity to stratify patients into prognostic stage groups. Tumour size categories were as follows: T1 (≤2 cm), T2 (>2 cm and ≤5 cm), T3 (>5 cm and ≤10 cm), and T4 (>10 cm). Nodal status was defined as N0 (no regional lymph node metastasis or unknown status) or N1 (regional lymph node metastasis). Distant metastasis was defined as M0 (absent) or M1 (present). Mitotic rate was graded as low (≤5 mitoses per 5 mm² or per 50 high-power fields [HPFs]) or high (>5 mitoses per 5 mm² or per 50 HPFs).

Stage groupings were as follows:

Gastric GISTs:

Small intestine and colorectal GISTs:

The gastric staging system also applies to omental GISTs, while small intestinal staging also applies to oesophageal, colorectal, mesenteric, and peritoneal GISTs.

Statistical analyses were performed using IBM SPSS Statistics version 25.0 (IBM Corp., Armonk, NY, USA) and R software version 4.4.2. Descriptive statistics for sex, age, race, primary site, tumour size, mitotic count, N stage, modified NIH criteria, and TNM stage were summarized as frequencies and percentages (%). Univariate survival analysis was conducted using the Kaplan–Meier method and log-rank tests. Multivariate analysis was performed using the Cox proportional-hazards regression model to identify independent risk factors associated with CSS. The proportional-hazards assumption for the Cox models was tested using Schoenfeld residuals and found to be satisfied. Kaplan–Meier analysis was also used to estimate 5-year CSS rates across different subgroups defined according to the TNM staging system and the modified NIH criteria in the three cohorts, with pairwise comparisons between adjacent subgroups evaluated via log-rank tests. Univariate Cox regression analysis was used to calculate hazard ratios (HRs) for each subgroup within the TNM staging system and modified NIH criteria. The prognostic performance of the TNM staging system and the modified NIH criteria was compared using the area under the receiver operating characteristic curve (AUC), concordance index (C-index), Akaike information criterion, and Bayesian information criterion. The C-index and its 95% confidence intervals (CIs) were calculated using 1,000 bootstrap resamples. This bootstrap procedure served as internal validation to assess the robustness of model discrimination. Differences in C-index between models were tested using the DeLong method. A P-value < 0.05 was considered statistically significant.

We retrospectively analysed the data of 3,034 patients with GISTs who had not undergone preoperative neoadjuvant therapy, according to the SEER database. Of these, 2,106 patients had gastric GISTs and 928 had small intestine and colorectal GISTs. All the patients underwent tumour-related surgery and had complete data that could be used for analysis with the modified NIH criteria and TNM staging system. Univariate analysis showed that age, primary site, tumour size, mitotic count, N stage, NIH criteria, and TNM stage were prognostic factors for patients with GISTs. The proportion of patients in each subgroup, and 5-year CSS rates are shown in Table 1 . This study protocol was approved by the Ethics Committee of Fujian Cancer Hospital (Fuzhou, China). The SEER database is publicly available and provides data of de-identified cases. As the analysis used anonymous clinical data, the requirement of written informed consent was waived.

Variables with P<0.05 in the univariate analyses were included in the multivariate Cox regression analysis. We discovered that age (HR, 1.914; 95% CI, 1.401–2.615; P<0.001) and TNM stage (HR, 1.540; 95% CI, 1.161–2.043; P=0.003) were independent prognostic factors for CSS in patients with GISTs. Primary site, tumor size, mitotic count, N stage, and NIH criteria were not independent prognostic factors ( Figure 2 ). The lack of significance of tumor size and mitotic count may be explained by their incorporation into the TNM staging system, which likely led to collinearity and loss of independent predictive power in the multivariate model.

The median follow-up duration for all patients was 58 months (range, 1–119 months). Of the 3034 patients included in our analysis, 198 died of GISTs. The 3- and 5-year CSS rates were 95.8% and 93.5%, respectively. Kaplan–Meier survival analysis showed significant differences in cancer-specific survival rates among very low-, low-, intermediate-, and high-risk groups in the modified NIH criteria ( Figure 3A ). The 5-year CSS rates of these risk groups were 95.4%, 97.9%, 95.6%, and 87.8%, respectively. Significant differences were also seen in CSS between stages I, II, and IIIA in the TNM staging system. However, no significant differences were found between stages IIIA, IIIB, and IV ( Figure 3B ). The 5-year CSS rates for stages I, II, IIIA, IIIB, and IV were 96.6%, 94.2%, 85.7%, 81.1%, and 85.7%, respectively. According to the modified NIH criteria, the HRs of low-risk, intermediate-risk, and high-risk groups were 0.349 (95% CI, 0.185–0.656), 0.869 (95% CI, 0.478–1.580), and 2.737 (95% CI, 1.745–4.293), respectively, compared with the very low-risk group ( Figure 3C ). According to the TNM staging system, the HRs for stages II, IIIA, IIIB, and IV were 2.149 (95% CI, 1.426–3.240), 5.542 (95% CI, 3.709–8.280), 6.365 (95% CI, 4.386–9.235), and 8.042 (95% CI, 3.835–16.862), respectively, compared with stage I ( Figure 3D ).

Of the 2,106 patients with gastric GISTs, 121 died of GISTs, and the 3-and 5-year CSS rates were 96.1% and 94.0%, respectively. Kaplan–Meier survival analysis showed a significant difference in CSS between the very low-risk, low-risk, intermediate-risk, and high-risk groups in the modified NIH criteria ( Figure 4A ); however, no significant difference was found in CSS between the very low-risk and intermediate-risk groups. The 5-year CSS rates for very low, low, intermediate, and high-risk groups were 95.6%, 98.0%, 95.6%, and 85.5%, respectively. Kaplan–Meier survival analysis showed a significant difference in CSS between stages II and IIIA in the TNM staging system, but no difference was observed between any other adjacent pair of stages ( Figure 4B ). The 5-year CSS rates for stages IA, IB, II, IIIA, IIIB, and IV were 97.1%, 94.9%, 93.2%, 83.3%, 76.6%, and 81.9%, respectively. According to the modified NIH criteria, the HRs of low-risk, intermediate-risk, and high-risk groups were 0.352 (95% CI, 0.168–0.738), 0.928 (95% CI, 0.489–1.759), and 3.303 (95% CI, 1.946–5.608), respectively, compared with the very low-risk group ( Figure 4C ). According to the TNM staging system, the HRs for stages IB, II, IIIA, IIIB, and IV were 1.683 (95% CI, 0.934–3.031), 2.824 (95% CI, 1.633–4.881), 7.485 (95% CI, 4.408–12.709), 8.025 (95% CI, 4.540–14.187), and 12.395 (95% CI, 4.796–32.029), respectively, compared with stage IA ( Figure 4D ).

Of the 928 patients with small intestine and colorectal GISTs, 77 died of GISTs, and the 3-and 5-year CSS rates were 95.1% and 92.4%, respectively. In the small intestine and colorectal GISTs cohort, only two cases were classified as intermediate risk based on the modified NIH standard analysis. Kaplan–Meier survival analysis showed a significant difference in CSS only between the low-risk and high-risk groups in the modified NIH criteria ( Figure 5A ), while there was no significant difference in CSS between the very low-risk and low-risk groups. The 5-year CSS rates in the very low-, low-, and high-risk groups were 94.3%, 97.8%, and 89.9%, respectively. Kaplan–Meier survival analysis showed that, in the TNM staging system, significant differences in CSS occurred between stages II and IIIA and stages IIIB and IV but not between the other stages ( Figure 5B ). The 5-year CSS rates for stages I, II, IIIA, IIIB, and IV were 96.9%, 95.3%, 88.9%, 83.4%, and 89.2%, respectively. According to the modified NIH criteria, the HRs of the low-risk and high-risk groups compared to the very low-risk group were 0.318 (95% CI, 0.092–1.101) and 2.002, (95% CI, 0.807–4.966), respectively ( Figure 5C ). According to the TNM staging system, the HRs for stages II, IIIA, IIIB, and IV to stage I were 1.973 (95% CI, 0.886–4.393), 4.800 (95% CI, 2.154–10.696), 6.577 (95% CI, 3.248–13.321), and 6.280 (95% CI, 1.725–22.861), respectively ( Figure 5D ).

In the entire GISTs cohort, the prognostic value of the TNM staging system (AUC, 0.715; 95% CI, 0.675–0.755) was similar to that of the modified NIH criteria (AUC, 0.711; 95% CI, 0.677–0.744) ( Figure 6A ). In the gastric GISTs cohort, the TNM staging system (AUC, 0.721; 95% CI, 0.670–0.772) was also comparable to the modified NIH criteria (AUC, 0.726; 95% CI, 0.680–0.771) in terms of prognostic value ( Figure 6B ). In the small intestine and colorectal GISTs cohort, the prognostic discriminatory ability of the TNM staging system (AUC, 0.713; 95% CI, 0.654–0.771) was better than that of the modified NIH criteria (AUC, 0.645; 95% CI, 0.591–0.699) ( Figure 6C ). In all cohorts, the C-index of the NIH criteria (all GISTs: 0.70 [95% CI, 0.67–0.73]; gastric GISTs: 0.72 [95% CI, 0.68–0.76]) was similar to that of the TNM stage (all GISTs: 0.69 [95% CI, 0.65–0.73]; gastric GISTs: 0.69 [95% CI, 0.64–0.74]), while the C-index of the NIH criteria (small intestine and colorectal GISTs: 0.62 [95% CI, 0.58–0.66]) was significantly lower than that of the TNM stage (small intestine and colorectal GISTs: 0.69 [95% CI, 0.62–0.74]) (P<0.001). In the small intestine and colorectal GIST cohort, both the Akaike information criterion and Bayesian information criterion of the TNM stage were lower than those of the NIH criteria ( Table 2 ).