Patientsbaseline characteristics and treatment patterns
A total of 1379 GIST patientsmet the eligibility criteriain this process.Patients' demographic and clinical characteristics across treatment categories aresummarized in Table 1. Of the analyzed patients, 53.0% received surgery alone, followed by chemotherapy and surgery (32.2%), chemotherapy alone (10.2%), and no therapy (4.6%). The majority of cases were white (67.4%), female (52.5%), insured (97.0%), and married (57.6%) individuals. Over half of the patients were identified with early-stage AJCC tumors (50.5%) and lower mitotic index (73.2%). Approximately 70.2% of patients were located in the stomach, with lower prevalence of small intestine (26.3%) and other sites (3.5%) including colorectum, esophagus, peritoneum or retroperitoneum, and appendix. In the whole study set, they were far less likely to happen regional lymph node metastasis (3.6%) and distant metastasis (13.3%). As to treatment patterns, surgery alone was still the widest choice whatever the age, race, gender, marital and insurance status changed, while patients with T4, N1, or M1stage tumors more often underwent surgery plus chemotherapy, as well as tumors originated from other sites or with higher mitotic index and grade. In addition, characteristics of the patients in the training cohort (n=951) and validation cohort (n=428) are listed in Table S1, and there were no significant differences observed across the two sets (P>0.05).
Polytomous logistic regression
The results of the polytomous logistic regression are shown in Table 2. Compared to older counterparts, patients aged 65-74 years had higher possibility to receive therapy(aOR: 0.415, 95% CI: 0.235-0.734),and no differences were detected for any of the treatment categories in the light ofgender, race, or insured status. We can also notice that patients with tumor location in the stomach or small intestine were connectedwith greater odds of receiving surgeryfor alltreatment categories except surgery plus chemotherapy when compared with tumors located at other sites. Moreover, for all treatment categories, patients without distant metastases were more likely to undergo surgery alone than those with M1 stage tumors (all p<0.05). Using T4 stage as a reference, patients presented with T2 stage were prone to choose surgery alone, while T1, T3, and T1 stage patients, respectively in the combined category and chemotherapy alone category, still had the same choice.
Treatment patterns on survival in different groups
Survival curves (Figure S1) and corresponding adjusted hazard ratios (Table 3) were presented to describe the results of survival analysisstratified by age acrosstreatment categories, and P values for paired comparison of treatment methods are listed in Table S2. Patients aged ≥75 years who underwent surgery alone had a median survival of 69 months, chemotherapy alone 51 months, surgery plus chemotherapy 60 months, and no therapy 51 months. While among patients aged 65-74 years, the median OS was not reached (NR), 49, 83, and NR, respectively. The adjusted HRs further revealed that chemotherapy alone had a remarkably higher risk of death than those receiving surgery alone (HR: 2.773, 95% CI: 1.451-5.299, P=0.002), while patients receiving no therapyled to a poor survival compared with the surgery alone group for patients more than 75 years old (HR: 2.075, 95% CI: 1.136-3.793, P=0.018).
Univariate and multivariate analysis for OS
Univariate and multivariate analyses were performed to identify predictors of survival among the 951 patients in the training set. As shown in Table 4, the univariate analysis demonstrated that except for the race and T stage, all other variables were remarkably associated with OS (P<0.05). Then, factors with P<0.10 were further entered into the multivariable model, and we can find that age, gender, marital status, insurance status, location, T stage, M stage, mitotic index, and treatment were recognized as significant independent prognostic factors for elderly GIST patient survival (P<0.05).
Construction and validation of the predictive model
Figure 2 displays the nomogram to predict 3- and 5-year OS. The C-indexes for OS prediction in the training set and validation set were 0.771 (95% CI: 0.734-0.808) and 0.761 (95% CI: 0.710-0.812), respectively, indicating more accurate ability in prognosis predicting. Of importance, the calibration plots revealed that the predicted 3- and 5-year OS were in excellent agreement with actual survival in both the training cohort and validation cohort (Figure 3). In addition, the integrated AUC for nomogram showed more powerful efficacy of discrimination in survival prediction compared with that of AJCC TNM 8th edition (nomogram: 0.775 vs 0.638 P<0.001 and 0.748 vs 0.670 P=0.032, respectively for training and validation cohorts, Figure 4). We also conducted a DCA, and results illustrated that applying this nomogram to predict 3-year and 5-year OS would be better than the newest TNM stage (Figure S2), suggesting greater net benefits (i.e. higher clinical applicability) in this predictive model. Furthermore, we obtained the optimal cut-off value determined by the X-tile program for the two cohorts. In the training set, patients were divided into low-risk group (n=428, median OS: not reached), middle-risk group (n=355, median OS: 81 months), and high-risk group (n=168, median OS: 38 months), which was similar in the validation set (median OS: not reached, 66 months, and 40 months, respectively). The Kaplan-Meier curves in both cohorts showed significant differences (P<0.05, Figure5), indicating that patients who presented with lower risk had a stronger correlation with the reduction of overall death.
Development of webserver
Additionally, in order to make it easier for clinicians and researchers to predict the survival probability of individual patients, we developed a more accurate web calculator, which was available at https://sizhao.shinyapps.io/elderly_gist/ (Figure S3). By inputting relevant clinical features, the online version of our nomogram can generate corresponding figures and tables, avoiding errors caused by manual measurement.