The baseline characteristics of the whole study cohort are presented in Table 1. In general, a total of 2,705 lung ASC patients were identified in the SEER database. For these patients, the median age at diagnosis was 69 years (IQR: 61-76). A larger proportion of patients were aged above 65 years (1,776, 65.7%), male (1,437, 53.1%), married (1,590, 58.8%), and white (2,248, 83.1%). The majority of tumours were located in the upper lobe (1,651, 61.0%) and on the right side (1,564, 57.8%). Most patients were diagnosed with histological grade III (64.7%), followed by grade II (31.7%), grade IV (2.3%), and grade I (1.3%). The distribution of AJCC stage was as follows: 42.2% had stage I, 12.3% had stage II, 23.4% had stage III, and 22.1% had stage IV. A total of 67.2% of patients received surgical treatment.
The median follow-up of the whole study cohort was 21 months (IQR: 8-52). In total, 1,895 (70.1%) patients died throughout the whole follow-up period, of whom 1,535 (81.0%) died due to lung cancer and 362 (19.0%) died due to non-lung cancer causes. The 3-year and 5-year cumulative incidences of LC-SM and OCSM by different clinicopathological characteristics are displayed in Table 1, and the corresponding CIF curves are presented in Figure 2. Overall, the 3-year and 5-year LC-SM rates were 49.6% (CI: 47.7%-51.5%) and 55.8% (CI: 53.8%-57.8%), respectively, while the 3-year and 5-year OCSM rates were 8.2% (CI: 7.1%-9.2%) and 11.8% (CI: 10.5%-13.1%).
Subsequently, both univariate and multivariate competing risk models were adopted to evaluate the LC-SM of lung ASC patients. In univariate analysis, male sex, unmarried status, black race, main bronchus, advanced TNM stage, advanced histological grade, and surgical treatment were related to significantly higher incidences of LC-SM, whereas there were no significant differences for age and tumour laterality (Figure 3). In multivariate analysis, age, sex, surgery, T stage, N stage, and M stage were independent predictive factors for LC-SM (Table 2). In detail, increasing age was associated with an increased probability of LCSM. Male sex was related to a significantly higher likelihood of LCSM (1.26, CI: 1.10-1.43), while surgery was related to a significantly lower likelihood of LC-SM (0.45, CI: 0.37-0.53). Compared with patients with T1, advanced T-stage patients were more likely to face LCSM, with SHRs of 1.44 (1.21-1.72), 2.24 (1.72-2.92), and 1.99 (1.59-2.49) for T2, T3, and T4 patients, respectively. A similar phenomenon was observed among advanced N-stage patients compared with N0 patients, with SHRs of 1.52 (1.26-1.84), 1.57 (1.32-1.87), and 1.51 (1.12-2.03) for N1, N2, and N3 patients, respectively.
A nomogram on the basis of the competing risk models was developed to calculate the 3-year and 5-year cumulative LC-SM probabilities (Figure 4). For each patient, first locate the values of different variables on the corresponding rows and then draw vertical lines pointing to the “Points” row to obtain corresponding scores. For instance, for a male patient, by drawing a vertical line straight up to the “Point” row, we would obtain approximately 28 points. Similarly, this process is performed for the other variables. By adding up these scores, a total score can be obtained and is located on the “Total Points” row. Subsequently, a vertical line can be drawn straight down to acquire the 3-year or 5-year cumulative death probabilities. For example, if the total score was 100, the corresponding 3-year and 5-year probabilities of LC-SM would be approximately 30% and 36%, respectively.
The calibration curves accompanied by C-indexes are displayed in Figure 5. As shown in Figure 5, the calibration curves are close to the 45-degree diagonal line, indicating that the developed nomogram is well calibrated (good agreement between the observed mortality probability and the predicted mortality probability). Additionally, the 3-year and 5-year C-indexes for the nomogram predicting the probabilities of LC-SM were 0.83 (CI, 0.78-0.87) and 0.82 (CI, 0.73-0.90) for the training cohort, and 0.79 (CI, 0.75-0.84) and 0.79 (CI, 0.71-0.88) for the validation cohort, respectively, which indicated superb model discrimination. The ten-fold cross validation C-indexes are shown in Table 3. The adjusted 3-year and 5-year C-indexes were 0.81 (CI, 0.80-0.83) and 0.81 (CI, 0.80-0.83), respectively. Overall, the 3-year or 5-year C-indexes of the cross validation were almost equal to those of the training set or validation set, which indicated robust model performance.
The outcomes of DCA are shown in Figure 6A, which shows that the clinical net benefit gained from the competing risk model was higher than that in the hypothetical non-screening or all-screening scenarios, when the threshold probabilities were within the range of 0.24-0.89 and 0.25-0.91 for 3-year and 5-year LCSM, respectively. According to the tertile values (117.1 and 180.5) of the nomogram-based scores derived from the training cohort, the patients were categorized into high-risk, medium-risk, and low-risk groups in the training cohort and validation cohort. As displayed in Figure 6B-6C, the high-risk group had the highest probabilities of LC-SM, followed by the medium-risk group and the low-risk group in the training cohort and validation cohort (both P<0.0001). Therefore, when using the nomogram as a predictive tool, clinicians could successfully discriminate among different risk groups.