Patient characteristics
According to the screening criteria, a total of 6,770 T1N0Mx patients were in the study predicting distant metastasis and 428 T1N0 patients with DM were in the study predicting CSS. In all T1N0 patients, most proportions were found in 60-79 years old, white, male, right colon, tumor size ≤ 3cm, moderately differentiated, adenocarcinoma, CEA negative and absence of DM. And in patients with DM, most cases were found to be associated with 60-79 years old, white, male, right colon, 3 < tumor size ≤ 5cm, moderately differentiated, adenocarcinoma, CEA positive, absence of surgery and presence of chemotherapy. The baseline characteristics of patients after 3: 1 ratio randomly stratified were calculated in Tables 1 and 2.
Construction and validation of nomogram to predict DM probability
In order to further explore the risk factors for DM in T1N0 patients, univariate and multivariate logistic regression analyses were performed to determine the independent risk factors for DM step by step. In univariate analysis, the candidate predictors for the model were age, race, sex, tumor size, tumor site, grade, histology and CEA. All the predictors except for sex, tumor site and histology were significantly different between subgroups, which were then further analyzed by multivariate logistic regression model. And the results indicated that age (OR = 0.647, 95%CI = 0.243-1.719 for 40-59 years old, P = 0.382; OR =0.332, 95%CI = 0.126-0.874 for 60-79 years old, P = 0.026; OR = 0.202, 95%CI = 0.073-0.557, P = 0.002 for ≥ 80 years old; using < 40 years old as the reference), tumor size (OR = 0.092, 95%CI = 0.064-0.134 for 3cm < tumor size ≤ 5 cm, P < 0.001; OR = 0.654, 95%CI = 0.446-0.958 for tumor size > 5cm, P = 0.029; using tumor size ≤ 3 cm as the reference), grade (OR = 2.323, 95%CI = 1.511-3.574 for moderately differentiated, P < 0.001; OR = 5.686, 95%CI = 3.236-9.99 for poorly differentiated, P < 0.001; OR = 7.159, 95%CI = 2.462-20.821 for undifferentiated, P < 0.001; using well differentiated as the reference), CEA level (OR = 0.054, 95%CI = 0.040-0.074 for CEA negative, P < 0.001, using CEA positive as the reference) were independent risk factors in predicting the occurrence of DM (Table 3).
Based on the independent risk factors in the multivariate analysis, we construct a nomogram to predict DM in T1N0Mx patients (Figure 1). The AUCs for development cohort and validation cohort were 0.901 (95%CI = 0.879-0.922) and 0.899 (95%CI =0.865-0.940), respectively (Figure 2). The calibration curves (development cohort: S: p = 0.712; validation cohort: S: p = 0.681) showed the relatively satisfactory prediction accuracy of the nomogram (Figure 3). In addition, the DCA curve also indicated good clinical practicability in both cohorts (Figure 4).
Construction and validation of nomogram to predict CSS in T1N0 patients with DM
After analyzing the risk factors of DM in T1N0Mx patients, we also explored CSS in DM patients using Kaplan-Meier method and Cox regression model. Univariate analysis revealed that sex and grade were not important factors for CSS in DM patients. The risk factors in the univariate analysis were further analyzed by Cox multivariate regression model. And the results indicated that age (HR = 0.587, 95%CI = 0.272-11.264 for < 40 years old, P = 0.174; HR =0.439, 95%CI = 0.296-0.651 for 40-59 years old, P < 0.001; HR = 0.603, 95%CI = 0.418-0.871 for 60-79 years old, P = 0.007; using ≥ 80 years old as the reference), histology (HR = 1.487, 95%CI = 0.794-2.784 for mucinous, P = 0.215; HR = 4.682, 95%CI = 1.672-13.115 for other, P = 0.003; using adenocarcinoma as the reference), surgery (HR = 3.450, 95%CI = 2.313-5.145 for no, P < 0.001; using yes as the reference), chemotherapy (HR = 2.032, 95%CI = 1.512-2731 for no, P < 0.001; using yes as the reference), CEA level (HR = 0.454, 95%CI = 0.293-0.703 for CEA negative, P < 0.001, using CEA positive as the reference) were independent prognosticators in predicting CSS with DM patients (Table4).
We constructed a nomogram to predict 1-, 2- and 3-year survival in DM patients with T1N0 colon cancer, incorporating age, histology, CEA, surgery and chemotherapy (Figure 5). The C-indices of the development and validation cohort were 0.718 (95%CI=0.639-0.737) and 0.712 (95%CI=0.681-0.743). The area under the ROC curves of the CSS nomogram were shown in Figure 6. The AUCs of the nomogram at 1-, 2-, and 3-year were 0.763 (95%CI=0.744-0.782), 0.794 (95%CI=0.775-0.813), and 0.822 (95%CI=0.803-0.841) for the development cohort, and 0.785 (95%CI=0.754-0.816), 0.748 (95%CI=0.717-0.779) and 0.896 (95%CI=0.865-0.927) for the validation cohort. The calibration plot showed a satisfactory predictive accuracy between 1-, 2-, and 3-year predicted CSS and observed CSS in both cohorts (Figure 7). In addition, clinical impact curves were drawn based on DCA to help us more intuitively understand the significant value of the nomogram model (Figure 8).
Using the nomogram derived scores, all DM patients were classified into two subgroup low-risk (risk score ≤ 171) and high-risk groups (risk score > 171) by the X-tile program (Figure 9). And we found there were significant differences in Kaplan-Meier curves between the high risk and low risk groups in development cohort (P < 0.001) and validation cohort (P < 0.001) (Figure 10).