196304 early stage BC patients from 2010 to 2016 were obtained for the final analysis and divided into two groups at a ratio of nine to one: training cohort (N=176671) and validation cohort (N=19630). The baseline characteristics of the two groups were summarized in Table 1 and there was no significant difference between them (p>0.05). The median age at diagnosis was 60 years (IQRs, 51-70 years). Among the whole population, nearly a half of the patients (47.5%) were younger than 60 years old. Moderate differentiation (Grade II) (44.4%) accounted for the highest proportion, followed by poor differentiation (Grade III-IV) (30.3%), and well differentiation (Grade I) (25.3%). Small tumors were prevailing in early-stage BC patients. 62.7% of the tumor sizes were smaller than 2 centimeter (cm) and only 2.8% of tumors were larger than 5cm. Most patients (75.1%) were categorized as Luminal A subtype (hormone receptor [HR]+/ human Epidermal growth factor receptor-2 [HER2]-), followed by Triple Negative subtype (HR-/HER2-) (10.8%), Luminal B subtype (HR+/HER2+) (10.2%) and HER2 Enriched subtype (HR-/HER2+) (3.9%).
The median follow-up time was 41 months (IQRs, 24-60 months). 12417/196304 (6.3%) patients died: 5628/12417 (45.3%) as a result of cancer-specific deaths and 6789/12417 (54.7%) as a result of other-cause-specific deaths. The 3-, 4-, and 5-year overall survival rates, cancer-specific mortalities and other-cause-specific mortalities which were stratified by age, grade, tumor size, subtype and surgery were listed in Table 2.
On multivariable Cox regression (Table 3), older age (p<0.001), poorer differentiation (Grade II vs. Grade I; hazard ratio [HR], 1.135; 95% confidence interval [CI], 1.075-1.199; p<0.001; Grade III-IV vs. Grade I; HR, 1.703; 95% CI, 1.602-1.809; p<0.001), larger tumor size (p<0.001), triple negative subtype (vs. luminal B subtype; HR, 1.859; 95% CI, 1.724-2.004; p<0.001), HER2 enriched subtype (vs. luminal B subtype; HR, 1.167; 95% CI, 1.048-1.300; p=0.005),and absence of surgery (vs. surgery; HR, 3.428; 95% CI, 3.277-3.641; p<0.001) were significantly associated with a poorer OS. For CSM and OCSM, Fine and Gray’s competing risk analysis was used and the following factors were validated (Table 3): older age, poorer differentiation, larger tumor size, triple negative subtype, HER2 enriched subtype and absence of surgery for CSM; and older age, larger tumor size and absence of surgery for OCSM. Age was a strong predictive factor and more obvious on OCSM (p<0.001). The OCSM were significantly elevated in patients with age increasing: compared with the young patients, the elderly patients carried higher competing risks (60-65 years vs. <60 years; subdistribution hazard ratio [SHR], 2.561, 95% CI, 2.283-2.873; p<0.001; 66-70 years vs. <60 years; SHR, 4.182, 95% CI, 3.742-4.674; p<0.001; 71-75 years vs. <60 years; SHR, 6.727, 95% CI, 6.049-7.482; p<0.001; 76-80 years vs. <60 years; SHR, 12.008, 95% CI, 10.855-13.284; p<0.001; 81-85 years vs. <60 years; SHR, 22.167, 95% CI, 20.112-24.433; p<0.001; >85 years vs. <60 years; SHR, 39.263, 95% CI, 35.634-43.262; p<0.001). The impact of surgery was more prominent on CSM. Increasing tumor size was correlated with OS and CSM and slightly correlated with OCSM. Notably, grade and subtype differences were more predominant for CSM than for OCSM. The Kaplan-Meier curves for OS and cumulative incidence curves for CSM and OCSM were presented in Figure 2.
Five validated variables were incorporated to develop the prognostic nomogram: age, grade, tumor size, subtype and surgery of primary site (Figure 3). Thus, the probability of 3-, 4-, and 5-year OS, CSS and OCSS could be predicted by summing up the scores of each selected variable (higher total points, worse prognosis), helping to identify patients with high risk of cancer or non-cancer causes of death. The nomogram demonstrated considerably strong discriminative ability, with a concordance index (C-index) of 0.801 (95% CI, 0.795-0.806; p=0.003) for the OS model (using Cox regression analysis), 0.830 (95% CI, 0.824-0.836; p=0.003) for the CSM and 0.806 (95% CI, 0.798-0.814; p=0.004) for the OCSM model (using Fine and Gray’s competing risk analysis). Calibration plots presented high conformance between the nomogram-predicted probabilities and the observed probabilities in both the training cohort and the validation cohort (Figure 4).
The discriminatory capacity of the nomogram was evaluated by calculating the AUC values (Figure 5). The AUC values for predicting 3-, 4-, and 5-year OS rates were 80.2%, 79.5% and 78.7%, respectively. As for the prediction of the 3-, 4-, and 5-year CSM, the AUC values were 83.0%, 81.7% and 80.3%, respectively. While the AUC values were 81.3%, 80.8% and 81.7%, respectively for the 3-, 4-, and 5-year OCSM.
According to the C-index and AUC values, the model predicting CSM and OCSM which using Fine and Gray’s competing risk analysis was more precision than that of predicting OS.
Furthermore, to further evaluate the discrimination of the model, the validation cohort was stratified into three groups by the predicted probability of the group calculated from the nomogram: low risk group, middle risk group, and high risk group. Patients of the high risk group had substantially lower OS rates and higher CSM and OCSM compared with patients of low risk group and middle risk group (p<0.001) (Figure 6).