3.1 Baseline characteristics of 345 patients with AC
The characteristics of the 345 AC patients are presented in Table S1. The median age was 46.4 years, and most patients were in stage I (89.6%). Of the 345 patients, there were 96 (27.8%) with pattern A, 90 (26.1%) with pattern B and 159 (46.1%) with pattern C. The median follow-up period was 102 (36-168) months, during which 27 (7.8%) patients died and 32 (9.3%) patients experienced recurrence. In terms of the recurrence site, there were 16 patients with initial recurrence in the local region and 16 patients with recurrence in the distant region.
3.2 Prognostic value of the Silva classification in AC patients
The clinicopathological factors of the 345 patients were compared according to different Silva patterns (Table 1). There was no positive parametrial involvement, LVSI or perineural invasion (PNI) in patients with pattern A. Compared to patients with pattern A, patients with pattern B or C had higher FIGO stages (p<0.001), larger tumor sizes (p<0.001), and deeper stromal invasion (p<0.001) and showed a higher frequency of positive LNs (p<0.001), positive surgical margins (p=0.043), positive parametrial involvement (p<0.001), positive LVSI (p<0.001) and positive PNI (p=0.002). In addition, patients with pattern B or C were more likely to receive postoperative adjuvant therapy (p<0.001) and undergo laparotomy (p=0.033) than those with pattern A.
We then compared the survival outcome according to different Silva patterns. Only 1 (1%) patient with pattern A experienced recurrence, while 3 (3.3%) with pattern B and 28 (17.6%) with pattern C experienced recurrence. In terms of death, 1 (1%) pattern A patient, 3 (3.3%) pattern B patients and 23 (14.5%) pattern C patients died during the follow-up period. The results of Kaplan-Meier survival analysis showed that patients with pattern C had the worst prognosis. The 3-year RFS rates for patterns A, B and C were 100%, 96.2% and 83.1%, respectively (p<0.001), while the 3-year overall survival rates for patterns A, B and C were 100%, 95.8% and 86.7% (p<0.001) (Fig 1). The univariate Cox analysis also showed the same findings (RFS: HR, 9.06 [3.177, 25.836], p<0.001; OS: HR, 7.406 [2.56, 21.421], p<0.001), which again indicated that as the Silva grade increased, the patient's prognosis worsened.
3.3 Significant intermediate-risk factors in AC patients
In the cohort of 345 AC patients, univariate Cox analysis showed that in addition to Silva pattern (RFS, p<0.001; OS, p=0.002), the following 13 factors were significantly associated with RFS and OS (Table 2): age (RFS, p=0.009; OS, p=0.047), FIGO stage (RFS, p<0.001; OS, p=0.003), adjuvant therapy (RFS, p=0.012; OS, p=0.026), surgical approach (RFS, p<0.019; OS, p=0.008), transfusion (RFS, p=0.015; OS, p=0.005), LN metastasis (RFS, p<0.001; OS, p<0.001), LN metastasis site (RFS, p<0.001; OS, p<0.001), surgical margin (RFS, p=0.001; OS, p<0.001), parametrial involvement (RFS, p<0.001; OS, p<0.001), tumor size (RFS, p=0.001; OS, p=0.002), LVSI (RFS, p<0.001; OS, p<0.001), DSI (RFS, p<0.001; OS, p<0.001) and PNI (RFS, p=0.005; OS, p=0.001). Among these 14 factors, LN, surgical margin and parametrial involvement were identified as high-risk factors according to NCCN guidelines. Thus, we further excluded patients who had any one of the above three high-risk factors and focused on the remaining 254 patients as the intermediate-risk group in the following analysis.
We then divided the remaining risk factors based on different cutoff values. For example, age was categorized as 40, 50 or 60 years old. Silva pattern was categorized as Silva B+C or Silva C. Tumor size was categorized as 2, 2.5, 3, 3.5, 4, 4.5 or 5 cm. DSI was categorized to >1/3 DSI or >2/3 DSI. LVSI was categorized as mild LVSI or substantial LVSI (Table S2). Univariate Cox analysis showed that 5 variables had a p-value of less than 0.05: Silva C, ≥3 cm, ≥3.5 cm, DSI>2/3 and >mild LVSI. In addition, considering the important prognostic value of Silva pattern, we also added Silva B+C, which had a critical p-value of 0.057, as a potential intermediate-risk variable. Thus, we focused on these 6 variables (Silva C, Silva B+C, ≥3 cm, ≥3.5 cm, DSI>2/3 and >mild LVSI) (Table 3), which are also considered as 4 intermediate-risk factors (Silva pattern, diameter, DSI and LVSI), for the following study.
3.4 Establishment of a novel Silva-based model specific for intermediate-risk AC patients
To explore a unique Silva-based model specific for intermediate-risk AC patients to better guide postoperative adjuvant therapy, we established various models using different combinations of the above 4 intermediate-risk factors (also considered the above 6 variables: Silva C, Silva B+C, ≥3 cm, ≥3.5 cm, DSI>2/3 and >mild LVSI), which exhibited a significant association with RFS, and compared the model performance with that of the traditional Sedlis criteria. First, we established 12 models via various combinations of the four intermediate-risk factors, including the above 6 variables (Table 4). Interestingly, most four-factor models had superior performance to the traditional Sedlis criteria. Of note, among the 12 four-factor models (Fig 2), model 6 (any 3 of the 4 factors: Silva C, ≥3 cm, DSI >2/3, and >mild LVSI) showed the best recurrence prediction performance, the highest chi-square score (24.262), and the highest area under the curve (AUC) value (0.761), and was significant in univariate Cox analysis (HR, 10.792 [3.31, 35.19], p<0.001).
Moreover, given the best performance of model 6 (any 3 of the 4 factors: Silva C, ≥3 cm, DSI >2/3, and >mild LVSI), we wanted to further determine which combination in model 6 showed the best predictive ability for recurrence. To accomplish this, we compared the 4 combinations in model 6 based on any 3 of the 4 factors. The results showed that when patients met the three factors of Silva C, ≥3 cm, and DSI >2/3, the model exhibited the best discriminating ability for recurrence (Table S3) (Fig S4).
We also established 30 three-factor models (Table S5, Fig S6) and 16 two-factor models (Table S7, Fig S8) by replacing one or two of the three intermediate risk factors (tumor size, LVSI, and DSI) with Silva patern. The results showed that, despite being inferior to the four-factor models, most three-factor or two-factor models were better than the Sedlis criteria. This again demonstrated the important recurrence prediction value of Silva pattern.