Patient characteristics
During the study period, a total of 301 patients with CLMs underwent RFA, from which 101 patients were excluded due to insufficient data (n=41) or repeated RFA (n=60) (Figure 1). The final study cohort comprised 200 hospitalized patients with 410 CLMs. The average age of these patients was 61.0 ±10.5 years (range, 27–89 years), with males accounting for 63.0% (Table 1). The most common primary tumor location was the sigmoid colon (65/200), followed by the rectum (57/200). Approximately half of the patients presented with synchronous CLMs (106/200), and 72 (36.0%) patients had extrahepatic metastases, most frequently in the lung. Comorbidities were present in 101 patients (50.5%), with hypertension being the most prevalent. Pre-RFA laboratory examinations revealed elevated carcinoembryonic antigen levels (> 30 µg/mL) in 45 patients (22.5%) and elevated carbohydrate antigen 199 levels (> 37 U/mL) in 55 patients (27.5%). Fifty-five patients (27.5%) had undergone prior liver resection, and 172 (86.0%) had received chemotherapy before RFA. Major complications related to RFA occurred in 22 cases (11.0%), primarily involving pleural effusion and bleeding. In our study, iRFA was noted in 67 CLMs (16.3%). Among the patients, 246 tumors from 120 patients were randomly assigned to the development cohort, with an iRFA incidence of 37 tumors (15.0%). The internal validation cohort included 164 tumors (from 80 patients) with an iRFA incidence of 30 tumors (18.8%). No significant differences in baseline characteristics were observed between the two cohorts (Table 1).
Univariate and multivariate analysis
In the development cohort of 246 CLMs, the average minimal ablative margin was 5.8 ± 2.5 mm, and the average ∆CT value was 21.8 ± 12.3 Hu. Ninety-one tumors (37.0%) were located near critical organs or structures, 32 (13.0%) CLMs were in perivascular locations, 53 (21.5%) CLMs had a diameter ≥ 20 mm, and 129 (52.4%) CLMs were subcapsular metastases. Univariate analysis demonstrated significant differences in perivascular tumor location, tumor size (≥20 mm), and minimal ablative margin between the iRFA and cRFA groups (all P < 0.001) (Table 2).
To ensure important predictor variables were not overlooked, subcapsular tumor location (P=0.199) and ∆CT value (P=0.154) were also included alongside the aforementioned three independent variables as potential risk factors in the multivariate analysis. Results from the MLR model (Table 2) revealed that perivascular tumor location (odds ratio [OR] = 3.164, 95% CI: 1.201–8.333, P=0.02), tumor size ≥ 20 mm (OR = 5.639, 95% CI: 2.547–12.487, P<0.001), and minimal ablative margins (OR = 0.607, 95% CI: 0.498–0.753, P<0.001) independently predicted iRFA. The omnibus test indicated that the current model significantly outperformed the null model (χ2 = 77.974, P<0.001). Additionally, the Hosmer–Lemeshow test demonstrated good model fit (χ2 = 4.449, P=0.815). The predictive model's performance in both the development and internal validation cohorts is illustrated in Figure 2, with corresponding AUC values of 0.884 (95% CI: 0.819–0.950, P<0.001) and 0.857 (95% CI: 0.785–0.930, P<0.001), respectively.
Sensitivity analysis
The goodness-of-fit of the GLM is presented in Supplementary Table 2. According to the GLM results, a minimal ablative margin (OR = 0.620, 95% CI: 0.479–0.801, P<0.001) was identified as a protective factor against iRFA, while perivascular tumor location (OR = 3.214, 95% CI: 1.860–8.543, P=0.025) and tumor size ≥ 20 mm (OR = 5.095, 95% CI: 1.815–14.300, P=0.002) were significant risk factors for iRFA. Sensitivity analyses indicated that the β regression coefficient and OR of the GLM remained stable, confirming the robustness of our findings compared to those from the MLR.
OS and NIHM
Follow-up was completed for 164 hospitalized patients (338 CLMs), with 36 patients lost to follow-up. The median follow-up period was 22 months. Among the patients, 50 (30.5%) died, and the cumulative OS rates at 1, 3, and 5 years after RFA were 92.2%, 59.7%, and 41.2%, respectively. Before RFA, there were no significant differences in extrahepatic metastasis (P = 0.468) between patients in the cRFA and iRFA groups. The estimated median survival time, calculated using the Kaplan–Meier method, was 45 months (95% CI: 29.348–60.652) for the entire cohort, 74 months (95% CI: 48.212–99.788) for the cRFA group, and 35 months (95% CI: 22.954–47.046) for the iRFA group. The log-rank test results indicated a significant difference in cumulative OS between the two groups (χ2 = 12.269, P<0.001) (Fig. 3A), with a hazard ratio of 2.54 (95% CI: 1.341–4.814).
During the follow-up period, a total of 98 cases (59.8%) of patients were observed to have experienced NIHM, and appropriate anti-tumor treatment strategies were implemented based on their condition, including repeated RFA (n=62), transcatheter arterial chemoembolization (n=9), stereotactic body radiotherapy (n=2), surgical resection (n=8), and other comprehensive treatments (n=17). The cumulative incidence of NIHM at 1 and 3 years after RFA was estimated at 44.2% and 65.0%, respectively. All NIHM occurrences were identified within 48 months, with five patients showing no NIHM during the follow-up period exceeding 5 years after cRFA. The median time without NIHM, estimated using Kaplan–Meier analysis, was 16 months (95% CI: 7.868–24.132) overall. In the cRFA group, NIHM was observed in 58 (50.9%) patients, with a median NIHM-free period of 31 months (95% CI: 15.818–46.182). In contrast, NIHM was detected in 40 patients in the iRFA group (80%), with a median time without NIHM of 5 months (95% CI: 3.359–6.641). The difference between the two groups was significant (χ2 = 26.811, P<0.001) (Fig. 3B), with a hazard ratio of 2.679 (95% CI: 1.64–4.377).