结直肠恶性肿瘤术后早期肝转移的多因素分析及建立风险预测模型

Multi-Factor Analysis and Risk Prediction Model of Early Postoperative Liver Metastasis of Colorectal Malignancies

摘要: 目的：探究结直肠恶性肿瘤行根治术后发生早期肝转移的影响因素，并且建立早期肝转移的风险预测模型。方法：回顾性分析黑龙江省第二肿瘤医院5年间(2018年1月~2023年1月) 230例罹患结直肠恶性肿瘤且行结直肠手术患者的一般临床资料，包括围手术期、术后病理、门诊及再次住院随访资料。根据是否发生早期肝转移分为早期转移组(n = 30例)和未转移组(n = 200例)，早期肝转移定为术后6个月内。统计学应用SPSS 23.0软件进行数据分析，单因素及多因素采用COX回归分析影响结直肠恶性肿瘤术后早期肝转移发生的独立危险因素，应用独立危险因素构建列线图模型预测患者早期肝转移风险，并绘制感受性曲线(SC, sensitivity curve)评价线性列线图的预测能力。校正曲线验证风险模型的预测效能。结果：230例患者早期肝转移发生率为13%。单因素COX分析早期转移组患者术前癌胚抗原(CEA) ≥ 5 μg/L、区域淋巴结清扫数量、手术方法、低分化肿瘤、肿瘤最大横截面 ≥ 5 cm、T分期(T_3~4)、N分期(N_1~2)比例高于未转移组，差异具有统计学意义(P < 0.05)；多因素COX回归分析显示，肿瘤最大横截面积 ≥ 5 cm、T分期(T_3~4)、N分期(N_1~2)、低分化肿瘤为影响结直肠恶性肿瘤术后早期肝转移发生的独立危险因素(P < 0.05)。构建结直肠恶性肿瘤术后早期肝转移线性列线图风险预测模型，肿瘤最大横截面积 ≥ 5 cm、T分期(T_3~4)、N分期(N_1~2)、低分化肿瘤的相应得分总和对应早期肝转移风险值。绘制感受性曲线(SC)评价线性列线图，结果显示曲线下面积(area of under curve, AUC)为0.953 (95%CI: 0.9086~0.9964)。校正曲线显示，该线性列线图风险模型的预测概率和实际结直肠恶性肿瘤术后早期肝转移概率具有良好的一致性(P > 0.05)。结论：肿瘤最大横截面积 ≥ 5cm、T分期(T_3~4)、N分期(N_1~2)、低分化肿瘤是结直肠恶性肿瘤术后早期肝转移发生的独立风险因素，且构建的线性列线图预测模型对结直肠恶性肿瘤术后早期肝转移的预测效能评价良好。

Abstract: Objective: To explore the influencing factors of early liver metastasis after radical resection of colorectal malignancies and establish a risk prediction model for early liver metastasis. Methods: General clinical data, including perioperative, postoperative pathology, outpatient and re-hospitalization follow-up data, of 230 patients with colorectal malignant tumors who underwent colorectal surgery in The Second Cancer Hospital of Heilongjiang Province in the past 5 years (January 2018 to January 2023) were retrospectively analyzed. The patients were divided into early metastatic group (n = 30 cases) and non-metastatic group (n = 200 cases) according to whether early liver metastasis occurred. Early liver metastasis was defined as 6 months after surgery. SPSS 23.0 statistical software was applied for data analysis, COX regression was used to analyze the independent risk factors affecting the occurrence of early postoperative liver metastasis of colorectal malignancy for both single factor and multiple factors, independent risk factor for constructing nomogram model is applied to forecast early liver metastasis in patients with risk, and the sensitivity curve (SC) was drawn to evaluate the predictive power of linear nomogram. The correction curve verifies the predictive efficiency of the risk model. Results: The incidence of early liver metastasis was 13% of 230 patients. The proportion of preoperative carcinoembryonic antigen (CEA) ≥ 5 μg/L, number of regional lymph node dissection, surgical method, poorly differentiated tumor, maximum tumor cross-section ≥ 5 cm, T stage (T_3~4) and N stage (N_1~2) in early metastatic group was higher than that in non-metastatic group by univariate COX analysis, and the differences were statistically significant (P < 0.05). Multivariate COX regression analysis showed that the maximum cross-sectional area of tumor ≥5cm, T stage (T_3~4), N stage (N_1~2) and poorly differentiated tumor were independent risk factors for the occurrence of early liver metastasis after colorectal malignancy (P < 0.05). A linear nematic risk prediction model for early liver metastasis after colorectal malignancy was established. The maximum cross-sectional area of tumor ≥ 5 cm, the sum of the corresponding scores of T stage (T_3~4), N stage (N_1~2) and poorly differentiated tumor corresponded to the risk value of early liver metastasis. The sensitivity curve (SC) was drawn to evaluate the linear nomogram, and the results showed that the area of under curve (AUC) was 0.953 (95%CI: 0.9086~0.9964). The calibration curve showed that the prediction probability of the linear nomogram risk model was in good agreement with the actual probability of early liver metastasis after colorectal malignancy (P > 0.05). Conclusion: The maximum cross-sectional area of tumor ≥ 5 cm, T stage (T_3~4), N stage (N_1~2), and poorly differentiated tumor are independent risk factors for early liver metastasis after colorectal malignancy, and the constructed linear nomogram prediction model has good predictive efficacy for early liver metastasis after colorectal malignancy.