基于LASSO-Logistic回归构建三叉神经痛射频术后复发列线图预测模型

Construction of a Nomogram Prediction Model for Recurrence after Radiofrequency Thermocoagulation for Trigeminal Neuralgia Based on LASSO-Logistic Regression

摘要:

目的：分析三叉神经痛(TN)患者射频热凝术(RFT)后疼痛复发的危险因素,并构建列线图预测模型。方法：回顾性纳入2020年3月至2022年1月在南京大学医学院附属鼓楼医院疼痛科住院接受RFT治疗的243例TN患者的临床资料,按7：3的比例随机分配至训练组(n=170)与测试组(n=73)。根据术后2年随访时疼痛是否复发,将患者分为复发组和无复发组(对照组),采用LASSO-Logistic回归筛选疼痛复发的独立危险因素,并构建列线图风险预测模型。结果：72例(29.6%)患者术后疼痛复发。通过LASSO回归变量筛选及Logistic回归分析表明,年龄、疼痛类型、术前药物治疗有效性是术后2年疼痛复发的影响因素(P<0.05),并基于此构建了列线图预测模型。该模型在训练组中的受试者工作特征曲线下面积(AUC)为0.757(95%置信区间：0.676-0.838)；在测试组中为0.772(95%置信区间：0.634-0.911)。校准曲线斜率趋近于1,Hosmer-Lemeshow拟合优度检验显示p> 0.05,表明模型拟合度较好。决策曲线分析(DCA)表明,当阈值概率为0.10-0.95时,患者临床净获益率较高。结论：为早期识别与筛查三叉神经射频手术后的复发高风险患者,构建并验证了一个列线图预测模型。该模型纳入年龄、疼痛类型以及药物治疗反应性等作为预测指标。

Abstract:

Objective: To analyze the risk factors for pain recurrence in patients with trigeminal neuralgia (TN) after radiofrequency thermocoagulation (RFT) and construct a nomogram prediction model. Methods: The clinical data of 243 TN patients who were hospitalized in the Department of Pain, Nanjing Drum Tower Hospital and underwent RFT treatment from March 2020 to January 2022 were retrospectively included. Patients were randomly divided into the training group (n=170) and the test group (n=73) at a 7:3 ratio. Based on pain recurrence status during a 2-year postoperative follow-up, patients were categorized into recurrence and non-recurrence (control) groups. LASSO-Logistic regression was employed to screen the independent risk factors for pain recurrence, and a nomogram prediction model was constructed. Results: Pain recurrence occurred in 72 patients (29.6%) after the operation. LASSO regression variable screening and logistic analysis revealed that age, type of pain, and the effectiveness of preoperative drug treatment were significant factors of pain recurrence 2 years after the operation (P < 0.05), and a nomogram prediction model was developed based on these factors. The model demonstrated an area under the curve (AUC) of 0.757 (95% CI: 0.676-0.838) in the training group and 0.772 (95% CI: 0.634-0.911) in the test group. The slope of the calibration curve approached 1, and the Hosmer-Lemeshow test showed p > 0.05, indicating a good model fit. Decision curve analysis (DCA) demonstrated that patients had a relatively high clinical net benefit rate when the threshold probability ranged from 0.10 to 0.95. Conclusion: A nomogram prediction model was constructed and verified for the early identification and screening of patients at high risk of recurrence after trigeminal nerve radiofrequency surgery. This model incorporates age, pain type, and the responsiveness to drug treatment as predictive indicators.