De Langhe Sofie, De Meerleer Gert, De Ruyck Kim, Ost Piet, Fonteyne Valérie, De Neve Wilfried, Thierens Hubert
Department of Basic Medical Sciences, Ghent University, Belgium.
Department of Radiotherapy, Ghent University Hospital, Belgium.
Radiother Oncol. 2014 Jul;112(1):95-9. doi: 10.1016/j.radonc.2014.04.005. Epub 2014 Jun 17.
To develop predictive models for late radiation-induced hematuria and nocturia allowing a patient individualized estimation of pre-treatment risk.
We studied 262 PCa patients treated with curative intensity modulated radiotherapy to the intact prostate or prostate bed. A total of 372 variables were used for prediction modeling, among which 343 genetic variations. Toxicity was scored using an in-house developed toxicity scale. Predictor selection is achieved by the EMLasso procedure, a penalized logistic regression method with an EM algorithm handling missing data and crossvalidation avoiding overfit. Model performance was expressed by the area under the curve (AUC) and by sensitivity and specificity.
Variables of the model predicting late hematuria (36/262) are bladder volume receiving ⩾75 Gy, prostatic transurethral resection and four polymorphisms. (AUC = 0.80, sensitivity = 83.3%, specificity = 61.5%). The AUC drops to 0.67 when the genetic markers are left out. The model that predicts for late nocturia (29/262) contains the minimal clinical target volume (CTV) dose, the CTV volume and three polymorphisms (AUC = 0.76, sensitivity = 75.9%, specify = 67.4%). This model is a better predictor for nocturia compared to the nongenetic model (AUC of 0.60).
We were able to develop models that predict for the occurrence of late radiation-induced hematuria and nocturia, including genetic factors which might improve the prediction of late urinary toxicity.
建立晚期放射性血尿和夜尿症的预测模型，以便对患者治疗前风险进行个体化评估。
我们研究了262例接受前列腺完整组织或前列腺床根治性调强放疗的前列腺癌患者。共使用372个变量进行预测建模，其中343个为基因变异。使用内部开发的毒性量表对毒性进行评分。通过EMLasso程序进行预测变量选择，这是一种带有EM算法处理缺失数据和交叉验证以避免过拟合的惩罚逻辑回归方法。模型性能用曲线下面积（AUC）、敏感性和特异性表示。
预测晚期血尿（36/262）的模型变量包括接受≥75 Gy照射的膀胱体积、经尿道前列腺切除术和四种多态性（AUC = 0.80，敏感性 = 83.3%，特异性 = 61.5%）。若不考虑基因标记，AUC降至0.67。预测晚期夜尿症（29/262）的模型包含最小临床靶体积（CTV）剂量、CTV体积和三种多态性（AUC = 0.76，敏感性 = 75.9%，特异性 = 67.4%）。与非基因模型（AUC为0.60）相比，该模型对夜尿症的预测效果更好。
我们能够建立预测晚期放射性血尿和夜尿症发生的模型，包括可能改善晚期泌尿毒性预测的基因因素。
