Ogawa Shimpei, Hida Jin-Ichi, Ike Hideyuki, Kinugasa Tetsushi, Ota Mitsuyoshi, Shinto Eiji, Itabashi Michio, Okamoto Takahiro, Yamamoto Masakazu, Sugihara Kenichi, Watanabe Toshiaki
Department of Surgery, Institute of Gastroenterology, Tokyo Women's Medical University School of Medicine, 8-1, Kawada-cho, Shinjuku-ku, Tokyo, 162-8666, Japan.
Department of Surgery II, Tokyo Women's Medical University School of Medicine, 8-1, Kawada-cho, Shinjuku-ku, Tokyo, 162-8666, Japan.
Int J Colorectal Dis. 2017 Oct;32(10):1479-1487. doi: 10.1007/s00384-017-2874-9. Epub 2017 Jul 31.
The goal of the study was to examine prediction of lateral pelvic lymph node (LPLN) metastasis from lower rectal cancer using a logistic model including risk factors for LPLN metastasis and magnetic resonance imaging (MRI) clinical LPLN (cLPLN) status, compared to prediction based on MRI alone.
The subjects were 272 patients with lower rectal cancer who underwent MRI prior to mesorectal excision combined with LPLN dissection (LPLD) at six institutes. No patients received neoadjuvant therapy. Prediction models for right and left pathological LPLN (pLPLN) metastasis were developed using cLPLN status, histopathological grade, and perirectal lymph node (PRLN) status. For evaluation, data for patients with left LPLD were substituted into the right-side equation and vice versa.
Left LPLN metastasis was predicted using the right-side model with accuracy of 86.5%, sensitivity 56.4%, specificity 92.7%, positive predictive value (PPV) 61.1%, and negative predictive value (NPV) 91.2%, while these data using MRI cLPLN status alone were 80.4, 76.9, 81.2, 45.5, and 94.5%, respectively. Similarly, right LPLN metastasis was predicted using the left-side equation with accuracy of 83.8%, sensitivity 57.8%, specificity 90.4%, PPV 60.5%, and NPV 89.4%, and the equivalent data using MRI alone were 78.4, 68.9, 80.8, 47.7, and 91.1%, respectively. The AUCs for the right- and left-side equations were significantly higher than the equivalent AUCs for MRI cLPLN status alone.
A logistic model including risk factors for LPLN metastasis and MRI findings had significantly better performance for prediction of LPLN metastasis compared with a model based on MRI findings alone.
本研究的目的是使用一个逻辑模型来检验低位直肠癌侧方盆腔淋巴结（LPLN）转移的预测情况，该模型纳入了LPLN转移的危险因素以及磁共振成像（MRI）临床LPLN（cLPLN）状态，并与仅基于MRI的预测进行比较。
研究对象为272例低位直肠癌患者，这些患者在6家机构接受了直肠系膜切除联合LPLN清扫术（LPLD）之前进行了MRI检查。所有患者均未接受新辅助治疗。利用cLPLN状态、组织病理学分级和直肠周围淋巴结（PRLN）状态建立了右侧和左侧病理性LPLN（pLPLN）转移的预测模型。为了进行评估，将左侧LPLD患者的数据代入右侧方程，反之亦然。
使用右侧模型预测左侧LPLN转移的准确率为86.5%，灵敏度为56.4%，特异度为92.7%，阳性预测值（PPV）为61.1%，阴性预测值（NPV）为91.2%，而仅使用MRI cLPLN状态时这些数据分别为80.4%、76.9%、81.2%、45.5%和94.5%。同样，使用左侧方程预测右侧LPLN转移的准确率为83.8%，灵敏度为57.8%，特异度为90.4%，PPV为60.5%，NPV为89.4%，仅使用MRI时的相应数据分别为78.4%、68.9%、80.8%、47.7%和91.1%。右侧和左侧方程的曲线下面积（AUC）显著高于仅基于MRI cLPLN状态的等效AUC。
与仅基于MRI结果的模型相比，纳入LPLN转移危险因素和MRI结果的逻辑模型在预测LPLN转移方面具有显著更好的性能。
