Imaizumi Takahiro, Yokota Takashi, Funakoshi Kouta, Yasuda Kazushi, Hattori Akiko, Morohashi Akemi, Kusakabe Tatsumi, Shojima Masumi, Nagamine Sayoko, Nakano Toshiaki, Huang Yong, Morinaga Hiroshi, Ohta Miki, Nagashima Satomi, Inoue Ryusuke, Nakamura Naoki, Ota Hideki, Maruyama Tatsuya, Gobara Hideo, Endoh Akira, Ando Masahiko, Shiratori Yoshimune, Maruyama Shoichi
Department of Nephrology, Nagoya University Graduate School of Medicine, 65 Tsurumai-cho, Showa-ku, Nagoya, Aichi, 464-8550, Japan.
Department of Advanced Medicine, Nagoya University Hospital, Nagoya, Japan.
Clin Exp Nephrol. 2025 May;29(5):650-661. doi: 10.1007/s10157-024-02614-3. Epub 2025 Jan 6.
Identifying patients on dialysis among those with an estimated glomerular filtration rate (eGFR) < 15 mL/min/1.73 m remains challenging. To facilitate clinical research in advanced chronic kidney disease (CKD) using electronic health records, we aimed to develop algorithms to identify dialysis patients using laboratory data obtained in routine practice.
We collected clinical data of patients with an eGFR < 15 mL/min/1.73 m from six clinical research core hospitals across Japan: four hospitals for the derivation cohort and two for the validation cohort. The candidate factors for the classification models were identified using logistic regression with stepwise backward selection. To ensure transplant patients were not included in the non-dialysis population, we excluded individuals with the disease code Z94.0.
We collected data from 1142 patients, with 640 (56%) currently undergoing hemodialysis or peritoneal dialysis (PD), including 426 of 763 patients in the derivation cohort and 214 of 379 patients in the validation cohort. The prescription of PD solutions perfectly identified patients undergoing dialysis. After excluding patients prescribed PD solutions, seven laboratory parameters were included in the algorithm. The areas under the receiver operation characteristic curve were 0.95 and 0.98 and the positive and negative predictive values were 90.9% and 91.4% in the derivation cohort and 96.2% and 94.6% in the validation cohort, respectively. The calibrations were almost linear.
We identified patients on dialysis among those with an eGFR < 15 ml/min/1.73 m. This study paves the way for database research in nephrology, especially for patients with non-dialysis-dependent advanced CKD.
在估算肾小球滤过率（eGFR）<15ml/min/1.73m²的患者中识别透析患者仍然具有挑战性。为了利用电子健康记录促进晚期慢性肾脏病（CKD）的临床研究，我们旨在开发算法，通过常规实践中获得的实验室数据识别透析患者。
我们收集了来自日本六家临床研究核心医院的eGFR<15ml/min/1.73m²患者的临床数据：四家医院用于推导队列，两家用于验证队列。使用逐步向后选择的逻辑回归确定分类模型的候选因素。为确保移植患者不被纳入非透析人群，我们排除了疾病编码为Z94.0的个体。
我们收集了1142例患者的数据，其中640例（56%）目前正在接受血液透析或腹膜透析（PD），包括推导队列中763例患者中的426例和验证队列中379例患者中的214例。PD溶液的处方能够完美识别正在接受透析的患者。排除开具PD溶液的患者后，算法纳入了七个实验室参数。推导队列中受试者操作特征曲线下面积分别为0.95和0.98，阳性和阴性预测值分别为90.9%和91.4%；验证队列中分别为96.2%和94.6%。校准几乎呈线性。
我们在eGFR<15ml/min/1.73m²的患者中识别出了透析患者。本研究为肾脏病学的数据库研究铺平了道路，尤其是对于非透析依赖的晚期CKD患者。
