Wan Elizabeth R, Iancu Daniela, Ashton Emma, Siew Keith, Mohidin Barian, Sung Chih-Chien, Nagano China, Bockenhauer Detlef, Lin Shih-Hua, Nozu Kandai, Walsh Stephen B
Department of Renal Medicine, University College London, London, UK.
North East Thames Regional Genetics Service Laboratories, Great Ormond Street Hospital for Children National Health Service Foundation Trust, London, UK.
Kidney Int Rep. 2022 Dec 24;8(3):556-565. doi: 10.1016/j.ekir.2022.12.008. eCollection 2023 Mar.
Clinically distinguishing patients with the inherited salt-losing tubulopathies (SLTs), Gitelman or Bartter syndrome (GS or BS) from other causes of hypokalemia (LK) patients is difficult, and genotyping is costly. We decided to identify clinical characteristics that differentiate SLTs from LK.
A total of 66 hypokalemic patients with possible SLTs were recruited to a prospective observational cohort study at the University College London Renal Tubular Clinic, London. All patients were genotyped for pathogenic variants in genes which cause SLTs; 39 patients had pathogenic variants in genes causing SLTs. We obtained similar data sets from cohorts in Taipei and Kobe, as follows: the combined data set comprised 419 patients; 291 had genetically confirmed SLT. London and Taipei data sets were combined to train machine learning (ML) algorithms, which were then tested on the Kobe data set.
Single biochemical variables (e.g., plasma renin) were significantly, but inconsistently, different between SLTs and LK in all cohorts. A decision table algorithm using serum bicarbonate and urinary sodium excretion (FE) achieved a classification accuracy of 74%. This was superior to all the single biochemical variables identified previously.
ML algorithms can differentiate true SLT in the context of a specialist clinic with some accuracy. However, based on routine biochemistry, the accuracy is insufficient to make genotyping redundant.
临床上很难将患有遗传性失盐性肾小管病（SLTs）、吉特曼综合征或巴特综合征（GS或BS）的患者与其他低钾血症（LK）病因的患者区分开来，而且基因分型成本高昂。我们决定确定能够将SLTs与LK区分开来的临床特征。
伦敦大学学院肾小管诊所对总共66例可能患有SLTs的低钾血症患者进行了一项前瞻性观察队列研究。所有患者都对导致SLTs的基因中的致病变异进行了基因分型；39例患者在导致SLTs的基因中有致病变异。我们从台北和神户的队列中获得了类似的数据集，具体如下：合并后的数据集包括419例患者；291例经基因证实患有SLT。将伦敦和台北的数据集合并以训练机器学习（ML）算法，然后在神户数据集上进行测试。
在所有队列中，单一生化变量（如血浆肾素）在SLTs和LK之间存在显著但不一致的差异。使用血清碳酸氢盐和尿钠排泄（FE）的决策表算法实现了74%的分类准确率。这优于先前确定的所有单一生化变量。
在专科诊所的背景下，ML算法能够以一定的准确率区分真正的SLT。然而，基于常规生化检查，其准确率不足以使基因分型变得多余。
