# Integrating machine learning for advanced analysis of bioelectrical impedance parameters in children with nephrotic syndrome

## TL;DR

This study explores using machine learning with bioelectrical impedance analysis to help diagnose nephrotic syndrome in children, showing promising accuracy but limited specificity.

## Contribution

The novel use of machine learning to analyze bioelectrical impedance data for diagnosing nephrotic syndrome in children.

## Key findings

- The ML model achieved an AUC of 0.84 for identifying nephrotic syndrome in children.
- Key features included resistance, impedance, and phase angle at 50 kHz, along with age, height, and sex.
- The model had high sensitivity (92%) but low specificity (22%), limiting its clinical utility.

## Abstract

Nephrotic syndrome (NS) in children, characterised kidney-related protein leakage and peripheral oedema, remains challenging to assess. Bioelectrical impedance analysis (BIA) provides indices of body water (oedema), and analysis with machine learning (ML) may improve clinical care. We tested an ML model to identify NS in children compared with healthy children.

This cross-sectional study included children with active NS in the acute phase (aNS group) recruited from the Department of Paediatrics and Adolescent Medicine, Aarhus University Hospital, Denmark. Anonymised MF-BIA data from frequencies between 5 and 1000 kHz were analysed using the web-based ML platform JustAddDataBio (JADBio)® to identify potential biomarkers for improved diagnosis.

Eight children with aNS and 38 healthy children of similar ages were included. The ML software employed ridge logistic regression with the penalty hyperparameter lambda = 0.001 and a selected threshold of 0.81 by JADBio. The best model achieved an area under the curve (AUC) of 0.84 [95% confidence interval (CI): 0.72;0.94]. The software selected the following features: height, age, resistance at 50 kHz, impedance at 50 kHz, the characteristic frequency, phase angle at 50 kHz, and sex. The model demonstrated a statistically significant true positive classification rate of 0.92 (92%) [CI: 0.88;0.96] and a specificity of 0.22 (22%) [CI: 0.08;0.36].

Applying ML-supported evaluation of BIA affirmed diagnostics. However, low specificity limits clinical applications. A larger population of patients and inclusion of additional biomarkers may be needed to develop a more acceptable model.