Integrating machine learning for advanced analysis of bioelectrical impedance parameters in children with nephrotic syndrome: phase angle, impedance ratio, and cell membrane capacitance

Integrating machine learning for advanced analysis of bioelectrical impedance parameters in children with nephrotic syndrome: phase angle, impedance ratio, and cell membrane capacitance | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Integrating machine learning for advanced analysis of bioelectrical impedance parameters in children with nephrotic syndrome: phase angle, impedance ratio, and cell membrane capacitance Josephine Reinert Quist, Leigh C Ward, Lars Jødal, René Frydensbjerg Andersen, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7197037/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Nephrotic syndrome (NS) in children, entailing kidney-related protein leakage and peripheral oedema, remains difficult to assess. Bioelectrical impedance analysis (BIA) provides several body composition measures, and integration of machine learning (ML) may improve clinical care. We tested an ML model to identify NS in children, compared with healthy children. Methods This was a cross-sectional study, conducted on children with active NS in the acute phase (aNS group) included from the Department of Paediatrics and Adolescent Medicine, Aarhus University Hospital, Denmark. Anonymized MF-BIA data from frequences between 5-1000 kHz were added to the JustAddDataBio (JADBio)®, a web-based ML platform for analysing potential biomarkers to diagnose. Results Eight children with aNS and 38 age-matched healthy children were included. The ML software employed a ridge logistic regression with the penalty hyperparameter lambda = 0.001, with a selected threshold of 0.81 by JADBio, and the area under the curve (AUC) was 0.84 [95% confidence interval (CI): 0.72;0.94] as the best model. 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 had a statistically significant true positive classification of a healthy child of 0.92 (92%) [CI: 0.88;0.96], and a specificity of 0.22 (22%) [CI: 0.08;0.36]. Conclusion Applying an ML-supported evaluation of BIA improved diagnostics. A low specificity limits the clinical application. To obtain a more acceptable model, a larger population of patients and the inclusion of more biomarkers may be needed. Bioelectrical impedance electric capacitance machine learning nephrotic syndrome children Figures Figure 1 Figure 2 Introduction Nephrotic syndrome (NS) is a relatively rare disease, incidence 2–7 per 100,000 children [ 1 ] characterised by excessive renal protein loss, leading to hypoalbuminemia and oedema [ 2 ]. It is associated with extended hospitalisation [ 3 ]. Conventionally, the degree of oedema is estimated from clinical evaluation [ 4 ] and body weight measurement, and a weight increase is assumed to primarily indicate accumulation of extracellular water (ECW). This approach has limitations, especially over extended periods when weight changes due to factors other than changes in ECW become significant [ 5 ]. In the search for a practical, cost-effective, and non-invasive technique to evaluate routine oedema in children with NS, bioelectrical impedance analysis (BIA) emerges as a promising tool [ 6 , 7 ]. BIA offers characteristics ideally suited for clinical routine use, such as safety, speed, simplicity, and portability [ 8 ]. BIA operates on the principle that the flow of alternating current through the body varies in different tissues. Tissues rich in water and electrolytes exhibit high conductivity and low impedance (Z, ohm), while fat and bone, with low conductivity, show high impedance. Impedance comprises two components: resistance (R) and capacitive reactance (X C ). The body’s resistance is inversely related to its water content (R ~ 1/body water), which is the foundation for estimating body water volumes with BIA [ 9 ]. Many BIA devices measure only at a single frequency (SF-BIA) or a few, multiple frequencies (MF-BIA). The bioimpedance spectroscopy (BIS) technique measures a broad spectrum of frequencies and employs the Cole model for data analysis, where coordinate points (R, X C ) are plotted and fitted with a semicircular arc to describe the data. From BIA measurements, clinical body parameters such as ECW, total body water (TBW), and fat-free mass and fat mass can be estimated using prediction models derived from healthy individuals. Importantly, these estimates rely on assumptions that can vary among study populations; the prediction equations typically vary among BIA devices from different manufacturers, and only some manufacturers openly reveal the specifics of the prediction equations used by their devices. The use of the measured electrical properties is increasingly being considered as they are openly described and are independent of a particular manufacturer’s BIA device. Such parameters are called “raw” or “derived raw” impedance parameters and act as proxies for their physiological correlates. We use the term “derived raw” to reflect that these parameters are not directly measured data but are the results from basic calculations made upon the measured data. Four important derived raw impedance parameters are the phase angle (PhA), impedance ratio (IR, ratio of impedances measured at different frequencies, typically a high and low frequency), cell membrane capacitance (C m ) and resistance. PhA, IR, and C m are composite parameters derived from various raw data, whereas R directly reflects the opposition to the flow of electric current through the body’s tissues, influenced predominantly by the body’s total water content and tissue composition. Each of these may be valuable for assessing the clinical state of NS patients` body water distribution. PhA is associated with tissue hydration, cell membrane quality and cellular mass [ 10 – 12 ], although its exact biological significance remains a question of ongoing research [ 13 ]. IR is another promising parameter that potentially indicates oedema or overall health and reflects fluid distribution between TBW and the ECW. Lastly, C m is a parameter that promisingly provides insights into cell membrane function, size, hydration status and overall tissue composition [ 14 ]. Although ongoing research is being conducted into the potential information content of these parameters, especially PhA in adults, there is a scarcity of studies focusing on paediatric populations [ 15 ]. Integrating machine learning (ML) software’s capabilities includes efficient processing of large datasets, precise pattern recognition, and creating predictive models [ 16 ]. This increases clinicians’ ability to extract intricate insights from impedance measurements, enhancing their understanding of parameters like PhA, IR, and C m in NS. ML’s notable speed and minimising the mean prediction error may ensure timely and informed decision-making, facilitating more effective patient care and treatment strategies. The aim of this clinical pilot study was to assess the feasibility of employing ML software with the bioimpedance outcomes PhA, IR, C m , and R, with the perspective to identify and differentiate children with aNS effectively, to contribute to further characterization of the body composition. Materials and Methods Study design This study was conducted as a single-centre cross-sectional study and was reported according to the STREAM-URO network guidelines [ 17 ]. Study subjects Children with active NS in the acute phase (aNS patient group) in the age range of 3 to 10 years old, were included from the Department of Paediatrics and Adolescent Medicine, Aarhus University Hospital, Denmark. Before treatment with prednisolone and diuretics was initiated, blood samples, blood pressures and impedance measurements were collected from the patient group. The 38 healthy controls included were selected from a previously established reference cohort of children, chosen to span a representative range of age and sex, but were not individually matched to the patients with NS. Patient data were drawn from a previously described cohort [ 18 ]. BIA device and parameters Whole-body impedance was measured with electrodes placed on the wrist and ankle in pairs. A Xitron 4200, HYDRA BIS device (Xitron Technologies, San Diego, CA, USA) was used and was tested weekly with an electronic verification module (TS4201) according to the manufacturer’s instructions. The BIS device performed measurements of the electrical parameters, including impedance (Z), resistance (R), and capacitive reactance (X C ), all measured in ohms (Ω) at fifty different frequencies in the range from 5 to 1000 kHz. Impedance (Z) is the length of the vector to an individual point, \(\:Z=\sqrt{{R}^{2}+{X}_{C}^{2}}\) . PhA is the angle between the vector and the horizontal axis. For measurements at a given frequency, the phase angle is calculated as PhA = tan −1 (X C /R) ∙ 180°/π. The result depends on the frequency, but PhA is often reported at 50 kHz frequency [ 2 , 3 ]. This requires only SF-BIA (the single frequency being 50 kHz). When full-spectrum data are available (BIS), PhA is sometimes also considered for the characteristic frequency f c , corresponding to the frequency of maximum reactance. Impedance ratio, IR, compares impedance at a high frequency with the impedance at a low frequency. Measurements at high frequencies are dominated by TBW, while measurements at low frequencies are dominated by ECW; typically, the frequencies used for IR are 200 and 5 kHz [ 19 ]. It is calculated as IR 200/5 = R 200 /R 5 and has no unit [ 3 ]. Determination of IR requires at least two measurements, i.e., MF-BIA or BIS. The cellular membrane capacitance, C m , was calculated from the Cole model based on the spectrum of frequencies [ 19 ]. The unit of C m is farad, F. Determination of C m requires knowledge of a spectrum of frequencies, i.e., BIS not SF-BIA or MF-BIA. Data acquisition Before conducting impedance measurements, trained personnel recorded weight and height in duplicate. Weight, measured on digital scales with light clothing, was recorded to the nearest 0.1 kg, while height, measured without shoes using a stadiometer, was recorded to the nearest 0.5 cm. The BIA measurements adhered to standardised procedures, with participants refraining from intense physical exercise four hours before the study but not required to fast. Details of the protocol can be found in a previous publication [ 6 ] and in the study from which the subject was initially enrolled [ 17 ]. Data were extracted using the ImpediMed SFB7 Multi-Frequency Analysis software (Bioimp Version 5.4.0.3, Brisbane QLD, Australia) from measurements of 50 kHz and at the characteristic frequency ( f c ) where capacitive reactance is at its maximum value. These measurements were used to calculate specific parameters within the software. In this study, the parameters analysed at 50 kHz were resistance (R 50 ), capacitive reactance (X C50 ), impedance (Z 50 ), and phase angle (PhA 50 ). The parameters analysed at the subject-specific characteristic frequency were resistance (R f c ), capacitive reactance (X C f c ), impedance (Z f c ) and phase angle (PhA f c ). In addition, C m was also analysed. Each subject was measured three times in one setting, and the mean values from the three measurements were used as individual data points for input to the software. Finally, the clinical data sex, height, weight, and age, were included. Machine learning software Just Add Data Bio (JADBio)® (3154 Glendale Blvd Los Angeles, CA 90039 − 1830, USA) [ 20 ] is a web-based machine learning platform for analysing potential biomarkers to diagnose and estimate the prognosis of a disease. It is a fully automated machine learning software developed for a small sample population with many data points for each subject. It was designed for predictive modelling and exploring the possible application of machine learning methods with minimal entry-level skills typically required for developing such models. When the data have been fed into the JADBio software, it starts locating the features (e.g., variables that help determine the investigated outcome), and the software returns the developed model and information related to the model. It comprises several algorithms for the standard phases when making a machine learning model: data preprocessing, data transformation, data imputation, feature selection, and predictive modelling. The label of interest is to predict disease activity. The label “false” predicts a child without aNS, whereas “true” predicts a child with aNS. Data were divided into training and testing cohorts. In data pre-processing, mean and mode imputation of missing values is performed. Zero variance features are removed. Continuous features are standardised to zero mean and standard deviation (SD) of 1. The imputation of missing data on a variable replaces a missing value computed from an estimate of the distribution of this variable [ 21 ]. If imputation is not possible, missing data and the reasons behind it will be disclosed. Feature selection algorithms, least absolute shrinkage and selection operator (LASSO) regularised regression, and statistical equivalent signatures were investigated to determine the minimal size predictive feature subset. Predictive modelling algorithms tested are decision trees, ridge logistic regression, random forests, support vector machines, cox regression and random survival forests [ 20 ]. Removal of features tested for a machine learning model is based on clinical intuition and discussion in the research group. This was chosen based on assumptions that increase the chance for causal inference [ 22 ]. Performance estimations are based on generalised cross-validation, which determines the hyperparameters, where the latter are external settings for a model, influencing its performance (e.g., learning rates and regularisation strengths). Cross-validation is a technique where the sample size is proportionally divided into K folds of equal size. K refers to the number of groups a given data sample will split into where R-repeated cross-validation indicates that the procedure runs R times with different partitions (refer to the divisions or subsets into which the dataset is divided) to folds to reduce the variance in the estimation. The final model is estimated with a bootstrap bias correction to adjust p-values and to remove bias due to the multiple tries. Bootstrapping, a statistical technique involving resampling with replacement from observed data, estimates uncertainty in a sample statistic. In the final model estimation, a bootstrap bias correction is applied to adjust p-values, mitigate bias resulting from multiple attempts, and increase the risk for individual error [ 23 ]. The working model is presented by sensitivity, specificity, accuracy, receiver operating characteristic (ROC) curves for correct classification of children with or without aNS, false positive rate, and true positive rate. This model can be changed with different thresholds for the ROC. Predictive performance is defined as the average accuracy across the test folds in a repeated 9-fold cross-validation. Individual conditional expectation (ICE) plots visualise the effect of a given variable for a specific model. The final analysis is evaluated by the area under the curve (AUC) and CI for the best model. Results Patient characteristics Data were collected from 8 children with aNS, seven boys and one girl, and 38, 23 boys and 15 girls, age-matched healthy controls (HC group) from a previous study reported by Brantlov et al. [ 18 ]. 5 of the 8 children with aNS experienced remission. Determined by no protein from a urine sample on three consecutive days. These 5 children out of the original 8 were re-measured and sub-grouped in a remission group (NSr). Results are presented as mean ± standard deviation (SD), after test for normality, using Q-Q plots and the Shapiro-Wilk test. Patients with aNS had an age of 6.7 ± 3.1 years, the subgroup patients with NSr had an age of 7.7 ± 3.8 years, and the HC had an age of 7.5 ± 2.2 years. No significant differences were observed in body weight between groups of aNS and HC. Characteristics of all the subjects enrolled are summarised in Table 1 . Compared to HC, children with aNS had higher study weight despite similar age and height, while NSr showed intermediate values closer to controls. Bioimpedance data are presented in Table 2 . Table 1 Characteristics of enrolled subjects. Parameter aNS aNS* NSr HC Sex (M/F) 7/1 4/1 4/1 23/15 Age (years) 6.9 ± 3.1 6.8 ± 3.1 7.7 ± 3.8 7.5 ± 2.2 Study weight (kg) 31.3 ± 17.1 28.5 ± 9.6 28.3 ± 10.4 25.3 ± 6.2 Height (cm) 120.7 ± 21.1 120.3 ± 21.8 126.1 ± 26.2 126.4 ± 14.2 BMI (kg/m 2 ) 20.1 ± 4.6 19.1 ± 1.5 17.3 ± 2.0 15.6 ± 1.2 aNS: active nephrotic syndrome; aNS*: patients who entered remission; NSr: patients remeasured at remission; HC: healthy controls. Table 2 Bioimpedance data of enrolled subjects. Parameter aNS aNS* NSr HC R E (ohm) 447.1 ± 48.9 420.2 ± 43.5 752.6 ± 71.7 816.3 ± 73.8 R I (ohm) 1871.6 ± 182.3 1919.7 ± 176.2 1799.5 ± 239.5 1922.3 ± 224.0 R INF (ohm) 359.7 ± 33.3 344.1 ± 31.6 528.6 ± 46.0 572.3 ± 53.2 C m (nF) 0.53 ± 0.21 0.47 ± 0.13 0.82 ± 0.34 0.68 ± 0.20 R E : resistance at 0 kHz (resistance of extracellular water); R I : resistance of intracellular water; R INF : resistance at infinite frequency (resistance of total body water); aNS: active nephrotic syndrome; aNS*: patients who entered remission; NSr: patients remeasured at remission; HC: healthy controls. Data abstraction BIA data were obtained from the 8 patients with aNS. Only 5 who had a remission and were remeasured and included in the software analysis as “healthy” together with the HC group. The three patients who were not in remission/unknown status and consequently not re-measured and therefore not included in the NSr group had either experienced repeated relapses (two children) or been transferred to another hospital (one child). Initially, JADBio first ML model identified weight as the only influential feature, prompting its exclusion from the dataset based on a discussion in the group before the second model execution. This was to decrease the bias provided by the weight feature, which would obscure other potential features for a model. Model specification For classifying the eight children with aNS compared to the 38 HC, by bootstrapping the real-world data, a large number of models (271,530 models with 3017 configurations) were trained and explored using an extensive tuning effort. The best model for the overall results performance (e.g. Precision, true positive rate etc.) with the chosen threshold was ridge logistic regression with the penalty hyperparameter lambda = 0.001, with an AUC of 0.84 [CI: 0.72;0.94] and a threshold of 0.81. See Fig. 1 for the AUC and ROC plot. The average Matthews correlation was 0.36 [CI 0.1;0.6]. Features were selected based on the LASSO feature selection penalty = 0.0. Features selected were height, age, X C50 , Z 50 . Z f c , PhA 50 and sex. The percentage drop in predictive performance accuracy when a feature is removed from the model is height 23% [CI 15.8;31.7], age 13.7% [6.6;21.4], X C 5.3% [1.9;9.3], and Z 50 0.6% [CI 0;2.8]. See Fig. 2 . PhA had a percentage drop of 0% [CI 0;0], and sex dropped 0% [CI 0;2.3]. See Table 3 . There was a repeated 9-fold cross-validation (CV) without falling, and max repeats = 20. Internal accuracy achieved by using only height and age was 70.6% [CI 61.6;80.3]; when adding the X C , the internal accuracy increased to 100% [CI 100;100]. See Table 4 . The reported 100% Internal accuracy refers to the internal accuracy achieved in specific cross-validation folds during model development. This metric reflects performance within training-derived partitions and does not represent the model's generalisation ability to unseen data. When applied to a held-out test set, the final model achieved an area under the curve (AUC) of 0.84. Table 3 The percentage decrease when excluding features from the prediction model. Feature Value added (%) Lower Confidence Interval (%) Upper Confidence Interval (%) Ht (cm) 23.4 15.8 31.7 Age (month) 13.7 6.6 21.4 X C50 (ohm) 5.3 1.9 9.3 Z 50 (ohm) 0.6 0 2.8 Z f c (ohm) 0 0 0 PhA 50 (degrees) 0 0 0 Sex 0 0 2.3 X C50 : reactance at 50 kHz; Z 50 : impedance at 50 kHz; Z f c : impedance at characteristic frequency; PhA 50 : phase angle as 50 kHz. Table 4 Predictive performance when including one by one parameter in the model. Feature Predictive Performance Value (%) Lower Confidence Interval (%) Upper Confidence Interval (%) Ht 70.6 61.6 80.3 Age 70.6 61.6 80.3 X C50 100 100 100 Z 50 100 99 100 Z f c 100 99.4 100 PhA 50 100 97.7 100 Sex 100 100 100 X C50 : reactance at 50 kHz; Z 50 : impedance at 50 kHz; Z f c : impedance at characteristic frequency; PhA 50 : phase angle as 50 kHz. If a child has aNS, the model has a true positive rate of 0.92 [CI 0.88;0.96]. The model’s precision, correct positives out of the total positive population, was 0.84 [CI 0.80;0.89], and its specificity, how many correct negatives were predicted, was 0.22 [CI 0.08;0.36]. Model evaluation The mean values of R were significantly lower for the children with aNS compared to the HC group when using traditional statistical methods such as unpaired two-tailed t-tests and in the reference material [ 18 ]. The model evaluated and included the specific features regarding the BIA available measurements X C , Z and PhA, but not the C m . The ML software tested the C m but was not included in the final model. Discussion In this study of eight children with aNS, an automatic ML software was able to construct a model with high accuracy for classifying the healthy controls, but with low precision for classifying the children with aNS. The model could discriminate between the two classes, aNS and HC, with an AUC of 84%. To the authors’ knowledge, this report is the first to demonstrate the connection between alterations in disease status and raw impedance parameters in children with aNS, analysed with commercially available automatic ML software for biological data. Other researchers have used ML to employ models predicting disease status in children with aNS and clinical data, such as Kou et al. [ 24 ], who constructed a multivariate logistic regression analysis with the four clinical variables: erythrocyte sedimentation rate, suppressor T cells, D-dimer and beta2-microglobulin, which correlated with steroid-resistance aNS in children with an AUC of 87%. Ye et al. [ 25 ] constructed a prediction model in children to evaluate the internal conditions and disease status of patients with steroid-resistant aNS. They used 8 clinical variables such as erythrocyte sedimentation rate, urine occult blood, percentage of neutrophils, immunoglobulin A, cholesterol, vinculin autoantibody, etc., and the model had an accuracy of 94%. Ye et al. and Kou et al. used clinical variables and not BIA outcomes and had a total of 91 and 111 subjects with aNS and steroid-resistant aNS recruited. Another study group used a dataset of 2,520 participants BIA data in a ML to optimize the intracellular fluid prediction with success in healthy control group [ 26 ]. Research in healthcare ML and artificial intelligence is rapidly gaining momentum, revealing potential applications across a range of medical fields [ 27 , 28 ], such as interpreting chest radiographs [ 23 ], detecting cancer in mammograms [ 24 ], or detecting arrhythmias [ 29 ]. Until now, very few well-validated ML models, usually developed from large datasets, have been deployed successfully in clinical practice [ 27 , 30 ]. The potential for faster data analysis and a more personalised treatment plans for patients with aNS could potentially be reached by employing ML and ultimately improving patient outcomes and optimising healthcare resources. As suggested in Birk et al for a more precise BIA measurements enhanced with ML in an Indian population [ 31 ], resulting in a more precise estimate of body composition parameters. This pilot study assessed the feasibility of automated ML software in patients suffering from severe oedema due to their disease. Ideally, ML software such as JADBio performs best with large datasets, up to millions, in variables. The low prevalence of paediatric NS, estimated to be around 2–5 per 100,000 [ 32 ], generally precludes accumulating large data sets. The small sample size was a limitation of this study, which poses a risk for over-fitting the model to the data. A larger population would increase the model’s ability to generalise, but this may not be achievable in real-life cohorts. While small cohorts may pose challenges due to limited data availability, it remains clinically important to prioritize research in this most common glomerular disease in childhood [ 33 ]. We excluded the body weight parameter for our model because body weight does not accurately estimate fluid volume or oedema in patients with aNS [ 5 ]. This introduces a potential bias towards overestimating the ML ability to help diagnose children with aNS. The study included only one girl with aNS, which is not ideal for the software to model the effects of sex. The control group of 15 girls mitigates this limitation. The ML model incorporated sex as a parameter but removing it would only reduce the model’s predictive performance by 0-2.3%. Since BIA was conducted before puberty, when body composition differences are minimal, it is less likely that sex is an important factor to consider when diagnosing NS. While internal accuracy reached 100% when using height, age, and XC50 alone in specific cross-validation configurations, the final model selected by the JADBio platform included additional features such as Z50, Zfc, PhA50, and sex. The model included PhA 50 , which showed a 0% drop in internal accuracy [CI 0;2.3] when excluded from the model. We used a non-aggressive feature solution due to the nature of this being a pilot study. This meant the model used all features that could contribute. This included sex and PhA 50 which may hold contextual or biological value. This was due to the application of a LASSO feature selection method, which allowed the inclusion of all features with marginal added value. This broader feature inclusion slightly reduced generalisation performance, as expected with small datasets. Another potential limitation was including BIA data with only three pre-determined frequencies (5, 50, and 200 kHz) and f c . JADBio used the mean of two pre-determined frequencies measured by the BIA. With the pre-determined frequencies, the software was able to employ a “working” model. Still, the potential for a more precise model could be reached by including the full range of impedance data from all measurement frequencies (50 within the range of 5 to 1000 kHz) or complete frequency spectrum data obtained from Cole modelling. This would harness the potential of automated ML software to analyse large data sets and potentially enhance the true positive rate. Finally, the omission of other biomarkers such as biochemical measurements or clinical symptoms limit the model and could be included in future studies. Our findings call for further research in a larger dataset of children with NS, e.g., through multicentre trials and/or data sharing, and to include additional impedance data (at more frequencies) and inclusion of clinical variables (e.g., blood chemistry). It could be considered removing some features, as the LASSO penalty was set to zero. This resulted in the inclusion of all features with the possibility to add value. Where both sex and PhA 50 contributed with no measurable improvement in predictive performance, meaning a 0% drop when excluded. As seen in Table 4 , adding more features beyond a certain point can introduce noise into the model, which can then potentially lowering its cross-validated performance accuracy. This showcases a known phenomenon in ML, where adding additional variables may disturb the model’s internal structure, causing it to re-evaluate previous classifications and perform worse. Further work is needed to ascertain the importance of BIA measurements in ML models relative to clinical variables. ML models are excellent for predicting an outcome, but do not imply a causal inference [ 22 ]. Additional studies will be needed in these larger datasets to develop a clinically useful prediction model and investigate the causal inference of using an ML model to predict children with NS and complications in sick children. Conclusion This cross-sectional study assessed the feasibility of automated ML software with the following BIA outcomes: PhA, IR, R, Z, X C and C m derived from impedance measurements obtained at specific frequencies. The ML software selected the features BIA parameters X C , Z, and PhA, and the clinical features height, age, and sex. A ridge logistic model detected and divided children with high true positive rate (91.8%) but a low specificity (21.7%) for the healthy controls and children with NSr. While this is too low to be acceptable in the clinical setting, these findings suggest that the automated ML software can distinguish the two groups and warrants further study in a larger cohort of children. JADBio is a promising tool for analysing complex biomedical data such as BIA, especially in small datasets. However, it can function as a “black box” in some situations, with limited insight into model construction and validation. This lack of transparency should be taken into account when interpreting results and evaluating their clinical relevance. If a larger study confirmed the present pilot observations, implementation of ML using BIA measures could potentially improve decision-making in selecting treatment pathways in children with NS. Abbreviations aNS Active nephrotic syndrome (study group) AUC Area under the ROC curve BIA Bioelectrical impedance analysis BIS Bioimpedance spectroscopy CI Confidence interval C m Cell membrane capacitance CV Cross validation ECW Extra-cellular water Ht Person height in cm HC Healthy controls (control group) ICE Individual conditional expectation IR Impedance ratio ICW Intra-cellular water JADBio Just add data bio LASSO Least absolute shrinkage and selection operator Machine learning ML, NS Nephrotic syndrome NSr Nephrotic syndrome in remission (study group) PhA Phase angle R Electrical resistance ROC Receiver operating characteristics SD Standard deviation STREAM-URO Standardized reporting of machine learning application in urology TBW Total body water X C Capacitive reactance Z impedance Declarations Ethics approval and consent to participate Written informed consent was obtained from the subjects’ parents or legal guardians before study enrolment. The study was performed following the Helsinki Declaration and was approved by the Central Denmark Region Committees on Health Research Ethics (case number: 1-10-72-17-12). Consent for publication Not applicable. Availability of data and materials The dataset used during the current study is available from the corresponding author on reasonable request at the following address: [email protected] Competing interests Author Leigh C Ward provides consultancy services to ImpediMed Ltd. ImpediMed Ltd was not involved in preparing this manuscript. Other authors have no conflicts of interest to declare concerning this work. Funding None. CLH received research funding from the Novo Nordisk Foundation (grant no. NNF22OC0074080). Authors’ contributions JQ, with assistance from SB, drafted and approved the research protocol. RFA helped with the enrolment of NS patients. JQ undertook the data analysis with assistance from LCW, LJ, and SB and prepared the manuscript. All authors participated in critically interpreting results, edited the manuscript, and approved its final content. References Tapia C, Bashir K (2025) Nephrotic Syndrome. StatPearls. StatPearls Publishing Copyright © 2025. StatPearls Publishing LLC.. Treasure Island (FL) Brantlov S, Andersen T, Jødal L, Rittig S, Lange A (2016) Bioimpedance Spectroscopy in Healthy Children. J Clin Eng 41:33–39 Norman K, Stobäus N, Pirlich M, Bosy-Westphal A (2012) Bioelectrical phase angle and impedance vector analysis–clinical relevance and applicability of impedance parameters. Clin Nutr 31:854–861 Gipson DS, Pal M, Desmond H, Anderson C, Walsh L, Trachtman H, Massengill SF, Gipson P, Rao PS, Thurman J, Kopp J, Kamil E, Lamothe J, Mariani LH, LaFleur P, Vento S, O’Shaughnessy M, Farag YMK, Simon C, Carlozzi NE (2023) Developing an Edema Clinician-Reported Outcome Measure for Nephrotic Syndrome. Glomerular Dis 3:132–139 Mulasi U, Kuchnia AJ, Cole AJ, Earthman CP (2015) Bioimpedance at the bedside: Current applications, limitations, and opportunities. Nutr Clin Pract 30:180–193 Brantlov S, Jødal L, Lange A, Rittig S, Ward LC (2017) Standardisation of bioelectrical impedance analysis for the estimation of body composition in healthy paediatric populations: a systematic review. J Med Eng Technol 41:460–479 Brantlov S, Ward LC, Jødal L, Rittig S, Lange A (2017) Critical factors and their impact on bioelectrical impedance analysis in children: a review. J Med Eng Technol 41:22–35 Gartner A, Maire B, Delpeuch F, Sarda P, Dupuy RP, Rieu D (1992) Importance of electrode position in bioelectrical impedance analysis. Am J Clin Nutr 56:1067–1068 Collins CT, Reid J, Makrides M, Lingwood BE, McPhee AJ, Morris SA, Gibson RA, Ward LC (2013) Prediction of body water compartments in preterm infants by bioelectrical impedance spectroscopy. Eur J Clin Nutr 67(Suppl 1):S47–53 VanderJagt DJ, Huang YS, Chuang LT, Bonnett C, Glew RH (2002) Phase angle and n-3 polyunsaturated fatty acids in sickle cell disease. Arch Dis Child 87:252–254 Nagano M, Suita S, Yamanouchi T (2000) The validity of bioelectrical impedance phase angle for nutritional assessment in children. J Pediatr Surg 35:1035–1039 Verheij R, Veenman CS, den Bakker JV, van Duyl WA (1996) Comments on the interpretation of tissue impedance measurements during hemodialysis. Blood Purif 14:8–14 Chertow GM, Jacobs DO, Lazarus JM, Lew NL, Lowrie EG (1997) Phase angle predicts survival in hemodialysis patients. J Ren Nutr 7:204–207 Garr Barry V, Peterson CM, Gower BA (2020) Membrane Capacitance from a Bioimpedance Approach: Associations with Insulin Resistance in Relatively Healthy Adults. Obes (Silver Spring) 28:2184–2191 Bosy-Westphal A, Danielzik S, Dörhöfer RP, Later W, Wiese S, Müller MJ (2006) Phase angle from bioelectrical impedance analysis: population reference values by age, sex, and body mass index. JPEN J Parenter Enter Nutr 30:309–316 Nematollahi MA, Jahangiri S, Asadollahi A, Salimi M, Dehghan A, Mashayekh M, Roshanzamir M, Gholamabbas G, Alizadehsani R, Bazrafshan M, Bazrafshan H (2023) Bazrafshan drissi H, Shariful Islam SM Body composition predicts hypertension using machine learning methods: a cohort study. Scientific Reports 13:6885 Kwong JCC, McLoughlin LC, Haider M, Goldenberg MG, Erdman L, Rickard M, Lorenzo AJ, Hung AJ, Farcas M, Goldenberg L, Nguan C, Braga LH, Mamdani M, Goldenberg A, Kulkarni GS (2021) Standardized Reporting of Machine Learning Applications in Urology: The STREAM-URO Framework. Eur Urol Focus 7:672–682 Brantlov S, Jødal L, Andersen RF, Lange A, Rittig S, Ward LC (2019) An evaluation of phase angle, bioelectrical impedance vector analysis and impedance ratio for the assessment of disease status in children with nephrotic syndrome. BMC Nephrol 20:331 Rinninella ECM, Addolorato G, Triarico S, Ruggiero A et al (2018) Phase Angle and impedance ratio: Two specular ways to analyze body composition. Ann Clin Nutr 1:1003 Tsamardinos I, Charonyktakis P, Papoutsoglou G, Borboudakis G, Lakiotaki K, Zenklusen JC, Juhl H, Chatzaki E, Lagani V (2022) Just Add Data: automated predictive modeling for knowledge discovery and feature selection. npj Precision Oncol 6:38 Donders AR, van der Heijden GJ, Stijnen T, Moons KG (2006) Review: a gentle introduction to imputation of missing values. J Clin Epidemiol 59:1087–1091 Hernán MA, Hsu J, Healy B (2019) A Second Chance to Get Causal Inference Right: A Classification of Data Science Tasks. CHANCE 32:42–49 Henderson AR (2005) The bootstrap: a technique for data-driven statistics. Using computer-intensive analyses to explore experimental data. Clin Chim Acta 359:1–26 Kou M, Wu F, Qu XY, Wang H, Guo XT, Yang YY, Zhao LJ (2023) [Establishment and validation of clinical prediction model for steroid-resistant nephrotic syndrome in children]. Zhonghua Er Ke Za Zhi 61:333–338 Ye Q, Li Y, Liu H, Mao J, Jiang H (2023) Machine learning models for predicting steroid-resistant of nephrotic syndrome. Front Immunol 14:1090241 Novera JN, Ali ME, Mia MMA, Haque A, Fahim T, Jahan A (2025) I. Machine Learning-Driven Optimization of Bioelectrical Impedance Analysis for Intracellular Fluid Prediction. International conference on engineering research, innovation and education, Shahjalal university of science and technology, sylhet, Bangladesh Kelly CJ, Karthikesalingam A, Suleyman M, Corrado G, King D (2019) Key challenges for delivering clinical impact with artificial intelligence. BMC Med 17:195 Gupta S, Vargas A, Saulnier G, Newell J, Faaborg-Andersen C, Kelley RS (2021) Uterine bioimpedance combined with artificial intelligence as a means of cancer detection. J Med Eng Technol 45:606–613 Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, Ng AY (2019) Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med 25:65–69 Vistisen ST, Pollard TJ, Harris S, Lauritsen SM (2022) Artificial intelligence in the clinical setting: Towards actual implementation of reliable outcome predictions. Eur J Anaesthesiol 39:729–732 Birk N, Kulkarni B, Bhogadi S, Aggarwal A, Walia GK, Gupta V, Rani U, Mahajan H, Kinra S, Mallinson PAC (2025) Machine learning-based equations for improved body composition estimation in Indian adults. PLOS Digit Health 4:e0000671 Tamura H (2021) Trends in pediatric nephrotic syndrome. World J Nephrol 10:88–100 Sanjad SA, Ulinski T, Aoun B (2021) Editorial: Nephrotic Syndrome in Children. Front Pead 9 Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7197037","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":490749225,"identity":"4fd31821-5f2b-41cc-aae9-fe68c6c99f47","order_by":0,"name":"Josephine Reinert Quist","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA70lEQVRIiWNgGAWjYBACAwbmBjCDjYGB8QGQ5uEjrIURpoWZ2QCkhY1oLQwMzGwSUOvwA3P2xuYXPxisE/v4zx+r/JpjJwO07uGjG3i0WPYcbLPsYUhPbJNIZrstuy0Z6DA2Y+McfA67kdhmwMNwGKiFme225DZmoBYeNmm8Wu4/bDP8A9LCf5itWHJbPRFabjA2PwbbwpDMxvhx22EitJxJbGOWMUg3BvrFWJpx23EeNmZCfjl++PDHNxXWsvP7Dz78+HNbtT0/e/PDx/i0AAEwOgyYwSxmHjCJXzlYyQeYMsYfhFWPglEwCkbBCAQA26xDZ88rHlkAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0001-9655-1138","institution":"Aarhus University Hospital Department of Hepatology and Gastroenterology: Aarhus Universitetshospital Afdeling Lever- Mave- og Tarmsygdomme","correspondingAuthor":true,"prefix":"","firstName":"Josephine","middleName":"Reinert","lastName":"Quist","suffix":""},{"id":490749226,"identity":"86f1a67b-5dba-491a-af35-0b8b9214c9b5","order_by":1,"name":"Leigh C Ward","email":"","orcid":"","institution":"The University of Queensland School of Chemistry and Molecular Biosciences","correspondingAuthor":false,"prefix":"","firstName":"Leigh","middleName":"C","lastName":"Ward","suffix":""},{"id":490749227,"identity":"17eb60af-54e5-4f6c-aa51-9c62284ab9d8","order_by":2,"name":"Lars Jødal","email":"","orcid":"","institution":"Aalborg University Hospital: Aalborg Universitetshospital","correspondingAuthor":false,"prefix":"","firstName":"Lars","middleName":"","lastName":"Jødal","suffix":""},{"id":490749228,"identity":"2684d01f-40eb-40ef-ae2b-8f9e40fa6c8f","order_by":3,"name":"René Frydensbjerg Andersen","email":"","orcid":"","institution":"Aarhus University Hospital: Aarhus Universitetshospital","correspondingAuthor":false,"prefix":"","firstName":"René","middleName":"Frydensbjerg","lastName":"Andersen","suffix":""},{"id":490749229,"identity":"5c5151dc-83af-46f7-881f-814c2ad2c501","order_by":4,"name":"Christian Lodberg Hvas","email":"","orcid":"","institution":"Aarhus University Hospital Department of Hepatology and Gastroenterology: Aarhus Universitetshospital Afdeling Lever- Mave- og Tarmsygdomme","correspondingAuthor":false,"prefix":"","firstName":"Christian","middleName":"Lodberg","lastName":"Hvas","suffix":""},{"id":490749230,"identity":"37246f6a-be80-43af-8ab0-414aabec89f3","order_by":5,"name":"Steven Brantlov","email":"","orcid":"","institution":"Aarhus University Hospital: Aarhus Universitetshospital","correspondingAuthor":false,"prefix":"","firstName":"Steven","middleName":"","lastName":"Brantlov","suffix":""}],"badges":[],"createdAt":"2025-07-23 13:47:36","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7197037/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7197037/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":88000177,"identity":"411831f9-d439-40bf-8316-3d4221a2888e","added_by":"auto","created_at":"2025-07-31 10:18:51","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":27332,"visible":true,"origin":"","legend":"\u003cp\u003eA) AUC curve illustrated with an AUC = 0.84, meaning that in 84% of the cases it can distinguish between healthy and patients. B) ROC plot with the threshold of 0.82, meaning if the model has an output \u0026gt;0.80, it is a positive case.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7197037/v1/917717df88a15529ac98f3e0.png"},{"id":88000179,"identity":"300d42c4-6b5b-4b2e-908d-b64566965a9f","added_by":"auto","created_at":"2025-07-31 10:18:51","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":31044,"visible":true,"origin":"","legend":"\u003cp\u003eA) Illustrative chart for the drop in the internal accuracy described by JADBio as “predictive performance” percentages if that particurlar feature were dropped. The largest drop in internal accuracy occurred if height was dropped, with up to 23.4%, and the lowest drop if gender was excluded from the model. B) Illustrative chart for the internal accuracy achievable by using features. If all the features were used, the internal accuracy would increase to 100%.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7197037/v1/bfd25ddf61f7afd0636e7407.png"},{"id":88828289,"identity":"63476fff-fc68-48ba-95d2-1287077c4d1d","added_by":"auto","created_at":"2025-08-11 20:30:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":893237,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7197037/v1/f03c64f0-29e5-48d7-86d4-2f25eb409407.pdf"}],"financialInterests":"","formattedTitle":"Integrating machine learning for advanced analysis of bioelectrical impedance parameters in children with nephrotic syndrome: phase angle, impedance ratio, and cell membrane capacitance","fulltext":[{"header":"Introduction","content":"\u003cp\u003eNephrotic syndrome (NS) is a relatively rare disease, incidence 2\u0026ndash;7 per 100,000 children [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] characterised by excessive renal protein loss, leading to hypoalbuminemia and oedema [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. It is associated with extended hospitalisation [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Conventionally, the degree of oedema is estimated from clinical evaluation [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] and body weight measurement, and a weight increase is assumed to primarily indicate accumulation of extracellular water (ECW). This approach has limitations, especially over extended periods when weight changes due to factors other than changes in ECW become significant [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIn the search for a practical, cost-effective, and non-invasive technique to evaluate routine oedema in children with NS, bioelectrical impedance analysis (BIA) emerges as a promising tool [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. BIA offers characteristics ideally suited for clinical routine use, such as safety, speed, simplicity, and portability [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. BIA operates on the principle that the flow of alternating current through the body varies in different tissues. Tissues rich in water and electrolytes exhibit high conductivity and low impedance (Z, ohm), while fat and bone, with low conductivity, show high impedance. Impedance comprises two components: resistance (R) and capacitive reactance (X\u003csub\u003eC\u003c/sub\u003e). The body\u0026rsquo;s resistance is inversely related to its water content (R\u0026thinsp;~\u0026thinsp;1/body water), which is the foundation for estimating body water volumes with BIA [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. Many BIA devices measure only at a single frequency (SF-BIA) or a few, multiple frequencies (MF-BIA). The bioimpedance spectroscopy (BIS) technique measures a broad spectrum of frequencies and employs the Cole model for data analysis, where coordinate points (R, X\u003csub\u003eC\u003c/sub\u003e) are plotted and fitted with a semicircular arc to describe the data.\u003c/p\u003e\u003cp\u003eFrom BIA measurements, clinical body parameters such as ECW, total body water (TBW), and fat-free mass and fat mass can be estimated using prediction models derived from healthy individuals. Importantly, these estimates rely on assumptions that can vary among study populations; the prediction equations typically vary among BIA devices from different manufacturers, and only some manufacturers openly reveal the specifics of the prediction equations used by their devices.\u003c/p\u003e\u003cp\u003eThe use of the measured electrical properties is increasingly being considered as they are openly described and are independent of a particular manufacturer\u0026rsquo;s BIA device. Such parameters are called \u0026ldquo;raw\u0026rdquo; or \u0026ldquo;derived raw\u0026rdquo; impedance parameters and act as proxies for their physiological correlates. We use the term \u0026ldquo;derived raw\u0026rdquo; to reflect that these parameters are not directly measured data but are the results from basic calculations made upon the measured data.\u003c/p\u003e\u003cp\u003eFour important derived raw impedance parameters are the phase angle (PhA), impedance ratio (IR, ratio of impedances measured at different frequencies, typically a high and low frequency), cell membrane capacitance (C\u003csub\u003em\u003c/sub\u003e) and resistance. PhA, IR, and C\u003csub\u003em\u003c/sub\u003e are composite parameters derived from various raw data, whereas R directly reflects the opposition to the flow of electric current through the body\u0026rsquo;s tissues, influenced predominantly by the body\u0026rsquo;s total water content and tissue composition. Each of these may be valuable for assessing the clinical state of NS patients` body water distribution. PhA is associated with tissue hydration, cell membrane quality and cellular mass [\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e], although its exact biological significance remains a question of ongoing research [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. IR is another promising parameter that potentially indicates oedema or overall health and reflects fluid distribution between TBW and the ECW. Lastly, C\u003csub\u003em\u003c/sub\u003e is a parameter that promisingly provides insights into cell membrane function, size, hydration status and overall tissue composition [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Although ongoing research is being conducted into the potential information content of these parameters, especially PhA in adults, there is a scarcity of studies focusing on paediatric populations [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eIntegrating machine learning (ML) software\u0026rsquo;s capabilities includes efficient processing of large datasets, precise pattern recognition, and creating predictive models [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. This increases clinicians\u0026rsquo; ability to extract intricate insights from impedance measurements, enhancing their understanding of parameters like PhA, IR, and C\u003csub\u003em\u003c/sub\u003e in NS. ML\u0026rsquo;s notable speed and minimising the mean prediction error may ensure timely and informed decision-making, facilitating more effective patient care and treatment strategies.\u003c/p\u003e\u003cp\u003eThe aim of this clinical pilot study was to assess the feasibility of employing ML software with the bioimpedance outcomes PhA, IR, C\u003csub\u003em\u003c/sub\u003e, and R, with the perspective to identify and differentiate children with aNS effectively, to contribute to further characterization of the body composition.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003e\u003cb\u003eStudy design\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThis study was conducted as a single-centre cross-sectional study and was reported according to the STREAM-URO network guidelines [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cb\u003eStudy subjects\u003c/b\u003e\u003c/p\u003e\u003cp\u003eChildren with active NS in the acute phase (aNS patient group) in the age range of 3 to 10 years old, were included from the Department of Paediatrics and Adolescent Medicine, Aarhus University Hospital, Denmark. Before treatment with prednisolone and diuretics was initiated, blood samples, blood pressures and impedance measurements were collected from the patient group. The 38 healthy controls included were selected from a previously established reference cohort of children, chosen to span a representative range of age and sex, but were not individually matched to the patients with NS. Patient data were drawn from a previously described cohort [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e\u003cp\u003e\u003cb\u003eBIA device and parameters\u003c/b\u003e\u003c/p\u003e\u003cp\u003eWhole-body impedance was measured with electrodes placed on the wrist and ankle in pairs. A Xitron 4200, HYDRA BIS device (Xitron Technologies, San Diego, CA, USA) was used and was tested weekly with an electronic verification module (TS4201) according to the manufacturer\u0026rsquo;s instructions. The BIS device performed measurements of the electrical parameters, including impedance (Z), resistance (R), and capacitive reactance (X\u003csub\u003eC\u003c/sub\u003e), all measured in ohms (Ω) at fifty different frequencies in the range from 5 to 1000 kHz.\u003c/p\u003e\u003cp\u003eImpedance (Z) is the length of the vector to an individual point, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:Z=\\sqrt{{R}^{2}+{X}_{C}^{2}}\\)\u003c/span\u003e\u003c/span\u003e. PhA is the angle between the vector and the horizontal axis. For measurements at a given frequency, the phase angle is calculated as PhA\u0026thinsp;=\u0026thinsp;tan\u003csup\u003e\u0026minus;1\u003c/sup\u003e(X\u003csub\u003eC\u003c/sub\u003e/R) ∙ 180\u0026deg;/π. The result depends on the frequency, but PhA is often reported at 50 kHz frequency [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. This requires only SF-BIA (the single frequency being 50 kHz). When full-spectrum data are available (BIS), PhA is sometimes also considered for the characteristic frequency \u003cem\u003ef\u003c/em\u003e\u003csub\u003ec\u003c/sub\u003e, corresponding to the frequency of maximum reactance.\u003c/p\u003e\u003cp\u003eImpedance ratio, IR, compares impedance at a high frequency with the impedance at a low frequency. Measurements at high frequencies are dominated by TBW, while measurements at low frequencies are dominated by ECW; typically, the frequencies used for IR are 200 and 5 kHz [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. It is calculated as IR\u003csub\u003e200/5\u003c/sub\u003e = R\u003csub\u003e200\u003c/sub\u003e/R\u003csub\u003e5\u003c/sub\u003e and has no unit [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Determination of IR requires at least two measurements, i.e., MF-BIA or BIS.\u003c/p\u003e\u003cp\u003eThe cellular membrane capacitance, C\u003csub\u003em\u003c/sub\u003e, was calculated from the Cole model based on the spectrum of frequencies [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. The unit of C\u003csub\u003em\u003c/sub\u003e is farad, F. Determination of C\u003csub\u003em\u003c/sub\u003e requires knowledge of a spectrum of frequencies, i.e., BIS not SF-BIA or MF-BIA.\u003c/p\u003e\u003cp\u003e\u003cb\u003eData acquisition\u003c/b\u003e\u003c/p\u003e\u003cp\u003eBefore conducting impedance measurements, trained personnel recorded weight and height in duplicate. Weight, measured on digital scales with light clothing, was recorded to the nearest 0.1 kg, while height, measured without shoes using a stadiometer, was recorded to the nearest 0.5 cm.\u003c/p\u003e\u003cp\u003eThe BIA measurements adhered to standardised procedures, with participants refraining from intense physical exercise four hours before the study but not required to fast. Details of the protocol can be found in a previous publication [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] and in the study from which the subject was initially enrolled [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Data were extracted using the ImpediMed SFB7 Multi-Frequency Analysis software (Bioimp Version 5.4.0.3, Brisbane QLD, Australia) from measurements of 50 kHz and at the characteristic frequency (\u003cem\u003ef\u003c/em\u003e\u003csub\u003ec\u003c/sub\u003e) where capacitive reactance is at its maximum value. These measurements were used to calculate specific parameters within the software.\u003c/p\u003e\u003cp\u003eIn this study, the parameters analysed at 50 kHz were resistance (R\u003csub\u003e50\u003c/sub\u003e), capacitive reactance (X\u003csub\u003eC50\u003c/sub\u003e), impedance (Z\u003csub\u003e50\u003c/sub\u003e), and phase angle (PhA\u003csub\u003e50\u003c/sub\u003e). The parameters analysed at the subject-specific characteristic frequency were resistance (R\u003csub\u003e\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e), capacitive reactance (X\u003csub\u003eC\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e), impedance (Z\u003csub\u003e\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e) and phase angle (PhA\u003csub\u003e\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e). In addition, C\u003csub\u003em\u003c/sub\u003e was also analysed. Each subject was measured three times in one setting, and the mean values from the three measurements were used as individual data points for input to the software. Finally, the clinical data sex, height, weight, and age, were included.\u003c/p\u003e\u003cp\u003e\u003cb\u003eMachine learning software\u003c/b\u003e\u003c/p\u003e\u003cp\u003eJust Add Data Bio (JADBio)\u0026reg; (3154 Glendale Blvd Los Angeles, CA 90039\u0026thinsp;\u0026minus;\u0026thinsp;1830, USA) [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] is a web-based machine learning platform for analysing potential biomarkers to diagnose and estimate the prognosis of a disease. It is a fully automated machine learning software developed for a small sample population with many data points for each subject. It was designed for predictive modelling and exploring the possible application of machine learning methods with minimal entry-level skills typically required for developing such models. When the data have been fed into the JADBio software, it starts locating the features (e.g., variables that help determine the investigated outcome), and the software returns the developed model and information related to the model. It comprises several algorithms for the standard phases when making a machine learning model: data preprocessing, data transformation, data imputation, feature selection, and predictive modelling. The label of interest is to predict disease activity. The label \u0026ldquo;false\u0026rdquo; predicts a child without aNS, whereas \u0026ldquo;true\u0026rdquo; predicts a child with aNS. Data were divided into training and testing cohorts.\u003c/p\u003e\u003cp\u003eIn data pre-processing, mean and mode imputation of missing values is performed. Zero variance features are removed. Continuous features are standardised to zero mean and standard deviation (SD) of 1. The imputation of missing data on a variable replaces a missing value computed from an estimate of the distribution of this variable [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. If imputation is not possible, missing data and the reasons behind it will be disclosed. Feature selection algorithms, least absolute shrinkage and selection operator (LASSO) regularised regression, and statistical equivalent signatures were investigated to determine the minimal size predictive feature subset. Predictive modelling algorithms tested are decision trees, ridge logistic regression, random forests, support vector machines, cox regression and random survival forests [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Removal of features tested for a machine learning model is based on clinical intuition and discussion in the research group. This was chosen based on assumptions that increase the chance for causal inference [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Performance estimations are based on generalised cross-validation, which determines the hyperparameters, where the latter are external settings for a model, influencing its performance (e.g., learning rates and regularisation strengths). Cross-validation is a technique where the sample size is proportionally divided into K folds of equal size. K refers to the number of groups a given data sample will split into where R-repeated cross-validation indicates that the procedure runs R times with different partitions (refer to the divisions or subsets into which the dataset is divided) to folds to reduce the variance in the estimation. The final model is estimated with a bootstrap bias correction to adjust p-values and to remove bias due to the multiple tries. Bootstrapping, a statistical technique involving resampling with replacement from observed data, estimates uncertainty in a sample statistic. In the final model estimation, a bootstrap bias correction is applied to adjust p-values, mitigate bias resulting from multiple attempts, and increase the risk for individual error [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The working model is presented by sensitivity, specificity, accuracy, receiver operating characteristic (ROC) curves for correct classification of children with or without aNS, false positive rate, and true positive rate. This model can be changed with different thresholds for the ROC. Predictive performance is defined as the average accuracy across the test folds in a repeated 9-fold cross-validation. Individual conditional expectation (ICE) plots visualise the effect of a given variable for a specific model. The final analysis is evaluated by the area under the curve (AUC) and CI for the best model.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003cb\u003ePatient characteristics\u003c/b\u003e\u003c/p\u003e\u003cp\u003eData were collected from 8 children with aNS, seven boys and one girl, and 38, 23 boys and 15 girls, age-matched healthy controls (HC group) from a previous study reported by Brantlov et al. [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. 5 of the 8 children with aNS experienced remission. Determined by no protein from a urine sample on three consecutive days. These 5 children out of the original 8 were re-measured and sub-grouped in a remission group (NSr). Results are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviation (SD), after test for normality, using Q-Q plots and the Shapiro-Wilk test. Patients with aNS had an age of 6.7 \u0026plusmn; 3.1 years, the subgroup patients with NSr had an age of 7.7 \u0026plusmn; 3.8 years, and the HC had an age of 7.5 \u0026plusmn; 2.2 years. No significant differences were observed in body weight between groups of aNS and HC. Characteristics of all the subjects enrolled are summarised in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Compared to HC, children with aNS had higher study weight despite similar age and height, while NSr showed intermediate values closer to controls. Bioimpedance data are presented in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eCharacteristics of enrolled subjects.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eaNS\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eaNS*\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNSr\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHC\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSex (M/F)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e7/1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e4/1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e4/1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e23/15\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eAge (years)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e6.9\u0026thinsp;\u0026plusmn;\u0026thinsp;3.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6.8\u0026thinsp;\u0026plusmn;\u0026thinsp;3.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e7.7\u0026thinsp;\u0026plusmn;\u0026thinsp;3.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e7.5\u0026thinsp;\u0026plusmn;\u0026thinsp;2.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eStudy weight (kg)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e31.3\u0026thinsp;\u0026plusmn;\u0026thinsp;17.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e28.5\u0026thinsp;\u0026plusmn;\u0026thinsp;9.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e28.3\u0026thinsp;\u0026plusmn;\u0026thinsp;10.4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e25.3\u0026thinsp;\u0026plusmn;\u0026thinsp;6.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eHeight (cm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e120.7\u0026thinsp;\u0026plusmn;\u0026thinsp;21.1\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e120.3\u0026thinsp;\u0026plusmn;\u0026thinsp;21.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e126.1\u0026thinsp;\u0026plusmn;\u0026thinsp;26.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e126.4\u0026thinsp;\u0026plusmn;\u0026thinsp;14.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eBMI (kg/m\u003csup\u003e2\u003c/sup\u003e)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e20.1\u0026thinsp;\u0026plusmn;\u0026thinsp;4.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e19.1\u0026thinsp;\u0026plusmn;\u0026thinsp;1.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e17.3\u0026thinsp;\u0026plusmn;\u0026thinsp;2.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c5\"\u003e\u003cp\u003e15.6\u0026thinsp;\u0026plusmn;\u0026thinsp;1.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eaNS: active nephrotic syndrome; aNS*: patients who entered remission; NSr: patients remeasured at remission; HC: healthy controls.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eBioimpedance data of enrolled subjects.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"5\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eParameter\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eaNS\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eaNS*\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eNSr\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c5\"\u003e\u003cp\u003eHC\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR\u003csub\u003eE\u003c/sub\u003e\u0026nbsp;(ohm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e447.1\u0026thinsp;\u0026plusmn;\u0026thinsp;48.9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e420.2\u0026thinsp;\u0026plusmn;\u0026thinsp;43.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e752.6\u0026thinsp;\u0026plusmn;\u0026thinsp;71.7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e816.3\u0026thinsp;\u0026plusmn;\u0026thinsp;73.8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR\u003csub\u003eI\u003c/sub\u003e\u0026nbsp;(ohm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e1871.6\u0026thinsp;\u0026plusmn;\u0026thinsp;182.3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e1919.7\u0026thinsp;\u0026plusmn;\u0026thinsp;176.2\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e1799.5\u0026thinsp;\u0026plusmn;\u0026thinsp;239.5\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e1922.3\u0026thinsp;\u0026plusmn;\u0026thinsp;224.0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eR\u003csub\u003eINF\u003c/sub\u003e\u0026nbsp;(ohm)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e359.7\u0026thinsp;\u0026plusmn;\u0026thinsp;33.3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e344.1\u0026thinsp;\u0026plusmn;\u0026thinsp;31.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e528.6\u0026thinsp;\u0026plusmn;\u0026thinsp;46.0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e572.3\u0026thinsp;\u0026plusmn;\u0026thinsp;53.2\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eC\u003csub\u003em\u003c/sub\u003e\u0026nbsp;(nF)\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c2\"\u003e\u003cp\u003e0.53\u0026thinsp;\u0026plusmn;\u0026thinsp;0.21\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e\u003cp\u003e0.47\u0026thinsp;\u0026plusmn;\u0026thinsp;0.13\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e\u003cp\u003e0.82\u0026thinsp;\u0026plusmn;\u0026thinsp;0.34\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e\u003cp\u003e0.68\u0026thinsp;\u0026plusmn;\u0026thinsp;0.20\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"5\"\u003eR\u003csub\u003eE\u003c/sub\u003e: resistance at 0 kHz (resistance of extracellular water); R\u003csub\u003eI\u003c/sub\u003e: resistance of intracellular water; R\u003csub\u003eINF\u003c/sub\u003e: resistance at infinite frequency (resistance of total body water); aNS: active nephrotic syndrome; aNS*: patients who entered remission; NSr: patients remeasured at remission; HC: healthy controls.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\u003cb\u003eData abstraction\u003c/b\u003e\u003c/p\u003e\u003cp\u003eBIA data were obtained from the 8 patients with aNS. Only 5 who had a remission and were remeasured and included in the software analysis as \u0026ldquo;healthy\u0026rdquo; together with the HC group. The three patients who were not in remission/unknown status and consequently not re-measured and therefore not included in the NSr group had either experienced repeated relapses (two children) or been transferred to another hospital (one child).\u003c/p\u003e\u003cp\u003eInitially, JADBio first ML model identified weight as the only influential feature, prompting its exclusion from the dataset based on a discussion in the group before the second model execution. This was to decrease the bias provided by the weight feature, which would obscure other potential features for a model.\u003c/p\u003e\u003cp\u003e\u003cb\u003eModel specification\u003c/b\u003e\u003c/p\u003e\u003cp\u003eFor classifying the eight children with aNS compared to the 38 HC, by bootstrapping the real-world data, a large number of models (271,530 models with 3017 configurations) were trained and explored using an extensive tuning effort. The best model for the overall results performance (e.g. Precision, true positive rate etc.) with the chosen threshold was ridge logistic regression with the penalty hyperparameter lambda\u0026thinsp;=\u0026thinsp;0.001, with an AUC of 0.84 [CI: 0.72;0.94] and a threshold of 0.81. See Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e1\u003c/span\u003e for the AUC and ROC plot. The average Matthews correlation was 0.36 [CI 0.1;0.6].\u003c/p\u003e\u003cp\u003eFeatures were selected based on the LASSO feature selection penalty\u0026thinsp;=\u0026thinsp;0.0. Features selected were height, age, X\u003csub\u003eC50\u003c/sub\u003e, Z\u003csub\u003e50\u003c/sub\u003e. Z\u003csub\u003e\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e, PhA\u003csub\u003e50\u003c/sub\u003e and sex. The percentage drop in predictive performance accuracy when a feature is removed from the model is height 23% [CI 15.8;31.7], age 13.7% [6.6;21.4], X\u003csub\u003eC\u003c/sub\u003e 5.3% [1.9;9.3], and Z\u003csub\u003e50\u003c/sub\u003e 0.6% [CI 0;2.8]. See Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e2\u003c/span\u003e. PhA had a percentage drop of 0% [CI 0;0], and sex dropped 0% [CI 0;2.3]. See Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. There was a repeated 9-fold cross-validation (CV) without falling, and max repeats\u0026thinsp;=\u0026thinsp;20. Internal accuracy achieved by using only height and age was 70.6% [CI 61.6;80.3]; when adding the X\u003csub\u003eC\u003c/sub\u003e, the internal accuracy increased to 100% [CI 100;100]. See Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. The reported 100% Internal accuracy refers to the internal accuracy achieved in specific cross-validation folds during model development. This metric reflects performance within training-derived partitions and does not represent the model's generalisation ability to unseen data. When applied to a held-out test set, the final model achieved an area under the curve (AUC) of 0.84.\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eThe percentage decrease when excluding features from the prediction model.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFeature\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eValue added (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLower Confidence Interval (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eUpper Confidence Interval (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eHt (cm)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e23.4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e15.8\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e31.7\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAge (month)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e13.7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e6.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e21.4\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eX\u003c/b\u003e\u003csub\u003e\u003cb\u003eC50\u003c/b\u003e\u003c/sub\u003e \u003cb\u003e(ohm)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e5.3\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e1.9\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e9.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eZ\u003c/b\u003e\u003csub\u003e\u003cb\u003e50\u003c/b\u003e\u003c/sub\u003e \u003cb\u003e(ohm)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.8\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eZ\u003c/b\u003e\u003csub\u003e\u003cb\u003ef\u003c/b\u003e\u003cb\u003ec\u003c/b\u003e\u003c/sub\u003e \u003cb\u003e(ohm)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePhA\u003c/b\u003e\u003csub\u003e\u003cb\u003e50\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e\u003cp\u003e\u003cb\u003e(degrees)\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e0\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e2.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"4\"\u003eX\u003csub\u003eC50\u003c/sub\u003e: reactance at 50 kHz; Z\u003csub\u003e50\u003c/sub\u003e: impedance at 50 kHz; Z\u003csub\u003e\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e: impedance at characteristic frequency; PhA\u003csub\u003e50\u003c/sub\u003e: phase angle as 50 kHz.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003ePredictive performance when including one by one parameter in the model.\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"4\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eFeature\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003ePredictive Performance\u003c/p\u003e\u003cp\u003eValue (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c3\"\u003e\u003cp\u003eLower Confidence Interval (%)\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c4\"\u003e\u003cp\u003eUpper Confidence Interval (%)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eHt\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e70.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e61.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e80.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eAge\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e70.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e61.6\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e80.3\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eX\u003c/b\u003e\u003csub\u003e\u003cb\u003eC50\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eZ\u003c/b\u003e\u003csub\u003e\u003cb\u003e50\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e99\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eZ\u003c/b\u003e\u003csub\u003e\u003cb\u003ef\u003c/b\u003e\u003cb\u003ec\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e99.4\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003ePhA\u003c/b\u003e\u003csub\u003e\u003cb\u003e50\u003c/b\u003e\u003c/sub\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e97.7\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003e\u003cb\u003eSex\u003c/b\u003e\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c3\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c4\"\u003e\u003cp\u003e100\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003ctfoot\u003e\u003ctr\u003e\u003ctd colspan=\"4\"\u003eX\u003csub\u003eC50\u003c/sub\u003e: reactance at 50 kHz; Z\u003csub\u003e50\u003c/sub\u003e: impedance at 50 kHz; Z\u003csub\u003e\u003cem\u003ef\u003c/em\u003ec\u003c/sub\u003e: impedance at characteristic frequency; PhA\u003csub\u003e50\u003c/sub\u003e: phase angle as 50 kHz.\u003c/td\u003e\u003c/tr\u003e\u003c/tfoot\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eIf a child has aNS, the model has a true positive rate of 0.92 [CI 0.88;0.96]. The model\u0026rsquo;s precision, correct positives out of the total positive population, was 0.84 [CI 0.80;0.89], and its specificity, how many correct negatives were predicted, was 0.22 [CI 0.08;0.36].\u003c/p\u003e\u003cp\u003e\u003cb\u003eModel evaluation\u003c/b\u003e\u003c/p\u003e\u003cp\u003eThe mean values of R were significantly lower for the children with aNS compared to the HC group when using traditional statistical methods such as unpaired two-tailed t-tests and in the reference material [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. The model evaluated and included the specific features regarding the BIA available measurements X\u003csub\u003eC\u003c/sub\u003e, Z and PhA, but not the C\u003csub\u003em\u003c/sub\u003e. The ML software tested the C\u003csub\u003em\u003c/sub\u003e but was not included in the final model.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this study of eight children with aNS, an automatic ML software was able to construct a model with high accuracy for classifying the healthy controls, but with low precision for classifying the children with aNS. The model could discriminate between the two classes, aNS and HC, with an AUC of 84%.\u003c/p\u003e\u003cp\u003eTo the authors\u0026rsquo; knowledge, this report is the first to demonstrate the connection between alterations in disease status and raw impedance parameters in children with aNS, analysed with commercially available automatic ML software for biological data. Other researchers have used ML to employ models predicting disease status in children with aNS and clinical data, such as Kou et al. [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], who constructed a multivariate logistic regression analysis with the four clinical variables: erythrocyte sedimentation rate, suppressor T cells, D-dimer and beta2-microglobulin, which correlated with steroid-resistance aNS in children with an AUC of 87%. Ye et al. [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e] constructed a prediction model in children to evaluate the internal conditions and disease status of patients with steroid-resistant aNS. They used 8 clinical variables such as erythrocyte sedimentation rate, urine occult blood, percentage of neutrophils, immunoglobulin A, cholesterol, vinculin autoantibody, etc., and the model had an accuracy of 94%. Ye et al. and Kou et al. used clinical variables and not BIA outcomes and had a total of 91 and 111 subjects with aNS and steroid-resistant aNS recruited. Another study group used a dataset of 2,520 participants BIA data in a ML to optimize the intracellular fluid prediction with success in healthy control group [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eResearch in healthcare ML and artificial intelligence is rapidly gaining momentum, revealing potential applications across a range of medical fields [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e], such as interpreting chest radiographs [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], detecting cancer in mammograms [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], or detecting arrhythmias [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Until now, very few well-validated ML models, usually developed from large datasets, have been deployed successfully in clinical practice [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. The potential for faster data analysis and a more personalised treatment plans for patients with aNS could potentially be reached by employing ML and ultimately improving patient outcomes and optimising healthcare resources. As suggested in Birk et al for a more precise BIA measurements enhanced with ML in an Indian population [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], resulting in a more precise estimate of body composition parameters. This pilot study assessed the feasibility of automated ML software in patients suffering from severe oedema due to their disease. Ideally, ML software such as JADBio performs best with large datasets, up to millions, in variables. The low prevalence of paediatric NS, estimated to be around 2\u0026ndash;5 per 100,000 [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], generally precludes accumulating large data sets. The small sample size was a limitation of this study, which poses a risk for over-fitting the model to the data. A larger population would increase the model\u0026rsquo;s ability to generalise, but this may not be achievable in real-life cohorts. While small cohorts may pose challenges due to limited data availability, it remains clinically important to prioritize research in this most common glomerular disease in childhood [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e\u003cp\u003eWe excluded the body weight parameter for our model because body weight does not accurately estimate fluid volume or oedema in patients with aNS [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. This introduces a potential bias towards overestimating the ML ability to help diagnose children with aNS. The study included only one girl with aNS, which is not ideal for the software to model the effects of sex. The control group of 15 girls mitigates this limitation. The ML model incorporated sex as a parameter but removing it would only reduce the model\u0026rsquo;s predictive performance by 0-2.3%. Since BIA was conducted before puberty, when body composition differences are minimal, it is less likely that sex is an important factor to consider when diagnosing NS. While internal accuracy reached 100% when using height, age, and XC50 alone in specific cross-validation configurations, the final model selected by the JADBio platform included additional features such as Z50, Zfc, PhA50, and sex. The model included PhA\u003csub\u003e50\u003c/sub\u003e, which showed a 0% drop in internal accuracy [CI 0;2.3] when excluded from the model. We used a non-aggressive feature solution due to the nature of this being a pilot study. This meant the model used all features that could contribute. This included sex and PhA\u003csub\u003e50\u003c/sub\u003e which may hold contextual or biological value. This was due to the application of a LASSO feature selection method, which allowed the inclusion of all features with marginal added value. This broader feature inclusion slightly reduced generalisation performance, as expected with small datasets.\u003c/p\u003e\u003cp\u003eAnother potential limitation was including BIA data with only three pre-determined frequencies (5, 50, and 200 kHz) and \u003cem\u003ef\u003c/em\u003e\u003csub\u003ec\u003c/sub\u003e. JADBio used the mean of two pre-determined frequencies measured by the BIA. With the pre-determined frequencies, the software was able to employ a \u0026ldquo;working\u0026rdquo; model. Still, the potential for a more precise model could be reached by including the full range of impedance data from all measurement frequencies (50 within the range of 5 to 1000 kHz) or complete frequency spectrum data obtained from Cole modelling. This would harness the potential of automated ML software to analyse large data sets and potentially enhance the true positive rate. Finally, the omission of other biomarkers such as biochemical measurements or clinical symptoms limit the model and could be included in future studies.\u003c/p\u003e\u003cp\u003eOur findings call for further research in a larger dataset of children with NS, e.g., through multicentre trials and/or data sharing, and to include additional impedance data (at more frequencies) and inclusion of clinical variables (e.g., blood chemistry). It could be considered removing some features, as the LASSO penalty was set to zero. This resulted in the inclusion of all features with the possibility to add value. Where both sex and PhA\u003csub\u003e50\u003c/sub\u003e contributed with no measurable improvement in predictive performance, meaning a 0% drop when excluded. As seen in Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, adding more features beyond a certain point can introduce noise into the model, which can then potentially lowering its cross-validated performance accuracy. This showcases a known phenomenon in ML, where adding additional variables may disturb the model\u0026rsquo;s internal structure, causing it to re-evaluate previous classifications and perform worse.\u003c/p\u003e\u003cp\u003eFurther work is needed to ascertain the importance of BIA measurements in ML models relative to clinical variables. ML models are excellent for predicting an outcome, but do not imply a causal inference [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Additional studies will be needed in these larger datasets to develop a clinically useful prediction model and investigate the causal inference of using an ML model to predict children with NS and complications in sick children.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis cross-sectional study assessed the feasibility of automated ML software with the following BIA outcomes: PhA, IR, R, Z, X\u003csub\u003eC\u003c/sub\u003e and C\u003csub\u003em\u003c/sub\u003e derived from impedance measurements obtained at specific frequencies. The ML software selected the features BIA parameters X\u003csub\u003eC\u003c/sub\u003e, Z, and PhA, and the clinical features height, age, and sex. A ridge logistic model detected and divided children with high true positive rate (91.8%) but a low specificity (21.7%) for the healthy controls and children with NSr. While this is too low to be acceptable in the clinical setting, these findings suggest that the automated ML software can distinguish the two groups and warrants further study in a larger cohort of children. JADBio is a promising tool for analysing complex biomedical data such as BIA, especially in small datasets. However, it can function as a \u0026ldquo;black box\u0026rdquo; in some situations, with limited insight into model construction and validation. This lack of transparency should be taken into account when interpreting results and evaluating their clinical relevance. If a larger study confirmed the present pilot observations, implementation of ML using BIA measures could potentially improve decision-making in selecting treatment pathways in children with NS.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eaNS\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eActive nephrotic syndrome (study group)\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eAUC\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eArea under the ROC curve\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eBIA\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eBioelectrical impedance analysis\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eBIS\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eBioimpedance spectroscopy\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eCI\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eConfidence interval\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eC\u003csub\u003em\u003c/sub\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eCell membrane capacitance\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eCV\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eCross validation\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eECW\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eExtra-cellular water\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eHt\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ePerson height in cm\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eHC\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eHealthy controls (control group)\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eICE\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eIndividual conditional expectation\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eIR\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eImpedance ratio\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eICW\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eIntra-cellular water\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eJADBio\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eJust add data bio\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eLASSO\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eLeast absolute shrinkage and selection operator\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eMachine learning\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eML, NS\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eNephrotic syndrome\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eNSr\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eNephrotic syndrome in remission (study group)\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003ePhA\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003ePhase angle\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eR\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eElectrical resistance\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eROC\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eReceiver operating characteristics\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eSD\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eStandard deviation\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eSTREAM-URO\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eStandardized reporting of machine learning application in urology\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eTBW\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eTotal body water\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eX\u003csub\u003eC\u003c/sub\u003e\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eCapacitive reactance\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003cdiv class=\"DefinitionListEntry\"\u003e\u003cdiv class=\"Term\"\u003eZ\u003c/div\u003e\u003cdiv class=\"Description\"\u003e\u003cp\u003eimpedance\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eEthics approval and consent to participate\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWritten informed consent was obtained from the subjects\u0026rsquo; parents or legal guardians before study enrolment. The study was performed following the Helsinki Declaration and was approved by the Central Denmark Region Committees on Health Research Ethics (case number: 1-10-72-17-12).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset used during the current study is available from the corresponding author on reasonable request at the following address: [email protected]\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eCompeting interests\u0026nbsp;\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eAuthor Leigh C Ward provides consultancy services to ImpediMed Ltd. ImpediMed Ltd was not involved in preparing this manuscript. Other authors have no conflicts of interest to declare concerning this work.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eNone. CLH received research funding from the Novo Nordisk Foundation (grant no. NNF22OC0074080).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/em\u003e\u003c/p\u003e\n\u003cp\u003eJQ, with assistance from SB, drafted and approved the research protocol. RFA helped with the enrolment of NS patients. JQ undertook the data analysis with assistance from LCW, LJ, and SB and prepared the manuscript. All authors participated in critically interpreting results, edited the manuscript, and approved its final content.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eTapia C, Bashir K (2025) Nephrotic Syndrome. StatPearls. StatPearls Publishing Copyright \u0026copy; 2025. StatPearls Publishing LLC.. Treasure Island (FL)\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrantlov S, Andersen T, J\u0026oslash;dal L, Rittig S, Lange A (2016) Bioimpedance Spectroscopy in Healthy Children. J Clin Eng 41:33\u0026ndash;39\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNorman K, Stob\u0026auml;us N, Pirlich M, Bosy-Westphal A (2012) Bioelectrical phase angle and impedance vector analysis\u0026ndash;clinical relevance and applicability of impedance parameters. Clin Nutr 31:854\u0026ndash;861\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGipson DS, Pal M, Desmond H, Anderson C, Walsh L, Trachtman H, Massengill SF, Gipson P, Rao PS, Thurman J, Kopp J, Kamil E, Lamothe J, Mariani LH, LaFleur P, Vento S, O\u0026rsquo;Shaughnessy M, Farag YMK, Simon C, Carlozzi NE (2023) Developing an Edema Clinician-Reported Outcome Measure for Nephrotic Syndrome. Glomerular Dis 3:132\u0026ndash;139\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMulasi U, Kuchnia AJ, Cole AJ, Earthman CP (2015) Bioimpedance at the bedside: Current applications, limitations, and opportunities. Nutr Clin Pract 30:180\u0026ndash;193\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrantlov S, J\u0026oslash;dal L, Lange A, Rittig S, Ward LC (2017) Standardisation of bioelectrical impedance analysis for the estimation of body composition in healthy paediatric populations: a systematic review. J Med Eng Technol 41:460\u0026ndash;479\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrantlov S, Ward LC, J\u0026oslash;dal L, Rittig S, Lange A (2017) Critical factors and their impact on bioelectrical impedance analysis in children: a review. J Med Eng Technol 41:22\u0026ndash;35\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGartner A, Maire B, Delpeuch F, Sarda P, Dupuy RP, Rieu D (1992) Importance of electrode position in bioelectrical impedance analysis. Am J Clin Nutr 56:1067\u0026ndash;1068\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCollins CT, Reid J, Makrides M, Lingwood BE, McPhee AJ, Morris SA, Gibson RA, Ward LC (2013) Prediction of body water compartments in preterm infants by bioelectrical impedance spectroscopy. Eur J Clin Nutr 67(Suppl 1):S47\u0026ndash;53\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVanderJagt DJ, Huang YS, Chuang LT, Bonnett C, Glew RH (2002) Phase angle and n-3 polyunsaturated fatty acids in sickle cell disease. Arch Dis Child 87:252\u0026ndash;254\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNagano M, Suita S, Yamanouchi T (2000) The validity of bioelectrical impedance phase angle for nutritional assessment in children. J Pediatr Surg 35:1035\u0026ndash;1039\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVerheij R, Veenman CS, den Bakker JV, van Duyl WA (1996) Comments on the interpretation of tissue impedance measurements during hemodialysis. Blood Purif 14:8\u0026ndash;14\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChertow GM, Jacobs DO, Lazarus JM, Lew NL, Lowrie EG (1997) Phase angle predicts survival in hemodialysis patients. J Ren Nutr 7:204\u0026ndash;207\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGarr Barry V, Peterson CM, Gower BA (2020) Membrane Capacitance from a Bioimpedance Approach: Associations with Insulin Resistance in Relatively Healthy Adults. Obes (Silver Spring) 28:2184\u0026ndash;2191\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBosy-Westphal A, Danielzik S, D\u0026ouml;rh\u0026ouml;fer RP, Later W, Wiese S, M\u0026uuml;ller MJ (2006) Phase angle from bioelectrical impedance analysis: population reference values by age, sex, and body mass index. JPEN J Parenter Enter Nutr 30:309\u0026ndash;316\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNematollahi MA, Jahangiri S, Asadollahi A, Salimi M, Dehghan A, Mashayekh M, Roshanzamir M, Gholamabbas G, Alizadehsani R, Bazrafshan M, Bazrafshan H (2023) Bazrafshan drissi H, Shariful Islam SM Body composition predicts hypertension using machine learning methods: a cohort study. Scientific Reports 13:6885\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKwong JCC, McLoughlin LC, Haider M, Goldenberg MG, Erdman L, Rickard M, Lorenzo AJ, Hung AJ, Farcas M, Goldenberg L, Nguan C, Braga LH, Mamdani M, Goldenberg A, Kulkarni GS (2021) Standardized Reporting of Machine Learning Applications in Urology: The STREAM-URO Framework. Eur Urol Focus 7:672\u0026ndash;682\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrantlov S, J\u0026oslash;dal L, Andersen RF, Lange A, Rittig S, Ward LC (2019) An evaluation of phase angle, bioelectrical impedance vector analysis and impedance ratio for the assessment of disease status in children with nephrotic syndrome. BMC Nephrol 20:331\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRinninella ECM, Addolorato G, Triarico S, Ruggiero A et al (2018) Phase Angle and impedance ratio: Two specular ways to analyze body composition. Ann Clin Nutr 1:1003\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTsamardinos I, Charonyktakis P, Papoutsoglou G, Borboudakis G, Lakiotaki K, Zenklusen JC, Juhl H, Chatzaki E, Lagani V (2022) Just Add Data: automated predictive modeling for knowledge discovery and feature selection. npj Precision Oncol 6:38\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDonders AR, van der Heijden GJ, Stijnen T, Moons KG (2006) Review: a gentle introduction to imputation of missing values. J Clin Epidemiol 59:1087\u0026ndash;1091\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHern\u0026aacute;n MA, Hsu J, Healy B (2019) A Second Chance to Get Causal Inference Right: A Classification of Data Science Tasks. CHANCE 32:42\u0026ndash;49\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHenderson AR (2005) The bootstrap: a technique for data-driven statistics. Using computer-intensive analyses to explore experimental data. Clin Chim Acta 359:1\u0026ndash;26\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKou M, Wu F, Qu XY, Wang H, Guo XT, Yang YY, Zhao LJ (2023) [Establishment and validation of clinical prediction model for steroid-resistant nephrotic syndrome in children]. Zhonghua Er Ke Za Zhi 61:333\u0026ndash;338\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYe Q, Li Y, Liu H, Mao J, Jiang H (2023) Machine learning models for predicting steroid-resistant of nephrotic syndrome. Front Immunol 14:1090241\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNovera JN, Ali ME, Mia MMA, Haque A, Fahim T, Jahan A (2025) I. Machine Learning-Driven Optimization of Bioelectrical Impedance Analysis for Intracellular Fluid Prediction. International conference on engineering research, innovation and education, Shahjalal university of science and technology, sylhet, Bangladesh\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKelly CJ, Karthikesalingam A, Suleyman M, Corrado G, King D (2019) Key challenges for delivering clinical impact with artificial intelligence. BMC Med 17:195\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGupta S, Vargas A, Saulnier G, Newell J, Faaborg-Andersen C, Kelley RS (2021) Uterine bioimpedance combined with artificial intelligence as a means of cancer detection. J Med Eng Technol 45:606\u0026ndash;613\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, Ng AY (2019) Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med 25:65\u0026ndash;69\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVistisen ST, Pollard TJ, Harris S, Lauritsen SM (2022) Artificial intelligence in the clinical setting: Towards actual implementation of reliable outcome predictions. Eur J Anaesthesiol 39:729\u0026ndash;732\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBirk N, Kulkarni B, Bhogadi S, Aggarwal A, Walia GK, Gupta V, Rani U, Mahajan H, Kinra S, Mallinson PAC (2025) Machine learning-based equations for improved body composition estimation in Indian adults. PLOS Digit Health 4:e0000671\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTamura H (2021) Trends in pediatric nephrotic syndrome. World J Nephrol 10:88\u0026ndash;100\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSanjad SA, Ulinski T, Aoun B (2021) Editorial: Nephrotic Syndrome in Children. Front Pead 9\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Bioelectrical impedance, electric capacitance, machine learning, nephrotic syndrome, children","lastPublishedDoi":"10.21203/rs.3.rs-7197037/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7197037/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e\u003cp\u003eNephrotic syndrome (NS) in children, entailing kidney-related protein leakage and peripheral oedema, remains difficult to assess. Bioelectrical impedance analysis (BIA) provides several body composition measures, and integration of machine learning (ML) may improve clinical care. We tested an ML model to identify NS in children, compared with healthy children.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e\u003cp\u003eThis was a cross-sectional study, conducted on children with active NS in the acute phase (aNS group) included from the Department of Paediatrics and Adolescent Medicine, Aarhus University Hospital, Denmark. Anonymized MF-BIA data from frequences between 5-1000 kHz were added to the JustAddDataBio (JADBio)\u0026reg;, a web-based ML platform for analysing potential biomarkers to diagnose.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e\u003cp\u003eEight children with aNS and 38 age-matched healthy children were included. The ML software employed a ridge logistic regression with the penalty hyperparameter lambda\u0026thinsp;=\u0026thinsp;0.001, with a selected threshold of 0.81 by JADBio, and the area under the curve (AUC) was 0.84 [95% confidence interval (CI): 0.72;0.94] as the best model. 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 had a statistically significant true positive classification of a healthy child of 0.92 (92%) [CI: 0.88;0.96], and a specificity of 0.22 (22%) [CI: 0.08;0.36].\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e\u003cp\u003eApplying an ML-supported evaluation of BIA improved diagnostics. A low specificity limits the clinical application. To obtain a more acceptable model, a larger population of patients and the inclusion of more biomarkers may be needed.\u003c/p\u003e","manuscriptTitle":"Integrating machine learning for advanced analysis of bioelectrical impedance parameters in children with nephrotic syndrome: phase angle, impedance ratio, and cell membrane capacitance","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-31 10:18:46","doi":"10.21203/rs.3.rs-7197037/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"6a868233-f4ee-4ae9-b983-671de9867c07","owner":[],"postedDate":"July 31st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-08-11T20:22:17+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-31 10:18:46","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7197037","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7197037","identity":"rs-7197037","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}