Vitamin D3 for Life

Live Enlightened

Research libraryVitamin D3 For Life
Study design and methods

Secondary relevance to vitamin D This paper mainly addresses another topic and also reports a vitamin D analysis, result or scientific synthesis.

This paper mainly addresses another topic and also reports a vitamin D analysis, result or scientific synthesis.

Why this paper is included

Supporting source evidence:

Most frequent predictors included: serum erythrocyte sedimentation rate (ESR) (n = 3), vitamin D-binding protein (VDBP) (n = 3), age (n = 2), serum D-dimer (n = 2), serum β2-microglobulin (n = 2), serum white blood cell (n = 2), serum cholesterol (n = 2), serum creatinine (n = 2), serum suppressor T cells (n = 2), urinary α-1-acid glycoprotein (n = 2), urinary retinol-binding protein (n = 2), urinary albumin (n = 2), urinary fetuin-A (n = 2), and urinary neutrophil gelatinase-associated lipocalin (n = 2).

Source: 3.1. Results of Prediction Models. View original paper

Relevance classification does not establish study quality, causation or editorial approval.

Study design Systematic review.

Systematic review

Uses explicit methods to identify, assess and summarise research on a defined question.

Meta-analysis reported · Blinding not applicable to this review · Research setting awaiting verification

How this paper was classified

Automatic source-based classification; awaiting editorial verification. This is not a quality score.

Prediction Models and Risk Factors for Steroid Resistance in Children with Nephrotic Syndrome: A Systematic Review and Meta-Analysis

Source: Article title.

Prediction Models and Risk Factors for Steroid Resistance in Children with Nephrotic Syndrome: A Systematic Review and Meta-Analysis

Source: Article title.

View supporting evidence in the original paper

What the study types mean
Systematic review
Uses explicit methods to identify, assess and summarise research on a defined question.
Scoping review
Maps the extent and characteristics of research on a topic.
Narrative review
Discusses existing research; systematic-review methods have not been established by this classification.
Review / evidence synthesis
Reviews existing research; the exact review methods need further assessment.
Randomised controlled trial
Assigns participants or groups to comparison interventions by chance.
Non-randomised intervention study
Examines an intervention without random allocation.
Cohort study
Examines exposures and outcomes over time within a defined group.
Case-control study
Compares people with an outcome against people without it, examining exposures.
Cross-sectional study
Examines a population at a particular point in time.
Observational study
Observes exposures and outcomes without assigning the exposure; the subtype is not established here.
Case report
Describes an individual case.
Case series
Describes a group of individual cases.
Secondary analysis
Analyses existing study data; this paper is not automatically another independent trial.
Laboratory study
Investigates biological processes in a laboratory setting.
Qualitative study
Examines experiences or meanings using qualitative methods.
Genetic instrumental-variable study
Uses genetic variation as an instrument to investigate relationships; this is not a randomised treatment trial.
Study protocol
Describes planned methods and is not evidence that the planned study has reported results.
Multiple reported designs
More than one design is explicitly reported; examine the supporting methods.
Study design awaiting verification
The available source does not establish a sufficiently clear study design.

Meta-analysis combines numerical results from studies. Blinding describes who was unaware of treatment allocation. Neither is a separate quality grade. Study design, vitamin D relevance and document format are independent filters.

Full studies

Prediction Models and Risk Factors for Steroid Resistance in Children with Nephrotic Syndrome: A Systematic Review and Meta-Analysis

Systematic review · Journal of Clinical Medicine · 2026

Original publication title

Prediction Models and Risk Factors for Steroid Resistance in Children with Nephrotic Syndrome: A Systematic Review and Meta-Analysis

Yuanhui Hu; Zehui Zhang; Sha Diao; Yannan Guo; Yangtingting Gao; Zheng Xu; Qilin Peng; Yao Xu; Zhenyan Bo; Linan Zeng; Liang Huang; Jingjing Chen; Yizhun Zhu; Hailong Li; Lingli Zhang

© 2026 by the authors.

Original publication · Publisher record · CC BY 4.0

Source wording retained. Formatting, section navigation and display-title punctuation adapted for this research record.

Authors’ conclusion

This systematic review summarized multivariable prediction models and independent risk factors for childhood SRNS. Although existing models reported apparent discrimination, most had a high risk of bias, lacked external validation, and included predictors that are not routinely available in clinical practice, indicating that their current clinical applicability remains very limited. The identified risk factors may provide preliminary clues for future candidate predictor selection, but their incremental predictive value and clinical feasibility require further evaluation. Future studies should prioritize clinically accessible predictors, rigorous model development, multi-center external validation, calibration assessment, decision curve analysis, and interpretable model presentation to improve clinical utility and generalizability.

An extract from the source publication. Read it in context.

At a glance

Study background

What background does the paper provide?

Pediatric primary nephrotic syndrome (PNS) is a common glomerular disorder marked by severe proteinuria, low serum protein, edema, and elevated lipid levels, with a reported prevalence between 1.15 and 16.9 per 100,000 [1]. Standard treatment involves glucocorticoids, yet around 10 to 20% of patients respond poorly [1], resulting in a diagnosis of steroid-resistant nephrotic syndrome (SRNS).

The response of children with PNS to glucocorticoid therapy is a critical factor in determining prognosis. By 10 years after diagnosis, the estimated risk of progression to chronic kidney disease among patients with steroid-sensitive nephrotic syndrome (SSNS) is below 5% [2], whereas the 10-year renal survival without progression to end-stage kidney disease among patients with SRNS has been reported to range between 35% and 51% [3]. No evidence supports the efficacy of long-term glucocorticoid therapy for SRNS patients; however, it may cause cumulative steroid-related adverse effects [4], and approximately 70% of SRNS patients may face a high risk of disease progression due to long-term ineffective hormone therapy [5].

The prediction of glucocorticoid response in nephrotic syndrome can assist clinicians in identifying potential hormone resistance in children with PNS and guide the development of optimal treatment strategies in clinical practice, thus shortening the time to urine protein remission and reducing both the length of hospital stay and treatment costs [6]. Furthermore, the guidelines from the Kidney Disease: Improving Global Outcomes (KDIGO) organization recommend assessing patient prognosis based on the initial response to glucocorticoid therapy to minimize unnecessary renal biopsies [7].

Although various SRNS prediction models have been developed, these models vary in methodology, performance, and applicability, and no study has systematically compared model performance across studies applied to SRNS. Additionally, the predictors across studies vary significantly, resulting in uncertainty regarding their clinical applicability.

Systematic reviews have highlighted numerous biomarkers available for predicting SRNS [8,9]; however, these studies often rely on univariate analysis or correlation-based approaches, which fail to identify predictors with robust associations. Additionally, many of the studies included in these reviews based their analyses on data measured at the time of SRNS diagnosis, which may not have accurately captured the predictive potential of biomarkers over time. This approach limited the ability to establish reliable prediction models, as it overlooked the dynamic nature of disease progression, and may have led to overestimation of associations that are not causally linked.

This study aims to comprehensively review existing multivariable prediction models for SRNS and assess their methodological quality. It also summarizes reported risk factors associated with SRNS to inform candidate predictor selection in future prediction model development.

Source: 1. Introduction Source: 1. Introduction Source: 1. Introduction Source: 1. Introduction Source: 1. Introduction Source: 1. Introduction

Study methods

What methods does the paper describe?

This review was prospectively registered with PROSPERO (CRD420250651875). Reporting adhered to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis for Systematic Reviews and Meta-Analysis [10], as well as the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [11]. The details of self-reports are provided in Supplementary Tables S1 and S2.

2.1. Searching Strategies

Two researchers (YH. Hu and ZH. Zhang) independently searched the China National Knowledge Infrastructure, the Chinese Biomedical Literature Database, Wanfang Database, VIP Database, PubMed, Embase, and Scopus databases. We searched for studies developing prediction models for SRNS or identifying associated risk factors. Search terms included “Nephrotic Syndrome”, “Resistan*”, “Steroid”, “Predict*”, and “Risk*”. The full search strategies are presented in Supplementary Table S3. Searches were conducted without language restrictions and included records from the inception of each database through 1 March 2025.

2.2. Inclusion and Exclusion Criteria

For predictive models, we included studies that described the development or validation of multivariable SRNS prediction models in children. Eligibility was determined according to the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis statement [12] for transparent reporting of multivariable prediction model studies.

Inclusion criteria include: (1) Study population: Children diagnosed with primary nephrotic syndrome, regardless of initial diagnosis or recurrence status, and with no restrictions on age, gender, or race; SRNS was defined as persistent proteinuria after 4 to 8 weeks of adequate full-dose prednisone or prednisolone therapy [1,4]; (2) Intervention: Each included study reported the development or external validation of a prediction model for glucocorticoid response in pediatric patients with nephrotic syndrome; (3) Comparison: Not performed, as no predictive model has been widely validated across multiple studies; (4) Outcome: Glucocorticoid treatment response; (5) Study design: We included cohort studies (prospective or retrospective), case-control studies, randomized controlled trials, and cross-sectional studies in this review; (6) Time: All predictors must be measured prior to the assessment of glucocorticoid treatment response, specifically at the initial diagnosis of nephrotic syndrome.

Exclusion criteria include: (1) Studies solely focusing on genetic or renal biopsy indicators, due to the KDIGO guidelines explicitly stating that routine genetic testing or renal biopsy is not recommended for newly diagnosed patients [4]; (2) Studies collecting or measuring variables after confirmation of glucocorticoid treatment response; (3) Preprints; (4) duplicate publications and studies were not available.

For risk factors, we included studies that examined predictors of glucocorticoid response in children with primary nephrotic syndrome using multivariable regression analysis following univariate screening. Like the predictive models, risk factors must also be assessed before the measurement of glucocorticoid treatment response. No limitations were applied regarding study design, and the exclusion criteria were the same as those used for prediction model studies.

2.3. Selection of Reviews

Two investigators (YH. Hu and ZH. Zhang) independently screened titles and abstracts, and subsequently reviewed the full texts of potentially eligible studies to assess which articles would be included. Discrepancies were resolved by a third researcher (HL. Li) through a consensus discussion.

2.4. Data Extraction

Two authors (YH. Hu and ZH. Zhang) independently performed data extraction, and any discrepancies were resolved by a third author (HL. Li). The following data were extracted from all included studies: first author, publication year, country/region, and model name (where applicable). Corresponding authors were contacted to obtain missing information/data. For predictive model research, a standardized form was constructed using the CHARMS checklist [13]. The extracted information from studies reporting model development included study design, study population, outcome measures, candidate predictors, modeling approach, internal validation method, numbers of patients and events, handling of missing and continuous variables, number and type of predictors in the final model, model presentation, and performance metrics (classification, discrimination, calibration, and clinical utility).

Classification performance was assessed by sensitivity, specificity, positive predictive value, and negative predictive value; discrimination was measured using the C-index; calibration was examined with the Hosmer-Lemeshow test and calibration plots; and clinical utility was evaluated through decision curve analysis and net benefit metrics. For external validation studies, extracted information included study population, number of patients and events, and measures of predictive performance. For studies reporting multiple models, information was extracted independently for each individual model. When studies reported both overall and subgroup-specific model performance, only overall population analyses were extracted.

For risk factor studies, the extracted information included the study design, study population, number of patients, analysis methods, and the results of the analysis.

2.5. Assessment of Methodological Quality of Included Studies

Two authors (YH. Hu and ZH. Zhang) independently evaluated the risk of bias for the included prediction models using the Prediction Model Risk of Bias Assessment Tool (PROBAST) [14], and similarly applied the Newcastle-Ottawa Scale [15] to the observational studies. Any discrepancies between reviewers were resolved through discussion with a third author (HL. Li).

2.6. Data Synthesis and Analysis

Baseline characteristics were summarized using mean ± standard deviation or median (interquartile range, IQR) for continuous variables, and expressed as percentages for categorical variables. For prediction models that had been validated in more than two independent datasets, we conducted a random-effects meta-analysis using the area under the curve (AUC) and 95% confidence intervals (CI) as the primary metrics. A 95% CI including 0.5 was considered indicative of insufficient discrimination. For risk factors, we applied a random-effects meta-analysis to combine odds ratios (OR) with their corresponding 95% CI.

Heterogeneity (I2) was assessed only for pooled analyses. Funnel plots were generated to evaluate publication bias when ten or more studies reported comparable outcomes. Two-tailed statistical tests were conducted with a significance level of p < 0.05. All analyses were performed using SPSS version 18.0 and RevMan version 5.3.

2.7. Sensitivity Analysis

Sensitivity analyses were planned separately for prediction model studies and risk factor studies. For prediction model studies, if the same model was externally validated in more than five independent datasets, sensitivity analysis would be conducted by excluding studies at high risk of bias and re-pooling the model performance estimates. For risk factor studies, if at least five studies reported the same risk factor with comparable definitions, sensitivity analysis would be conducted by excluding studies at high risk of bias and re-pooling the effect estimates.

Source: 2. Materials and Methods Source: 2. Materials and Methods Source: 2. Materials and Methods Source: 2. Materials and Methods Source: 2. Materials and Methods Source: 2. Materials and Methods Source: 2. Materials and Methods Source: 2. Materials and Methods

Results: Results of Prediction Models

Thirteen studies encompassed the development of 17 prediction models and the external validation of one model. Model development occurred primarily in China (n = 9), followed by the United States (n = 6). All development and validation utilized retrospective electronic health records from hospitalized cohorts. Across development cohorts, median sample size was 56 (IQR 41.5 to 116.5), with 30 events per variable (IQR 15.5 to 38.5) and 8.5 predictors per model (IQR 4.5 to 15.3). The external validation cohort contained 50 patients with 20 events and the events per variable was 2.5.

Eight studies evaluated patients’ glucocorticoid response based on whether urinary protein turned negative after 4 weeks of glucocorticoid administration (Table 1). Forty four predictors were documented. Most frequent predictors included: serum erythrocyte sedimentation rate (ESR) (n = 3), vitamin D-binding protein (VDBP) (n = 3), age (n = 2), serum D-dimer (n = 2), serum β2-microglobulin (n = 2), serum white blood cell (n = 2), serum cholesterol (n = 2), serum creatinine (n = 2), serum suppressor T cells (n = 2), urinary α-1-acid glycoprotein (n = 2), urinary retinol-binding protein (n = 2), urinary albumin (n = 2), urinary fetuin-A (n = 2), and urinary neutrophil gelatinase-associated lipocalin (n = 2). Supplementary Table S4 summarizes the predictors and performance of each model.

Models were predominantly developed using logistic regression (n = 16, 94.1%), followed by machine learning approaches (n = 1, 5.9%). Two models were respectively internally validated using bootstrap sampling and cross-validation methods. No model addressed missing values via interpolation. Most models (n = 7, 41.2%) directly excluded samples with missing data. All studies reported AUC values ranging from 0.75 to 0.88. Given that no predictive model has been externally validated in more than two cohorts, meta-analysis of AUC was not feasible.

Four models (23.5%) reported calibration via Hosmer-Lemeshow tests, while decision curve analysis evaluated clinical utility in two models (11.8%). The most common model presentation formats were mathematical equations (n = 6, 35.3%), followed by nomograms (n = 2, 11.8%) and web-based calculators (n = 2, 11.8%). One model that underwent external validation within the same healthcare system demonstrated a sensitivity of 0.9, specificity of 0.97, and accuracy of 0.94.

Overall, 76.92% of the models exhibited a high risk of bias (Supplementary Table S5), with the primary source of high risk of bias being the analytical domain (76.92%), particularly due to inadequate sample sizes (46.15%) or improper handling of missing data (38.46%). Details of the PROBAST risk-of-bias assessment are provided in Supplementary Table S6. These findings indicate that the reported model performance estimates are likely to be unstable and potentially optimistic.

Table 1

Basic characteristics of the included studies on SRNS prediction models.

StudyCountryStudy AimCohort TypeSR Cases/Total PatientsAge (Years)Outcome Definition
Agrawal 2021 [16]USADEHR14/407.0 (mean)NP-7 weeks
Agrawal 2020 [17]USADEHR13/378.7 (mean)NP-7 weeks
Bennett 2017 [18]USADEHR20/5012.3 (mean)NP-8 weeks
Burlaka 2022 [19]UkraineDEHR30/5611.0 (mean)NP-6 weeks
Gooding 2020 [20]USADEHR14/399.5 (mean)NP-6 weeks
Hong 2022 [21]ChinaDEHR41/1743.0 (median)NP-4 weeks
Jiang 2019 [22]ChinaDEHR16/795.5 (mean)NP-4 weeks
Kou 2023 [23]ChinaDEHR65/1113.2 (median)NP-4 weeks
Ling 2019 [24]ChinaDEHR44/1225.8 (mean)NP-4 weeks
Liu 2023 [25]ChinaDEHR31/1579.0 (mean)NP-4 weeks
Wang 2012 [26]ChinaDEHR19/434.5 (mean)NP-4 weeks
Ye 2023 [5]ChinaD, EVEHR34/91, 20/504.0 (median)NP-4 weeks
Zhang 2015 [27]ChinaDEHR36/1856.0 (median)NP-4 weeks

Footnote: D: derivation; EHR: electronic health record; EV: external validation; NP: Negative Proteinuria.

Source: 3.1. Results of Prediction Models Source: 3.1. Results of Prediction Models Source: 3.1. Results of Prediction Models Source: 3.1. Results of Prediction Models Source: 3.1. Results of Prediction Models

Results: Results of Risk Factors

The 19 included studies were published between 2012 and 2024, with the majority conducted in China (n = 8). The total number of patients across all studies was 2514. The study designs primarily included case-control studies (n = 15), retrospective cohort studies (n = 1), and prospective cohort studies (n = 3). All studies employed multivariate logistic regression to identify independent risk factors for SRNS. Ten studies evaluated patients’ glucocorticoid response based on whether urinary protein turned negative after 4 weeks of glucocorticoid administration (Table 2).

A total of 52 risk factors were analyzed across all studies, with 22 independent risk factors ultimately identified (Figure 2). The corresponding forest plots are presented in Supplementary Figure S1, and the definition of each pooled risk factor is provided after the relevant forest plot. Only a small subset of these independent risk factors could be quantitatively pooled; the remainder were reported descriptively. The 22 independent risk factors included: demographic/underlying disease-related factors (n = 3): age, low birth weight, hypertension; biochemical indicators (n = 9): white blood cells, cholesterol, ESR, albumin, D-dimer, globulin, proteinuria, decreased urine output, hematuria; immune/inflammatory factors (n = 4): suppressor T cells, immunoglobulin E, immunoglobulin M, immunoglobulin G; metabolic biomarkers (n = 6): polyligand proteoglynan-1, glutamine, endothelin-1, retinol-binding protein, α1-microglobulin, β2-microglobulin.

Overall, among the 19 observational studies, 14 were assessed as high-quality and 5 as medium-quality (Supplementary Table S7). None of the pooled analyses included 10 or more studies; therefore, funnel plots were not generated and publication bias could not be formally assessed.

Table 2

Basic characteristics of the included studies on SRNS-associated risk factors.

StudyCountryStudy DesignNumber of ParticipantsOutcome DefinitionsIndependent Risk Factors
Agrawal 2021 [16]USAR40NP-7 weeksSerum: interleukin-7, interleukin-9, monocyte chemoattractant protein 1
Agrawal 2020 [17]USAR37NP-7 weeksSerum: adiponectin, matrix metalloproteinase 2, vitamin D binding protein
Cicek 2024 [28]TurkeyR162NP-8 weeksAge
Urine: microscopic hematuria
Cuzzoni 2019 [29]ItalyP62NP-4 weeksAge
Serum: macrophage migration inhibitory factor
Gooding 2020 [20]USAR39NP-6 weeksAge
Serum: glutamine
Hong 2022 [21]ChinaC174NP-4 weeksSerum: white blood cell
Urine: retinol-binding protein
Imbusi 2020 [30]EthiopiaC63NP-8 weeksAge (>6), hypertension
Serum: cholesterol
Kifle 2020 [31]EthiopiaC85NP-4 weeksUrine: gross hematuria, urine output
Konstantelos 2019 [32]CanadaP377Not reportedlow birth weight
Kou 2023 [23]ChinaC111NP-4 weeksSerum: β-microglobulin, D-dimer, erythrocyte sedimentation rate, suppressor T cells
Li 2024 [33]ChinaC510NP-8 weeksPathology, age
Serum: 25-hydroxyvitamin D, CD4+ T cells, creatinine, D-dimer
Urine: proteinuria
Ling 2019 [24]ChinaC122NP-4 weeksSerum: immunoglobulin E, immunoglobulin G
Liu 2023 [25]ChinaC157NP-4 weeksSerum: endothelin-1, polyligandproteoglynan-1
Rehman 2022 [34]PakistanR135Not reportedHypertension
Urine: hematuria
Salah 2023 [35]EgyptP60NP-8 weeksAge
Serum: white blood cell
Udagawa 2021 [36]JapanC80NP-4 weeksGender
Serum: immunoglobulin M
Wang 2012 [26]ChinaC43NP-4 weeksSerum: globulin
Urine: α1-antitrypsin/creatinine, αl-microglobulin
Yin 2010 [37]ChinaC72NP-4 weeksAge
Serum: albumin, D-dimer
Zhang 2015 [27]ChinaC185NP-4 weeksAge (>6.5)
Urine: microscopic hematuria, proteinuria,

Footnote: C: Case-control study; NP: Negative Proteinuria; P: Prospective cohort study; R: Retrospective cohort study.

Source: 3.2. Results of Risk Factors Source: 3.2. Results of Risk Factors Source: 3.2. Results of Risk Factors Source: 3.2. Results of Risk Factors

Results: Results of Sensitivity Analysis

Sensitivity analyses were not conducted because the prespecified criteria were not met: prediction models had insufficient external validation studies, and age, the only risk factor reported in at least five studies, included no high-risk-of-bias study for exclusion.

Source: 3.3. Results of Sensitivity Analysis

Results: Cross-Comparison and Analysis

Cross-referencing revealed 17 variables serving as both predictive factors and independent risk factors. In addition, five variables were reported as independent risk factors but have not been incorporated into the published multivariable prediction models, including low birth weight, reduced urine output, hypertension, serum albumin, and serum immunoglobulin M (Figure 3).

Source: 3.4. Cross-Comparison and Analysis

Discussion

This systematic review evaluated 17 predictive models for childhood SRNS and identified 22 associated independent risk factors. Despite the reported apparent discriminative performance (AUC range: 0.75 to 0.88; AUC > 0.70 indicating acceptable discrimination [38]), these estimates should be interpreted with substantial caution because most models were at high risk of bias, were developed in small retrospective cohorts, and lacked adequate internal or external validation. First, only two models underwent internal validation and one model underwent external validation, which may lead to overestimation of model performance due to overfitting [39] and limit the generalizability of models across populations with different characteristics [40].

Therefore, the reported AUC values should not be interpreted as evidence of reliable clinical prediction; the reliability, robustness, and clinical applicability of existing models remain very limited at present. Second, missing data were often inadequately handled (frequently by complete-case analysis), which may bias estimates and reduce predictive accuracy [41]. Third, calibration was reported in fewer than one-quarter of models, and decision-curve analysis was rarely performed, constraining assessment of clinical usefulness [42]. Fourth, decision curve analysis quantifies clinical net benefit post-implementation [43], yet only two studies implemented this analysis.

Finally, many models lacked a clear presentation format, which may impede translation into routine practice [44].

This study found that ESR and VDBP measured at disease onset were the most frequently included predictors in SRNS prediction models. In nephrotic syndrome, ESR is often markedly elevated at presentation, largely due to increased circulating acute-phase reactants (e.g., fibrinogen and other pro-aggregatory proteins) and altered plasma protein composition that promotes erythrocyte aggregation [45,46]. Such elevated baseline ESR may reflect a greater inflammatory burden and more active immune dysregulation at disease onset, underlying its potential predictive value.

VDBP may reflect urinary protein loss, tubular injury, or inflammation-related renal damage, which may explain why it has been evaluated as a potential predictor [47,48].

Among variables frequently incorporated during model development, approximately 50% are not readily obtainable through routine clinical laboratory testing. This requirement for specialized reagent procurement consequently constrains their applicability in resource-limited settings, including primary care facilities.

Risk factors should not be considered equivalent to predictors in clinical prediction models, but they may provide clues for identifying candidate predictors that require further validation. This review found that five factors: low birth weight, reduced urine output, hypertension, serum albumin, and serum immunoglobulin M: were reported as independent risk factors but have not yet been incorporated into published multivariable prediction models. Due to the observational design and risk of bias of included studies, these associations are hypothesis-generating. Low birth weight is a well-established risk factor for SRNS, as studies have shown that the incidence of SRNS in low-birth-weight children is 3.78 times that in children with normal birth weight; for every 100 g increase in birth weight, the risk of SRNS decreases by approximately 8% [32].

A potential explanation is that low-birth-weight children, due to incomplete renal development and reduced nephron number, are prone to compensatory impairment of renal function, thereby increasing the risk of SRNS and exhibiting a higher recurrence rate and more complications [49,50,51]. Hypertension is another clinical marker that commonly accompanies nephrotic syndrome. It results from fluid retention due to glomerular damage, which, in turn, exacerbates kidney injury and promotes a vicious cycle of further renal deterioration [52,53].

Reduced urine output may serve as an early warning sign, indicating a decline in renal filtration capacity and worsening renal function, both of which may increase the risk of SRNS progression [31]. These markers not only reflect the severity of NS at diagnosis but also provide critical insight into the potential for steroid resistance, making them useful for early stratification. Evidence from the included multivariable analyses suggested an association between higher serum albumin levels and SRNS. This finding should be interpreted cautiously because the direction and mechanism of this association remain uncertain.

Serum albumin may reflect differences in disease phenotype, proteinuria selectivity [7,54], or underlying histopathology [55], but its value as a candidate predictor requires further validation. In this review, higher serum IgM was identified as an independent risk factor for SRNS. Although biopsy-derived predictors were outside our eligibility criteria, biopsy-based studies have reported that glomerular immunoglobulin M correlates with the severity of kidney injury, suggesting an elevated risk of disease recurrence and protracted remission periods.

Furthermore, children with immunoglobulin M positivity demonstrate enhanced therapeutic responsiveness to cyclosporine treatment, which provides further evidence for the role of immune factors in steroid resistance [56,57]. Finally, descriptive analyses in several studies identified additional potential independent risk factors, including pathological types, peripheral blood CD4+T cell levels, and serum macrophage migration inhibitory factor, which were not represented among predictors in the published multivariable models included in this review.

The primary strength of this study is its systematic synthesis and evaluation of childhood SRNS prediction models, as well as the identification of independent risk factors associated with SRNS development, thereby providing an evidence-based reference for optimizing future prediction models. However, this study has several limitations. First, most prediction model studies were at high risk of bias, mainly due to limitations in the analysis domain, including small sample sizes, limited validation, and inadequate handling of missing data. Therefore, the reported model performance and identified predictors should be interpreted cautiously.

Second, due to the scarcity of externally validated prediction models, meta-analysis and sensitivity analysis of model performance could not be performed. Third, relatively few studies were available for each risk factor, and most associations were supported by only one or two studies, limiting the robustness of pooled estimates and precluding sensitivity analyses. Finally, substantial heterogeneity in SRNS definitions, populations, laboratory methods, and adjustment models may have affected the reliability of the synthesized findings.

Source: 4. Discussion Source: 4. Discussion Source: 4. Discussion Source: 4. Discussion Source: 4. Discussion

Results

A total of 2264 records were retrieved. After screening titles and abstracts, 23 unique studies were included, comprising 13 predictive model studies and 19 risk factor studies with an overlap of nine studies between the two categories. This overlap is attributed to the fact that these nine studies developed predictive models by adopting multivariate analysis methods to screen and incorporate key risk factors.

The literature screening process is presented in Figure 1.

3.1. Results of Prediction Models

Thirteen studies encompassed the development of 17 prediction models and the external validation of one model. Model development occurred primarily in China (n = 9), followed by the United States (n = 6). All development and validation utilized retrospective electronic health records from hospitalized cohorts. Across development cohorts, median sample size was 56 (IQR 41.5 to 116.5), with 30 events per variable (IQR 15.5 to 38.5) and 8.5 predictors per model (IQR 4.5 to 15.3). The external validation cohort contained 50 patients with 20 events and the events per variable was 2.5.

Eight studies evaluated patients’ glucocorticoid response based on whether urinary protein turned negative after 4 weeks of glucocorticoid administration (Table 1). Forty four predictors were documented. Most frequent predictors included: serum erythrocyte sedimentation rate (ESR) (n = 3), vitamin D-binding protein (VDBP) (n = 3), age (n = 2), serum D-dimer (n = 2), serum β2-microglobulin (n = 2), serum white blood cell (n = 2), serum cholesterol (n = 2), serum creatinine (n = 2), serum suppressor T cells (n = 2), urinary α-1-acid glycoprotein (n = 2), urinary retinol-binding protein (n = 2), urinary albumin (n = 2), urinary fetuin-A (n = 2), and urinary neutrophil gelatinase-associated lipocalin (n = 2). Supplementary Table S4 summarizes the predictors and performance of each model.

Models were predominantly developed using logistic regression (n = 16, 94.1%), followed by machine learning approaches (n = 1, 5.9%). Two models were respectively internally validated using bootstrap sampling and cross-validation methods. No model addressed missing values via interpolation. Most models (n = 7, 41.2%) directly excluded samples with missing data. All studies reported AUC values ranging from 0.75 to 0.88. Given that no predictive model has been externally validated in more than two cohorts, meta-analysis of AUC was not feasible.

Four models (23.5%) reported calibration via Hosmer-Lemeshow tests, while decision curve analysis evaluated clinical utility in two models (11.8%). The most common model presentation formats were mathematical equations (n = 6, 35.3%), followed by nomograms (n = 2, 11.8%) and web-based calculators (n = 2, 11.8%). One model that underwent external validation within the same healthcare system demonstrated a sensitivity of 0.9, specificity of 0.97, and accuracy of 0.94.

Overall, 76.92% of the models exhibited a high risk of bias (Supplementary Table S5), with the primary source of high risk of bias being the analytical domain (76.92%), particularly due to inadequate sample sizes (46.15%) or improper handling of missing data (38.46%). Details of the PROBAST risk-of-bias assessment are provided in Supplementary Table S6. These findings indicate that the reported model performance estimates are likely to be unstable and potentially optimistic.

3.2. Results of Risk Factors

The 19 included studies were published between 2012 and 2024, with the majority conducted in China (n = 8). The total number of patients across all studies was 2514. The study designs primarily included case-control studies (n = 15), retrospective cohort studies (n = 1), and prospective cohort studies (n = 3). All studies employed multivariate logistic regression to identify independent risk factors for SRNS. Ten studies evaluated patients’ glucocorticoid response based on whether urinary protein turned negative after 4 weeks of glucocorticoid administration (Table 2).

A total of 52 risk factors were analyzed across all studies, with 22 independent risk factors ultimately identified (Figure 2). The corresponding forest plots are presented in Supplementary Figure S1, and the definition of each pooled risk factor is provided after the relevant forest plot. Only a small subset of these independent risk factors could be quantitatively pooled; the remainder were reported descriptively. The 22 independent risk factors included: demographic/underlying disease-related factors (n = 3): age, low birth weight, hypertension; biochemical indicators (n = 9): white blood cells, cholesterol, ESR, albumin, D-dimer, globulin, proteinuria, decreased urine output, hematuria; immune/inflammatory factors (n = 4): suppressor T cells, immunoglobulin E, immunoglobulin M, immunoglobulin G; metabolic biomarkers (n = 6): polyligand proteoglynan-1, glutamine, endothelin-1, retinol-binding protein, α1-microglobulin, β2-microglobulin.

Overall, among the 19 observational studies, 14 were assessed as high-quality and 5 as medium-quality (Supplementary Table S7). None of the pooled analyses included 10 or more studies; therefore, funnel plots were not generated and publication bias could not be formally assessed.

3.3. Results of Sensitivity Analysis

Sensitivity analyses were not conducted because the prespecified criteria were not met: prediction models had insufficient external validation studies, and age, the only risk factor reported in at least five studies, included no high-risk-of-bias study for exclusion.

3.4. Cross-Comparison and Analysis

Cross-referencing revealed 17 variables serving as both predictive factors and independent risk factors. In addition, five variables were reported as independent risk factors but have not been incorporated into the published multivariable prediction models, including low birth weight, reduced urine output, hypertension, serum albumin, and serum immunoglobulin M (Figure 3).

Table 1

Basic characteristics of the included studies on SRNS prediction models.

StudyCountryStudy AimCohort TypeSR Cases/Total PatientsAge (Years)Outcome Definition
Agrawal 2021 [16]USADEHR14/407.0 (mean)NP-7 weeks
Agrawal 2020 [17]USADEHR13/378.7 (mean)NP-7 weeks
Bennett 2017 [18]USADEHR20/5012.3 (mean)NP-8 weeks
Burlaka 2022 [19]UkraineDEHR30/5611.0 (mean)NP-6 weeks
Gooding 2020 [20]USADEHR14/399.5 (mean)NP-6 weeks
Hong 2022 [21]ChinaDEHR41/1743.0 (median)NP-4 weeks
Jiang 2019 [22]ChinaDEHR16/795.5 (mean)NP-4 weeks
Kou 2023 [23]ChinaDEHR65/1113.2 (median)NP-4 weeks
Ling 2019 [24]ChinaDEHR44/1225.8 (mean)NP-4 weeks
Liu 2023 [25]ChinaDEHR31/1579.0 (mean)NP-4 weeks
Wang 2012 [26]ChinaDEHR19/434.5 (mean)NP-4 weeks
Ye 2023 [5]ChinaD, EVEHR34/91, 20/504.0 (median)NP-4 weeks
Zhang 2015 [27]ChinaDEHR36/1856.0 (median)NP-4 weeks

Footnote: D: derivation; EHR: electronic health record; EV: external validation; NP: Negative Proteinuria.

Table 2

Basic characteristics of the included studies on SRNS-associated risk factors.

StudyCountryStudy DesignNumber of ParticipantsOutcome DefinitionsIndependent Risk Factors
Agrawal 2021 [16]USAR40NP-7 weeksSerum: interleukin-7, interleukin-9, monocyte chemoattractant protein 1
Agrawal 2020 [17]USAR37NP-7 weeksSerum: adiponectin, matrix metalloproteinase 2, vitamin D binding protein
Cicek 2024 [28]TurkeyR162NP-8 weeksAge
Urine: microscopic hematuria
Cuzzoni 2019 [29]ItalyP62NP-4 weeksAge
Serum: macrophage migration inhibitory factor
Gooding 2020 [20]USAR39NP-6 weeksAge
Serum: glutamine
Hong 2022 [21]ChinaC174NP-4 weeksSerum: white blood cell
Urine: retinol-binding protein
Imbusi 2020 [30]EthiopiaC63NP-8 weeksAge (>6), hypertension
Serum: cholesterol
Kifle 2020 [31]EthiopiaC85NP-4 weeksUrine: gross hematuria, urine output
Konstantelos 2019 [32]CanadaP377Not reportedlow birth weight
Kou 2023 [23]ChinaC111NP-4 weeksSerum: β-microglobulin, D-dimer, erythrocyte sedimentation rate, suppressor T cells
Li 2024 [33]ChinaC510NP-8 weeksPathology, age
Serum: 25-hydroxyvitamin D, CD4+ T cells, creatinine, D-dimer
Urine: proteinuria
Ling 2019 [24]ChinaC122NP-4 weeksSerum: immunoglobulin E, immunoglobulin G
Liu 2023 [25]ChinaC157NP-4 weeksSerum: endothelin-1, polyligandproteoglynan-1
Rehman 2022 [34]PakistanR135Not reportedHypertension
Urine: hematuria
Salah 2023 [35]EgyptP60NP-8 weeksAge
Serum: white blood cell
Udagawa 2021 [36]JapanC80NP-4 weeksGender
Serum: immunoglobulin M
Wang 2012 [26]ChinaC43NP-4 weeksSerum: globulin
Urine: α1-antitrypsin/creatinine, αl-microglobulin
Yin 2010 [37]ChinaC72NP-4 weeksAge
Serum: albumin, D-dimer
Zhang 2015 [27]ChinaC185NP-4 weeksAge (>6.5)
Urine: microscopic hematuria, proteinuria,

Footnote: C: Case-control study; NP: Negative Proteinuria; P: Prospective cohort study; R: Retrospective cohort study.

Source: 3. Results Source: 3. Results Source: 3. Results Source: 3. Results Source: 3. Results Source: 3. Results Source: 3. Results Source: 3. Results

Conclusions

This systematic review summarized multivariable prediction models and independent risk factors for childhood SRNS. Although existing models reported apparent discrimination, most had a high risk of bias, lacked external validation, and included predictors that are not routinely available in clinical practice, indicating that their current clinical applicability remains very limited. The identified risk factors may provide preliminary clues for future candidate predictor selection, but their incremental predictive value and clinical feasibility require further evaluation. Future studies should prioritize clinically accessible predictors, rigorous model development, multi-center external validation, calibration assessment, decision curve analysis, and interpretable model presentation to improve clinical utility and generalizability.

Source: 5. Conclusions
Authors’ abstract

Background:

Steroid resistance indicates poor prognosis in pediatric nephrotic syndrome, but predictive models and risk factors for steroid-resistant nephrotic syndrome (SRNS) remain poorly understood.

Methods:

We searched PubMed, Embase, Scopus, CNKI, SinoMed, Wanfang, and VIP (inception to 1 March 2025) for studies developing SRNS prediction models or identifying risk factors. Odds ratios and AUC were pooled using random-effects meta-analysis. Risk of bias was assessed with PROBAST and Newcastle-Ottawa Scale.

Results:

Out of 2264 studies, 23 were included. Prediction models were mainly developed using logistic regression (16/17, 94.1%). The most frequently reported predictors included erythrocyte sedimentation rate and vitamin D binding protein. The reported AUC ranged from 0.75 to 0.88. Only one model had undergone external validation with an accuracy of 0.94. A total of 22 independent risk factors were identified, five of which: low birth weight, decreased urine output, hypertension, serum albumin, and serum IgM: were not in existing models. In total, 76% of model studies and 26% of risk factor analyses were at high or moderate risk of bias.

Conclusions:

Existing SRNS prediction models reported apparent discrimination but had a high risk of bias and very limited external validation, which substantially restricts their current clinical applicability. Several relevant risk factors remain unincorporated. Future research should prioritize rigorous model development and multi-center external validation.

Read the full paper

1. Introduction

Pediatric primary nephrotic syndrome (PNS) is a common glomerular disorder marked by severe proteinuria, low serum protein, edema, and elevated lipid levels, with a reported prevalence between 1.15 and 16.9 per 100,000 [1]. Standard treatment involves glucocorticoids, yet around 10 to 20% of patients respond poorly [1], resulting in a diagnosis of steroid-resistant nephrotic syndrome (SRNS).

The response of children with PNS to glucocorticoid therapy is a critical factor in determining prognosis. By 10 years after diagnosis, the estimated risk of progression to chronic kidney disease among patients with steroid-sensitive nephrotic syndrome (SSNS) is below 5% [2], whereas the 10-year renal survival without progression to end-stage kidney disease among patients with SRNS has been reported to range between 35% and 51% [3]. No evidence supports the efficacy of long-term glucocorticoid therapy for SRNS patients; however, it may cause cumulative steroid-related adverse effects [4], and approximately 70% of SRNS patients may face a high risk of disease progression due to long-term ineffective hormone therapy [5].

The prediction of glucocorticoid response in nephrotic syndrome can assist clinicians in identifying potential hormone resistance in children with PNS and guide the development of optimal treatment strategies in clinical practice, thus shortening the time to urine protein remission and reducing both the length of hospital stay and treatment costs [6]. Furthermore, the guidelines from the Kidney Disease: Improving Global Outcomes (KDIGO) organization recommend assessing patient prognosis based on the initial response to glucocorticoid therapy to minimize unnecessary renal biopsies [7].

Although various SRNS prediction models have been developed, these models vary in methodology, performance, and applicability, and no study has systematically compared model performance across studies applied to SRNS. Additionally, the predictors across studies vary significantly, resulting in uncertainty regarding their clinical applicability.

Systematic reviews have highlighted numerous biomarkers available for predicting SRNS [8,9]; however, these studies often rely on univariate analysis or correlation-based approaches, which fail to identify predictors with robust associations. Additionally, many of the studies included in these reviews based their analyses on data measured at the time of SRNS diagnosis, which may not have accurately captured the predictive potential of biomarkers over time. This approach limited the ability to establish reliable prediction models, as it overlooked the dynamic nature of disease progression, and may have led to overestimation of associations that are not causally linked.

This study aims to comprehensively review existing multivariable prediction models for SRNS and assess their methodological quality. It also summarizes reported risk factors associated with SRNS to inform candidate predictor selection in future prediction model development.

2. Materials and Methods

This review was prospectively registered with PROSPERO (CRD420250651875). Reporting adhered to the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis for Systematic Reviews and Meta-Analysis [10], as well as the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [11]. The details of self-reports are provided in Supplementary Tables S1 and S2.

2.1. Searching Strategies

Two researchers (YH. Hu and ZH. Zhang) independently searched the China National Knowledge Infrastructure, the Chinese Biomedical Literature Database, Wanfang Database, VIP Database, PubMed, Embase, and Scopus databases. We searched for studies developing prediction models for SRNS or identifying associated risk factors. Search terms included “Nephrotic Syndrome”, “Resistan*”, “Steroid”, “Predict*”, and “Risk*”. The full search strategies are presented in Supplementary Table S3. Searches were conducted without language restrictions and included records from the inception of each database through 1 March 2025.

2.2. Inclusion and Exclusion Criteria

For predictive models, we included studies that described the development or validation of multivariable SRNS prediction models in children. Eligibility was determined according to the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis statement [12] for transparent reporting of multivariable prediction model studies.

Inclusion criteria include: (1) Study population: Children diagnosed with primary nephrotic syndrome, regardless of initial diagnosis or recurrence status, and with no restrictions on age, gender, or race; SRNS was defined as persistent proteinuria after 4 to 8 weeks of adequate full-dose prednisone or prednisolone therapy [1,4]; (2) Intervention: Each included study reported the development or external validation of a prediction model for glucocorticoid response in pediatric patients with nephrotic syndrome; (3) Comparison: Not performed, as no predictive model has been widely validated across multiple studies; (4) Outcome: Glucocorticoid treatment response; (5) Study design: We included cohort studies (prospective or retrospective), case-control studies, randomized controlled trials, and cross-sectional studies in this review; (6) Time: All predictors must be measured prior to the assessment of glucocorticoid treatment response, specifically at the initial diagnosis of nephrotic syndrome.

Exclusion criteria include: (1) Studies solely focusing on genetic or renal biopsy indicators, due to the KDIGO guidelines explicitly stating that routine genetic testing or renal biopsy is not recommended for newly diagnosed patients [4]; (2) Studies collecting or measuring variables after confirmation of glucocorticoid treatment response; (3) Preprints; (4) duplicate publications and studies were not available.

For risk factors, we included studies that examined predictors of glucocorticoid response in children with primary nephrotic syndrome using multivariable regression analysis following univariate screening. Like the predictive models, risk factors must also be assessed before the measurement of glucocorticoid treatment response. No limitations were applied regarding study design, and the exclusion criteria were the same as those used for prediction model studies.

2.3. Selection of Reviews

Two investigators (YH. Hu and ZH. Zhang) independently screened titles and abstracts, and subsequently reviewed the full texts of potentially eligible studies to assess which articles would be included. Discrepancies were resolved by a third researcher (HL. Li) through a consensus discussion.

2.4. Data Extraction

Two authors (YH. Hu and ZH. Zhang) independently performed data extraction, and any discrepancies were resolved by a third author (HL. Li). The following data were extracted from all included studies: first author, publication year, country/region, and model name (where applicable). Corresponding authors were contacted to obtain missing information/data. For predictive model research, a standardized form was constructed using the CHARMS checklist [13]. The extracted information from studies reporting model development included study design, study population, outcome measures, candidate predictors, modeling approach, internal validation method, numbers of patients and events, handling of missing and continuous variables, number and type of predictors in the final model, model presentation, and performance metrics (classification, discrimination, calibration, and clinical utility).

Classification performance was assessed by sensitivity, specificity, positive predictive value, and negative predictive value; discrimination was measured using the C-index; calibration was examined with the Hosmer-Lemeshow test and calibration plots; and clinical utility was evaluated through decision curve analysis and net benefit metrics. For external validation studies, extracted information included study population, number of patients and events, and measures of predictive performance. For studies reporting multiple models, information was extracted independently for each individual model. When studies reported both overall and subgroup-specific model performance, only overall population analyses were extracted.

For risk factor studies, the extracted information included the study design, study population, number of patients, analysis methods, and the results of the analysis.

2.5. Assessment of Methodological Quality of Included Studies

Two authors (YH. Hu and ZH. Zhang) independently evaluated the risk of bias for the included prediction models using the Prediction Model Risk of Bias Assessment Tool (PROBAST) [14], and similarly applied the Newcastle-Ottawa Scale [15] to the observational studies. Any discrepancies between reviewers were resolved through discussion with a third author (HL. Li).

2.6. Data Synthesis and Analysis

Baseline characteristics were summarized using mean ± standard deviation or median (interquartile range, IQR) for continuous variables, and expressed as percentages for categorical variables. For prediction models that had been validated in more than two independent datasets, we conducted a random-effects meta-analysis using the area under the curve (AUC) and 95% confidence intervals (CI) as the primary metrics. A 95% CI including 0.5 was considered indicative of insufficient discrimination. For risk factors, we applied a random-effects meta-analysis to combine odds ratios (OR) with their corresponding 95% CI.

Heterogeneity (I2) was assessed only for pooled analyses. Funnel plots were generated to evaluate publication bias when ten or more studies reported comparable outcomes. Two-tailed statistical tests were conducted with a significance level of p < 0.05. All analyses were performed using SPSS version 18.0 and RevMan version 5.3.

2.7. Sensitivity Analysis

Sensitivity analyses were planned separately for prediction model studies and risk factor studies. For prediction model studies, if the same model was externally validated in more than five independent datasets, sensitivity analysis would be conducted by excluding studies at high risk of bias and re-pooling the model performance estimates. For risk factor studies, if at least five studies reported the same risk factor with comparable definitions, sensitivity analysis would be conducted by excluding studies at high risk of bias and re-pooling the effect estimates.

3. Results

A total of 2264 records were retrieved. After screening titles and abstracts, 23 unique studies were included, comprising 13 predictive model studies and 19 risk factor studies with an overlap of nine studies between the two categories. This overlap is attributed to the fact that these nine studies developed predictive models by adopting multivariate analysis methods to screen and incorporate key risk factors.

The literature screening process is presented in Figure 1.

3.1. Results of Prediction Models

Thirteen studies encompassed the development of 17 prediction models and the external validation of one model. Model development occurred primarily in China (n = 9), followed by the United States (n = 6). All development and validation utilized retrospective electronic health records from hospitalized cohorts. Across development cohorts, median sample size was 56 (IQR 41.5 to 116.5), with 30 events per variable (IQR 15.5 to 38.5) and 8.5 predictors per model (IQR 4.5 to 15.3). The external validation cohort contained 50 patients with 20 events and the events per variable was 2.5.

Eight studies evaluated patients’ glucocorticoid response based on whether urinary protein turned negative after 4 weeks of glucocorticoid administration (Table 1). Forty four predictors were documented. Most frequent predictors included: serum erythrocyte sedimentation rate (ESR) (n = 3), vitamin D-binding protein (VDBP) (n = 3), age (n = 2), serum D-dimer (n = 2), serum β2-microglobulin (n = 2), serum white blood cell (n = 2), serum cholesterol (n = 2), serum creatinine (n = 2), serum suppressor T cells (n = 2), urinary α-1-acid glycoprotein (n = 2), urinary retinol-binding protein (n = 2), urinary albumin (n = 2), urinary fetuin-A (n = 2), and urinary neutrophil gelatinase-associated lipocalin (n = 2). Supplementary Table S4 summarizes the predictors and performance of each model.

Models were predominantly developed using logistic regression (n = 16, 94.1%), followed by machine learning approaches (n = 1, 5.9%). Two models were respectively internally validated using bootstrap sampling and cross-validation methods. No model addressed missing values via interpolation. Most models (n = 7, 41.2%) directly excluded samples with missing data. All studies reported AUC values ranging from 0.75 to 0.88. Given that no predictive model has been externally validated in more than two cohorts, meta-analysis of AUC was not feasible.

Four models (23.5%) reported calibration via Hosmer-Lemeshow tests, while decision curve analysis evaluated clinical utility in two models (11.8%). The most common model presentation formats were mathematical equations (n = 6, 35.3%), followed by nomograms (n = 2, 11.8%) and web-based calculators (n = 2, 11.8%). One model that underwent external validation within the same healthcare system demonstrated a sensitivity of 0.9, specificity of 0.97, and accuracy of 0.94.

Overall, 76.92% of the models exhibited a high risk of bias (Supplementary Table S5), with the primary source of high risk of bias being the analytical domain (76.92%), particularly due to inadequate sample sizes (46.15%) or improper handling of missing data (38.46%). Details of the PROBAST risk-of-bias assessment are provided in Supplementary Table S6. These findings indicate that the reported model performance estimates are likely to be unstable and potentially optimistic.

3.2. Results of Risk Factors

The 19 included studies were published between 2012 and 2024, with the majority conducted in China (n = 8). The total number of patients across all studies was 2514. The study designs primarily included case-control studies (n = 15), retrospective cohort studies (n = 1), and prospective cohort studies (n = 3). All studies employed multivariate logistic regression to identify independent risk factors for SRNS. Ten studies evaluated patients’ glucocorticoid response based on whether urinary protein turned negative after 4 weeks of glucocorticoid administration (Table 2).

A total of 52 risk factors were analyzed across all studies, with 22 independent risk factors ultimately identified (Figure 2). The corresponding forest plots are presented in Supplementary Figure S1, and the definition of each pooled risk factor is provided after the relevant forest plot. Only a small subset of these independent risk factors could be quantitatively pooled; the remainder were reported descriptively. The 22 independent risk factors included: demographic/underlying disease-related factors (n = 3): age, low birth weight, hypertension; biochemical indicators (n = 9): white blood cells, cholesterol, ESR, albumin, D-dimer, globulin, proteinuria, decreased urine output, hematuria; immune/inflammatory factors (n = 4): suppressor T cells, immunoglobulin E, immunoglobulin M, immunoglobulin G; metabolic biomarkers (n = 6): polyligand proteoglynan-1, glutamine, endothelin-1, retinol-binding protein, α1-microglobulin, β2-microglobulin.

Overall, among the 19 observational studies, 14 were assessed as high-quality and 5 as medium-quality (Supplementary Table S7). None of the pooled analyses included 10 or more studies; therefore, funnel plots were not generated and publication bias could not be formally assessed.

3.3. Results of Sensitivity Analysis

Sensitivity analyses were not conducted because the prespecified criteria were not met: prediction models had insufficient external validation studies, and age, the only risk factor reported in at least five studies, included no high-risk-of-bias study for exclusion.

3.4. Cross-Comparison and Analysis

Cross-referencing revealed 17 variables serving as both predictive factors and independent risk factors. In addition, five variables were reported as independent risk factors but have not been incorporated into the published multivariable prediction models, including low birth weight, reduced urine output, hypertension, serum albumin, and serum immunoglobulin M (Figure 3).

4. Discussion

This systematic review evaluated 17 predictive models for childhood SRNS and identified 22 associated independent risk factors. Despite the reported apparent discriminative performance (AUC range: 0.75 to 0.88; AUC > 0.70 indicating acceptable discrimination [38]), these estimates should be interpreted with substantial caution because most models were at high risk of bias, were developed in small retrospective cohorts, and lacked adequate internal or external validation. First, only two models underwent internal validation and one model underwent external validation, which may lead to overestimation of model performance due to overfitting [39] and limit the generalizability of models across populations with different characteristics [40].

Therefore, the reported AUC values should not be interpreted as evidence of reliable clinical prediction; the reliability, robustness, and clinical applicability of existing models remain very limited at present. Second, missing data were often inadequately handled (frequently by complete-case analysis), which may bias estimates and reduce predictive accuracy [41]. Third, calibration was reported in fewer than one-quarter of models, and decision-curve analysis was rarely performed, constraining assessment of clinical usefulness [42]. Fourth, decision curve analysis quantifies clinical net benefit post-implementation [43], yet only two studies implemented this analysis.

Finally, many models lacked a clear presentation format, which may impede translation into routine practice [44].

This study found that ESR and VDBP measured at disease onset were the most frequently included predictors in SRNS prediction models. In nephrotic syndrome, ESR is often markedly elevated at presentation, largely due to increased circulating acute-phase reactants (e.g., fibrinogen and other pro-aggregatory proteins) and altered plasma protein composition that promotes erythrocyte aggregation [45,46]. Such elevated baseline ESR may reflect a greater inflammatory burden and more active immune dysregulation at disease onset, underlying its potential predictive value.

VDBP may reflect urinary protein loss, tubular injury, or inflammation-related renal damage, which may explain why it has been evaluated as a potential predictor [47,48].

Among variables frequently incorporated during model development, approximately 50% are not readily obtainable through routine clinical laboratory testing. This requirement for specialized reagent procurement consequently constrains their applicability in resource-limited settings, including primary care facilities.

Risk factors should not be considered equivalent to predictors in clinical prediction models, but they may provide clues for identifying candidate predictors that require further validation. This review found that five factors: low birth weight, reduced urine output, hypertension, serum albumin, and serum immunoglobulin M: were reported as independent risk factors but have not yet been incorporated into published multivariable prediction models. Due to the observational design and risk of bias of included studies, these associations are hypothesis-generating. Low birth weight is a well-established risk factor for SRNS, as studies have shown that the incidence of SRNS in low-birth-weight children is 3.78 times that in children with normal birth weight; for every 100 g increase in birth weight, the risk of SRNS decreases by approximately 8% [32].

A potential explanation is that low-birth-weight children, due to incomplete renal development and reduced nephron number, are prone to compensatory impairment of renal function, thereby increasing the risk of SRNS and exhibiting a higher recurrence rate and more complications [49,50,51]. Hypertension is another clinical marker that commonly accompanies nephrotic syndrome. It results from fluid retention due to glomerular damage, which, in turn, exacerbates kidney injury and promotes a vicious cycle of further renal deterioration [52,53].

Reduced urine output may serve as an early warning sign, indicating a decline in renal filtration capacity and worsening renal function, both of which may increase the risk of SRNS progression [31]. These markers not only reflect the severity of NS at diagnosis but also provide critical insight into the potential for steroid resistance, making them useful for early stratification. Evidence from the included multivariable analyses suggested an association between higher serum albumin levels and SRNS. This finding should be interpreted cautiously because the direction and mechanism of this association remain uncertain.

Serum albumin may reflect differences in disease phenotype, proteinuria selectivity [7,54], or underlying histopathology [55], but its value as a candidate predictor requires further validation. In this review, higher serum IgM was identified as an independent risk factor for SRNS. Although biopsy-derived predictors were outside our eligibility criteria, biopsy-based studies have reported that glomerular immunoglobulin M correlates with the severity of kidney injury, suggesting an elevated risk of disease recurrence and protracted remission periods.

Furthermore, children with immunoglobulin M positivity demonstrate enhanced therapeutic responsiveness to cyclosporine treatment, which provides further evidence for the role of immune factors in steroid resistance [56,57]. Finally, descriptive analyses in several studies identified additional potential independent risk factors, including pathological types, peripheral blood CD4+T cell levels, and serum macrophage migration inhibitory factor, which were not represented among predictors in the published multivariable models included in this review.

The primary strength of this study is its systematic synthesis and evaluation of childhood SRNS prediction models, as well as the identification of independent risk factors associated with SRNS development, thereby providing an evidence-based reference for optimizing future prediction models. However, this study has several limitations. First, most prediction model studies were at high risk of bias, mainly due to limitations in the analysis domain, including small sample sizes, limited validation, and inadequate handling of missing data. Therefore, the reported model performance and identified predictors should be interpreted cautiously.

Second, due to the scarcity of externally validated prediction models, meta-analysis and sensitivity analysis of model performance could not be performed. Third, relatively few studies were available for each risk factor, and most associations were supported by only one or two studies, limiting the robustness of pooled estimates and precluding sensitivity analyses. Finally, substantial heterogeneity in SRNS definitions, populations, laboratory methods, and adjustment models may have affected the reliability of the synthesized findings.

5. Conclusions

This systematic review summarized multivariable prediction models and independent risk factors for childhood SRNS. Although existing models reported apparent discrimination, most had a high risk of bias, lacked external validation, and included predictors that are not routinely available in clinical practice, indicating that their current clinical applicability remains very limited. The identified risk factors may provide preliminary clues for future candidate predictor selection, but their incremental predictive value and clinical feasibility require further evaluation. Future studies should prioritize clinically accessible predictors, rigorous model development, multi-center external validation, calibration assessment, decision curve analysis, and interpretable model presentation to improve clinical utility and generalizability.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/jcm15124438/s1. (1) Supplementary Table S1: TRIPOD-SRMA Checklist for reporting systematic reviews of prediction model studies; (2) Supplementary Table S2: PRISMA Checklist for reporting systematic reviews and Meta-analysis; (3) Supplementary Table S3: Search Strategy; (4) Supplementary Table S4: Outcomes of studies reporting on prediction models; (5) Supplementary Table S5: Risk of bias and applicability assessment for each PROBAST domain; (6) Supplementary Table S6: Details of PROBAST risk-of-bias assessment; (7) Supplementary Table S7: Risk of bias and applicability assessment for each NOS domain; (8) Supplementary Figure S1: Forest plots for pooled analyses of risk factors associated with steroid-resistant nephrotic syndrome in children.

Supplement file at the original source: open the original paper. This file is not included in this text import.

Author Contributions

Conceptualization: Y.H., H.L., L.Z. (Lingli Zhang); Methodology: Y.H., Z.Z., L.Z. (Linan Zeng), L.Z. (Lingli Zhang); Validation: Y.H., Z.Z., S.D., Z.B., H.L.; Formal Analysis: Y.H.; Investigation: Z.Z., S.D., Y.G. (Yannan Guo), Y.G. (Yangtingting Gao), Z.X., Q.P., Y.X., L.H., J.C.; Resources: L.Z. (Linan Zeng); Data Curation: Y.H., Z.Z., S.D., Y.G. (Yannan Guo), Y.G. (Yangtingting Gao), Z.X., Q.P., Y.X., J.C.; Writing: Original Draft: Y.H., Z.Z.; Writing: Review and Editing: Y.H., Z.Z., S.D., Y.G. (Yannan Guo), Y.G. (Yangtingting Gao), Z.X., Q.P., Y.X., Z.B., L.Z. (Linan Zeng), L.H., J.C., Y.Z., H.L., L.Z. (Lingli Zhang); Visualization: Y.H.; Supervision: Y.Z., H.L., L.Z. (Lingli Zhang); Project Administration: L.H., L.Z. (Lingli Zhang); Funding Acquisition: H.L., L.Z. (Lingli Zhang).

All authors have read and agreed to the published version of the manuscript.

Data Availability Statement

The datasets used and analyzed during the current study are available in the Supplementary Materials.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

AUC: area under the curve; CHARMS: Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies; CI: Confidence Interval; ESR: erythrocyte sedimentation rate; IQR: Interquartile Range; KDIGO: Kidney Disease: Improving Global Outcomes; OR: odds ratios; PNS: Primary Nephrotic Syndrome; PROBAST: prediction model risk of bias assessment tool; SRNS: Steroid-Resistant Nephrotic Syndrome; TRIPOD: Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis; VDBP: vitamin D-binding protein.

References

1. Noone D.G. Iijima K. Parekh R.J.T.L. Idiopathic nephrotic syndrome in children Lancet 2018 392 61 74 10.1016/S0140-6736(18)30536-1 29910038
2. Mendonça A.C. Oliveira E.A. Fróes B.P. Faria L.D. Pinto J.S. Nogueira M.M. Lima G.O. Resende P.I. Assis N.S. Simões E.S.A.C. A predictive model of progressive chronic kidney disease in idiopathic nephrotic syndrome Pediatr. Nephrol. 2015 30 2011 2020 10.1007/s00467-015-3136-0 26084617
3. Trautmann A. Schnaidt S. Lipska-Ziętkiewicz B.S. Bodria M. Ozaltin F. Emma F. Anarat A. Melk A. Azocar M. Oh J. Long-Term Outcome of Steroid-Resistant Nephrotic Syndrome in Children J. Am. Soc. Nephrol. 2017 28 3055 3065 10.1681/ASN.2016101121 28566477 PMC5619960
4. Trautmann A. Vivarelli M. Samuel S. Gipson D. Sinha A. Schaefer F. Hui N.K. Boyer O. Saleem M.A. Feltran L. IPNA clinical practice recommendations for the diagnosis and management of children with steroid-resistant nephrotic syndrome Pediatr. Nephrol. 2020 35 1529 1561 10.1007/s00467-020-04519-1 32382828 PMC7316686
5. Ye Q. Li Y. Liu H. Mao J. Jiang H. Machine learning models for predicting steroid-resistant of nephrotic syndrome Front. Immunol. 2023 14 1090241 10.3389/fimmu.2023.1090241 36776850 PMC9911108
6. Zaorska K. Zawierucha P. Świerczewska M. Ostalska-Nowicka D. Zachwieja J. Nowicki M. Prediction of steroid resistance and steroid dependence in nephrotic syndrome children J. Transl. Med. 2021 19 130 10.1186/s12967-021-02790-w 33785019 PMC8011118
7. Floege J. Gibson K.L. Vivarelli M. Liew A. Radhakrishnan J. Rovin B.H. KDIGO 2025 Clinical Practice Guideline for the Management of Nephrotic Syndrome in Children Kidney Int. 2025 107 S241 S289 10.1016/j.kint.2024.11.007 40254391
8. May C.J. Ford N.P. Welsh G.I. Saleem M.A. Biomarkers to predict or measure steroid resistance in idiopathic nephrotic syndrome: A systematic review PLoS ONE 2025 20 e0312232 10.1371/journal.pone.0312232 39946431 PMC11824968
9. Lee J.M. Ahn Y.H. Lim S.H. Kang H.G. Biomarkers predicting treatment-response in nephrotic syndrome of children: A systematic review Child. Kidney Dis. 2021 25 92 111 10.3339/jkspn.2021.25.2.92
10. Snell K.I.E. Levis B. Damen J.A.A. Dhiman P. Debray T.P.A. Hooft L. Reitsma J.B. Moons K.G.M. Collins G.S. Riley R.D. Transparent reporting of multivariable prediction models for individual prognosis or diagnosis: Checklist for systematic reviews and meta-analyses (TRIPOD-SRMA) BMJ (Clin. Res. Ed.) 2023 381 e073538 10.1136/bmj-2022-073538 37137496 PMC10155050
11. Page M.J. McKenzie J.E. Bossuyt P.M. Boutron I. Hoffmann T.C. Mulrow C.D. Shamseer L. Tetzlaff J.M. Akl E.A. Brennan S.E. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews BMJ (Clin. Res. Ed.) 2021 372 n71 10.1136/bmj.n71 PMC8005924 33782057
12. Collins G.S. Reitsma J.B. Altman D.G. Moons K.G. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement BMJ (Clin. Res. Ed.) 2015 350 g7594 10.1161/CIRCULATIONAHA.114.014508 25569120
13. Moons K.G. de Groot J.A. Bouwmeester W. Vergouwe Y. Mallett S. Altman D.G. Reitsma J.B. Collins G.S. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: The CHARMS checklist PLoS Med. 2014 11 e1001744 10.1371/journal.pmed.1001744 25314315 PMC4196729
14. Wolff R.F. Moons K.G.M. Riley R.D. Whiting P.F. Westwood M. Collins G.S. Reitsma J.B. Kleijnen J. Mallett S. PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies Ann. Intern. Med. 2019 170 51 58 10.7326/M18-1376 30596875
15. Wells G.A. Shea B. O’Connell D. Peterson J. Welch V. Losos M. Tugwell P. The Newcastle-Ottawa Scale (NOS) for Assessing the Quality of Nonrandomised Studies in Meta-Analyses Available online: https://ohri.ca/en/who-we-are/core-facilities-and-platforms/ottawa-methods-centre/newcastle-ottawa-scale(accessed on 15 March 2012)
16. Agrawal S. Brier M.E. Kerlin B.A. Smoyer W.E. Plasma Cytokine Profiling to Predict Steroid Resistance in Pediatric Nephrotic Syndrome Kidney Int. Rep. 2021 6 785 795 10.1016/j.ekir.2020.12.027 33732993 PMC7938200
17. Agrawal S. Merchant M.L. Kino J. Li M. Wilkey D.W. Gaweda A.E. Brier M.E. Chanley M.A. Gooding J.R. Sumner S.J. Predicting and Defining Steroid Resistance in Pediatric Nephrotic Syndrome Using Plasma Proteomics Kidney Int. Rep. 2020 5 66 80 10.1016/j.ekir.2019.09.009 31922062 PMC6943770
18. Bennett M.R. Pleasant L. Haffner C. Ma Q. Haffey W.D. Ying J. Wagner M. Greis K.D. Devarajan P. A Novel Biomarker Panel to Identify Steroid Resistance in Childhood Idiopathic Nephrotic Syndrome Biomark. Insights 2017 12 1177271917695832 10.1177/1177271917695832 28469399 PMC5391984
19. Burlaka I. Mityuryayeva I. Bagdasarova I. Clinical and Apoptotic Factors Defining and Predicting Steroid Resistance in Nephrotic Syndrome in Children Glob. Pediatr. Health 2022 9 2333794x221085392 10.1177/2333794X221085392 35342775 PMC8943305
20. Gooding J.R. Agrawal S. McRitchie S. Acuff Z. Merchant M.L. Klein J.B. Smoyer W.E. Sumner S.J. Predicting and Defining Steroid Resistance in Pediatric Nephrotic Syndrome Using Plasma Metabolomics Kidney Int. Rep. 2020 5 81 93 10.1016/j.ekir.2019.09.010 31922063 PMC6943762
21. Hong M.Z. Zhong P.Q. Chen M.Y. Chen P.S. Laboratory indicators predict the efficacy of glucocorticoid therapy in pediatric primary nephrotic syndrome J. Trop. Med. 2022 22 1082 1087
22. Jiang Y. Zhang B.L. Wang W.H. Clinical significance of detection of urine renal injury biomarkers in children with primary nephrotic syndrome Chin. J. Appl. Clin. Pediatr. 2019 34 1326 1330
23. Kou M. Wu F. Qu X.Y. Wang H. Guo X.T. Yang Y.Y. Zhao L.J. Establishment and validation of clinical prediction model for steroid-resistant nephrotic syndrome in children Chin. J. Pediatr. 2023 61 333 338 10.3760/cma.j.cn112140-20220924-00837 37011979
24. Ling C. Chen Z. Fan J.F. Sun Q. Meng Q. Hua L. Liu X.R. Value of serum lgG combined with IgE in predicting steroid therapy response in children with primary nephrotic syndrome Chin. J. Nephrol. 2019 35 835 840
25. Liu Q. Zhu J. Tang G.Y. Jiang P.Y. Serum levels of ET-1 and SDC-1 in children with nephrotic syndrome and their correlation with hormone therapy response Tianjin Med. J. 2023 51 413 417
26. Wang Y.Y. DIng G.X. Yuan Y.G. Bao H.Y. Chen Y. Zhao F. Han Y. Zhang A.H. Huang S.M. Detection of urinary α1-antitrypsin for predicting response to glucocorticoid therapy in children with primary nephrotic syndrome Chin. J. Nephrol. 2012 28 857 862
27. Zhang B.L. Liu T. Lin S.X. Wang W.H. Liu Y. Liu Y. Wu X. Wang X. Liu Z. Analysis of risk factors for steroid resistance in children with primary nephritic syndrome and discussion on the prediction model of steroid resistance Chin. J. Nephrol. 2015 31 414 418
28. Cicek N. Yıldız N. Guven S. Kaya M. Gokce I. Alpay H. Clinical Predictors of Steroid Resistance in Childhood Nephrotic Syndrome Clin. Pediatr. 2024 63 1300 1307 10.1177/00099228231219109 38142361
29. Cuzzoni E. Franca R. De Iudicibus S. Marcuzzi A. Lucafò M. Pelin M. Favretto D. Monti E. Morello W. Ghio L. MIF plasma level as a possible tool to predict steroid responsiveness in children with idiopathic nephrotic syndrome Eur. J. Clin. Pharmacol. 2019 75 1675 1683 10.1007/s00228-019-02749-3 31463578
30. Imbusi E.A. Ekanem P.E. Gebrearegay H. Ambaye M. Tesfahunegn A. Nyaga K. Ekanem R. Peter N. Steroid response pattern among children with nephrotic syndrome in Northern Ethiopia Nephro-Urol. Mon. 2020 12 e106995 10.5812/numonthly.106995
31. Kifle M. Shimelis D. Predictors of resistance to steroids in pediatric nephrotic syndrome at a tertiary hospital, Addis Ababa Ethiop. J. Pediatr. Child Health 2020 15 5 15
32. Konstantelos N. Banh T. Patel V. Vasilevska-Ristovska J. Borges K. Hussain-Shamsy N. Noone D. Hebert D. Radhakrishnan S. Licht C.P.B. Association of low birth weight and prematurity with clinical outcomes of childhood nephrotic syndrome: A prospective cohort study Pediatr. Nephrol. 2019 34 1599 1605 10.1007/s00467-019-04255-1 30976899
33. Li J.Z. Wang Y. Influencing factors and hormone resistance of 510 children with nephrotic syndrome J. Public Health Prev. Med. 2024 35 79 82
34. Rehman M. Ali A. Ehsan A. Aziz M. Khatri S. Hashmi S. Can Steroid Response in Idiopathic Childhood Nephrotic Syndrome be Predicted? A Single Center Quasi-Experimental Study Pak. Armed Forces Med. J. 2022 72 980 10.51253/pafmj.v72i3.6346
35. Salah D.M. Aoun A.H. Fahmy B.S. Zeid A. Fahmy Y.A. Does Urinary Vitamin D-Binding Protein Have a Role in the Prediction of Steroid Resistance in Nephrotic Syndrome? A Cohort Study on Egyptian Children J. Compr. Pediatr. 2023 14 e130133 10.5812/compreped-130133
36. Udagawa T. Matsuyama Y. Okutsu M. Motoyoshi Y. Okada M. Tada N. Kikuchi E. Shimoda M. Kanamori T. Omori T. Association between Immunoglobulin M and Steroid Resistance in Children with Nephrotic Syndrome: A Retrospective Multicenter Study in Japan Kidney360 2021 2 487 493 10.34067/KID.0004432020 35369027 PMC8785997
37. Yin L. Zhou W. Sun H. Jin Y.L. Zhou Z.Y. Risk Factors of Steroid Resistance in Children with Primary Nephrotic Syndrome Chin. J. Appl. Clin. Pediatr. 2010 25 1330 1332,1356
38. White N. Parsons R. Collins G. Barnett A. Evidence of questionable research practices in clinical prediction models BMC Med. 2023 21 339 10.1186/s12916-023-03048-6 37667344 PMC10478406
39. Moons K.G. Kengne A.P. Woodward M. Royston P. Vergouwe Y. Altman D.G. Grobbee D.E. Risk prediction models: I. Development, internal validation, and assessing the incremental value of a new (bio)marker Heart (Br. Card. Soc.) 2012 98 683 690 10.1136/heartjnl-2011-301246 22397945
40. Moons K.G. Kengne A.P. Grobbee D.E. Royston P. Vergouwe Y. Altman D.G. Woodward M. Risk prediction models: II. External validation, model updating, and impact assessment Heart (Br. Card. Soc.) 2012 98 691 698 10.1136/heartjnl-2011-301247 22397946
41. Little R.J. Missing Data Analysis Annu. Rev. Clin. Psychol. 2024 20 149 173 10.1146/annurev-clinpsy-080822-051727 38346291
42. Alba A.C. Agoritsas T. Walsh M. Hanna S. Iorio A. Devereaux P.J. McGinn T. Guyatt G. Discrimination and Calibration of Clinical Prediction Models: Users’ Guides to the Medical Literature JAMA 2017 318 1377 1384 10.1001/jama.2017.12126 29049590
43. Vickers A.J. Van Calster B. Steyerberg E.W. Net benefit approaches to the evaluation of prediction models, molecular markers, and diagnostic tests BMJ (Clin. Res. Ed.) 2016 352 i6 10.1136/bmj.i6 PMC4724785 26810254
44. Bonnett L.J. Snell K.I.E. Collins G.S. Riley R.D. Guide to presenting clinical prediction models for use in clinical settings BMJ (Clin. Res. Ed.) 2019 365 l737 10.1136/bmj.l737 30995987
45. Fabry T.L. Mechanism of erythrocyte aggregation and sedimentation Blood 1987 70 1572 1576 10.1182/blood.V70.5.1572.1572 3663946
46. Higuchi M. Watanabe N. Determination of the erythrocyte sedimentation rate using the hematocrit-corrected aggregation index and mean corpuscular volume J. Clin. Lab. Anal. 2023 37 e24877 10.1002/jcla.24877 37003619 PMC10156103
47. Bouillon R. Schuit F. Antonio L. Rastinejad F. Vitamin D Binding Protein: A Historic Overview Front. Endocrinol. 2019 10 910 10.3389/fendo.2019.00910 31998239 PMC6965021
48. Mirković K. Doorenbos C.R. Dam W.A. Lambers Heerspink H.J. Slagman M.C. Nauta F.L. Kramer A.B. Gansevoort R.T. van den Born J. Navis G. Urinary vitamin D binding protein: A potential novel marker of renal interstitial inflammation and fibrosis PLoS ONE 2013 8 e55887 10.1371/journal.pone.0055887 23409077 PMC3569442
49. Teeninga N. Schreuder M.F. Bökenkamp A. Delemarre-van de Waal H.A. van Wijk J.A. Influence of low birth weight on minimal change nephrotic syndrome in children, including a meta-analysis Nephrol. Dial. Transplant. Off. Publ. Eur. Dial. Transpl. Assoc.: Eur. Ren. Assoc. 2008 23 1615 1620 10.1093/ndt/gfm829 18065792
50. Plank C. Ostreicher I. Dittrich K. Waldherr R. Voigt M. Amann K. Rascher W. Dötsch J. Low birth weight, but not postnatal weight gain, aggravates the course of nephrotic syndrome Pediatr. Nephrol. 2007 22 1881 1889 10.1007/s00467-007-0597-9 17874138
51. Conti G. De Vivo D. Fede C. Arasi S. Alibrandi A. Chimenz R. Santoro D. Low birth weight is a conditioning factor for podocyte alteration and steroid dependance in children with nephrotic syndrome J. Nephrol. 2018 31 411 415 10.1007/s40620-018-0473-7 29350347
52. Shatat I.F. Becton L.J. Woroniecki R.P. Hypertension in Childhood Nephrotic Syndrome Front. Pediatr. 2019 7 287 10.3389/fped.2019.00287 31380323 PMC6646680
53. Skrzypczyk P. Wabik A.M. Deja A. Ofiara A. Szyszka M. Panczyk-tomaszewska M. The prevalence and risk factors for arterial hypertension in children with idiopathic nephrotic syndrome J. Hypertens. 2024 42 e207 10.1097/01.hjh.0001021548.17111.52
54. Tojo A. Mechanism Underlying Selective Albuminuria in Minimal Change Nephrotic Syndrome Int. J. Nephrol. 2019 2019 5859102 10.1155/2019/5859102 31781392 PMC6874928
55. Haeri H.H. Eisermann J. Schimm H. Büscher A. Hoyer P. Hinderberger D. Profound Changes in Functional Structure and Dynamics of Serum Albumin in Children with Nephrotic Syndrome: An Exploratory Research Study J. Med. Chem. 2023 66 12115 12129 10.1021/acs.jmedchem.3c00680 37648246 PMC10510392
56. Swartz S.J. Eldin K.W. Hicks M.J. Feig D.I. Minimal change disease with IgM+ immunofluorescence: A subtype of nephrotic syndrome Pediatr. Nephrol. 2009 24 1187 1192 10.1007/s00467-009-1130-0 19219463
57. Kanemoto K. Ito H. Anzai M. Matsumura C. Kurayama H. Clinical significance of IgM and C1q deposition in the mesangium in pediatric idiopathic nephrotic syndrome J. Nephrol. 2013 26 306 314 10.5301/jn.5000133 22641570
Figure 1

Flow diagram of study selection.

Figure image at the original source: open the original paper. This file is not included in this text import.
Figure 2

Associations between reported risk factors and SRNS (summary of multivariable results). Footnote: Risk factors of demographic/underlying disease, biochemical indicators, immune/inflammatory factors and metabolic biomarkers. Prognostic factor (number of studies) (right column): OR and 95% CI from pooled analysis; Boldface indicates statistically significant associations (p < 0.05); For continuous variables, the effect estimates correspond to the unit of increase as reported in the original studies; for categorical variables, the cutoff/definition is indicated in the label.

Figure image at the original source: open the original paper. This file is not included in this text import.
Figure 3

Overview of included risk factors and predictors. Footnote: (ANot reportedD): Variables included exclusively of risk factors and predictors in Demographic/underlying disease, Biochemical indicators, Immune/inflammatory factors and Metabolic biomarkers; (E): Variables reported as both risk factors and predictors (overlapping variables). S: measured in serum; U: measured in urine. α-1-AGP1: α-1 acid glycoprotein 1; α-1-AGP2: α-1 acid glycoprotein 1; α-1-BG: α-1-B glycoprotein; α-2-MCG: α-2 macroglobulin; AST: aspartate aminotransferase; Cr: Creatinine; ESR: erythrocyte sedimentation rate; FDP: fibrin degradation products; IgA: Immunoglobulin A; IgE: Immunoglobulin E; IgG: Immunoglobulin G; IgM: Immunoglobulin M; IL-7: interleukin-7; IL-9: interleukin-9; MMP-2: matrix metalloproteinase 2; MCP-1: monocyte chemoattractant protein 1; NAG: N-acetyl-β-D-glucosaminidase; Neu%: percentage of neutrophils; NGAL: neutrophil gelatinase-associated lipocalin; RBP: retinol-binding protein; SDC-1: polyligand proteoglynan-1; TBG: thyroxine-binding globulin; VDBP: vitamin D binding protein.

Figure image at the original source: open the original paper. This file is not included in this text import.
Table 1

Basic characteristics of the included studies on SRNS prediction models.

StudyCountryStudy AimCohort TypeSR Cases/Total PatientsAge (Years)Outcome Definition
Agrawal 2021 [16]USADEHR14/407.0 (mean)NP-7 weeks
Agrawal 2020 [17]USADEHR13/378.7 (mean)NP-7 weeks
Bennett 2017 [18]USADEHR20/5012.3 (mean)NP-8 weeks
Burlaka 2022 [19]UkraineDEHR30/5611.0 (mean)NP-6 weeks
Gooding 2020 [20]USADEHR14/399.5 (mean)NP-6 weeks
Hong 2022 [21]ChinaDEHR41/1743.0 (median)NP-4 weeks
Jiang 2019 [22]ChinaDEHR16/795.5 (mean)NP-4 weeks
Kou 2023 [23]ChinaDEHR65/1113.2 (median)NP-4 weeks
Ling 2019 [24]ChinaDEHR44/1225.8 (mean)NP-4 weeks
Liu 2023 [25]ChinaDEHR31/1579.0 (mean)NP-4 weeks
Wang 2012 [26]ChinaDEHR19/434.5 (mean)NP-4 weeks
Ye 2023 [5]ChinaD, EVEHR34/91, 20/504.0 (median)NP-4 weeks
Zhang 2015 [27]ChinaDEHR36/1856.0 (median)NP-4 weeks

Footnote: D: derivation; EHR: electronic health record; EV: external validation; NP: Negative Proteinuria.

Table 2

Basic characteristics of the included studies on SRNS-associated risk factors.

StudyCountryStudy DesignNumber of ParticipantsOutcome DefinitionsIndependent Risk Factors
Agrawal 2021 [16]USAR40NP-7 weeksSerum: interleukin-7, interleukin-9, monocyte chemoattractant protein 1
Agrawal 2020 [17]USAR37NP-7 weeksSerum: adiponectin, matrix metalloproteinase 2, vitamin D binding protein
Cicek 2024 [28]TurkeyR162NP-8 weeksAge
Urine: microscopic hematuria
Cuzzoni 2019 [29]ItalyP62NP-4 weeksAge
Serum: macrophage migration inhibitory factor
Gooding 2020 [20]USAR39NP-6 weeksAge
Serum: glutamine
Hong 2022 [21]ChinaC174NP-4 weeksSerum: white blood cell
Urine: retinol-binding protein
Imbusi 2020 [30]EthiopiaC63NP-8 weeksAge (>6), hypertension
Serum: cholesterol
Kifle 2020 [31]EthiopiaC85NP-4 weeksUrine: gross hematuria, urine output
Konstantelos 2019 [32]CanadaP377Not reportedlow birth weight
Kou 2023 [23]ChinaC111NP-4 weeksSerum: β-microglobulin, D-dimer, erythrocyte sedimentation rate, suppressor T cells
Li 2024 [33]ChinaC510NP-8 weeksPathology, age
Serum: 25-hydroxyvitamin D, CD4+ T cells, creatinine, D-dimer
Urine: proteinuria
Ling 2019 [24]ChinaC122NP-4 weeksSerum: immunoglobulin E, immunoglobulin G
Liu 2023 [25]ChinaC157NP-4 weeksSerum: endothelin-1, polyligandproteoglynan-1
Rehman 2022 [34]PakistanR135Not reportedHypertension
Urine: hematuria
Salah 2023 [35]EgyptP60NP-8 weeksAge
Serum: white blood cell
Udagawa 2021 [36]JapanC80NP-4 weeksGender
Serum: immunoglobulin M
Wang 2012 [26]ChinaC43NP-4 weeksSerum: globulin
Urine: α1-antitrypsin/creatinine, αl-microglobulin
Yin 2010 [37]ChinaC72NP-4 weeksAge
Serum: albumin, D-dimer
Zhang 2015 [27]ChinaC185NP-4 weeksAge (>6.5)
Urine: microscopic hematuria, proteinuria,

Footnote: C: Case-control study; NP: Negative Proteinuria; P: Prospective cohort study; R: Retrospective cohort study.

Publication and source details

Original title: Prediction Models and Risk Factors for Steroid Resistance in Children with Nephrotic Syndrome: A Systematic Review and Meta-Analysis

Yuanhui Hu; Zehui Zhang; Sha Diao; Yannan Guo; Yangtingting Gao; Zheng Xu; Qilin Peng; Yao Xu; Zhenyan Bo; Linan Zeng; Liang Huang; Jingjing Chen; Yizhun Zhu; Hailong Li; Lingli Zhang

Journal of Clinical Medicine · 2026

* Correspondence: lhl_stat@scu.edu.cn (H.L.); zhanglingli@scu.edu.cn (L.Z.); Tel.: +86-28-85503205 (H.L. & L.Z.)© 2026 by the authors.2026https://creativecommons.org/licenses/by/4.0/Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. National Natural Science Foundation of Chinahttps://ror.org/01h0zpd94 72574158The Key Research Project of Sichuan Province2024YFFK0082This study was supported by the National Natural Science Foundation of China (No. 72574158) and The Key Research Project of Sichuan Province (No. 2024YFFK0082).

Repository JATS snapshot; version as supplied

Source retrieved: 2026-11-10T22:23:02.806758+00:00

DOI: 10.3390/jcm15124438 · PMC13301261