Letter to the editors: ‘‘Outcomes and prognostic factors in childhood-onset steroid-resistant nephrotic syndrome’’

Yun Cheng, Jinlin LiuPediatric 2025 · DOI · OpenAlex

Introduction

This article presents a formal methodological critique of a recently published study by Zerkowitz et al. (2025) concerning childhood-onset steroid-resistant nephrotic syndrome (SRNS). While identifying prognostic factors for rare conditions is vital for clinical management, the authors of this letter argue that the original study's findings may be compromised by significant statistical "overfitting."

The Problem: Statistical Overfitting in Clinical Research

In clinical research, multivariate regression models are frequently employed to identify variables (prognostic factors) that predict disease progression or treatment outcomes. However, the reliability of these models is intrinsically linked to the ratio of "events" (outcomes such as remission or organ failure) to the number of predictor variables included in the model.

A widely accepted heuristic in medical statistics is the "one in ten" rule. This principle suggests that for every ten outcome events, no more than one predictor variable should be included in a multivariate model to maintain stability and predictive power. When this ratio is ignored, the model risks overfitting—capturing random noise within a specific small dataset rather than identifying true underlying clinical patterns.

Methodological Analysis

The authors performed a meta-methodological analysis of the data reported by Zerkowitz et al. by comparing the number of recorded events against the number of variables tested in their multivariate models. The analysis focused on two primary outcomes:

  1. Remission Analysis: The original study included 32 patients.
  2. Kidney Failure Analysis: The original study recorded 23 events.

By benchmarking these figures against established statistical literature (e.g., Riley et al., 2019; Pate et al., 2023), the authors evaluated the minimum sample sizes required for stable multivariable prediction.

Key Findings: A Fourfold Excess in Variables

The critique reveals a severe departure from standard statistical practices in the original study. The following table summarizes the mismatch between the sample size and the model complexity:

Analysis Type Number of Events Recommended Variables (Max) Actual Variables Used
Remission 32 patients $\le 3$ 12
Kidney Failure 23 events $\le 3$ 14

In the remission analysis, the original researchers used 12 predictors for only 32 patients—a fourfold excess over the recommended limit. For kidney failure, 14 factors were tested against only 23 events. This high "dimensionality" relative to the sample size suggests that the identified prognostic factors are highly susceptible to being artifacts of the specific dataset rather than generalizable medical truths.

Conclusion and Future Directions

The authors conclude that the prognostic factors identified by Zerkowitz et al. cannot be considered definitive until they are validated in a much larger, independent cohort. The "curse of dimensionality"—the desire to test numerous clinical and genetic factors simultaneously—often undermines the scientific validity of small-scale studies.

Future research into SRNS must prioritize larger sample sizes or more focused variable selection to ensure that the complex relationships between genetics and clinical outcomes are reliably mapped. Without such rigor, clinical decisions based on these models may be premature.

Summary updated

BibTeX
@inproceedings{cheng2025letter,
  author = {Yun Cheng and Jinlin Liu},
  title = {{Letter to the editors: ‘‘Outcomes and prognostic factors in childhood-onset steroid-resistant nephrotic syndrome’’}},
  booktitle = {Pediatric},
  year = {2025},
  doi = {10.1007/s00467-025-06944-6},
}
Figure