Letter: Fecal VOCs in Lynch Syndrome: Sample Stability, Specificity and Statistical Validation

The study by Jacobs et al. [1] explores the potential of fecal volatile organic compounds (VOCs) as a non-invasive biomarker to optimise colonoscopy intervals for patients with Lynch syndrome (LS). The finding regarding the normalisation of VOC patterns after polypectomy was particularly compelling. It adds a dynamic dimension to biomarker research in this field. However, we have several inquiries regarding the methodology and clinical interpretation. First, regarding sample storage, the authors noted that fecal samples were stored at −20°C for a median duration of 1 year. They cited a study on urine VOC stability to justify this timeframe [2]. However, stool harbours a complex and active microbiome that differs significantly from urine. Microbial fermentation and VOC production may persist at −20°C. Standard recommendations usually suggest storing the fecal metabolome at −80°C to effectively halt enzymatic activity [3]. We wonder if the authors performed stability experiments specifically on the fecal matrix at −20°C. This is essential to ensure that the observed VOC profiles reflect the host's pathological state rather than microbial alterations caused by storage. Second, while the sensitivity for advanced neoplasia (AN) was impressive at 89–100%, the specificity was notably lower at 54%–58%. In the context of LS surveillance, the primary objective of a biomarker is to safely delay invasive colonoscopy. A specificity of approximately 55% implies a high false-positive rate. Consequently, nearly half of the patients without AN would still be flagged as requiring colonoscopy. Given the potential anxiety associated with a ‘positive’ cancer screening result in this high-risk population, did the authors simulate the actual reduction in colonoscopy burden [4]? It is crucial to ascertain whether the clinical utility outweighs the potential psychological burden and the resources required to test the large portion of the healthy cohort who screen positive. Finally, the study employed 10-fold cross-validation on a dataset where the target class of interest (CRC + advanced adenoma) comprised only 9 out of 200 individuals. In such a highly imbalanced dataset with an AN prevalence of roughly 4.5%, cross-validation can sometimes yield overly optimistic performance estimates without careful stratification [5]. Furthermore, the risk of model overfitting is substantial when facing high-dimensional data from FAIMS/GC-IMS alongside a paucity of positive cases. Could the authors clarify whether any data augmentation techniques, such as SMOTE, were used to address class imbalance? Additionally, did the authors consider using nested cross-validation to provide a more unbiased assessment of the model's generalisability? We hope these perspectives contribute to the further optimisation of non-invasive surveillance strategies for families with Lynch syndrome. Xidi Yang: conceptualization, writing – original draft, writing – review and editing. Yue Wang: conceptualization, writing – original draft. Fa Tian: conceptualization, writing – original draft. The authors have nothing to report. The authors declare no conflicts of interest. This article is linked to van Liere et al. paper. To view this article, visit https://doi.org/10.1111/apt.18328. The authors have nothing to report.