Systemic Inflammatory Markers and Quantitative Stool Culture in Machine Learning Classification of Inflammatory Bowel Disease: An Exploratory Case–Control Study
Microorganisms · 09.11.2026 · 14 (11) · 2592
DOI: 10.3390/microorganisms14112592 ↗Kurz gefasst
Fachartikel in der Zeitschrift Microorganisms, online erschienen am 09.11.2026. Verfasst von Luana Alexandrescu und 14 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Machine learning (ML) studies of the gut microbiota rarely test whether microbial data add anything to routine clinical and inflammatory markers. We examined this question in inflammatory bowel disease (IBD) using quantitative stool culture. In this exploratory case–control study, 52 adults with IBD and 34 hospital controls underwent clinical, anthropometric, bioimpedance, biochemical and inflammatory assessment, together with quantitative stool culture interpreted against laboratory reference intervals. Six classifiers were evaluated with repeated nested cross-validation (5 folds × 10 repeats) using culture-only (23), clinical-only (78) and integrated (101) feature sets, and were compared with single inflammatory markers and a three-marker logistic model. Calibration, decision curves and both impurity-based and permutation importance were assessed. After false discovery rate correction, the IBD group had higher C-reactive protein (CRP), high-sensitivity CRP and interleukin-6 (IL-6), higher body mass index, and lower fasting glucose and dietary fibre intake. Culture differences, such as more frequent overgrowth above reference intervals (90.4% vs. 64.7%, q = 0.056), did not survive correction. Gradient boosting reached an AUC of 0.912 [95% CI 0.855–0.969] on the integrated set. The clinical-only model performed similarly (AUC 0.903 [95% CI 0.841–0.964], DeLong p = 0.367), and a logistic model on CRP, hs-CRP and IL-6 discriminated comparably (AUC 0.885 [95% CI 0.808–0.962]). The culture-only model discriminated weakly (AUC 0.637 [95% CI 0.508–0.765]) and was not distinguishable from chance in a label-permutation test (p = 0.25). Within IBD, the treatment-response model (AUC 0.789 [95% CI 0.655–0.924]) did not outperform disease severity alone (AUC 0.87), because all responders had mild disease. In this cohort, discrimination between IBD and controls came from systemic inflammatory markers, and quantitative stool culture added no measurable value. These results do not support diagnostic or screening use; studies including faecal calprotectin, symptomatic controls and external validation are needed.
Zitieren
APA
Alexandrescu, L., Alexandrescu, D. M., Serbanescu, L., Twakor, A. N., Suliman, I. L., Rusu, E., Panculescu, F. G., Tofolean, I. T., Condur, L. M., Aftenie, C., Tocia, C., Docu, S. A., Leopa, N., Mihu, M., & Tofolean, D. E. (2026). Systemic Inflammatory Markers and Quantitative Stool Culture in Machine Learning Classification of Inflammatory Bowel Disease: An Exploratory Case–Control Study. Microorganisms, 14(11), 2592. https://doi.org/10.3390/microorganisms14112592
Vancouver
Alexandrescu L, Alexandrescu DM, Serbanescu L, Twakor AN, Suliman IL, Rusu E, et al. Systemic Inflammatory Markers and Quantitative Stool Culture in Machine Learning Classification of Inflammatory Bowel Disease: An Exploratory Case–Control Study. Microorganisms. 2026;14(11):2592. doi:10.3390/microorganisms14112592
BibTeX
@article{Alexandrescu2026,
author = {Alexandrescu, Luana and Alexandrescu, Daria Maria and Serbanescu, Lucian and Twakor, Andreea Nelson and Suliman, Ioana Livia and Rusu, Elena and Panculescu, Florin Gabriel and Tofolean, Ioan Tiberiu and Condur, Laura Maria and Aftenie, Cristina and Tocia, Cristina and Docu, Silviu Axelerad and Leopa, Nicoleta and Mihu, Miruna and Tofolean, Doina Ecaterina},
title = {Systemic Inflammatory Markers and Quantitative Stool Culture in Machine Learning Classification of Inflammatory Bowel Disease: An Exploratory Case–Control Study},
journal = {Microorganisms},
year = {2026},
volume = {14},
number = {11},
pages = {2592},
doi = {10.3390/microorganisms14112592}
}
EUVIMED