Key points are not available for this paper at this time.
Objective This study aims to characterize the GM in LUAD patients and develop and validate a GM-based diagnostic model for LUAD. Methods In this prospective, randomized, multi-center study, the GM was characterized, and an LUAD classifier was developed using a training cohort of 175 early-stage LUAD patients and 107 healthy controls. The model was further validated in a test cohort, two independent external cohorts from Jiangsu and Hainan, and an advanced LUAD cohort. Additional ML models were also developed and compared to assess their predictive performance. Results LUAD patients exhibited reduced microbial diversity and significantly altered microbial composition compared to healthy controls. The phylum Verrucomicrobia and 13 genera, including Enterococcus and Akkermansia , were more abundant in the LUAD group, while 5 phyla, such as Fusobacteria and Cyanobacteria, and 17 genera, including Lactobacillus and Weissella , were enriched in the control group. Using random forest (RF), eight operational taxonomic units were identified as the optimal subset, achieving an area under the curve (AUC) of 0.998 in the training cohort and maintaining high accuracy in the test cohort (AUC = 96.9%). The model also demonstrated robust performance in two independent cohorts from Jiangsu (AUC = 97.6%) and Hainan (AUC = 82.9%), with strong diagnostic potential for advanced LUAD. Among five common models, the RF model exhibited the highest diagnostic accuracy. Conclusions This study provides a comprehensive characterization of the gut microbiome in LUAD and develops a diagnostic model based on microbial biomarkers, which is validated across regionally diverse cohorts, highlighting its potential as a reliable and non-invasive screening tool for LUAD.
Zhang et al. (Thu,) studied this question.