Development of a novel diagnostic tool for community-acquired pneumonia based on microbial community characteristics.

BACKGROUND

The lung microbiota plays a key role in pneumonia. A correct diagnosis in the early stage helps treatment and limits severity progression of community-acquired pneumonia (CAP).

METHODS

There were 42 patients with CAP and 20 controls. Amplicon bacterial 16S sequencing from bronchoalveolar lavage fluid samples revealed indicator genera by linear discriminant analysis effect size, random forest and ROC analyses. The abundance of bacteria was combined with laboratory inflammatory indexes to construct logistic regression models for CAP diagnosis. The final predictive model with a lower Akaike information criterion (AIC) was internally validated by bootstrap resampling.

RESULTS

After confirming lower respiratory tract (LRT) microbiome dysbiosis in CAP patients, we identified five indicator genera: Moraxella (M), Prevotella_1 (P), Lautropia (L), Neisseria (N), and Fusobacterium (F). Their combined effect was incorporated by calculation of the joint prediction probability (PRE) based on their relative abundances as PRE = 1 / (1 + EXP (-4.39 + 18.96 M + 9.30P - 5.38 L + 0.89 N + 0.70F)). A nomogram was constructed for CAP diagnosis that incorporated PRE with serum levels of CRP (cut-off ≥5.23 mg/L) and Neutrophil % (cut-off ≥66.05%). After internal validation, this nomogram showed satisfactory diagnostic performance with the bootstrap-corrected AUC of 0.927 and a well-fitting calibration curve.

CONCLUSIONS

The LRT microbiome was altered in CAP patients. Indicator genera were shown to have the potential to serve as diagnosis biomarkers in the constructed predictive model. These results may provide further clues for the microbial pathogenesis, optimized diagnostics and new therapeutic targets of CAP.