Quantitative interpretation models for targeted next-generation sequencing in lower respiratory tract infections: a
Chuwei Jing1, Yuchen Ding1, Ji Zhou1
1Department of Respiratory and Critical Care Medicine, Jiangsu Province Hospital/Nanjing Medical University First Affiliated Hospital, Nanjing, Jiangsu, 210029, China.
Background:
Lower respiratory tract infections (LRTIs) represent a significant global health burden. While targeted next-generation sequencing (tNGS) offers potential advantages for pathogen detection, its clinical implementation is hindered by the absence of validated quantitative interpretation criteria for pathogen discrimination.
Methods:
We conducted a multicenter prospective study of 631 patients with suspected LRTIs across five intensive care units in eastern China from January 2022 to March 2025. Bronchoalveolar lavage fluid specimens underwent concurrent tNGS and conventional microbiological testing (CMT). Expert group A established the reference standard by classifying patients into LRTI/non-LRTI categories and identifying clinically significant pathogens based on comprehensive clinical criteria. Expert group B, blinded to tNGS quantitative data, provided qualitative interpretation based solely on detected microorganisms to eliminate any influence from quantitative parameters. Expert group C, blinded to all tNGS data, provided interpretation based on conventional microbiological testing combined with clinical manifestations. Quantitative diagnostic models incorporating reads per kilobase per million mapped reads (RPKM) and pathogen copy numbers were developed using a training cohort (n = 420) and validated in an independent cohort (n = 211).
Results:
Of 631 patients, 358 (56.7%) met the diagnostic criteria for LRTI. Polymicrobial infections were identified in 77 patients, with the majority co-infected with Acinetobacter baumannii and Pseudomonas aeruginosa. tNGS demonstrated enhanced detection of Gram-negative bacteria, Candida species and Pneumocystis jirovecii, while CMT showed better detection for Aspergillus species. The quantitative models demonstrated excellent discriminatory performance for bacterial pathogens. The sensitivity and specificity for conventional microbiological testing alone were 58.7% and 74.7%. Adding clinical manifestations to CMT resulted in a sensitivity of 68.8% and specificity of 72.0%. In comparison, qualitative tNGS achieved a sensitivity of 78.5% and a specificity of 76.6%, while the model-based algorithm demonstrated the highest diagnostic accuracy with a sensitivity of 82.4% and a specificity of 85.0%. For antimicrobial resistance prediction, tNGS achieved moderate accuracy (AUC 0.715) with high concordance for key antimicrobial resistance markers including KPC, NDM, OXA-48 and mecA.
Conclusion:
We developed and validated quantitative models for tNGS-based pathogen detection in LRTIs, enabling precise discrimination between pathogenic and background organisms. These models represent a significant step forward in the clinical application of tNGS for LRTI diagnosis and antimicrobial resistance detection.
