Abstract: Objective: To compare the performance of metagenomic next-generation sequencing(mNGS) and conventional pathogen detection in identifying causative pathogens and characterizing pathogen spectrum in children with severe pneumonia admitted to the pediatric intensive care unit(PICU), using a clinical composite reference standard(CRS) as the reference standard for determining causative pathogens, and to evaluate the additional diagnostic value of mNGS in identifying mixed infections and in CRS-negative cases on conventional testing. Methods: A single-center retrospective study was conducted. A total of 84 children with severe pneumonia admitted to the PICU of the First Affiliated Hospital of Guangxi Medical University from July 2020 to July 2023 were included: 44 in the bronchoalveolar lavage fluid(BALF) group and 40 in the sputum group. All children underwent mNGS of respiratory samples and concurrent conventional pathogen detection during the same infectious episode. Causative pathogens were determined according to the CRS. The detection rate of causative pathogens, diagnostic performance, pathogen spectrum composition, mixed infection rate, and incremental detection rate in CRS-negative cases on conventional testing were compared between mNGS and conventional pathogen detection. McNemar's test and Kappa test were used for statistical analysis. Results: A total of 84 children with severe pneumonia in the PICU were included, and all of them received antibiotic therapy within 72 h before mNGS sample submission. In the overall analysis, the CRS-confirmed causative pathogen positivity rate of mNGS was 94.0%(79/84), which was significantly higher than the CRS-confirmed positivity rate for conventional pathogen detection(33.3%, 28/84)(P<0.001), with low agreement between the two methods(Kappa=0.061). Using CRS as the reference standard, the sensitivity, specificity, positive predictive value, negative predictive value, and accuracy of raw mNGS detection were 100.0%, 20.0%, 95.2%, 100.0%, and 95.2%, respectively. The corresponding values for conventional pathogen detection that was CRS-positive were 35.4%, 100.0%, 100.0%, 8.9%, and 39.3%, respectively. In this study, only 1 case was negative on raw mNGS detection, and it was also classified as negative by the CRS. In the BALF group, the positivity rate of mNGS for CRS-confirmed causative pathogens was 97.7%(43/44), which was higher than that of conventional pathogen detection(38.6%, 17/44)(P<0.001). Pathogen spectrum analysis showed that among the causative pathogen categories determined by the CRS, bacteria accounted for the highest detection proportion(65.5%), followed by viruses(60.7%), special pathogens(34.5%), and fungi(32.1%), and the proportion of mixed infections was 69.0%. Among CRS-negative cases on conventional pathogen detection(n=56), mNGS newly identified causative pathogens in 51 cases, with an incremental detection rate of 91.1%; of these, 29 cases(56.9%, 29/51) underwent adjustment of anti-infective therapy after mNGS results were reported. Conclusion: In the context of widespread empirical antimicrobial therapy among children with severe pneumonia in the PICU, mNGS shows high performance for identifying CRS-confirmed causative pathogens and favorable incremental diagnostic value in mixed infection identification, complementing pathogen spectra, and pathogen identification in CRS-negative cases on conventional testing. Thus, mNGS can serve as an important complementary method to conventional pathogen detection.