Based on the established inclusion and exclusion criteria, a total of 113 patients were retrospectively enrolled. According to the mNGS and CMT results, 44 patients (38.94%) were classified into the Antimicrobial resistance test positive (AMRT+) group, while the remaining 69 patients were belonged to the Antimicrobial resistance test negative (AMRT-) group. The demographic and baseline characteristics of the enrolled patients were detailed in Table 1. No significant differences were observed between the two groups regarding gender and age. However, the body mass index (BMI) of AMRT + patents was slightly lower than that of AMRT- patients (14.80 ± 2.58 vs. 16.12 ± 3.53, P = 0.0411). In comparison to AMRT- patients, a significantly higher proportion of AMRT + cases exhibited immunosuppression (50% vs. 28.99%, P = 0.0242) and had underlying disease (70.45% vs. 40.58%, P = 0.0019), with leukemia being predominant among them. Most clinical features did not differ significantly between both groups, except for a notable increase of C-reaction protein (CRP) level among AMRT + children compared to the AMRT- cases (67.69 ± 80.16 vs. 29.18 ± 46, P = 0.0025). Scores measuring disease severity varied significantly between groups: pediatric sequential organ failure assessment (pSOFA) and pediatric risk of mortality (PRISM) were higher in the AMRT + group, whereas pediatric critical illness score (PCIS) were elevated in the AMRT- group (P < 0.05). Additionally, both length of hospital stay (LOHS) and ICU duration were significantly longer for those in AMRT + group compared to those in AMRT- group. However, mortality rates did not show any significant difference between these two groups.

Among these patients, 28 experienced multi-site infections. Pulmonary infections (n = 79) were most prevalent, followed by bloodstream infections (n = 29) and central nervous system infections (n = 25). Furthermore, a markedly higher incidence rate of pulmonary infection (88.64% vs. 57.97%, P = 0.0005) and central nervous system infection (45.45% vs. 7.25%, P < 0.0001) was noted within the AMRT + group compared to the AMRT- group (Table 2). The most frequently observed disease types included severe pneumonia (n = 53), sepsis (n = 34), septic shock (n = 25), pneumonia (n = 24) and encephalitis (n = 19), with incidence of these first three conditions being substantially higher in AMRT + group compared to AMRT- group (P < 0.05). Conversely, the proportion of encephalitis was found to be significantly greater in the AMRT- group than in AMRT + group (23.19% vs. 6.82%, P = 0.0233).

The distribution of samples used for mNGS testing is illustrated in Supplementary Fig. 1, with BALF being the most common sample, followed by CSF and blood. The pathogen detection rate in AMRT + samples was significantly higher than that in AMRT- samples, irrespective of the detection method employed (Fig. 1A). Within the AMRT + group, no significant differences were noted in pathogen detection rates across various methods. Conversely, in the AMRT- group, the positive culture rate was markedly lower compared to those of mNGS and CMT (P < 0.0001) (Fig. 1A). In the AMRT + Group, the consistency rate (complete match + partial match) between mNGS and culture reached an impressive 75%. However, this proportion was only 1.45% within the AMRT- group (Fig. 1B). When compared to CMT, the consistency of pathogen detection via mNGS was found to be 81.82% in the AMRT + group and merely 21.74% in the AMRT- group (Fig. 1C). Compared to culture and CMT results, the sensitivity of mNGS was 89.74% and 90.70% in AMRT + group, and was 90.01% and 70.59% in AMRT- group (Fig. 1D).

The diagnose performance of mNGS, culture and CMT for pathogens in AMRT + and AMRT- patients. (A) The positive rate of mNGS, culture and CMT for pathogens in AMRT + and AMRT- patients. (B) The pathogen consistency between mNGS and culture for AMRT + and AMRT- patients. (C) The pathogen consistency between mNGS and CMT for AMRT + and AMRT- patients. (D) 2 × 2 contingency tables comparing the performance of mNGS relative to clinical testing

Significant differences were also observed regarding pathogen types detected between both groups. The most common Gram-positive bacteria in both groups were Streptococcus and Staphylococcus. Gram-positive bacterial detection rates were higher than that of Gram-negative bacteria within the AMRT- group. In contrast, a greater variety of Gram-negative bacteria were detected in the AMRT + group, mainly Klebsiella, Pseudomonas and A. baumannii (Fig. 2A). Candida albicans was the most common detected fungus in both groups, and rare fungi such as Rhizopus microsporus and Cryptococcus neoformans were also detected in the AMRT + group (Fig. 2B). Human herpesvirus, Human parainfluenza virus and rhinovirus are the most common detected viruses. Notably, more viral types were identified within AMRT- samples compared to those from AMRT + patients (Fig. 2C).

The pathogen profiles for the AMRT + and AMRT- patients. (A) Spectrum of detected bacteria. (B) Spectrum of detected fungi. (C) Spectrum of detected virus

The distribution of drug resistance identified in pathogens was systematically summarized and analyzed. Gram-negative bacteria exhibit a broader spectrum of resistance, predominantly resistant to cephalosporins, β-lactams/β-Lactamase inhibitors, carbapenems and sulfonamides (Fig. 3). The resistance profiles of the three most prevalent Gram-negative pathogens to these four classes of antibiotics demonstrated significant variability. K.pneumoniae displayed highly levels of resistance to cephalosporins, β-lactams/β-Lactamase inhibitors, and sulfonamides, with rates recorded at 46.15%, 39.29% and 44.44%, respectively; it showed slightly lower resistance to carbapenems at a rate of 21.05%. P. aeruginosa exhibited higher resistance rates to carbapenems (33.33%), but lower rates for cephalosporins, β-lactams/β-Lactamase inhibitors and sulfonamides, with rates of 0%, 5.26% and 0%, respectively. A. baumannii demonstrated nearly complete resistance to cephalosporins (91.67%), carbapenems (100%) and sulfonamides (100%), alongside a resistance rate of 66.67% against β-lactams/β-Lactamase inhibitors. In contrast, Gram-positive bacteria primarily exhibited resistance towards antibiotics such as penicillins, marcrolides, and lincosamides. Among the antibiotics with sample counts exceeding 22 (50%) in the AST analysis, Levofloxacin, Trimethoprim and sulfamethoxazole, Cefoperazone and Sulbactam, Ceftazidime, Piperacillin and Tazobactam, Imipenem, Cefepime, Tigecycline, Meropenem, Amikacin and Ceftriaxone had a resistance rate of 16.67%, 60%, 33.33%, 46.67%, 41.38%, 41.38%, 37.93%, 0%, 41.67%, 8.70% and 50%, respectively. The drug resistance of pathogens in 2022 and 2023 is highly consistent (Supplementary Fig. 2). Furthermore, we analyzed changes in bacteria resistance to cephalosporins,β-lactams/β-Lactamase inhibitors, carbapenems and sulfonamides over different years, and results showed that K.pneumoniae showed lowest resistant to cephalosporins and β-lactams/β-Lactamase inhibitors in 2023; P. aeruginosa only showed 14.29% resistance to β-lactams/β-Lactamase inhibitors in 2023, and showed same resistant rate to carbapenems in 2022 and 2023; the resistance rate of A. baumannii to carbapenems and sulfonamides was 100% in 2022–2024, and the resistance rate to cephalosporins andβ-lactams/β-Lactamase inhibitors was higher than 50%, with the lowest levels occurring in 2023.

Antibiotics resistance rate of the AST results. Left panel: heatmap of antibiotics resistance rate by AST; Right panel: AST result for each antibiotic. Gray box means no AST test was performed

A total of 76 drug-resistant genes were detected via mNGS (Supplementary Fig. 4). The most frequently detected gene was Mex, followed by ade, mdt, OXA, Mux, and Opm, which identified for more than 10 times. Using AST results as the gold standard we evaluated mNGS’s performance in predicting antibiotic resistances across three classes: cephalosporins, carbapenems and sulfonamides. The sensitivity for predicting carbapenems resistance surpassed that for other categories (66.67% vs. 60.00% vs. 15.00%), and the specificity showed highest value to sulfonamides (92.31%). Furthermore, the accuracy in predicting carbapenems resistance also exceeded that for the other two categories (65.60% vs. 51.28% vs. 45.45%) (Table 3).