Patients and sample characteristics

As shown in Fig. 1A, a retrospective study included 510 samples from 451 patients, consisting of 451 BALF samples. Among these, 59 patients underwent both BALF and lung biopsy tissue testing. The clinical and demographic characteristics of the patients are detailed in Table 1, with 320 male patients (71.00%) and 131 female patients (29.00%), and a median age of 63 years. The levels of CRP (median 36.05 mg/L), PCT (median 0.50 ng/mL), and neutrophil percentage (76.45%) were elevated compared to normal levels, lymphocyte percentage (11.35%) and CD4+ T cell count (242.27 cells/µL) were decreased, indicating overall elevated infection markers and reduced cellular immunity. Chronic lung diseases represented the highest proportion at 29.3% (132 cases). Eighty-two cases (18.2%) involved mixed underlying diseases, and the patients had a relatively older age and multiple underlying conditions, with 16 cases (3.50%) at risk of aspiration. Integrating all pathogen results, clinical pharmacotherapy responses, and the medical team’s diagnoses, 436 out of 451 patients suspected of LRTIs were ultimately diagnosed with infections, while 15 were finally diagnosed with non-infected diseases., including tumour (n = 7), organizing pneumonia (n = 3), COPD (n = 2), bronchial asthma (n = 1),pneumoconiosis (n = 1), bronchiectasis (n = 1).Among the infected patients, 69.80% had bacterial infections (n = 315), 5.10% had fungal infections (n = 23), 2.90% had viral infections (n = 13), and 18.80% had mixed infections (n = 85). In mixed infections, 11 patients were found to have concurrent viral and fungal infections, with 8 cases testing positive for SARS-CoV-2, 2 cases with Aspergillus infection, 4 cases with Candida infection, and 2 cases with Pneumocystis jirovecii infection. Additionally, 3 cases tested positive for influenza virus, with 2 cases also showing Aspergillus fumigatus infection and 1 case with Pneumocystis jirovecii. There were 14 cases of viral and bacterial co-infections, primarily involving SARS-CoV-2, influenza A virus, and human parainfluenza virus. One case involved bacterial and atypical pathogen co-infection, with Orientia tsutsugamushi and Streptococcus pneumoniae detected by mNGS. There were 47 cases of bacterial and fungal co-infections. A total of 12 patients had concurrent bacterial, fungal, and viral infections, with 10 cases positive for SARS-CoV-2 and 2 cases positive for influenza A. These cases predominantly involved Aspergillus fumigatus, Candida, and Pneumocystis jirovecii, accompanied by high abundances of common bacteria associated with community-acquired pneumonia (CAP) and hospital-acquired pneumonia (HAP). Almost all patients with these multiple infections had underlying diseases. In terms of outcomes, 94.20% of patients (n = 425) showed improvement after treatment and were discharged, while 5.80% (n = 26) died. According to the mNGS test results, 52.80% of patients were positive for herpes virus (n = 238), while 47.20% (n = 213) were negative.

Table 1 Demographic and clinical characteristics of patients included in this study interquartile range[IQR]Fig. 1

A. Contribution of mNGS testing to the diagnosis of patients with lower respiratory tract infections. A.Enrollment details and study design. B. Diagnostic performance of mNGS and CMT for pathogen detection (upper panel) and diagnostic performance of mNGS in paired lavage fluid and lung tissue samples (lower panel). C. Bar graph comparing pathogen detection between mNGS and culturing, bacterial spectrum comparison between mNGS and CMT, fungal spectrum comparison between mNGS and CMT, and comparison of viruses identified by mNGS and CMT

The false positive and false negative rates of mNGS in infected patients were 2.29% (10/436) and 16.74% (73/436), respectively, with the predictive accuracy of mNGS lower than the 90% threshold reported in many articles. In non-infected patients, the true negative rate of mNGS was relatively high at 86.67%, but there were also two false positive cases, including detection of EBV and human herpes virus 7(HHV-7),and Candida albicans with co-infection of EBV, herpes simplex virus 1(HSV-1), and HHV-7. In this study, instances of false negative mNGS but true positive CMT results were primarily associated with infections involving Klebsiella pneumoniae (n = 3), Escherichia coli (n = 3), Staphylococcus aureus (n = 1), Pseudomonas aeruginosa (n = 1), Acinetobacter baumannii (n = 2), Candida albicans (n = 2), Mycoplasma pneumoniae (n = 1), influenza A virus (n = 1), and SARS-CoV-2 (n = 2). All clinical samples underwent testing using CMT, and the diagnostic performance of mNGS and CMT in LRTIs is illustrated in Fig. 1B. In BALF, mNGS demonstrated a sensitivity of 82.86% and a specificity of 52.00%, with positive predictive value and negative predictive value of 96.71% and 15.12%, respectively. In comparison, CMT showed a sensitivity of 44.42% and a specificity of 21.05%, with positive predictive value and negative predictive value of 79.55% and 5.19%, respectively. The overall specificity, sensitivity, positive predictive value, and negative predictive value of mNGS were significantly higher than those of traditional methods (χ2 = 152.15, p < 0.001). When examining paired samples, the sensitivity of mNGS in tissue and BALF was 57.41% vs. 84.21% and the specificity was 80.00% vs. 50.00%. The positive predictive values were 96.88% vs. 97.96%, and the negative predictive values were 14.81% vs. 10.00%. In general, the accuracy of lavage fluid was superior (83.05% vs. 59.32%) (χ2 = 11.58, p < 0.001).

The distribution of pathogens among patients

Figure 1C illustrates the pathogen distribution among patients in our facility (test results, not diagnostic results). Among bacterial pathogens, Klebsiella pneumoniae, Haemophilus influenzae, Streptococcus pneumoniae, Pseudomonas aeruginosa, Staphylococcus aureus, and Mycoplasma pneumoniae were the most commonly detected, identified through a combination of CMT and mNGS. Among fungal pathogens, Candida albicans, Aspergillus fumigatus, and Pneumocystis jirovecii exhibited relatively higher detection rates. Notably, Pichia kudriavzevii was only detected through culture but not identified by mNGS. Regarding viral pathogens, human herpesvirus 4, human herpesvirus 7, picornavirus, human herpesvirus 5, herpes simplex virus type 1, SARS-CoV-2, and influenza A virus showed relatively higher detection rates. SARS-CoV-2 and influenza A virus are typically detected using qPCR or antigen testing in CMT. It is noteworthy that none of the cases with identified herpesviruses in this retrospective analysis received specific antiviral treatment for herpesviruses. MNGS detected a broader range of pathogens compared to CMT. While most pathogens identified by CMT were also identifiable by mNGS, there were instances where CMT produced true positives whereas mNGS generated false negatives. Instances of this discrepancy were predominantly noted in the isolation of Klebsiella pneumoniae and Klebsiella oxytoca from BALF, where mNGS failed to identify the specific species. Additionally, two cases showed positive IgM for Mycoplasma pneumoniae in serum, resulting in the clinical diagnosis of Mycoplasma pneumonia, despite no Mycoplasma species being detected by mNGS. Moreover, Escherichia coli was cultured from BALF in one patient, whereas mNGS identified Escherichia coli but did not report it. The discrepancy could be due to the frequent presence of Escherichia coli as a contaminant in mNGS reagents and its extensive prevalence in wet laboratory reagents. A relatively high reporting threshold for Escherichia coli in mNGS may lead to its filtration and subsequent false negatives.

Differences between mNGS testing and BALF sampling

Fifty-nine cases underwent pathogen distribution testing in both tissue and BALF samples (Fig. 2A). Identified bacteria comprised Propionibacterium acnes, Staphylococcus aureus, Streptococcus pneumoniae, Lactococcus lactis, and yeast, typically normal flora of the skin surface or respiratory tract, not classified as pathogens. The detection rate of pathogens in BALF exceeded that in tissue, with 35 cases(59.32%) showing consistent results in both sample types, 20 cases(33.90%) with pathogen detection in BALF but not in tissue, and 4 cases with exclusive detection in tissue. Among the latter, two cases revealed Cryptococcus neoformans in the tissue, confirmed as pulmonary cryptococcosis through tissue pathology; one case featured Legionella pneumophila, and one case showed Pneumocystis jirovecii. The initial treatment plan was adjusted based on the test results for these four patients. Species detected in both BALF and tissue, including Streptococcus mitis, Haemophilus influenzae, Klebsiella pneumoniae, Prevotella melaninogenica, and Fusobacterium nucleatum., exhibited higher sequence numbers in BALF than in lung biopsy tissue. For one patient, definitive diagnosis remained elusive despite CMT, pathology, and mNGS testing. The patient was eventually diagnosed with bacterial pneumonia through imaging and clinical diagnostic treatment, yet the pathogen remained unidentified. False positives also emerged in lung biopsy tissue, exemplified by a case of lung abscess with pneumonia indicating Candida haemulonii in the lung tissue and Haemophilus influenzae in BALF. The pneumonia was ultimately treated without considering the potential involvement of Candida in the infection. Overall, BALF demonstrated higher sensitivity but lower specificity than tissue. The most frequently detected species in both BALF and tissue were EBV, followed by Fusobacterium nucleatum, Mycoplasma pneumoniae, Chlamydia psittaci, Haemophilusin fluenzae, Parvimonas micra, Porphyromona sendodontalis, Eubacterium brachy, Haemophilus parainfluenzae, CMV.

Fig. 2

Results of mNGS testing on BALF and synchronous lung tissue samples. (A) Heatmap of species sequencing numbers detected in lavage fluid and tissue samples (sequencing numbers were normalized). (B) Feature Sankey diagram of tissue detection. Frequency and relative abundance of herpesvirus infections

Among patients undergoing both tissue biopsy and lavage fluid testing, 18 cases (30.51%) received a diagnosis of lung abscess with pneumonia, all confirmed through mNGS testing. Bacterial infections were present in all 18 patients, including one case of mixed bacterial and fungal infection, and one case of mixed bacterial and SARS-CoV-2 infection. Lung abscesses often result from multiple infections from dental or oral flora, typically involving weak pathogens and difficult-to-culture anaerobic or facultatively anaerobic bacteria. Notable species encompassed Fusobacterium nucleatum., Prevotella spp., Streptococcus intermedius, Streptococcus anginosus, Porphyromonas gingivalis, Rothia mucilaginosa, Streptococcus mitis and Campylobacter showae. Pathological findings commonly revealed focal acute suppurative changes and chronic inflammation (Table 2).

Table 2 Pathologic findings and final clinical diagnosis of patients with tissue and BALF mNGS were done simultaneouslyDetection statistics of herpesviruses

Among the 451 samples subjected to mNGS, EBV was detected in 33.04% of individuals. The subsequent prevalent viruses included HHV-7 (91 occurrences, 20.18%), CMV (54 occurrences, 11.97%), HSV-1 (44 occurrences, 9.76%), and Herpesvirus-6B (HHV-6B) (17 occurrences, 3.77%) (Fig. 3.A). Scatter plots were created using the natural logarithm (Ln) of their relative abundances. Significant differences were observed in the relative abundances of HSV-1 compared to HHV-7 (p = 0.0061) and CMV (p = 0.0227). The median relative abundance of HHV-6B was the highest (3.03%), followed by EBV (1.02%), HSV-1 (1.02%), CMV (0.28%), and HHV-7 (0.13%) (Fig. 3B).

Fig. 3

Detection Frequency and Abundance of Human Herpesviruses. (A) Prevalence of the five human viruses in suspected lower respiratory tract infection patients at our center, arranged by their prevalence.(Left) Viral load is presented on the x-axis as the proportion of viral sequences detected per patient sample relative to all microorganisms in that sample, with values transformed to a logarithmic scale, the bar represents the median(Right). (B) Heatmap illustrated the detection rates of various subtypes of herpesviruses across patient populations with different underlying diseases

Subsequently, we tallied the instances of concurrent herpesvirus detection. Fluid samples from immunosuppressed hosts facilitate the easy identification of herpesviruses, leading us to conduct a statistical analysis of herpesvirus subtype variations based on the patients’ underlying diseases (Fig. 3B). EBV was the most commonly detected virus, with detection rates of 22.22% in patients with Cardiovascular and cerebrovascular diseases, 16.67% in Chronic lung diseases, 12.20% in Comorbidity with multiple underlying diseases, 10.00% in Immunosuppressed patients, 7.14% in Malignant tumor patients, 12.50% in patients with Metabolic diseases (diabetes), and 18.75% in patients at risk of aspiration. The detection rates of HHV-7 were 7.41% in patients with Cardiovascular and cerebrovascular diseases, 4.55% in Chronic lung diseases, 12.20% in Comorbidity with multiple underlying diseases, 30.00% in Immunosuppressed patients, 10.71% in Malignant tumor patients, and 4.17% in patients with Metabolic diseases (diabetes). It is notable that the proportion of immunosuppressed patients was relatively high among those with detected HHV-7. Patients with detected EBV exhibited the highest proportion of Cardiovascular and cerebrovascular diseases (33.32%) and Chronic lung diseases (25.77%). In patients with Comorbidity with multiple underlying diseases (n = 82), the simultaneous detection rate of EBV and HHV-7 was 10.98%, followed by Malignant tumor (n = 28, 10.71%) and Metabolic diseases (diabetes) (n = 48, 10.42%). The distribution of different herpesvirus subtypes among patients with other types of underlying diseases was relatively uniform.

Risk factors affecting adverse prognosis in patients with LRTIs

Out of 451 LRTIs patients, 425 cases (94.24%, 425/451) demonstrated improvement, while 26 cases (5.75%, 26/451) resulted in mortality. To investigate the risk factors contributing to mortality in LRTIs patients, we initially performed a baseline differential analysis of various outcome groups (Table 3). The Mortality group exhibited a lower lymphocyte percentage than the Improvement group (6.65 Vs. 11.80, p = 0.003). A significant disparity in the prevalence of underlying diseases between the two groups was observed(p < 0.001). Although the duration from admission to mNGS testing was lengthier in the Mortality group (4 days Vs. 2 days, P = 0.08), the variance was not statistically significant. The distribution of CMT diagnostic results varied significantly between the two groups (P = 0.02). However, factors such as CRP, PCT, WBC, neutrophil percentage, CD4+ T cell levels, age, gender, mNGS diagnostic outcomes, infection status, mixed infection, and whether the herpes virus did not reveal significant differences between the Improvement group and the Death group (P > 0.05) (Table 3). Subsequently, factors with P < 0.10, including Age, PCT, Days from admission to mNGS testing, Lymphocyte percentage, Underlying disease, CMT, and Herpesvirus positive, underwent further evaluation through logistic regression. Initially, univariate regression analysis was conducted, revealing that an extended time from admission to mNGS testing (P = 0.02, OR: 1.10; 95% CI: 1.01, 1.19), low lymphocyte percentage (P = 0.011, OR: 0.93; 95% CI: 0.88, 0.98), the presence of underlying diseases, particularly Chronic lung diseases (P = 0.001, OR: 0.11; 95% CI: 0.03, 0.42), Comorbidity with multiple underlying diseases (P = 0.007, OR: 0.13, 95% CI: 0.03, 0.58), false negative CMT test results (P = 0.015, OR: 0.30, 95% CI: 0.11, 0.79), and Herpesvirus positive (P = 0.089, OR: 0.48, 95% CI: 0.20, 1.12) may be associated with poor prognosis in LRTIs patients (Table 4). When the P value in univariate analysis is less than 0.1, a multivariate regression analysis was conducted. Chronic lung diseases (P = 0.024, OR: 0.18, 95% CI: 0.04, 0.79) and Comorbidity with multiple underlying diseases (P = 0.01, OR: 0.12, 95% CI: 0.02, 0.61) were significantly associated with adverse prognosis in LRTIs patients.

Table 3 Characteristics of basic information in the two groups of patients enrolled in this study with improved outcomes and mortalityTable 4 Univariate and Multivariate Logistic Regression Analysis of Risk Factors for Recovery in CAP Patients

Subsequently, a nomogram model was developed to predict mortality risk in LRTIs patients using four significant risk factors identified through univariate regression. The left breakpoint of each scoring line segment corresponds to 0 points. A 10.00% decrease in lymphocyte proportion results in a 6.25 points increase in the nomogram model score, with a maximum value corresponding to a point of 63. The breakpoints for “Underlying disease” in “Risk of aspiration” correspond to 0 points, “None” to 0.50 points, “Chronic lung diseases” to 81 points, and “Comorbidity with multiple underlying diseases” to 77.50 points. In “CMT,” “True negative” corresponds to 0 points, “False negative” to 80 points, and “True positive” to 89 points. “Herpes virus positive” corresponds to 11 points. The total score for all variables is 402 points. As depicted in Fig. 4A, a higher total score in the nomogram model indicates an increased risk of mortality for LRTIs patients. Furthermore, the calibration curve of the modeling set (Fig. 4B) exhibits no significant deviation from the reference line, signifying strong consistency between predicted and actual values. To assess the nomogram model’s accuracy in predicting mortality risk in patients with LRTIs, a nomogram ROC curve was generated (Fig. 4C) with an AUC value of 0.825 (P < 0.05, 95% CI: 0.75, 0.90). This model effectively predicts the likelihood of mortality risk, with a cutoff value of 0.77 (95% CI: 0.69, 0.83).

Fig. 4

Adverse prognostic risk factors in CAP patients. (A) Nomogram and scoring methodology for adverse prognostic factors in CAP patients. Each predictive variable’s value (the line after each variable) corresponds to a score (top row), which is then totaled to obtain the overall score, determining the corresponding predictive probability (bottom row). (B) Calibration analysis of the training set, where greater alignment with the reference line indicates more precise predictions. (C) Nomogram model-generated ROC curve for the modeling set in predicting mortality risk in CAP patients