Metagenomic next-generation sequencing (mNGS) guides targeted antibiotic therapy and reduces mortality in sepsis: a propensity-matched retrospective cohort study

Abstract

Objectives

To evaluate the impact of metagenomic next-generation sequencing (mNGS)-guided antimicrobial therapy on mortality in patients with sepsis.

Methods

A retrospective cohort study was conducted on 303 patients with sepsis admitted to the ICU between January 2021 and December 2022. Patients were divided into the mNGS group (93 cases) and the non-mNGS group (130 cases). The primary outcome was 28-day mortality. Statistical analysis included binary logistic regression, propensity score matching (PSM), and mediation analysis.

Results

A 1:1 PSM was performed for the original cohort (n = 303), yielding a matched cohort of 160 patients (n = 80 each group). In the matched cohort, antibiotics adjustments based on pathogenic results were more frequent and the 28-day mortality was lower in the mNGS group than in the non-mNGS group (P < 0.01). Binary logistic regression further confirmed significant associations of mNGS detection with a higher likelihood of antibiotic adjustment (odds ratio [OR] = 29.742, 95% CI [10.630, 83.170]) and shorter time to antibiotic adjustment (OR = 0.671, 95% CI [0.566, 0.795]), also with less skin and soft tissue infection (OR = 0.089, 95% CI [0.009, 0.880]) and abdominal infection (OR = 0.206, 95% CI [0.052, 0.812]). Each 1-point increase in the Sequential Organ Failure Assessment (SOFA) score was associated with a 16.9% higher risk of mortality (OR = 1.169, CI [1.054, 1.297]). The decision of antibiotic adjustment (OR = 0.252, CI [0.101, 0.627]) was negatively associated with 28-day mortality. The treatments of mechanical ventilation (OR = 4.260, CI [1.783, 10.181]) and continuous renal replacement therapy (CRRT) (OR = 4.253, CI [2.117, 8.541]) were positively related to the 28-day mortality. The mediation analysis indicated the total effect of mNGS detection on 28-day mortality was significant (Z = − 2.754, P = 0.006), while the indirect effect through antibiotics adjustment was not significant (Z = 0.367, P = 0.714).

Conclusion

The mNGS detection facilitates antibiotic adjustment and is associated with reduced 28-day mortality in septic ICU patients. These findings highlight its potential for broader clinical application.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-025-11854-x.

Introduction

Sepsis is a significant public health problem and a life-threatening condition characterized by organ dysfunction due to a dysregulated host response to infection [1, 2]. It is one of the leading causes of death in the intensive care unit (ICU) [3], with over 5 million sepsis-related deaths annually worldwide [4]. The prognosis of sepsis presents substantial challenge for clinicians [5]. Despite advancements in early identification and standardized management protocols, such as fluid resuscitation and hormone use, the early optimization of antibiotic therapy remains crucial for the successful treatment of sepsis [6].

Currently, microbiological culture is the most common method for identifying pathogenic microorganisms in sepsis [1]. However, its clinical applicability is limited due to the low culture positive rate for certain microorganisms and the longer culture cycle required, leading to delayed or inadequate targeted antimicrobial therapy [7, 8]. To address these limitations, metagenomic next-generation sequencing (mNGS) has gained widespread use due to its high sensitivity in pathogen detection [9]. Compared to traditional culture methods, mNGS is an innovative technology that combines high-throughput sequencing with bioinformatics analysis. Its advantages include comprehensive spectrum detection, semi-quantitative analysis (the number of gene sequences reflects the pathogen load), rapid detection time, guidance for antibiotic use, and accurate detection of bacteria, fungi, viruses, and parasites through DNA or RNA sequencing of clinical samples [10, 11].

The mNGS method provides a clear etiological diagnosis for patients with severe infections, enabling more targeted drug treatment [12]. Recent studies comparing mNGS with traditional diagnostic methods have demonstrated that mNGS can provide valuable diagnostic information and contribute to optimizing antibiotic treatment [13, 14]. Although mNGS technology has the potential to rapidly detect a wide range of pathogens, there is little clinical evidence to show whether its use improves therapeutic decisions and reduces mortality. Recent cohort studies of mechanically ventilated COVID-19 patients have reported increased multidrug-resistant organism infections, elevated mortality rates, and significant challenges in guiding antimicrobial therapy [15, 16]. These findings demonstrate an urgent need for rapid, accurate diagnostic approaches to enable targeted antibiotic stewardship in the ICU.

Therefore, this study assessed the impact of mNGS implementation on antimicrobial regimen modification and 28-day mortality in ICU-admitted sepsis patients.

Method

Patient selection

A retrospective cohort study was conducted, involving 303 patients with sepsis admitted to the ICU of Foshan Hospital of Traditional Chinese Medicine between January 2021 and December 2022. Based on predefined inclusion and exclusion criteria, 80 patients with incomplete clinical data were excluded.

Inclusion criteria: (1) Patients meet the Sepsis 3.0 criteria [2]: for patients with confirmed or suspected infection, the sepsis-related Sequential Organ Failure Assessment (SOFA) score is increased by 2 points from baseline; (2) Patients are aged more than 18 years old. Exclusion criteria: (1) patients with incomplete clinical data; (2) ICU length of stay < 3 days; (3) multiple organ dysfunction is difficult to be distinguished from non-infectious factors, such as immunological diseases, trauma, and cancer.

Patients were divided into two groups based on whether they underwent mNGS detection: the mNGS group and the non-mNGS group. Both groups underwent routine microbial culture based on the suspected infection site. Specifically, 93 patients in the mNGS group received mNGS detection, while 130 patients in the non-mNGS group did not detect (Fig. 1). Samples for mNGS detection included alveolar lavage fluid, urine, cerebrospinal fluid, peritoneal effusion, central venous blood, and secretions. If the mNGS report identified pathogen-derived gene sequences (viruses, rickettsiae, chlamydia, mycoplasma, bacteria, and fungi), the detection was considered positive. No novel gene segments were detected in this research.

Fig. 1.

A schematic of the study profile

Clinical data collection

Clinicopathological data and clinical courses of the patients were retrospectively collected from hospital medical records. Baseline characteristics collected included age, gender, medical history (e.g., hypertension, diabetes, coronary heart disease [CHD]), and surgical history, all retrieved from patients’ inpatient medical records. Clinical data during ICU hospitalization were collected, including the Acute Physiology And Chronic Health Evaluation (APACHE II [17]), SOFA score, white blood cell count, procalcitonin (PCT), platelet count, total bilirubin, serum creatinine, troponin I, N-terminal pro-brain natriuretic peptide (NT-proBNP), and antithrombin III (AT-III). Additional collected data included treatment measures: continuous renal replacement therapy (CRRT), mechanical ventilation (MV), and the use of vasoactive drugs. Infection sites included the pulmonary, urinary tract, brain, abdominal cavity, myocardium, the skin and soft tissue.

Bacterial culture samples were collected from the following sources based on infection site: alveolar lavage fluid (pulmonary infection), urine (urinary tract infection), cerebrospinal fluid (intracranial infection), peritoneal effusion (abdominal infection), central venous blood (endocardial infection), and secretions (skin and soft tissue infection or pelvic infection). Causative agent types were retrieved from traditional culture results. The time of antibiotics adjustment, the length of ICU stays, the cost of hospitalization, and 28-day mortality were all collected. Detailed distributions of pathogen types, infection sites, and treatments were presented in the supplementary table. The primary outcome measure was the 28-day mortality.

Statistical analysis

Data analysis was performed using SPSS software (version 29.0, IBM, Armonk, NY, USA). Normality of data distribution was assessed using the Kolmogorov-Smirnov test, with P > 0.05 indicating normality. Continuous variables with a normal distribution were presented as mean ± standard deviation and compared using independent samples t-test. Non-normally distributed data were summarized as median (interquartile range) and compared using the Mann-Whitney U test. Categorical variables were analyzed using Pearson’s chi-square test or Fisher’s exact test. Binary logistic regression was used for multivariate analysis. A P-value < 0.05 was considered statistically significant.

To minimize selection bias between the mNGS group and the non-mNGS group, PSM was conducted. Nearest-neighbor matching (1:1) was performed within a caliper width of 0.02 standard deviations without replacement, based on the estimated propensity scores. Given the limited sample size, robust methods were employed to conduct mediation analysis, examining whether mNGS detection influenced patient mortality through antibiotic adjustment. The indirect effects were estimated using the Bootstrap method with 5000 resamples, and 95% confidence intervals (CI) were calculated to assess the significance of the mediation effects. The significance of the indirect effect was determined by examining whether the 95% CI excluded zero.

Result

Clinical characteristics of the two groups before and after PSM

Before PSM, the number of patients with abdominal infection was higher in the non-mNGS group compared to the mNGS group (P < 0.01). A 1:1 PSM was conducted to match the two groups based on the following variables: age, gender, underlying diseases, APACHE II, SOFA score, infection sites, PCT level, pro-BNP level, and vasoactive drug use status. After matching, 80 patients who underwent mNGS detection were successfully propensity-matched with 80 patients who did not receive mNGS detection. All the included variables showed good balance between the two groups after matching. Notably, there were no significant differences in the length of ICU stay or hospitalization cost between the two groups either before or after PSM (all P > 0.05) (Table 1).

Table 1.

Comparison of clinical characteristics between mNGS group and non-mNGS group

Patient characteristic	Before PSM				After PSM
	mNGS group (N = 93)	Non-mNGS group (N = 130)	P-value	SMD	mNGS group (N = 80)	Non-mNGS group (N = 80)	P-value	SMD
Age (year)	71.00 (59.00, 80.00)	71.50 (57.25, 81.00)	0.98	0.07	72.00 (62.25, 81.25)	74.50 (58.25, 83.75)	0.58	−0.01
Gender			0.57	0.08			0.72	−0.1
Male, n (%)	60 (64.50)	79 (60.80)	/	/	24 (34.30)	46 (65.70)	/	/
Female, n (%)	33 (35.50)	51 (39.20)	/	/	26 (37.10)	44 (62.90)	/	/
Medical history
Hypertension, n (%)	49 (52.70)	57 (43.80)	0.19	0.18	38 (54.30)	36 (51.40)	0.73	−0.04
Diabetes mellitus, n (%)	26 (28.00)	40 (30.80)	0.65	−0.07	18 (25.70)	19 (27.10)	0.85	0.07
CHD, n (%)	24 (25.80)	27 (20.80)	0.38	0.12	19 (27.10)	17 (24.30)	0.70	−0.05
Surgery, n (%)	25 (26.90)	43 (33.10)	0.32	−0.13	19 (27.10)	18 (25.70)	0.85	−0.07
Disease and severity assessment score
APACHE II/score	22.00 (15.50, 26.00)	21.00 (16.00, 27.00)	0.76	−0.14	20.67 ± 7.51	22.59 ± 8.23	0.12	−0.02
SOFA/score	7.00 (5.00, 9.00)	7.00 (5.00, 10.00)	0.37	−0.16	7.50 (5.00, 10.00)	8.00 (5.00, 11.00)	0.52	−0.07
Time of ICU stays, n (day)	8 (7.00, 21.50)	10 (5.00, 15.75)	0.68	0.3	12.00 (7.00, 21.00)	11.00 (6.00, 17.75)	0.52	0.4
Hospitalization cost (dollar)	28996.38 (20217.96, 41711.59)	28377.11 (21025.48, 40281.33)	0.21	0.47	29450.92 (20243.32, 42528.55)	28934.93 (22804.93, 44542.15)	0.61	0.63
Infection sites
Pulmonary infection, n (%)	76(81.70)	99(76.20)	0.32	0.15	58(82.90)	58(82.90)	1	0.15
Urinary tract infection, n (%)	13(14.00)	20(15.40)	0.77	−0.03	10(14.30)	11(15.70)	0.81	0.03
Brain infection, n (%)	2(2.20)	1(0.80)	0.38	0.09	1(1.40)	1(1.40)	1	0.13
Abdominal infection, n (%)	3(3.20)	20(15.40)	0.003	−0.4	3(4.30)	5(7.10)	0.47	0
Endocardial infection, n (%)	1(1.10)	2(1.50)	0.77	−0.09	0	0	/	0
Skin and soft tissue infection, n (%)	1(1.10)	8(6.20)	0.06	−0.26	1(1.40)	5(7.10)	0.09	−0.43
Pelvic infection, n (%)	0(0)	2(1.50)	0.23	−0.21	0	0	/	−0.18

SMD Standardized mean difference, PSM Propensity score matching, CHD Coronary heart disease, APACHE II The acute physiology and chronic health assessments, SOFA Sequential organ failure assessment

Before PSM, antibiotic adjustment based on pathogen identification was more frequent in the mNGS group (P < 0.01), while both the time to antibiotic adjustment and the 28-day mortality were lower in the mNGS group compared to the non-mNGS group (P < 0.01). After PSM, these differences remained statistically significant, except for the time to antibiotic adjustment (Table 2).

Table 2.

Patient characteristics of both groups before and after PSM

	Before PSM			After PSM
	mNGS group (N = 93)	Non-mNGS group (N = 130)	P-value	mNGS group (N = 80)	Non-mNGS group (N = 80)	P-value
Time of antibiotic adjustment (day)	2.00 (2.00, 3.00)	7.50 (4.75, 9.25)	0.007	2.00 (2.00, 3.00)	4.50 (2.00, 8.00)	0.57
Antibiotics adjustment based on pathogenic results, n (%)	82 (88.20)	79 (60.80)	0.001	70 (87.50)	47 (58.80)	0.001
28-day mortality, n (%)	24 (25.8)	51 (39.2)	0.012	22 (27.5)	39 (48.8)	0.006

PSM Propensity score matching

Relationship of mNGS detection with antibiotic adjustment decision and timing

A binary logistic regression model was constructed to evaluate the association between mNGS detection and several independent factors, including gender, age, MV, CRRT, infection site, APACHE II score, SOFA score, antibiotic adjustment, and timing of antibiotic adjustment. The analysis revealed that positive mNGS detection was significantly associated with an increased likelihood of antibiotic adjustment (OR = 29.742, 95% CI [10.630, 83.170]). Conversely, it showed a negative association with skin and soft tissue infection (OR = 0.089, 95% CI [0.009, 0.880]), abdominal infection (OR = 0.206, 95% CI [0.052, 0.812]) and with the timing of antibiotic adjustment (OR = 0.671, 95% CI [0.566, 0.795]) (Fig. 2).

Fig. 2.

Binary regression analysis results of the mNGS detection before PSM

Factors associated with 28-day mortality of septic patients in ICU

Binary logistic regression analysis was performed to identify factors associated with 28-day mortality before PSM. The dependent variable was 28-day mortality. Independent variables included age, MV, CRRT, APACHE II score, SOFA score, the decision and timing of antibiotic adjustment. Each one-point increase in SOFA score was associated with a 16.9% higher risk of mortality (OR = 1.169, 95% CI [1.054, 1.297]). The decision to adjust antibiotics (OR = 0.252, 95% CI [0.101, 0.627]) was significantly negatively associated with 28-day mortality. Conversely, the use of MV (OR = 4.260, 95% CI [1.783, 10.181]) and CRRT (OR = 4.253, 95% CI [2.117, 8.541]) were positively related to the 28-day mortality (Fig. 3).

Fig. 3.

Binary regression analysis results of the 28-day mortality before PSM

Robust mediation analysis of the effect of mNGS detection on 28-day mortality through antibiotics adjustment

The path from mNGS detection to antibiotic adjustment was significant (Z = 4.052, P < 0.001), as was the path from mNGS detection to 28-day mortality (Z = − 2.508, P = 0.012). The total effect of mNGS detection on 28-day mortality was significant (Z = − 2.754, P = 0.006). However, the path from antibiotic adjustment to 28-day mortality was not significant (Z = 0.368, P = 0.713). Additionally, the indirect effect of mNGS detection on 28-day mortality through antibiotic adjustment was not significant (Z = 0.367, P = 0.714) (Table 3).

Table 3.

Robust mediation analysis of the effect of mNGS detection on 28-day mortality through antibiotics adjustment

Path	Estimate	SE	Z-value	P-value	95%CI
					Lower	Upper
Antibiotics adjustment →28-day mortality	0.054	0.146	0.368	0.713	−0.232	0.339
mNGS detection → 28-day mortality	−0.616	0.246	−2.508	0.012	−1.098	−0.135
mNGS detection → Antibiotics adjustment	0.929	0.229	4.052	< 0.001	0.48	1.379
mNGS detection → 28-day mortality indirect effects	0.05	0.136	0.367	0.714	−0.216	0.316
mNGS detection → 28-day mortality total effects	−0.566	0.206	−2.754	0.006	−0.969	−0.163

mNGS Metagenomic next-generation sequencing

Discussion

Our study systematically revealed that mNGS detection facilitates antibiotic regimen adjustment and is associated with reduced 28-day mortality among septic patients. without increasing hospitalization cost and length of ICU stay. The results of this study highlight the importance of mNGS detection in critical care medicine. Our findings not only support but also expand upon the existing literature regarding the potential benefits of mNGS in the treatment of sepsis. Previous studies have shown that the mNGS detection can rapidly identify pathogens, thereby more accurately guiding antibiotic use [18]. Therefore, mNGS detection is expected to become part of standard care procedures in the ICU to improve treatment outcomes and reduce mortality.

For the management of sepsis, clinical guidelines recommend a sequential approach: first identifying the site of infection, followed by collecting fluid specimens for routine culture, isolating and identifying the causative pathogens, and ultimately selecting targeted antimicrobial agents [19, 20]. In this context, mNGS detection offers a critical advantage: by rapidly identifying pathogens, which enables clinicians to adjust antibiotic regimens with greater precision and speed. This capability is widely recognized as a key contributor to the observed reduction in sepsis-related mortality [17, 18]. Consistent with this mechanism, our study further demonstrated that the decision to adjust antibiotics was significantly associated with both mNGS detection results and 28-day mortality among septic patients in the ICU. This finding aligns with evidence from other studies, such as the work by Zhang et al. [21], which highlighted that mNGS not only exhibits substantial potential for pathogen diagnosis in focal infections but also facilitates the more timely implementation of targeted therapeutic strategies, thereby reinforcing its value in optimizing sepsis care.

In contrast, conventional blood cultures have an approximate 20% positive rate and generally take three to five days for results [22]. Specific bacterial cultures, such as M. tuberculosis, require longer incubation periods; pathogens like Nocardia, Cryptococcus, and Brucella have low positive rates; and viruses cannot be cultured [23–25] often causing treatment delays. Rapid pathogen detection via mNGS enables timely and precise antibiotic adjustment, which may avoid unnecessary broad-spectrum antibiotic use, reduce the emergence of drug-resistant bacteria, and thus lower the risk of multidrug resistance-related complication and mortality [26].

However, confounding biases existed between the mNGS and non-mNGS groups, including age, gender, underlying diseases, APACHE II score, SOFA score, infection sites, PCT level, pro-BNP level, and vasoactive drug use. Previous studies have identified risk factors for sepsis-related mortality, such as age over 65 years, male sex, SOFA score, APACHE II score, renal failure, nosocomial infection, urinary infection, positive lactate and blood culture, increased NT-proBNP, and decreased AT-III [22–24]. Additionally, some studies suggest that the diagnosis and prognosis of sepsis in emergency settings should take different clinical criteria into account, based on the infection site [25]. Through the PSM analysis, excluding the aforementioned confounding bias, there was still a significant difference in the 28-day mortality of patients grouped by mNGS detection. These findings lay the groundwork for subsequent mediation analysis, indicating that mNGS detection may influence patient outcomes by affecting both the decision to adjust antibiotics and the timing of such adjustment.

Then, we further analyzed the relationships among mNGS detection, antibiotic adjustment, and 28-day mortality. Although the indirect effect of mNGS detection on 28-day mortality (mediated by antibiotic adjustment) was not significant, the total effect was. Notably, formal mediation analysis failed to identify antibiotic adjustment as a statistically significant mediator of the survival benefit, suggesting that other pathways may contribute to the observed mortality reduction. Previous studies have shown that mNGS detection can accurately detect pathogens and influence adjustment to clinical anti-infection regimens in sepsis [12, 27]. However, in certain special cases, such as infections occurring after antimicrobial therapy, mNGS diagnostic efficacy is inferior to that of traditional culture [28]. Additionally, mNGS has only been shown to have reduced susceptibility to antibiotic exposure within 24 h, longer exposure times have not been proven [29].

Although this study yielded important findings, it has several limitations. First, despite using PSM to control for observed confounders, the study remains susceptible to unmeasured confounding and selection bias. As a single-center study, our patient population and clinical practices may not be fully representative; additionally, the decision to perform mNGS could have been influenced by unmeasured factors, such as disease severity or clinician preference. Second, the timing of mNGS sampling relative to hospital admission and antibiotic initiation was not standardized, which may introduce bias. Earlier mNGS detection might lead to different outcomes compared with testing later in the clinical course. Similarly, we were unable to precisely control for the time from mNGS reporting to result acquisition and to subsequent antibiotic adjustment, both of which are critical factors affecting clinical outcomes. To address these limitations, further prospective, multicenter studies are needed to validate our results.

Conclusions

In summary, the implementation of mNGS testing was associated with a higher rate of antibiotic regimen adjustment and a significant reduction in 28-day mortality among septic ICU patients, which would not increase the financial burden on patients. These findings support the potential clinical utility of mNGS. However, the association between adjusting antibiotics based on mNGS test results and reducing the 28-day mortality rate has not yet been confirmed. The exact mechanisms underlying the mortality benefit requires further elucidation and broader validation.

Author contributions

YS designed the study and drafted the article; QX and SY collected the data; FQ and QG performed data analysis; QC translated the article; PC, YY, and SH performed the data extraction; ZW designed and approved the final manuscript.

Funding

This research was supported by the Guangdong Medical Research Foundation (No. A2024130 and No. A2023198); National Natural Science Foundation of China (No. 82374216); Research Project of Traditional Chinese Medicine Bureau of Guangdong Province (No. 20251363); Self-funded Scientific and Technological Innovation Projects of Foshan Science and Technology Bureau (No. 2220001005576, No. 2320001006680, No. 2320001006896); Guangdong Hospital of Traditional Chinese Medicine Special Research Project on Traditional Chinese Medicine Science and Technology (YN2022QN01); Foshan Medical Research Project (No. 20210001); Planned Science Technology Project of Guangzhou (SL2025A04J4201). The study was also supported by the Xiong Jibai Chinese medical master inheritance studio.

Data availability

The datasets generated during the current study are available in the Figshare repository: https://doi.org/10.6084/m9.figshare.28927757.

Declarations

Ethics approval and consent to participate

This study was approved by the Ethics Committee of Foshan Hospital of Traditional Chinese Medicine (approval number: 2021245, 17/09/2022). To protect the confidentiality of the patients, the study conduction was in accordance with the Helsinki Declaration. Owing to the retrospective nature of the study, a waiver of informed consent was granted by the Ethics Committee of the Foshan Hospital of Traditional Chinese Medicine.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.