Assessment of the Clinical Utility of Plasma Metagenomic Next-Generation Sequencing in a Pediatric Hospital Population

Metagenomic next-generation sequencing (mNGS) of plasma cell-free DNA (cfDNA) is commercially available, but its role in the workup of infectious diseases is unclear. To understand the clinical utility of plasma mNGS, we retrospectively reviewed patients tested at a pediatric institution over 2 years to evaluate the clinical relevance of the organism(s) identified and the impact on antimicrobial management. We also investigated the effect of pretest antimicrobials and interpretation of molecules of microbial cfDNA per microliter (MPM) of plasma.

ABSTRACT

Metagenomic next-generation sequencing (mNGS) of plasma cell-free DNA (cfDNA) is commercially available, but its role in the workup of infectious diseases is unclear. To understand the clinical utility of plasma mNGS, we retrospectively reviewed patients tested at a pediatric institution over 2 years to evaluate the clinical relevance of the organism(s) identified and the impact on antimicrobial management. We also investigated the effect of pretest antimicrobials and interpretation of molecules of microbial cfDNA per microliter (MPM) of plasma. Twenty-nine of 59 (49%) mNGS tests detected organism(s), and 28/51 (55%) organisms detected were clinically relevant. The median MPM of clinically relevant organisms was 1,533, versus 221 for irrelevant organisms (P = 0.01). mNGS test positive and negative percent agreements were 53% and 79%, respectively, and 50% of negative mNGS tests were true negatives. Fourteen percent of tests impacted clinical management by changing antimicrobial therapy. Immunocompromised status was the only patient characteristic that trended toward a significant clinical impact (P = 0.056). No patients with culture-negative endocarditis had organisms identified by mNGS. There were no significant differences in antimicrobial duration retest between tests with clinically relevant organism(s) and those that returned negative, nor were the MPMs different between pretreated and untreated organisms, suggesting that 10 days of antimicrobial therapy as observed in this cohort did not sterilize testing; however, no pretreated organisms identified resulted in a new diagnosis impacting clinical management. Plasma mNGS demonstrated higher utility for immunocompromised patients, but given the detection of many clinically irrelevant organisms (45%), cautious interpretation and infectious diseases consultation are prudent.

INTRODUCTION

Next-generation sequencing (NGS) describes high-throughput sequencing methods in which millions of DNA fragments can be independently and simultaneously sequenced. Cell-free DNA (cfDNA) in the bloodstream was first described in 1948 (1). cfDNA primarily originates from apoptotic human cells; inflammation, autoimmune disease, trauma, and cancer increase cfDNA levels (2, 3). NGS of cfDNA has been previously described for noninvasive diagnosis of fetal abnormalities (4–6), cancer monitoring (7–10), and transplant rejection (11–15). Its adoption in these fields raised the prospect of diagnosing infections through sequencing of microbial cfDNA by metagenomic NGS (mNGS) followed by bioinformatic taxonomic classification.

mNGS, sometimes called shotgun sequencing, has been applied to various clinical sample types, including cerebrospinal fluid, blood, respiratory samples, gastrointestinal fluid, and ocular fluid (16). mNGS testing is hypothesis free, unlike many contemporary molecular diagnostic infectious disease tests. Potential strengths include the ability to diagnose polymicrobial infections and quantitative reporting of cfDNA molecules detected. As blood traverses the entire body, it is hypothesized that even protected sites of infection may shed enough pathogen nucleic acid into blood for detection (17). This pathogen-agnostic method is in contrast to targeted nucleic acid amplification tests (NAAT) that use specific primers, limiting detection to suspected targets. Because the vast majority of mNGS cfDNA reads will reflect the human host, sample processing methods for human DNA depletion are needed, supplemented by postprocessing bioinformatic removal. Due to the amplification of background human DNA, mNGS is generally less sensitive than targeted approaches and requires greater sequencing depth for organism identification (18, 19).

A commercially available plasma cfDNA mNGS test from Karius, Inc. (Redwood City, CA), available since 2016, reports molecules of microbial cfDNA per microliter (MPM) of plasma. This laboratory is certified under the Clinical Laboratory Improvement Amendments of 1988, although the test has not been approved by the Federal Drug Administration. A recent company publication describes clinical and analytical test validation for detection of 1,250 human pathogens (20). The limit of detection of the Karius test is 41 MPM, and organisms are reported if cfDNA from the organism is detected at statistically significant levels relative to the results for negative controls run in parallel. For all reported organisms, a reference interval (MPM) is provided, based on abundances seen in samples from asymptomatic adult controls (20). The relationship between MPM and microbe concentrations (e.g., CFU) in blood is not well understood. Publications have described ongoing MPM detection for weeks after clearance of the organism on blood culture while on appropriate antimicrobial therapy (21).

Despite potential strengths of cfDNA detection by mNGS, notable limitations exist. One obvious limitation is that the test will not detect RNA viruses. Importantly, uncertainty remains regarding how to assess whether detected organism DNA (DNAemia) indicates a pathogen contributing to patient disease versus sample contamination or transient bacteremia from colonizing flora. In the clinical validation study by Karius, Inc. (20), 350 patients who presented with sepsis alert criteria were tested, and a diagnostic sensitivity of 92.9% and specificity of 62.7% were reported in comparison to a composite reference standard that included all microbiological data and clinical history (20). Sensitivity was 84.8% in comparison to the results of standard microbiological testing alone. A recent study of 100 plasma mNGS tests sent from a pediatric hospital determined a sensitivity and specificity of the test for detection of organisms that impacted clinical decision making of 92% and 64%, respectively (22).

At our hospital, clinicians have postulated that plasma mNGS may be useful in the following clinical scenarios: (i) culture-negative infections due to antibiotic pretreatment and/or fastidious or nonculturable organisms and (ii) deep-seated and difficult-to-sample infections, such as invasive fungal infections, pneumonia, or osteomyelitis. The purpose of this study was to assess test performance characteristics and explore how mNGS findings impacted clinical management.

MATERIALS AND METHODS

We retrospectively reviewed medical records of all patients for whom commercial plasma mNGS testing was sent at Boston Children’s Hospital from October 2017 through October 2019. This study was approved by our institutional review board. Tests required approval from the directors of the Infectious Diseases (ID) Diagnostic Laboratory, as well as an ID clinical consultation. The approval process involved a discussion about the utility of testing between the ID team and laboratory director when the diagnosis was not evident from initial testing. There were no fixed criteria, and this study was conducted to help inform institutional guideline development based on identification of patient subsets in which the test was found to be the most clinically impactful. We assessed patient demographics, underlying comorbidities, ordering team, site of infection, duration of antimicrobial use prior to test, final clinical diagnoses, and reported MPM if testing returned positive for any organism. Patients were classified as immunocompromised if they had an underlying immunodeficiency, malignancy on active chemotherapy, hematopoietic stem cell or solid organ transplant, or other condition requiring immunosuppression.

The clinical relevance of organisms identified from plasma mNGS was assessed relative to the final overall diagnosis (infection versus no infection). Presence of an infection was determined by the treating clinical team, and the determination incorporated the clinical presentation that prompted mNGS testing and all microbiologic testing performed (including mNGS findings). A subgroup of clinically relevant organisms was considered confirmed positive if they correlated with a non-mNGS microbiological result (e.g., PCR or culture); however, in some cases, the clinical team made diagnoses on the basis of clinical picture and mNGS findings (Table 1). These definitions of infection are consistent with prior studies that have evaluated the performance characteristics of mNGS (22, 23). In the absence of a gold standard for this novel technology, our composite reference standard nonetheless reflects how clinicians interpreted and acted on all results, and we surmise this is the most clinically meaningfully definition of “infection.” Clinical relevance and confirmed positives were determined by expert opinions of two pediatric ID physicians not involved in the patient’s care at time of testing (R.A.L. and F.A.), with a tie-breaker opinion of a third (T.S.S.) if discordant.

TABLE 1.

Scenarios for clinically relevant (true positive) and clinically irrelevant (false positive/negative) organisms

Scenario
Clinically relevant (true positive)
Confirmed positive and primary etiology of illness: e.g., patient septic from Enterococcus bacteremia on blood culture, which was also identified on mNGS testing
Confirmed positive but not primary reason for hospitalization/severe acute illness: e.g., HSV gingivostomatitis in patient septic from Pseudomonas bacteremia, but HSV (and Pseudomonas) identified on mNGS testing and verified by standard workup (PCR swab and blood culture, respectively)
Not confirmed positive but consistent with infectious diagnosis: e.g., F. necrophilum identified in mNGS testing in patient diagnosed with aspiration pneumonia, although standard microbiological workup did not identify this organism
Clinically irrelevant (false positive or negative)
Pathogens that are likely contaminants: e.g., S. epidermidis identified on mNGS but no evidence of bloodstream infection and concurrent blood cultures negative with no treatment
Pathogens that may reflect gastrointestinal/skin colonization with no obvious manifestation in the patient: e.g., N. sicca coidentified in patient with respiratory failure/sepsis from adenovirus, and not confirmed on blood culture or treated
Pathogens with no known clinical significance: e.g., virus with no known associated infectious clinical manifestation
Pathogens identified on mNGS that were discordant with final clinical diagnosis made on the basis of standard microbiological workup: e.g., E. coli and H. influenzae on mNGS in setting of Streptococcus gordonii endocarditis identified from blood culture and universal PCR of valve tissue

A novel aspect of our study was to assess the relationship of MPM to determination of a clinically relevant organism. We additionally considered whether there was antimicrobial use active against the organism by reviewing susceptibility data obtained via concurrent routine microbiological methods, when possible, and by assessing whether the patient improved clinically on empirical therapy, suggesting that it was appropriate.

We further evaluated the effect of mNGS testing on overall patient care to specifically assess the added value of plasma mNGS testing over standard microbiological workup, and we defined “clinical impact” if testing resulted in (i) new organism(s) with new targeted antimicrobial therapy, (ii) new organism(s) with deescalation of antibiotic therapy, or (iii) negative testing, thus motivating teams to deescalate antimicrobial therapy. Cases in which redundant organisms were identified on plasma mNGS and standard microbiological testing were only considered to have clinical impact if there was a change in antimicrobial management on the basis of the plasma mNGS result. For example, if the mNGS resulted in a diagnosis sooner than standard microbiological workup and affected antimicrobial management, this was considered to have a clinical impact. Clinical impact was adjudicated by the research team. Standard microbiological testing was defined as routine microbiological testing/NAAT performed either in our Infectious Diseases Diagnostic Laboratory or in reference laboratories. Logic gates of possible scenarios to determine clinical impact dependent on plasma mNGS, standard microbiological testing, and antimicrobial change are demonstrated in Table 2.

TABLE 2.

Possible scenarios for determining clinical impact

Plasma mNGS result	Standard microbiological testing	Antimicrobial change due to mNGS result	Clinical impact
			Details	Effect on treatment
−	−	−	Redundant information, antibiotics and clinical plan were not changed (no impact)	−
−	−	+	Clinical impact (e.g., deescalation of antimicrobial) if team used negative mNGS results to deescalate	+
−	+	−	No additional information (no impact)	−
−	+	+	Clinical impact (e.g., deescalation)a	+
+	−	−	Not-relevant organism (considered contamination or transient unrelated bacteremia)	−
+	−	+	Clinical impact (e.g., new diagnosis and targeted therapy)	+
+	+	−	Redundant information, antibiotics and clinical plan were not changed (e.g., known bacteria identified and no impact)	−
+	+	+	Clinical impact (e.g., different diagnosis and additional therapy)	+

^aExample clinical scenario: concern for contaminant from standard microbiological testing and negative plasma mNGS results are used to clinically confirm suspicion and antibiotics are deescalated.

Statistical analysis.

Demographic data were summarized using descriptive statistics. Test characteristics for mNGS findings were calculated using two different methods labeled as counting by test versus result (Fig. 1 and 2). In the absence of a reference gold standard for this novel technology, we report positive percent agreement (PPA; agreement between mNGS test and infection diagnosis determined by the providing clinical team based on all microbiological data and clinical history) and negative percent agreement (NPA; agreement between negative mNGS test and diagnosis of no infection as determined by the providing clinical team) instead of sensitivity and specificity. In addition, since positive and negative predictive values cannot be calculated when the natural prevalence of disease is altered (plasma mNGS was only sent for select high-risk patients), we report the proportion of true positives out of all positive mNGS findings and the proportion of true negatives out of all negative mNGS tests. Method 1 counted all mNGS results from one plasma sample as one test (n = 59). If the mNGS test sent identified a clinically relevant organism, whether or not the organism was a confirmed positive, the test result was considered a true positive. However, mNGS tests often identified multiple organisms, and in many of these instances, both clinically relevant and clinically irrelevant organisms (not related to any known or suspected infection in the patient) were reported. By method 1, the mNGS test would be classified as a true positive based on identification of a clinically relevant organism even if clinically irrelevant organism(s) were also identified. Method 1 therefore does not fully account for the “noise” of coidentified clinically irrelevant organisms. To account for this noise, we used method 2, where we counted each organism identified so each organism result was assessed independently (n = 81). Method 2 provides more granular detail for mNGS findings by separately assessing the clinical relevance of each organism identified.

FIG 1.

mNGS findings were counted by two separate methods, as illustrated, for assessment of test characteristics by plasma test sent (method 1) and by organism(s) detected (method 2).

FIG 2.

Test performance characteristics calculated by method 1 (each plasma test sent interpreted as a whole, n = 59) and method 2 (by organism) to discriminate noise in mNGS tests due to clinically irrelevant organisms coidentified with relevant pathogens. Infection was defined by composite reference method (provider interpretation of clinical history and all microbiological data, including mNGS findings). Box B was added to the usual 2-by-2 contingency table, as these are clinically irrelevant organism(s) identified in the setting of an infection diagnosed by non-mNGS findings (i.e., diagnosed by standard microbiological workup). They cannot be included in box D since mNGS identified the organism(s) and cannot be included in box C as the patient’s final diagnosis was infection. Nonetheless, these cases contribute to test performance and should not be dropped from calculations.

Comparative analysis was conducted by Fisher’s exact test or chi-square test as appropriate, and continuous data were compared using the Wilcoxon rank sum test and Kruskal-Wallis test for group medians. MPM performance in the determination of clinically relevant organisms was assessed by receiver operating characteristic (ROC) analysis and area under the curve (AUC). An optimal cutoff score was found using the Youden index. Statistical tests were performed using Stata 15.1 software (Stata Corporation, College Station, TX, USA) and GraphPad version 8 software (GraphPad Software, San Diego, CA, USA) with a P value of ≤0.05 as the significance threshold.

Data availability.

Data used for test performance and clinical impact calculation, including descriptions of all organisms, MPMs, and patient characteristics, are available in Data Set S1 in the supplemental material.

RESULTS

A total of 59 plasma NGS tests were sent on 54 patients during the study period. Figure 3 summarizes patient characteristics, ordering teams, primary sites of infection, and final diagnoses of patients. Of the 5 tests that were resent on patients, two revealed new diagnoses (one with clinical impact), and all tests were sent at least a month apart with new or worsening clinical symptoms. The most common final diagnosis of patients on whom plasma mNGS was sent was no clear diagnosis (e.g., prolonged fever that could be due to infection or drug fever but resolved without determination of specific etiology [25%]). Half of these patients were thought to ultimately have no infection at all, while the others were treated empirically for presumed infection. Autoimmune conditions were identified in 17% of patients and endocarditis in 14%. While cardiology teams ordered the second largest number of tests, no organisms were identified via mNGS on any of the culture-negative endocarditis cases, and redundant organisms were identified in three cases by standard microbiological workup. In one case of culture-positive endocarditis, plasma mNGS identified discordant organisms that were deemed clinically irrelevant; Escherichia coli and Haemophilus influenzae were identified on plasma mNGS, but PCR of the eventually explanted valve identified Streptococcus gordonii, which also grew from an initial blood culture and was preliminarily considered a possible contaminant. No ordering team, primary site of infection, underlying comorbidity, or final patient diagnosis was noted to have a statistically significant association with clinical impact.

FIG 3.

Plasma mNGS test and organism characteristics, clinical impact, and relevance. Patient characteristic P values assess association of dichotomized categorical variable versus clinical impact by Fisher’s exact test. *, P value to compare MPM medians by organism type did not include parasite as there was only one case. Ϯ, “No diagnosis” refers to no clear final diagnosis assigned by providers: 7 received empiric antimicrobials (assigned as infection), and 8 were ultimately considered to have no infection (no empiric antimicrobials).

Fifty-one organisms were identified from all testing combined (29 bacteria, 15 DNA viruses, 7 fungi, and 1 parasite), 55% of which were considered clinically relevant. Figure 3 summarizes the proportions of organisms identified that resulted in clinical impact or were determined to be redundant or clinically irrelevant. In eight cases, testing led to clinical impact with a change (addition or deescalation) in antimicrobial therapy. Seven of the eight cases were immunocompromised patients, and all of the five mNGS cases where a new organism was identified and a new diagnosis was made impacting clinical management were in immunocompromised hosts (Fig. S1 in the supplemental material). Underlying immunodeficiency and overall immunocompromised status were the only variables found to trend toward a significant clinical impact, although they did not reach our statistical threshold of 0.05 (P = 0.08 and P = 0.06, respectively). While unexpected false-positive and -negative test results could lead to unnecessary investigations or treatment, we did not observe this in our cohort.

The positive and negative percent agreements of plasma mNGS by test sent (method 1, n = 59) were 53% (95% confidence interval [CI], 36 to 68%) and 79% (95% CI, 54 to 94%), respectively, the proportion of true positives out of mNGS positives was 72% (95% CI, 53 to 87%), and the proportion of true negatives out of negative mNGS tests was 50% (95% CI, 31 to 69%) (Fig. 2). Eight mNGS tests (14%) identified only clinically irrelevant organisms, and five mNGS tests deemed clinically relevant coidentified irrelevant organisms. When each organism identified was analyzed independently (method 2, n = 81), the PPA and NPA were 46% and 75%, respectively, and the proportion of true positives out of all mNGS positives was 55% (Fig. 2).

Testing was collected after a median of 8 days into clinical workup and a median of 9 days of antimicrobial therapy, with a median turnaround time (from time of receipt of sample by testing laboratory to report) of 1 day, which is clinically actionable. For patients with plasma mNGS testing that returned negative in the setting of presumed infection treated empirically (“possibly sterilized” tests, n = 15), antimicrobial therapy had been administered for a median of 8 days (mean, 9.5; standard deviation, 8.9) prior to test collection. Surprisingly, we found that the duration of pretest therapy for patients with organisms detected on mNGS that should have been sterilized by the antimicrobial(s) in use (n = 27 organisms) was similar (median, 10 days of therapy [P = 0.59]; mean, 19; standard deviation, 30). For cases of presumed infection where both plasma mNGS and standard microbiological workup were negative, the majority of infections were deep-seated infections (4 pulmonary infections, 2 cases of osteomyelitis, 1 case of septic arthritis, 2 intrabdominal infections, and 1 case of sepsis); four patients were diagnosed with culture-negative endocarditis.

We also assessed the relationship of MPM to identification of a clinically relevant organism. The median MPM for clinically relevant organisms was 1,533 (interquartile range [IQR], 340 to 11,309), in contrast to that of clinically irrelevant organisms (median MPM, 221; IQR, 62 to 717), which was a statistically significant difference (P = 0.01). The median MPM for organisms with no pretest antimicrobial therapy active against the organism was 407 (IQR, 68 to 5,852), compared to an MPM of 527 for organisms with a covering antimicrobial (IQR, 215 to 6,267), which was not a statistically significant difference (P = 0.78). While median MPMs did vary by organism type (Fig. 3), the differences were not statistically significant (P = 0.48 for bacteria versus fungi versus virus). An ROC curve for MPM data to distinguish between clinically relevant and irrelevant organisms yielded an AUC of 0.75 (95% CI, 0.611 to 0.887). An optimal cutoff of 390 MPM by Youden index was 74% sensitive (95% CI, 55% to 87%) and 73% specific (95% CI, 52% to 87%), with a likelihood ratio of 2.7 (Fig. 4).

FIG 4.

(A) Comparison of distributions of MPM results for clinically relevant and irrelevant organisms (lines indicate median values). (B) Analysis of performance of MPM for distinction between clinically relevant and irrelevant organisms by receiver operating characteristic (ROC) curve.

DISCUSSION

In this study, we describe the clinical utilization of plasma mNGS testing at our clinical center and include novel assessments not described in other studies. The test characteristics of plasma mNGS testing at our hospital were considerably lower than results reported in the main clinical validation study led by the company (20), as well as in a recent retrospective description of another pediatric hospital experience (22). We surmise that the difference in test performance in part reflects a difference in how mNGS was applied, which was as a tertiary-level test sent in high-stakes scenarios where standard workup was unrevealing. At our institution, due to the considerable cost and unknown clinical utility, mNGS requires approval from the Infectious Diseases Diagnostic Laboratory Director and an ID consultation. We feel that our utilization likely reflects how many clinical centers would use plasma mNGS, in contrast to how this test was validated commercially as a sepsis screen in the emergency department (20). This is the first study to account for the noise of polymicrobial identification in plasma mNGS in assessment of test performance and to individually assess the clinical relevance of each organism, which substantially impacted the proportion of true positives out of mNGS positive results (72% for per-test assessment versus 55% for per-organism assessment). We also included patients with a discordant mNGS finding (where the final clinical diagnosis of infection was made from standard microbiological workup and was not consistent with the mNGS finding) as cases for our calculations, rather than excluding them, in order to provide the most realistic estimates of test performance. Our study uniquely defined additional clinical factors we hypothesized could be relevant to plasma mNGS yield, including days into disease course, pretest antimicrobial duration, and MPM interpretation.

This study illustrates how pretest probability affects testing utility, as the likelihood of plasma mNGS revealing an as-of-yet unidentified organism and new diagnosis after standard workup was low, particularly for immunocompetent patients. Many of our patients ultimately had a noninfectious diagnosis or a presumed infection treated empirically in the absence of microbiological data, which yielded higher false positives and negatives than prior studies. Negative mNGS results in patients with culture-negative infections (designated false negatives) also mostly involved protected sites of infection (pulmonary, intra-abdominal, or bone), which suggests that plasma mNGS may be an inadequate and at worst a misleading proxy for invasive microbiological sampling. Notably, the test had minimal yield for culture-negative endocarditis, despite the adjacency of cardiac valves to blood (only one endocarditis case underwent surgical management and had confirmed endocarditis on pathology, but all cases had presentations that met modified Duke’s criteria for endocarditis and improved on therapy). We additionally report that the clinical impact of tests through changes in antimicrobial therapy was low (14%), although notably, this was higher than another study that found that only 7% of tests led to a positive clinical impact (21).

A key overall finding was that the proportion of true negatives out of negative mNGS tests in our clinical practice was only 50%. While many providers wanted to use plasma mNGS to rule out an infection, we found that negative tests only predicted the absence of an infection as well as a coin flip, and, thus, in this setting, were a poor rule-out screening test. However, we did find a significant association between reported MPM and clinical relevance (Fig. 4), suggesting that high MPMs should make providers more confident that the result is meaningful.

Given that mNGS was sent several days into the disease course, we also wanted to address the possible impact of empirical pretest antimicrobials on plasma mNGS yield. While clearance of bloodstream pathogen cfDNA over time is expected, kinetics for specific pathogens will need to be elucidated as mNGS becomes more routine. Counterintuitively, we did not find significant differences in MPM values between organisms treated with an appropriate antimicrobial pretest and those that were untreated, even when only considering clinically relevant organisms (dismissing organisms that may have been contaminants and thus unaffected by antimicrobials). Furthermore, we did not find significant differences in antimicrobial duration between “possibly sterilized” mNGS tests and tests where an organism was identified with an active antimicrobial on-board. This suggests that pretest antimicrobial durations of 10 days (median) as observed in this cohort do not likely substantially affect sterilization of plasma mNGS. The ongoing detectable MPM may be related to slow-to-clear DNAemia from high pathogen burden even though organisms may have been appropriately killed on targeted therapy, a finding that is consistent with prior reports (22). Notably, no identified pretreated organisms resulted in a novel diagnosis that affected clinical management in our cohort.

Limitations of this study include a relatively small sample size, which in turn leads to a small number of patients in each relevant diagnostic subcategory (e.g., culture-negative endocarditis) and for establishment of the MPM cutoff in ROC analysis. Additionally, our comparator gold standard of the presence of infection was a composite assessment from the provider team, which included interpretation of all microbiological data, including mNGS findings. In the ideal scenario, we would have an independent reference standard of the test under evaluation, although there is precedent in the literature for assessing novel and possibly more sensitive technologies this way (23–25). In clinical practice, providers routinely incorporate the results of this test with other clinical data and, understanding the limitation that there is no reference standard for mNGS, our goal was to characterize provider response to findings in the context of all of the information available for the patient.

In summary, our major findings included lower test performance characteristics of plasma mNGS than prior literature suggests, with only half of the organisms identified as clinically relevant, emphasizing the need for ID consultation for interpretation. We found higher utility for immunocompromised patients and less value than expected for endocarditis. Additionally, although we expected that pretest antimicrobials would decrease the yield of plasma mNGS testing, after 10 days (median) of antimicrobial therapy, the MPM did not differ significantly between treated and untreated organisms, nor was overall detection compromised. Despite the insights gained in this study regarding plasma mNGS test performance and utility, further work will be required to understand how to optimally integrate this technology into the infectious diseases diagnostic workup.