mNGS vs Culture in ICU Patients: Sensitivity Data Compared
For ICU patients with suspected sepsis, every hour of delayed pathogen identification correlates with worsened outcomes. Yet traditional blood culture — the current diagnostic standard — has a positivity rate of roughly 25% in critically ill patients, leaving clinicians to manage life-threatening infections without a confirmed etiology. Metagenomic next-generation sequencing (mNGS) vs culture in the ICU is now being evaluated in head-to-head clinical studies, and the sensitivity gap is striking. In this symposium presentation, Dr. Zi-Lun Lai shared results from a 135-sample comparative study demonstrating that mNGS achieves more than double the positivity rate of conventional culture while also revealing the critical importance of host depletion and quality control metrics.
Watch the full webinar:
Key Takeaways
- mNGS achieved a 60.7% positive detection rate compared to 25.2% for conventional blood culture across 135 ICU samples — a 2.4-fold improvement.
- Positive concordance between mNGS and culture was 44.1%, improving to 62% when corroborated with additional clinical laboratory data.
- Among 48 samples that were mNGS-positive but culture-negative, 16 were confirmed by other diagnostic methods, highlighting cases that culture misses entirely.
- Samples with microbial reads per million (RPM) above 1,000 showed significantly higher pathogen detection rates and concordance with culture (67.5%).
- Effective host DNA depletion is essential — a negative correlation between human DNA percentage and microbial RPM confirms that reducing host background directly improves sensitivity.
Study Design — Head-to-Head Comparison in the ICU
The study enrolled patients from the ICU at a major hospital in Taiwan. The workflow followed a rigorous parallel design: when a patient’s conventional blood culture turned positive and was sent for BC ID2 panel testing, consented patients had an additional blood draw processed simultaneously through both standard culture and an mNGS workflow.
The mNGS protocol used approximately 3 mL of whole blood filtered through a Devin host depletion filter to remove white blood cells and reduce human DNA background. The filtrate was spiked with 10,000 copies per minute of an internal spike control, then processed through two-step centrifugation to obtain a pellet for genomic DNA purification. Libraries were prepared for Illumina sequencing at 150-base single-end reads.
The bioinformatics pipeline included quality control filtering, mapping against the human genome to remove residual host reads, and classification against a curated pathogen database covering approximately 1,500 species. The team developed two key metrics for result interpretation: microbial reads per million total QC reads (RPM) and a pathogen read percentage threshold set at three-fold or greater above baseline.
Pathogen Detection Rates — mNGS Sensitivity vs Culture
Concordance Analysis
Across 135 evaluable samples, the mNGS method detected pathogens in 60.7% of cases compared to just 25.2% for conventional culture. The raw positive concordance rate — where both methods identified the same organism — was 44.1%.
To assess whether mNGS-positive, culture-negative results represented true detections rather than false positives, the team cross-referenced these discordant cases against other available clinical test data. Of 48 such discordant samples, 16 were corroborated by independent diagnostic methods, raising the adjusted concordance rate to 62%. This finding directly demonstrates that mNGS detects clinically relevant pathogens that conventional culture misses.
When the analysis was restricted to samples with microbial RPM above 1,000, concordance improved further to 67.5%, suggesting that microbial RPM functions as a reliable quality indicator for result confidence.
The Role of Microbial RPM as a Quality Indicator
Dr. Lai presented stratified analyses showing that samples with higher microbial RPM demonstrated consistently better positive detection rates and stronger concordance with culture results. Conversely, samples below 1,000 RPM showed reduced sensitivity — even with adequate total sequencing depth. In one illustrative case (sample 33), microbial RPM was approximately 400 despite achieving 6 million total QC reads, and no pathogen was identified by mNGS despite a positive culture result.
This dissociation between total read depth and microbial RPM underscores that sequencing more reads does not compensate for insufficient microbial enrichment. The microbial fraction in the sample — determined primarily by sample type and host depletion efficiency — is the rate-limiting factor for mNGS sensitivity.
Host DNA Depletion Drives Assay Sensitivity
A central finding of the study was the negative correlation between human DNA read percentage and microbial RPM. Samples with higher proportions of human reads yielded lower microbial signal, directly reducing pathogen detection sensitivity. This relationship was consistent when comparing the study’s ICU samples against published benchmarks, confirming that the host depletion strategy — using the Devin filter — achieved effective reduction of human DNA in real clinical specimens, not just in controlled laboratory conditions.
The practical implication is clear: for blood-based mNGS in sepsis patients, host depletion is not optional — it is a prerequisite for achieving clinically meaningful sensitivity. Without adequate removal of human DNA, even deep sequencing produces insufficient microbial coverage to identify pathogens reliably.
Dr. Lai also noted that high microbial RPM combined with excessive total QC reads could paradoxically increase the risk of false positives, as demonstrated in specific cases. This highlights the need for balanced quality control parameters — both microbial enrichment and total read depth must be considered together during result interpretation.
Repeatability and Quality Control Metrics
To assess assay reproducibility, the team tested 11 samples in duplicate, including both library preparation resequencing and complete workflow replication. Results showed consistent human read percentages and microbial RPM across replicates, demonstrating good repeatability of the end-to-end workflow from filtration through sequencing.
However, the team noted that final pathogen detection results are influenced by both total QC read depth and microbial RPM, meaning that borderline samples near detection thresholds may produce variable results. Establishing minimum thresholds for both metrics — not just one — is essential for reliable clinical reporting.
The PaRTI-Seq analysis platform provides standardized bioinformatics for this type of mNGS result interpretation, including automated RPM calculation and threshold-based pathogen calling that addresses the quality control challenges identified in this study.
About the Speaker
Dr. Zi-Lun Lai is a researcher specializing in metagenomic diagnostics and bioinformatic analysis for infectious disease. His work focuses on evaluating mNGS workflows for pathogen identification in critically ill ICU patients.
