A machine learning and centrifugal microfluidics platform for bedside prediction of sepsis.

Summary

Using 586 derivation samples and validation across 3,178 independent patients, a six-gene expression signature (Sepset) predicted 24-hour clinical deterioration in suspected sepsis. An RT-PCR assay and an automated centrifugal microfluidic instrument achieved 94% and 92% sensitivity, respectively, with 89% specificity on the instrument.

Key Findings

• Defined a six-gene Sepset signature predicting 24-hour deterioration in suspected sepsis
• Validated performance in 3,178 independent patients from external cohorts
• RT-PCR Sepset test showed 94% sensitivity for worsening SOFA within 24 hours
• Automated centrifugal microfluidic device achieved 92% sensitivity and 89% specificity at bedside

Clinical Implications

If prospectively implemented, the Sepset platform could enable early escalation/de-escalation decisions within the first 24 hours, improving triage, ICU utilization, and timely antimicrobial and organ support.

Limitations

• Impact on patient-centered outcomes not yet tested in randomized implementation trials
• Potential site-specific biases in derivation cohort despite external validation

Future Directions

Conduct pragmatic implementation trials to assess time-to-antibiotics, ICU admission decisions, and mortality; evaluate performance across diverse health systems and pathogens; pursue regulatory approval and cost-effectiveness analyses.