AI Blood Tests for Sepsis: A New Era of Rapid Diagnostics

Leon Wirz
Oct 6
3 min read

Published in Nature Medicine, September 2025

Introduction

Every minute counts in the emergency room, especially when doctors suspect sepsis, a life-threatening condition where the body overreacts to infection. Yet the early symptoms often mimic mild viral illness, making diagnosis notoriously difficult. Traditional lab tests can take hours or even days, and clinicians often prescribe antibiotics “just in case,” fueling antibiotic resistance and healthcare costs.

Now, a study published in Nature Medicine presents a breakthrough: a blood test powered by artificial intelligence that can distinguish bacterial from viral infections, and even predict which patients will require critical care in the following week.

The Core Discovery

Researchers developed a machine-learning model that interprets gene activity (mRNA levels) in the patient’s blood. Instead of detecting the pathogen directly, the test reads the body’s immune response, identifying molecular patterns characteristic of bacterial or viral infections.

In a prospective, multicenter validation study involving 22 emergency departments across the United States, the AI-assisted test accurately identified bacterial vs viral infections and predicted severe outcomes such as organ failure or ICU admission over the next seven days.

How the Study was Conducted

The team collected whole-blood samples from over 4,000 patients presenting with suspected infection. They measured the expression of 29 specific mRNAs, previously identified as key immune markers, and fed these into a trained machine-learning classifier.

The classifier produced two key scores:

Bacterial vs viral likelihood, and
Risk of clinical deterioration or sepsis.

Results were benchmarked against physicians’ diagnoses, standard blood tests (CRP, procalcitonin), and patient outcomes. The AI model consistently outperformed traditional biomarkers, providing clinically actionable results within about one hour of sample collection.

Key Findings

High diagnostic accuracy: The model achieved >90 % sensitivity and specificity for distinguishing bacterial from viral infections.
Prognostic power: It identified patients likely to require ICU care within seven days.
Reduced diagnostic uncertainty: Physicians using the AI score made more confident and appropriate antibiotic decisions in pilot implementations.
Broad generalizability: Performance remained stable across age groups, hospitals, and comorbidities — a key step toward real-world deployment.

Limitations of the Study

While promising, the technology still faces hurdles:

Clinical integration: The test requires specialized instruments and trained personnel to process mRNA rapidly.
Cost-effectiveness not yet proven: Economic analyses and reimbursement frameworks are still pending.
Data bias: Most samples were collected in U.S. hospitals; validation in European and low-resource settings is still needed.
Interpretability: Although machine learning improves accuracy, clinicians must understand the reasoning behind results to trust AI in critical decisions.

Relevance for Switzerland

Sepsis remains a leading cause of in-hospital mortality — also in Swiss hospitals. Rapid, precise diagnostics could substantially reduce ICU admissions and inappropriate antibiotic use. Switzerland’s well-structured healthcare and insurance systems make it an ideal environment for evaluating such technology under value-based care models, where faster, more accurate testing can directly translate into lower downstream costs.

Moreover, Swiss biotech and med-tech sectors — especially in the Romandie region — are well positioned to collaborate on AI-driven molecular diagnostics, bridging cutting-edge computation with clinical translation.

Potential Impacts of a Successful Therapy

If implemented at scale, AI-guided blood diagnostics could:

Reduce unnecessary antibiotic use, curbing resistance and preserving drug efficacy.
Enable earlier sepsis detection, improving survival and reducing ICU burden.
Shorten hospital stays, leading to major cost savings.
Create new business models in diagnostic reimbursement and data-driven personalized care.

Risks

Algorithm drift: As pathogens evolve or hospital populations change, the AI model may need periodic recalibration.
Data privacy: Widespread molecular profiling raises new challenges for data governance.
Overreliance on AI: Clinical judgement must remain central, with AI serving as a decision support — not replacement.

Overall Assessment

This Nature Medicine study marks a pivotal step toward host-response diagnostics, shifting focus from identifying pathogens to decoding how the body reacts. With robust validation across multiple centers, the work shows that AI can translate molecular data into real-time clinical insight.

If cost-effective implementation follows, such diagnostics could redefine emergency medicine transforming how hospitals, insurers, and health systems manage infection risk.

What Comes Next

Future work will focus on:

Expanding validation to Europe and low-income settings.
Integrating the test into electronic health records for automated triage.
Economic and implementation studies to demonstrate real-world value.

If successful, AI blood diagnostics may soon become as routine as a CRP test — a quiet revolution in how medicine interprets illness before it’s visible.

Reference

Liesenfeld, O., Arora, S., Aufderheide, T.P. et al. Clinical validation of an AI-based blood testing device for diagnosis and prognosis of acute infection and sepis, Nat Med (2025) Link