blood cultures fail

the smallest patients

Problem statement

Neonatal sepsis kills an estimated 345,000 newborns per year globally — approximately 15% of all neonatal deaths. The incidence in low- and middle-income countries is 49-170 per 1,000 live births, compared to 1-4 per 1,000 in high-income countries. Blood culture, the diagnostic gold standard, is functionally useless in most LMIC neonatal units: 68% of septic neonates have bacteremia below 10 CFU/mL, the tiny blood volumes obtainable from newborns yield false-negative rates up to 60%, and results take 24-36 hours — far too slow for a condition where hours determine survival. In a study across 61 hospitals in four African countries, 70% of neonates received antibiotics but only 6% had a blood culture submitted.

Why this matters

Without diagnostic confirmation, clinicians in LMICs treat empirically: broad-spectrum antibiotics for every febrile neonate. This drives antimicrobial resistance — the WHO explicitly frames this TPP as an AMR intervention. An estimated 84% of neonatal infection deaths could be prevented through early diagnosis and appropriate management. The WHO's 2025 TPP defines two priority use cases — primary health care (where most neonates first present) and hospital-level care — but no existing diagnostic meets the specifications for either.

What’s been tried and why it hasn’t worked

C-reactive protein (CRP) is the most widely used biomarker but rises 6-12 hours after infection onset, missing early-onset sepsis when intervention is most critical. Procalcitonin (PCT) rises faster (2-4 hours) and has better sensitivity, but costs are prohibitive in resource-poor settings and diagnostic accuracy ranges widely (67-92%). Combined CRP + PCT panels improve sensitivity but neither achieves standalone diagnostic reliability. Molecular methods (PCR, next-generation sequencing) offer speed and sensitivity but require laboratory infrastructure, electricity, and trained operators unavailable in most LMIC primary care settings. The fundamental constraint is biological: neonatal sepsis presents with nonspecific signs, has low and variable bacterial loads in blood, and requires a test that works with <100 microliters of blood in settings with minimal infrastructure.

What would unlock progress

The WHO TPP specifies: turnaround time <30 minutes, sensitivity >=90%, blood volume <100 microliters, functional at both point-of-care and hospital levels. Meeting this requires either novel host-response biomarker panels that outperform CRP/PCT (likely multiplex combinations of cytokines, acute-phase proteins, and transcriptomic signatures) or microfluidic platforms that can detect pathogens directly from microliter blood volumes without culture. A student contribution could address the biomarker discovery side: systematically evaluating which combinations of commercially available biomarkers best discriminate bacterial sepsis from viral infection and non-infectious inflammation in neonates.

Entry points for student teams

A team could build a multiplex lateral flow prototype targeting 3-5 host-response biomarkers (e.g., CRP + PCT + IL-6 + IP-10) at neonatal volumes (<50 microliters from a heel prick) and validate it analytically on the bench — recombinant analytes spiked into a surrogate matrix at the concentrations the neonatal literature reports for septic and non-septic infants — characterizing limit of detection, cross-reactivity, and quantitative accuracy against the WHO TPP's stated bar of under 30 minutes, sensitivity at or above 90%, and under 100 microliters. The engineering challenge is maintaining quantitative accuracy across multiple analytes on a single low-cost strip, and that challenge is fully present on the bench. Benchmarking against clinical diagnosis in a neonatal unit is the study that ultimately matters, but neonates are the most protected research population and blood culture is itself an unreliable reference standard here, so the semester deliverable is that trial's design — reference-standard strategy, sample size, pre-registered analysis plan — written for a partner unit to execute. A third team needs only the published literature: extract reported biomarker distributions and ROC performance across the neonatal sepsis studies, model which combinations and cutoffs could actually reach the TPP's sensitivity target, and turn the result into a decision rule combining bedside signs with a rapid CRP — the antibiotic stewardship tool, derived from public data rather than from patient records no student team can obtain. Relevant disciplines: biomedical engineering, neonatology, immunology, data science.