A prominent talking point at this year’s ESCMID Global conference in Munich was the persistent diagnostic challenges surrounding neonatal sepsis. Following their symposium session at the congress, we caught up with Marieke Emonts, Professor of Pediatric Infectious Diseases at Newcastle University and the Great North Children's Hospital, and Mohan Pammi, Professor at Baylor College of Medicine, to discuss the potential impact of rapid diagnostic tools and AI-powered screening in this setting.
Neonatal sepsis remains a diagnostic challenge – where do current approaches fall short, particularly from a laboratory perspective?
Marieke Emonts: A key limitation is the small sample volumes we often have to work with, alongside assays that are not always as sensitive as needed. Clinical decision rules can help rule out bacterial infections, but they do not address viral causes.
Timing is also critical. For example, in suspected HSV infection, a CSF test performed within the first 24 hours of symptom onset may give a false-negative result, requiring repeat testing. Greater awareness of these limitations is important to avoid missed diagnoses and premature discontinuation of treatment.
Mohan Pammi: Laboratory diagnosis of infection still relies largely on blood cultures, which typically take 24–72 hours to yield results. During this time, babies are often started on antibiotics. If cultures are negative, this can lead to unnecessary antibiotic use, contributing to antimicrobial resistance (AMR) and disruption of the infant’s microbiome. If cultures are positive, an additional 24 hours may be needed for antimicrobial susceptibility results, meaning initial treatment may not be optimal.
How are emerging biomarkers improving the precision and timeliness of neonatal sepsis diagnosis?
MP: If new biomarkers are sufficiently accurate – sensitive enough to detect infections and specific enough to rule them out – they could reduce reliance on blood cultures. This would allow earlier diagnosis and help discontinue unnecessary antibiotics.
ME: Using single biomarkers alone is often not sufficient. Accuracy improves when multiple markers are combined – particularly those associated with viral infection alongside those linked to bacterial infection. However, these results must always be interpreted in the clinical context.
It is also important to consider differences in epidemiology. What is common in high-income settings may differ significantly in low- and middle-income countries, so diagnostic approaches need to be adapted to the population in which they are used.
While some biomarkers are already used in adults and are being introduced in pediatric care, implementation studies are still ongoing. Data in neonates are more limited. Importantly, children are not small adults, and neonates are not small children. Biomarker levels can vary significantly in early life due to factors such as maternal influence and ongoing physiological changes, including liver and renal function. Normal ranges may also differ between preterm and term infants, and across the first months of life.
These variations mean that biomarkers may perform differently depending on age, and their sensitivity – particularly in young neonates – can be limited. Although biomarkers may help distinguish bacterial from viral infections, they are likely to require age-specific algorithms and risk scores to improve diagnostic performance.
Before these tests can be widely implemented, large-scale studies involving substantial numbers of children and neonates will be needed to ensure reliability and clinical utility.
What role do rapid diagnostic tools play in distinguishing true infection from non-infectious inflammation in neonates?
MP: This is an important area of discussion. Currently, no biomarker can do this with sufficient accuracy. Some perform better than others, but none are reliable enough on their own. Most biomarkers also rise in response to inflammation in general, not just infection, which limits their specificity.
ME: Mohan is correct: at present, progress is still limited. Distinguishing bacteria from viral infection remains the primary goal, while identifying inflammation more broadly is less well developed. Although research is ongoing, it has not yet led to robust diagnostic tools.
One of the main challenges in biomarker development is moving from discovery to clinical use. Promising biomarker signatures need to be validated, translated into practical diagnostic assays, and then approved for use. This process involves multiple steps, including collaboration with industry to develop platforms or adapt existing ones for clinical implementation.
How can clinicians and laboratories better differentiate viral from bacterial infections in neonates?
MP: There is currently no biomarker that can reliably distinguish bacterial from viral infection. In practice, viral infections are usually identified using PCR panels. However, systemic viral infections can still trigger significant inflammation and raise multiple biomarkers, as seen in conditions such as disseminated HSV infection.
ME: At present, assessment still relies heavily on clinical parameters and established risk factors. In high-income settings, maternal group B streptococcal infection remains a key risk factor for early-onset sepsis, along with prolonged rupture of membranes. Clinical history is also important, including potential exposure to pathogens such as HSV, as Mohan mentioned.
It is important to recognize that the absence of typical signs – such as vesicles in HSV – does not exclude infection. Diagnosis therefore depends on a combination of clinical assessment, appropriate timing of tests, and use of biomarkers where available.
Awareness is also critical. Maintaining a broad differential and considering less obvious causes can help support earlier and more accurate diagnosis.
How is AI being integrated into neonatal sepsis screening, and what advantages does it offer over traditional diagnostic methods?
ME: Using routinely collected clinical data – particularly in NICU settings – can help detect early changes in a baby’s condition. This allows earlier identification of potential sepsis and supports prompt treatment, which may improve outcomes.
However, effective use of these approaches depends on how data are captured. Integration with electronic patient records is essential, and standardized data entry is key. Structured inputs, such as checkboxes, are more useful than free text, as they allow consistent tracking and analysis of clinical parameters. Regular, standardized monitoring is therefore critical to making these tools work in practice.
MP: The use of AI in neonatal sepsis diagnosis is still emerging, and it is not yet widely used in routine NICU practice. Although many predictive models have been published, few have been externally validated, which limits their generalizability.
As Marieke alluded to, real-time AI approaches, such as those using continuous data like heart rate (for example, HeRO monitoring), may become more widely adopted. However, predictive accuracy is critical. Without careful validation, there is a risk of increasing unnecessary sepsis evaluations.
Current research is focusing on integrating AI-based screening tools with real-time electronic medical record data, aiming to support clinical decision-making by identifying infants at higher risk of sepsis.
What are the key challenges in implementing these tools in routine clinical and laboratory workflows, particularly in neonatal settings?
MP: AI-based predictive models could be useful as clinical decision support tools to inform clinicians about sepsis risk. However, integrating these models into electronic medical records systems, such as EPIC, requires institutional approval and careful attention to data privacy. Despite these challenges, this is likely to be an important direction for future practice.
How can laboratories balance sensitivity and specificity in screening to avoid both missed cases and unnecessary treatment?
ME: This remains a difficult balance. It depends on what level of false negatives is considered acceptable. In practice, sensitivity is often prioritized over specificity to avoid missing cases, as missed infections can lead to serious outcomes, including mortality and long-term complications. As a result, some degree of overtreatment is often accepted.
There is ongoing debate about how many cases can be missed to reduce unnecessary treatment. For example, some tools, such as the Kaiser Permanente model, have seen limited use in certain settings because they may miss a proportion of cases that is considered too high.
Overall, biomarkers and predictive tools should be used as part of a broader diagnostic approach. They are not standalone solutions and need to be interpreted alongside clinical assessment and other investigations.
MP: Test selection is often driven by clinicians, who need to choose tools with appropriate sensitivity and specificity for each patient. Currently, no biomarker can replace blood cultures, but some can be used as complementary tests to provide earlier information while awaiting results.
Laboratories may have limited control over which tests are ordered, but they play an important role in supporting appropriate test use and interpretation.
What role does continuous monitoring and data integration play in improving early detection?
MP: Continuous data integration using AI is a promising approach, as it can provide dynamic insights over time rather than single time-point results. Systems like HeRO monitoring demonstrate this potential, and similar methods may become available in the near future.
However, new approaches need careful evaluation, ideally through randomized controlled trials, before they are adopted into clinical practice.
ME: AI offers potential by bringing together large volumes of clinical data and identifying early changes before a patient becomes critically unwell. This may allow earlier intervention and improved outcomes. However, it also raises the possibility of overtreatment, so overall patient outcomes must be carefully considered.
At the same time, there is a need to reduce unnecessary antibiotic use to limit the development of AMR, which remains a significant challenge. Common empirical treatments, such as ampicillin with gentamicin, are becoming less effective in some settings due to rising resistance. Increasing AMR in the community is particularly relevant, as these are often the first pathogens neonates are exposed to during birth.
Looking ahead, how do you see precision diagnostics – combining biomarkers and AI – shaping the future of neonatal sepsis care?
MP: Point-of-care tests that deliver results within 10-15 minutes would offer a clear clinical advantage. Real-time data monitoring, combined with integration into electronic medical records as decision support tools, could further assist clinicians in managing patients more effectively.
Rapid antimicrobial susceptibility testing is another important area, as it may help reduce the use of broad-spectrum antibiotics and limit the spread of AMR.
ME: In the near term, a key goal is to safely reduce antibiotic use – particularly by stopping treatment earlier when appropriate. This needs to be done step by step, ensuring that each change maintains patient safety before moving further.
There is also growing interest in developing multi-diagnostic tools that combine different markers and data sources. While distinguishing bacteria from viral infection is especially important in neonates, broader inflammatory conditions also need to be considered, although these differ from those seen in older children.
Applying these approaches in neonates will require well-designed studies, including early recruitment of infants when infection is suspected and follow-up to confirm diagnoses. Beyond identifying pathogens, there is potential to develop predictive tools that can identify which babies are likely to become more severely unwell. This could support earlier intervention and, ultimately, improve outcomes.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
