Antibiotic exposure for culture-negative early-onset sepsis in late-preterm and term newborns: an international study.

BACKGROUND: Early-life antibiotic exposure is disproportionately high compared to the burden of culture-proven early-onset sepsis (CP-EOS). We assessed the contribution of culture-negative cases to the overall antibiotic exposure in the first postnatal week. METHODS: We conducted a retrospective analysis across eleven countries in Europe, North America, and Australia. All late-preterm and term infants born between 2014 and 2018 who received intravenous antibiotics during the first postnatal week were classified as culture-negative cases treated for ≥5 days (CN ≥ 5d), culture-negative cases treated for <5 days (CN < 5d), or CP-EOS cases. RESULTS: Out of 757,979 infants, 21,703 (2.9%) received intravenous antibiotics. The number of infants classified as CN ≥ 5d, CN < 5d, and CP-EOS was 7996 (37%), 13,330 (61%), and 375 (1.7%). The incidence of CN ≥ 5d, CN < 5d, and CP-EOS was 10.6 (95% CI 10.3-10.8), 17.6 (95% CI 17.3-17.9), and 0.49 (95% CI 0.44-0.54) cases per 1000 livebirths. The median (IQR) number of antibiotic days administered for CN ≥ 5d, CN < 5d, and CP-EOS was 77 (77-78), 53 (52-53), and 5 (5-5) per 1000 livebirths. CONCLUSIONS: CN ≥ 5d substantially contributed to the overall antibiotic exposure, and was 21-fold more frequent than CP-EOS. Antimicrobial stewardship programs should focus on shortening antibiotic treatment for culture-negative cases. IMPACT: In a study of 757,979 infants born in high-income countries, we report a presumed culture-negative early-onset sepsis incidence of 10.6/1000 livebirths with an associated antibiotic exposure of 77 antibiotic days per 1000 livebirths. This study sheds light on the major contribution of presumed culture-negative early-onset sepsis to early-life antibiotic exposure. Given the diagnostic uncertainty surrounding culture-negative early-onset sepsis, the low mortality rate, and the disproportionate antibiotic exposure associated with this condition, our study emphasizes the importance of targeting culture-negative early-onset sepsis in antimicrobial stewardship programs.

Association between Congenital Anomalies and Late-Onset Bacterial Infections in Neonates Admitted to Neonatal Intensive Care Units in Australia and New Zealand: A Population-Based Cohort Study.

INTRODUCTION: Compromised neonatal intensive care unit neonates are at risk of acquiring late-onset infections (late-onset sepsis [LOS]). Neonates born with congenital anomalies (CAs) could have an additional LOS risk. METHODS: Utilising the population-based Australian and New Zealand Neonatal Network data from 2007 to 2017, bacterial LOS rates were determined in very preterm (VPT, <32 week), moderately preterm (MPT, 32-36 weeks), and term (FT, 37-41 weeks) neonates with or without CA. Stratified by major surgery, the association between CA and bacterial LOS was evaluated. RESULTS: Of 102,808 neonates, 37.7%, 32.8%, and 29.6% were born VPT, MPT, and FT, respectively. Among these, 3.4% VPT, 7.5% MPT, and 16.2% FT neonates had CA. VPT neonates had the highest LOS rate (11.1%), compared to MPT (1.8%) and FT (1.8%) neonates. LOS rates were higher in CA neonates than those without (8.2% versus 5.1% adjusted relative risk [aRR] 1.67, 95% confidence interval [CI]: 1.45-1.92). Neonates with surgery had a higher LOS rate (14.2%) than neonates without surgery (4.4%, p < 0.001). Among the neonates without surgery, CA neonates had consistently higher LOS rates than those without CA (VPT 14.3% vs. 9.6% [aRR 1.32, 95% CI: 1.11-1.57]; MPT 4% vs. 0.9% [aRR 4.45, 95% CI: 3.23-6.14]; and FT 2% vs. 0.7% [aRR 2.87, 95% CI: 1.97-4.18]). For the neonates with surgery, CAs were not associated with additional LOS risks. CONCLUSION: Overall, we reported higher rates of LOS in neonates with CA compared to those without CA. Regardless of gestation, CA was associated with an increased LOS risk among non-surgical neonates. Optimisation of infection prevention strategies for CA neonates should be explored. Future studies are needed to evaluate if the infection risk is caused by CA or associated complications.

Whole-of-Life Inclusion in Bayesian Adaptive Platform Clinical Trials.

Importance: There is a recognized unmet need for clinical trials to provide evidence-informed care for infants, children and adolescents. This Special Communication outlines the capacity of 3 distinct trial design strategies, sequential, parallel, and a unified adult-pediatric bayesian adaptive design, to incorporate children into clinical trials and transform this current state of evidence inequity. A unified adult-pediatric whole-of-life clinical trial is demonstrated through the Staphylococcus aureus Network Adaptive Platform (SNAP) trial. Observations: Bayesian methods provide a framework for synthesizing data in the form of a probability model that can be used in the design and analysis of a clinical trial. Three trial design strategies are compared: (1) a sequential adult-pediatric bayesian approach that involves a separate, deferred pediatric trial that incorporates existing adult trial data into the analysis model to potentially reduce the pediatric trial sample size; (2) a parallel adult-pediatric bayesian trial whereby separate pediatric enrollment occurs in a parallel trial, running alongside an adult randomized clinical trial; and (3) a unified adult-pediatric bayesian adaptive design that supports the enrollment of both children and adults simultaneously in a whole-of-life bayesian adaptive randomized clinical trial. The SNAP trial whole-of-life design uses a bayesian hierarchical model that allows information sharing (also known as borrowing) between trial age groups by linking intervention effects of children and adults, thereby improving inference in both groups. Conclusion and Relevance: Bayesian hierarchical models may provide more precision for estimates of safety and efficacy of treatments in trials with heterogenous populations compared to traditional methods of analysis. They facilitate the inclusion of children in clinical trials and a shift from children deemed therapeutic orphans to the vision of no child left behind in clinical trials to ensure evidence for clinical practice exists across the life course. The SNAP trial provides an example of a bayesian adaptive whole-of-life inclusion design that enhances trial population inclusivity and diversity overall, as well as generalizability and translation of findings into clinical practice.

Neonatal bacterial sepsis.

Neonatal sepsis remains one of the key challenges of neonatal medicine, and together with preterm birth, causes almost 50% of all deaths globally for children younger than 5 years. Compared with advances achieved for other serious neonatal and early childhood conditions globally, progress in reducing neonatal sepsis has been much slower, especially in low-resource settings that have the highest burden of neonatal sepsis morbidity and mortality. By contrast to sepsis in older patients, there is no universally accepted neonatal sepsis definition. This poses substantial challenges in clinical practice, research, and health-care management, and has direct practical implications, such as diagnostic inconsistency, heterogeneous data collection and surveillance, and inappropriate treatment, health-resource allocation, and education. As the clinical manifestation of neonatal sepsis is frequently non-specific and the current diagnostic standard blood culture has performance limitations, new improved diagnostic techniques are required to guide appropriate and warranted antimicrobial treatment. Although antimicrobial therapy and supportive care continue as principal components of neonatal sepsis therapy, refining basic neonatal care to prevent sepsis through education and quality improvement initiatives remains paramount.

Outcomes of interventions in neonatal sepsis: A systematic review of qualitative research.

BACKGROUND: While a systematic review exists detailing neonatal sepsis outcomes from clinical trials, there remains an absence of a qualitative systematic review capturing the perspectives of key stakeholders. OBJECTIVES: Our aim is to identify outcomes from qualitative research on any intervention to prevent or improve the outcomes of neonatal sepsis that are important to parents, other family members, healthcare providers, policymakers, and researchers as a part of the development of a core outcome set (COS) for neonatal sepsis. SEARCH STRATEGY: A literature search was carried out using MEDLINE, EMBASE, CINAHL, and PsycInfo databases. SELECTION CRITERIA: Publications describing qualitative data relating to neonatal sepsis outcomes were included. DATA COLLECTION AND ANALYSIS: Drawing on the concepts of thematic synthesis, texts related to outcomes were coded and grouped. These outcomes were then mapped to the domain headings of an existing model. MAIN RESULTS: Out of 6777 records screened, six studies were included. Overall, 19 outcomes were extracted from the included studies. The most frequently reported outcomes were those in the domains related to parents, healthcare workers and individual organ systemas such as gastrointestinal system. The remaining outcomes were classified under the headings of general outcomes, miscellaneous outcomes, survival, and infection. CONCLUSIONS: The outcomes identified in this review are different from those reported in neonatal sepsis clinical trials, thus highlighting the importance of incorporating qualitative studies into COS development to encapsulate all relevant stakeholders' perspectives.

Epidemiology and Outcomes of Neonatal Sepsis: Experience from a Tertiary Australian NICU.

INTRODUCTION: Neonatal sepsis is associated with significant mortality and morbidity. Low-middle-income countries are disproportionately affected, but late-onset sepsis (LOS) still occurs in up to 20% of infants <28 weeks in high-income countries. Understanding site-specific data is vital to guide management. METHODS: A retrospective cohort study was conducted at King Edward Memorial Hospital (KEMH), Perth. Infants admitted between January 2012 and June 2022 were included. Data were extracted from routine electronic databases. Incidence and aetiology of sepsis were determined and the association of sepsis with neonatal outcomes analysed. RESULTS: During the study period, 23,395 newborns were admitted with a median gestation of 37 weeks and birth weight of 2,800 g. There were 370 sepsis episodes in 350 infants; 102 were early-onset sepsis (EOS) (1.6 per 1,000 live births), predominantly Streptococcus agalactiae (35, 34.3%) and Escherichia coli (27, 26.5%); 268 were LOS (0.9 per 1,000 inpatient days), predominantly coagulase-negative staphylococci (CONS) (156, 57.6%) and E. coli (30, 11.1%). The incidence of LOS declined from 2012 to 2022 (p = 0.002). Infants with EOS had increased brain injury (25.7% vs. 4.1%; p = 0.002) and mortality (18.8% vs. 1.6%; p < 0.001). Those with LOS had increased hospital stay (median 95 vs. 15 days; p < 0.001), mortality (15.3% vs. 1.6%; p = 0.018), necrotising enterocolitis (NEC) (7.4% vs. 0.5%; p < 0.001), and chronic lung disease (CLD) (58.1% vs. 5.9%; p = 0.005). Infants <28 weeks with sepsis were at increased risk of neurodevelopmental impairment compared to those without infection (43.2% vs. 30.9%, p = 0.027). CONCLUSIONS: While we observed a reduction in LOS incidence, sepsis remains associated with higher mortality, and in survivors with longer hospital stay and increased risk of brain injury, NEC, CLD, and neurodevelopmental impairment.

Angiogenesis-associated pathways play critical roles in neonatal sepsis outcomes.

Neonatal sepsis is a major cause of childhood mortality. Limited diagnostic tools and mechanistic insights have hampered our abilities to develop prophylactic or therapeutic interventions. Biomarkers in human neonatal sepsis have been repeatedly identified as associated with dysregulation of angiopoietin signaling and altered arachidonic acid metabolism. We here provide the mechanistic evidence in support of the relevance for these observations. Angiopoetin-1 (Ang-1), which promotes vascular integrity, was decreased in blood plasma of human and murine septic newborns. In preclinical models, administration of Ang-1 provided prophylactic protection from septic death. Arachidonic acid metabolism appears to be functionally connected to Ang-1 via reactive oxygen species (ROS) with a direct role of nitric oxide (NO). Strengthening this intersection via oral administration of arachidonic acid and/or the NO donor L-arginine provided prophylactic as well as therapeutic protection from septic death while also increasing plasma Ang-1 levels among septic newborns. Our data highlight that targeting angiogenesis-associated pathways with interventions that increase Ang-1 activity directly or indirectly through ROS/eNOS provide promising avenues to prevent and/or treat severe neonatal sepsis.

Physicochemical compatibility of caffeine citrate and caffeine base injections with parenteral medications used in neonatal intensive care settings.

PURPOSE: To investigate the physicochemical compatibility of caffeine citrate and caffeine base injections with 43 secondary intravenous (IV) drugs used in Neonatal Intensive Care Unit (NICU) settings. METHODS: Caffeine citrate (20 mg/mL or 10 mg/mL) or caffeine base injection (10 mg/mL) were mixed in a volume ratio of 1:1 with the secondary drug solution to simulate Y-site co-administration procedures in NICUs. Physical compatibility was evaluated based on visual observation for 2 h, against a black and white background and under polarised light, for changes in colour, precipitation, haze and evolution of gas. Chemical compatibility was determined from caffeine concentration measurements, using a validated high-performance liquid chromatography assay. RESULTS: Six of the 43 secondary drugs tested (aciclovir, amphotericin (liposomal), furosemide, hydrocortisone, ibuprofen and ibuprofen lysine) were physically incompatible with caffeine citrate undiluted injection (20 mg/mL), at their high-end, clinically relevant concentrations for NICU settings. However, when tested at lower concentrations, hydrocortisone (1 mg/mL) was physicochemically compatible, whereas furosemide (0.2 mg/mL) was physically incompatible with caffeine citrate. The six drugs which showed physical incompatibility with caffeine citrate 20 mg/mL injection were also physically incompatible with caffeine citrate 10 mg/mL solution. All 43 secondary drugs tested were physicochemically compatible with caffeine base injection. CONCLUSIONS: Most secondary test drugs, except aciclovir, amphotericin (liposomal), furosemide, hydrocortisone, ibuprofen and ibuprofen lysine, were physicochemically compatible with caffeine citrate injection. Caffeine base injection was physicochemically compatible with all 43 test drugs tested.

The Physicochemical Compatibility of Sildenafil Injection with Parenteral Medications Used in Neonatal Intensive Care Settings.

Sildenafil is used to treat pulmonary hypertension in neonatal intensive care unit (NICU) settings. As multiple intravenous (IV) medications are co-administered in NICU settings, we sought to investigate the physicochemical compatibility of sildenafil with a range of IV drugs. Sildenafil 600 mcg/mL or 60 mcg/mL was mixed 1:1 with the secondary drug solution to simulate Y-site co-administration procedures. Physical compatibility was evaluated by visual observation against a black and white background and under polarized light for two hours for changes in colour, precipitation, haze and evolution of gas. Chemical compatibility was determined from sildenafil concentrations, using a validated, stability-indicating high-performance liquid chromatography assay. Sildenafil 600 mcg/mL was physicochemically compatible with 29 of the 45 drugs tested at 'high-end' clinical concentrations and physically incompatible with 16 drugs and six '2-in-1' parenteral nutrition solutions. Sildenafil 600 mcg/mL was compatible with lower, clinically relevant concentrations of calcium gluconate, heparin and hydrocortisone. Aciclovir, amoxicillin, ampicillin, ibuprofen lysine, indometacin, phenobarbitone and rifampicin were incompatible with sildenafil 600 mcg/mL, however these IV medications were compatible with sildenafil 60 mcg/mL. Sildenafil 600 mcg/mL and 60 mcg/mL were incompatible with amphotericin, flucloxacillin, furosemide, ibuprofen, meropenem and sodium bicarbonate. Sildenafil compatibility with commonly used syringe filters was also investigated. Sildenafil solution was compatible with nylon syringe filters, however, absorption/adsorption loss occurred with polyethersulfone and cellulose ester filters.

Development of a pharmaceutical science systematic review process using a semi-automated machine learning tool: Intravenous drug compatibility in the neonatal intensive care setting.

Our objective was to establish and test a machine learning-based screening process that would be applicable to systematic reviews in pharmaceutical sciences. We used the SPIDER (Sample, Phenomenon of Interest, Design, Evaluation, Research type) model, a broad search strategy, and a machine learning tool (Research Screener) to identify relevant references related to y-site compatibility of 95 intravenous drugs used in neonatal intensive care settings. Two independent reviewers conducted pilot studies, including manual screening and evaluation of Research Screener, and used the kappa-coefficient for inter-reviewer reliability. After initial deduplication of the search strategy results, 27 597 references were available for screening. Research Screener excluded 1735 references, including 451 duplicate titles and 1269 reports with no abstract/title, which were manually screened. The remainder (25 862) were subject to the machine learning screening process. All eligible articles for the systematic review were extracted from <10% of the references available for screening. Moderate inter-reviewer reliability was achieved, with kappa-coefficient ≥0.75. Overall, 324 references were subject to full-text reading and 118 were deemed relevant for the systematic review. Our study showed that a broad search strategy to optimize the literature captured for systematic reviews can be efficiently screened by the semi-automated machine learning tool, Research Screener.

Neonatal early-onset sepsis calculator: Impact on antibiotic use in a level II neonatal unit in Western Australia.

BACKGROUND: Overuse of empirical intravenous antibiotics in neonates in high-income countries (HICs) is well documented. The Kaiser Permanente neonatal early-onset sepsis (EOS) calculator is an evidence-based sepsis risk assessment tool that has demonstrated potential to reduce antibiotic usage in this population. The incidence of early-onset sepsis in most HICs is 0.4-0.8 per 1000 live births. The objective was to evaluate the calculator's impact on antibiotic rates and length of stay in a regional level II Special Care Nursery. METHODS: A single-centre retrospective cohort study compared antibiotic administration rates in the first 72 h in neonates ≥35 weeks gestation born during two 6-month periods in 2019 (pre-EOS calculator) and 2021 (post-EOS calculator). Electronic and paper case records were accessed to capture data. Continuous data were summarised using mean and standard deviation, and categorical data were summarized using frequency distributions. There were 951 (2019) and 1129 (2021) infants born during the study periods. RESULTS: Following implementation of the calculator, antibiotic exposure decreased from 13.7% to 4.7% of all neonates without reported negative outcomes. Mean length of stay for neonates born across the two periods decreased from 2.38 to 2.13 days. Indications for antibiotic use shifted more towards clinical condition and away from obstetric risk factors. There were no culture-proven cases of sepsis or readmissions with EOS in either period. CONCLUSION: Implementation of the EOS calculator significantly reduced exposure to antibiotics, without adverse outcomes.

Skin-Microbiome Assembly in Preterm Infants during the First Three Weeks of Life and Impact of Topical Coconut Oil Application.

The structure and function of infant skin is not fully developed until 34 weeks of gestation, and this immaturity is associated with risk of late-onset sepsis (LOS). Topical coconut oil improves preterm-infant skin integrity and may reduce LOS. However, data on early-life skin-microbiome succession and potential effects of emollient skin care in preterm infants are scarce. We therefore collected skin-microbiome samples from the ear, axilla, and groin on days 1, 7, 14, and 21 from preterm infants born <30 weeks of gestation as part of a randomized clinical trial of standard skin care vs. topical coconut oil. We found that within-sample microbiome diversity was highest on day 1 after birth, with a subsequent decline and emergence of Staphylococcus genus dominance from day 7. Moreover, microbiome assembly was less diverse in infants receiving coconut oil vs. standard skin care. Our study provides novel data on preterm-infant skin-microbiome composition and highlights the modifying potential of emollient skin care.

Protocol for the development of a core outcome set for neonatal sepsis (NESCOS).

Neonatal sepsis is a serious public health problem; however, there is substantial heterogeneity in the outcomes measured and reported in research evaluating the effectiveness of the treatments. Therefore, we aim to develop a Core Outcome Set (COS) for studies evaluating the effectiveness of treatments for neonatal sepsis. Since a systematic review of key outcomes from randomised trials of therapeutic interventions in neonatal sepsis was published recently, we will complement this with a qualitative systematic review of the key outcomes of neonatal sepsis identified by parents, other family members, parent representatives, healthcare providers, policymakers, and researchers. We will interpret the outcomes of both studies using a previously established framework. Stakeholders across three different groups i.e., (1) researchers, (2) healthcare providers, and (3) patients' parents/family members and parent representatives will rate the importance of the outcomes in an online Real-Time Delphi Survey. Afterwards, consensus meetings will be held to agree on the final COS through online discussions with key stakeholders. This COS is expected to minimize outcome heterogeneity in measurements and publications, improve comparability and synthesis, and decrease research waste.

Physical compatibility of lipid emulsions and intravenous medications used in neonatal intensive care settings.

OBJECTIVE: The purpose of this study was to investigate the physical compatibility of intravenous lipid emulsions with parenteral medications used in neonatal intensive care. METHODS: Lipid emulsion and drug solutions were combined 1:1 in glass vials, inspected for physical incompatibility at 0, 1 and 2 hours, and assessed on the basis of lipid droplet size at 0 and 2 hours after mixing. Intravenous fluid controls (Water for Injection, sodium chloride 0.9% w/v, glucose 5% w/v), positive controls (gentamicin, albumin), negative controls (metronidazole, paracetamol, vancomycin) and 21 previously untested drug combinations were evaluated. RESULTS: No phase separation, change in colour, gas production or other visible anomaly was observed. The between-run mean droplet diameter (MDD) for SMOFlipid20% alone (0.301±0.008 µm) was comparable to the lipid emulsion/intravenous fluid and lipid emulsion/drug solution combinations. In addition to gentamicin and albumin, caffeine citrate (20 mg/mL) was shown to be incompatible with the lipid emulsion. All other lipid:drug combinations were compatible, based on the MDD data. CONCLUSION: Intravenous lipid emulsions were found to be compatible with 20 parenteral medications, including antimicrobial agents, inotropes, anti-inflammatory drugs and caffeine base, in simulated Y-site conditions. The lipid emulsion was incompatible with caffeine citrate injection.

Health service utilisation for acute respiratory infections in infants graduating from the neonatal intensive care unit: a population-based cohort study.

BACKGROUND: Despite advances in neonatal intensive care, babies admitted to Neonatal Intensive Care Units (NICU) suffer from adverse outcomes. We aim to describe the longer-term respiratory infectious morbidity of infants discharged from NICU using state-wide population-based linked data in Western Australia. STUDY DESIGN: We used probabilistically linked population-based administrative data to analyse respiratory infection morbidity in a cohort of 23,784 infants admitted to the sole tertiary NICU, born 2002-2013 with follow up to 2015. We analysed incidence rates of secondary care episodes (emergency department presentations and hospitalisations) by acute respiratory infection (ARI) diagnosis, age, gestational age and presence of chronic lung disease (CLD). Poisson regression was used to investigate the differences in rates of ARI hospital admission between gestational age groups and those with CLD, after adjusting for age at hospital admission. RESULTS: From 177,367 child-years at risk (i.e., time that a child could experience an ARI outcome), the overall ARI hospitalisation rate for infants and children aged 0-8 years was 71.4/1000 (95% confidence interval, CI: 70.1, 72.6), with the highest rates in infants aged 0-5 months (242.9/1000). For ARI presentations to emergency departments, equivalent rates were 114/1000 (95% CI: 112.4, 115.5) and 337.6/1000, respectively. Bronchiolitis was the most common diagnosis among both types of secondary care, followed by upper respiratory tract infections. Extremely preterm infants (< 28 weeks gestation at birth) were 6.5 (95% CI: 6.0, 7.0) times more likely and those with CLD were 5.0 (95% CI: 4.7, 5.4) times more likely to be subsequently admitted for ARI than those in NICU who were not preterm or had CLD after adjusting for age at hospital admission. CONCLUSIONS: There is an ongoing burden of ARI in children who graduate from the NICU, especially those born extremely preterm, that persists into early childhood. Early life interventions to prevent respiratory infections in these children and understanding the lifelong impact of early ARI on later lung health are urgent priorities.

Less is more: Antibiotics at the beginning of life.

Antibiotic exposure at the beginning of life can lead to increased antimicrobial resistance and perturbations of the developing microbiome. Early-life microbiome disruption increases the risks of developing chronic diseases later in life. Fear of missing evolving neonatal sepsis is the key driver for antibiotic overtreatment early in life. Bias (a systemic deviation towards overtreatment) and noise (a random scatter) affect the decision-making process. In this perspective, we advocate for a factual approach quantifying the burden of treatment in relation to the burden of disease balancing antimicrobial stewardship and effective sepsis management.

Gene filtering strategies for machine learning guided biomarker discovery using neonatal sepsis RNA-seq data.

Machine learning (ML) algorithms are powerful tools that are increasingly being used for sepsis biomarker discovery in RNA-Seq data. RNA-Seq datasets contain multiple sources and types of noise (operator, technical and non-systematic) that may bias ML classification. Normalisation and independent gene filtering approaches described in RNA-Seq workflows account for some of this variability and are typically only targeted at differential expression analysis rather than ML applications. Pre-processing normalisation steps significantly reduce the number of variables in the data and thereby increase the power of statistical testing, but can potentially discard valuable and insightful classification features. A systematic assessment of applying transcript level filtering on the robustness and stability of ML based RNA-seq classification remains to be fully explored. In this report we examine the impact of filtering out low count transcripts and those with influential outliers read counts on downstream ML analysis for sepsis biomarker discovery using elastic net regularised logistic regression, L1-reguarlised support vector machines and random forests. We demonstrate that applying a systematic objective strategy for removal of uninformative and potentially biasing biomarkers representing up to 60% of transcripts in different sample size datasets, including two illustrative neonatal sepsis cohorts, leads to substantial improvements in classification performance, higher stability of the resulting gene signatures, and better agreement with previously reported sepsis biomarkers. We also demonstrate that the performance uplift from gene filtering depends on the ML classifier chosen, with L1-regularlised support vector machines showing the greatest performance improvements with our experimental data.