Sepsis, October 20 edition

This PRISMA-guided systematic review included 64 studies using MIMIC-III or eICU-CRD and found striking variation in sepsis detection rates, including 16.9%-42.2% within MIMIC-III and 13.9%-31.4% within eICU-CRD despite identical cohort definitions. Only 12 studies provided source code, and reporting was particularly incomplete for aggregation methods and missing-data handling. The findings indicate that implementation choices, rather than sampling error or study-level characteristics, largely determine automated sepsis labels.

Impact: This study identifies a fundamental reproducibility problem affecting sepsis epidemiology, machine-learning model development, and benchmarking: clinically identical definitions do not produce comparable labels when implementation details differ. Its recommendations for transparent methodological reporting and version-controlled code can directly improve the validity of future research.

Clinical Implications: Clinical informatics studies and sepsis prediction models should prespecify SOFA baselines, infection-detection rules, temporal windows, aggregation units, and missing-data procedures. Clinicians should interpret reported sepsis incidence and model performance cautiously when these details are not reproducible.

Future Directions: Future studies should adopt a standardized, machine-readable sepsis-label specification, preregister implementation choices, publish version-controlled code and dependency information, and validate automated labels against clinically adjudicated cohorts.

This prespecified process evaluation embedded in the multicentre ADAPT-Sepsis trial included 2,760 adults across 41 UK critical care units. Procalcitonin-guided protocols generated strong stop advice 1.7 days earlier than C-reactive protein-guided protocols, but clinician adherence was higher with C-reactive protein guidance by 6.8 percentage points. Non-adherence commonly reflected fixed microbiology treatment plans, illness severity, persistent fever, and elevated inflammatory markers.

Impact: The study addresses the implementation gap between an algorithm's theoretical ability to shorten antibiotic therapy and clinicians' willingness to act on early discontinuation advice. It provides practical evidence that a less aggressive biomarker signal may be followed more reliably, while procalcitonin may offer greater antibiotic-sparing potential if implementation support is strengthened.

Clinical Implications: Biomarker-guided antibiotic stewardship should incorporate clinician education, explicit escalation pathways, multidisciplinary agreement, and safeguards for patients with ongoing fever or severe illness. Procalcitonin-based algorithms should not be implemented as stand-alone automated stop orders without addressing clinician trust and contextual review.

Future Directions: Future implementation studies should test pragmatic, clinician-facing algorithms that combine biomarker results with illness severity, microbiology plans, fever trajectory, and explicit safety-net criteria. Cluster-randomized trials should determine whether improved adherence reduces antibiotic exposure without increasing relapse, mortality, or readmission.

Using nationwide Swedish register data from 1987 through 2022, this cohort study compared 55,906 pediatric sepsis hospitalizations with 93,274 nonsepsis hospitalizations involving organ dysfunction. Mortality remained higher in the sepsis cohort during the first five years of follow-up, with an adjusted hazard ratio of 1.17, and subsequent sepsis, infectious disease, and respiratory disease were more common. The findings extend the concept of post-sepsis burden to childhood and support structured survivorship care.

Impact: This is one of the largest population-based evaluations of long-term outcomes after pediatric sepsis and demonstrates that the consequences extend beyond the index hospitalization. The results support a shift from an acute-care-only model toward long-term surveillance and rehabilitation for pediatric sepsis survivors.

Clinical Implications: Children discharged after sepsis may benefit from structured follow-up addressing recurrent infection, respiratory disease, functional status, neurodevelopment, and family support. Pediatric services should consider post-sepsis pathways rather than ending care at hospital discharge.

Future Directions: Prospective survivorship cohorts should characterize neurodevelopmental, functional, respiratory, immunologic, and psychosocial outcomes after pediatric sepsis. Intervention studies should evaluate whether structured post-sepsis clinics, rehabilitation, vaccination strategies, and family-centered support improve long-term health.