Real-time Inflammation Panel for inpatient care: a before-after evaluation of post-analytic timeliness and outcomes at a tertiary center

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Background Post-analytic delays - between laboratory result availability and clinical action - remain a bottleneck in inpatient care. Prior alerting tools show variable benefit, often limited by workflow fit, routing, and duplicate notifications. We implemented a real-time, rule-based assistant to address these constraints. Objective To determine whether the “Inflammation Panel” shortens the interval from result-ready to treatment and improves outcomes. Materials and Methods Single-center, before-after evaluation at a tertiary hospital (Saint Petersburg, Russia). The assistant polls the lab store (around 2-s cadence), applies declarative rules (thresholds + trend + freshness + de-duplication) to CRP, WBC, troponin, and hemoglobin, and pushes role-routed alerts to team chats. Adults with eligible labs in contiguous pre- and post-deployment windows in 2025 were included (pre n = 2430, post n = 2560). Primary endpoint: time from result-ready to treatment start. Secondary endpoints: in-hospital mortality (sepsis, pneumonia), length of stay (LOS), and alert burden. Analyses used rank-based estimators for medians, adjusted logistic models, Kaplan-Meier displays, and interrupted time-series checks. System telemetry (poll cadence, dispatch latency, lab turnaround) was recorded. Results Among 4990 encounters, median result-to-treatment time decreased from 6.12 to 3.56 h. Sepsis mortality declined from 18.5% to 11.0% (OR 0.55, 95% CI 0.31–0.95); pneumonia mortality droped from 13.1% to 8.0% (OR 0.61, 95% CI 0.37–0.99). Mean legth of stay fell from 10.32 to 8.95 days (Δ -1.37). A Cox model indicated faster treatment post-deployment (HR 1.41, 95% CI 1.33–1.49). Alert burden averaged 0.23 alerts/patient-day with around 4% duplicate suppression. Against an operational composite, sensitivity was 0.88 and PPV 0.83. Telemetry showed around 2 s polling and around 2 s dispatch with lab processing median around 14 min. Conclusions A transparent, push-based assistant compressed the post-analytic interval and was associated with lower mortality and shorter LOS at modest alert burden. Declarative rules with trend context, freshness, and role-based routing, combined with continuous telemetry, offer a portable pattern for real-time clinical decision support.

Article activity feed