Early warning of infection in patients undergoing hematopoietic stem cell transplantation using heart rate variability and serum biomarkers.

Abstract

BACKGROUND\nEarly warning of infection is critical to reduce risk of deterioration and mortality, especially in neutropenic patients following hematopoietic stem cell transplantation (HCT). Given that heart rate variability (HRV) is a sensitive and early marker for infection and serum inflammatory biomarkers can have high specificity for infection, we hypothesized their combination may be useful for accurate early warning of infection.\n\n\nPURPOSE\nDevelop and evaluate a composite predictive model utilizing continuous HRV with daily serum biomarker measurements to provide risk stratification of future deterioration in HCT patients.\n\n\nMETHODS\n116 ambulatory outpatients about to undergo HCT consented to collection of prospective demographic, clinical (daily vital signs), HRV (continuous electrocardiogram (ECG) monitoring, laboratory (daily serum samples frozen -80°C) and infection outcome variables (defined as the time of escalation of antibiotics), all from 24 hours (h) pre-transplant until infection or 14 days post-transplant. Indications for antibiotic escalation were adjudicated as true infection or not by two blinded HCT clinicians. A composite time series of 8 HRV metrics was created for each patient and the probability of deterioration within the next 72h was estimated using logistic regression modelling of composite HRV and serum biomarkers using a rule-based Naïve-Bayes model, if the HRV-based probability exceeded a median threshold.\n\n\nRESULTS\n35(30%) patients withdrew within <24h due to intolerability of ECG monitoring, leaving 81 patients, of which 48(59%) had antibiotic escalation adjudicated as true infection. The combined HRV and biomarker (TNFα, IL6 and IL7) predictive model started increasing ∼48 hours on average prior to diagnosis of infection, could distinguish between high risk of impending infection (>90% incidence of subsequent infection within 72h), average risk (∼50%) and low risk (<10%), with an area under the receiver operating characteristic curve (AUC-ROC) of 0.87.\n\n\nCONCLUSION\nWe derived a predictive model using HRV and serum biomarker to predict being diagnosed with infection within 72h combined in patients at high risk of infection. As prophylactic predictive ECG monitoring and daily serum collection proved challenging for many patients, further refinement in measurement is necessary for further study.