Performance of routine surveillance diagnostics of external ventricular drain-associated infections in a critical care setting: a retrospective cohort study.

INTRODUCTION

External ventricular drains (EVDs) are crucial for treating neurocritically ill patients but are complicated by feared EVD-associated infections (EVDIs) in up to 35% of all inserted drains, contributing significantly to morbidity, mortality and account for a significant proportion of intensive care unit (ICU) antibiotic use. However, the lack of a universal definition for EVDIs leads to inconsistent diagnostic criteria across studies, with a concern of substantial overtreatment with broad-spectrum antibiotics. This study aimed to evaluate if current EVDI surveillance parameters can be optimized to better distinguish true from suspected EVDI.

METHODS

We conducted a retrospective cohort study at the Karolinska University Hospital ICU, including all patients treated with EVDs between 2006 and 2023, excluding patients with primary central nervous system (CNS) infections. EVDI surveillance included biweekly sampling and cultures from cerebrospinal fluid (CSF). Patients were categorized as no infection (NI), suspected infection (SI), or verified infection (VI) based on culture results and treatment status. We employed classification and regression analyses to identify predictors of VI.

RESULTS

Among 1,828 patients with EVDs, 29.8% were initiated on antibiotic treatment due to suspected infection and 4.1% were found to have culture confirmed infections. The main finding is that current accepted diagnostic parameters cannot distinguish aseptic inflammation from true EVDI. In multivariable logistic analysis the best models exhibited low accuracy, with a pseudo- of only 0.06. CSF lactate was the most important metric in a univariable setting, however with a cut-off of 8.9 mmol/L it showed low discrimintive ability and limited clinical utility.

CONCLUSIONS

In this study we evaluate current accepted EVDI surveillance methods in, to our knowledge, the largest cohort of paired samples to date. We find that current surveillance parameters cannot distinguish aseptic CNS inflammation from true EVDIs in an ICU setting. This contributes to a significant antibiotic overtreatment, with 25% of our entire cohort being unnecessarily initiated on broad-spectrum antibiotics, a number we expect can be generalized. We identify a large clinical problem with consequences on both a individual and population level, and recommend that future research focus on evaluating new techniques, such as fast bedside sequencing methods.