An improved laboratory protocol to assess subarachnoid haemorrhage in patients with negative cranial CT scan

Abstract

The laboratory analysis of cerebrospinal fluid (CSF) plays a key role in considering subarachnoid haemorrhage (SAH) in patients with clinical suspicion, but negative CT scan. Although the determination of the CSF bilirubin concentration generally provides high sensitivity, it was recently shown that specificity and positive predictive value are unacceptably low, limiting its use as a diagnostic tool. We intended to design and evaluate an improved laboratory protocol, which fulfills the requirement of better specificity without losing sensitivity. We present a procedure in which a "bili-excess" concentration is determined, which is the surplus CSF bilirubin measured after subtraction of an estimated upper limit for the individual patient. The latter is calculated from [bilirubin](serum), [albumin](serum) and [albumin](CSF), taking into account the propagation of analytical errors in the individual analyses. We investigated the applicability of direct absorption vs. derivative spectroscopy, thereby addressing the influence of various calibration methods. We evaluated our procedure in 92 CSF samples drawn from patients with (n=37) and without (n=55) clinical suspicion of SAH. In our study population, we show that specificity increases from 0.83 (95% CI, 0.74-0.91) to 1.00 (95% CI, 0.96-1.00) using the bili-excess concept, with an established upper limit for bili-excess of 0.11 micromol/L instead of the recommended use of an "uncorrected" CSF bilirubin upper limit of 0.20 micromol/L. Sensitivity in both cases is 1.00 (95% CI, 0.66-1.00). We demonstrate the merit of allowing for analytical imprecision in the bili-excess concept. We provide a quantitative procedure to explore the likelihood of (CT-negative) SAH independent of the absolute CSF bilirubin concentration by considering the "bili-excess" concentration per individual, using derivative spectroscopy to determine CSF bilirubin. This procedure led to an increase in specificity to 1.00 (95% CI, 0.96-1.00) in our study population.