Stroke is one of the most frequently mistriaged medical emergencies. As a result, significant efforts have been made to identify blood biomarkers that could aid in stroke recognition. Large numbers of studies have reported candidate assays with seemingly high levels of diagnostic performance; however, practices employed in many of these studies could limit generalizability. Thus, in this systematic review and meta-analysis, we sought to clarify whether any of the assays proposed to date have true potential for clinical use. To do this, we quantified the influence of a wide range of possible translational confounds on the diagnostic performance estimates published for 863 single analyte or multianalyte assays investigated for stroke recognition over the past 30 years. We found that up to 56% of the variance in reported levels of diagnostic performance could be directly attributed to latent study design and reporting factors, rather than true differences in the diagnostic capacity of the assays themselves. After correcting for these factors and further considering analytical measurability in the target use environment, not a single assay displayed evidence of clinical utility. Our results clarify the current state of the push towards a viable blood-based stroke diagnostic, and also provide a contextual framework that can be used to more critically evaluate the broader space of molecular biomarker research.
Smothers et al. (Thu,) studied this question.