Abstract Body: Background and Purpose:

Due to limitations associated with the symptom-based tools currently used for triage, up to 35% of strokes are missed at initial clinician contact in emergency medicine settings. As a result, there has been a decades-long push to identify blood biomarkers with utility for stroke recognition. Despite numerous studies reporting candidates with seemingly high levels of diagnostic performance, none have made their way into clinical use. It is possible this discrepancy is linked to translational limitations in how these prior biomarker discovery investigations have been designed and implemented. Thus, we performed a nontraditional systematic review and meta-analysis spanning 30 years of published acute stroke blood biomarker data in which we sought to explicitly quantify the effects of study design and reporting considerations on diagnostic performance estimates, and correct said estimates for such effects in order to provide better projections of how the candidate biomarkers that have been studied to date would perform in a true clinical use scenario.

Methods and Results:

A electronic database search was performed to identify previously reported human acute stroke blood biomarker investigations, which yielded 14,253 potential articles. 189 were included in our final analysis based on subsequent abstract and full-text screening. From each article, detailed methodological information for all reported diagnostic comparisons was manually extracted. Collectively, data from 705 different diagnostic comparisons involving 355 unique single analytes or multi-analyte panels were captured. The raw diagnostic performance estimates reported in the literature inferred that as high as 40% of previously investigated candidate biomarkers could be clinically useful for stroke recognition. However, multiple regression revealed that 47.2% of the variance in these diagnostic performance estimates could be attributed to a small number of confounding study design and reporting factors related to population sampling, timing of blood collection, blinding, and conflicts of interest. After using the model residuals to correct for these factors, not a single candidate biomarker displayed adequate evidence to suggest true clinical utility.

Conclusions:

Our results clarify the current state of the search for an acute stroke blood biomarker, and identify several translational limitations that need to addressed in future investigations if one is to be realized.