Neuroprognostication after cardiac arrest

Background pattern
In a series of 187 patients

Machine learning
Two studies have suggested that machine learning algorithms may prove to be useful for neuroprognostication provided that they take into account expected changes over time.

Pupillometry
In a study of 55 adult patients after cardiac arrest, the Neurological Pupil index (NPi) performed early (6 hours post-arrest) predicted poor outcome (CPC 3-5) with a specificity of 0.82, sensitivity of 0.60, and AUC of 0.72, with a false positive rate of 0.17 (95% CI 0.06-0.41). The constriction velocity at 6 hours <0.23 mm/s had an AUC of 0.78, perfect specificity but sensitivity was only 0.47, and had a false positive rate of 0 (0-0.18). The percent reactivity (%PLR) of <5% had an AUC of 0.75, specificity 0.94, sensitivity 0.45, and false positive rate of 0.06 (0.01-0.27). For NPi done early, if value was 0-3 86% had poor outcome, and if never <3 only 52% had poor outcome. If done during TTM, if value was 0-3 64% had poor outcome and if never <3 45% had poor outcome.

Routine labs
Higher HgBa1c after ROSC is associated with higher rates of unfavorable neurological outcome at 6 months (CPC 3-5) with OR 1.41 (95% CI 1.05-1.90) and higher neuron specific enolase (NSE) values.

Novel markers
A study of 299 proteins revealed several proteins which had predictive value for outcome of CPC 1-2 after cardiac arrest. Of those, the ones that collectively formed the best model included α-enolase, 14-3-3- protein ζ/δ, cofilin-1, and heat shock cognate 71 kDa protein.

Neurofilament light may prove to be beneficial, as it had the highest predictive value of several biomarkers (NSE, S100, GFAP, Tau, ubiqutin carboxyl hydrolase L1). Neurofilament light had AUC 0.92 compared with NSE of 0.84, and neurofilament light had higher sensitivity for poor outcome with comparable specificity to NSE.

Imaging
In a meta-analysis, loss of grey-white differentiation was very insensitive (sensitivity 0.44) for poor outcome (CPC variable in different studies, usually 3-5 but sometimes 4-5) but quite specific, with a false positive rate (FPR) of only 2-3%. MRI was more sensitive with sensitivity of DWI changes ranging from 0.65-0.83 depending on methodology (0.79 on average) but less specific, with FPR of 5-10%.

Disparities
Non-white patients have worsened outcomes, even after TTM.

Problems
Surveys suggest that approaches to prognostication are very inconsistent, and that educational efforts are needed to help standardize prognostication.