The pupillometer does not beat the torch after cardiac arrest

Design

bicentric analysis of prospective registries, predictive performance against Cerebral Performance Category at 3 months

Population

442 comatose adults after cardiac arrest with pupillometry in the first 48–72 hours, Lausanne and Paris, 2020–2024

Primary outcome

prognostic performance of quantitative pupillary light reflex and Neurological Pupil index for poor and good outcome

Effect

NPi ≤2: 99% specificity, 34% sensitivity for poor outcome; qualitative pupillary light reflex sensitivity 38.9% vs 23.6% and 19.0% for quantitative measures (p<0.001) at comparable specificity

Quantitative pupillometry has been adopted into post-arrest neuroprognostication on the reasoning that a number is better than a clinician's impression of whether a pupil reacted. This two-centre study analysed prospective registries from Lausanne and Paris: 442 comatose adults with pupillometry in the first 48 to 72 hours after cardiac arrest, with outcome at three months.

The correlations were strong and in the expected directions. Quantitative pupillary light reflex values were lower with early myoclonus, absent bilateral N20 evoked potentials, unreactive or epileptiform electroencephalogram, and suppressed background, and correlated with neuron-specific enolase at 48 hours. For poor outcome, a Neurological Pupil index of 2 or less was 99% specific at 34% sensitivity; a minimum quantitative reflex below 2% was 96.4% specific at 23.6% sensitivity.

Then the comparison that matters. Qualitative pupillary light reflex — the bedside examination — was more sensitive than either quantitative measure (38.9% against 23.6% and 19.0%, p<0.001) at comparable specificity. And no quantitative threshold predicted good outcome usefully: a maximum reflex above 25.5% reached 65.5% sensitivity at only 62.2% specificity. The device adds objectivity and reproducibility between observers. It does not add prognostic information over a torch.

• Keep pupillometry as one strand of multimodal prognostication, alongside evoked potentials, electroencephalogram and neuron-specific enolase — never alone.
• The value of the device is inter-observer reproducibility and documentation, not better prediction.
• Nothing here predicts good outcome, which is the question families actually ask; absence of a poor-outcome marker is not a favourable prognosis.
• Specificity of 99% at 34% sensitivity means the test is for ruling in a poor outcome, never for ruling one out.
• Where pupillometers are not available — which is most Indian intensive care units — careful serial bedside examination is not the inferior option this technology implies.

Why it matters

A device bought to improve prognostic accuracy turns out to improve consistency instead, which is a different and smaller claim.

Don't overread it

This was a registry analysis with outcomes at three months; self-fulfilling prophecy is a known hazard when the same markers inform withdrawal decisions and outcome.

The statistics, in plain English

High specificity with low sensitivity is exactly what a prognostication test should have: a false positive here means withdrawing care from someone who could have recovered, so 99% specificity is the non-negotiable property and 34% sensitivity is the price. But 99% specificity in 442 patients still means the upper confidence limit for false positives is not zero, which is why no single marker should ever stand alone. The correlations with neuron-specific enolase (Spearman rho −0.283 and −0.444) are moderate at best — these markers agree in direction, not in individual patients.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.