Adapting Pupillometry Preprocessing and Analysis Methods for Older Adults with Mild Cognitive Impairment: A Decision-by-Decision Approach
Year: 2026
Type: preprint
Abstract: Cognitive pupillometry guidelines have been developed and validated almost exclusively on young, healthy participants recorded at high sampling rates, leaving open whether their preprocessing and statistical defaults transfer to harder recording contexts. Older adults with mild cognitive impairment (MCI) illustrate this problem directly: senile miosis, elevated blink rates, and pharmacologically induced pupillary constriction depart from the assumptions existing pipelines were built on. Lower-cost, lower-sampling-rate eye trackers, a translationally important target for scalable clinical screening, compound this further. Using a pooled sample of 67 older adults with MCI from two independent cohorts, recorded at 33 Hz during a visual oddball task, we adapt and empirically justify, decision by decision, a cognitive-pupillometry pipeline for this harder context, evaluating each decision against the well-established Target versus Distractor effect. Some literature defaults held up unchanged once tested directly against this dataset; others did not transfer; and others had no applicable default at all and were derived empirically. The Target versus Distractor effect was recovered consistently across every preprocessing alternative tested and across complementary statistical approaches, generalized additive and linear mixed-effects models and trial-level features, whose choice we show depends on whether the question concerns the timing, the average size, or specific components of the effect. This paper's contribution is not a new pipeline or a clinical marker, but a transparent, reproducible account of which standard assumptions and defaults hold in a harder recording context, which require empirical re-derivation, and which have no existing literature anchor at all, illustrated with accompanying code.
Source: PsyArXiv (OSF Preprints)
Language: English
Subfield: Human-Computer Interaction
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7223287828
