Skip to contents

Gazepoint survival-analysis reproducibility checklist

Source: vignettes/articles/gaze-survival-reproducibility-checklist.Rmd
gaze-survival-reproducibility-checklist.Rmd

Purpose

This checklist documents what must be frozen when gp3tools is used as the Gazepoint-specific adapter to the vendor-neutral eyeprocess survival engine. It does not introduce any new estimator or censoring logic.

Gazepoint input contract

Record the Gazepoint export type, trial-window construction, participant/trial identifiers, coordinate transformation, event/fixation source, target AOI geometry, time origin, and gaze-quality rule. A missing target event is not sufficient evidence for censoring unless the observation window is complete and usable.

Adapter choices

Record the explicit cox_structure and aft_distribution supplied to run_gazepoint_latency_analysis(). Distinguish participant-clustered Cox from latent participant frailty and report the corresponding hazard-ratio interpretation. Record the AFT family and report time ratios when AFT results are presented.

Quality and censoring audit

Archive observed-event, right-censored, and review-required counts; censoring percentage by condition; ambiguous join failures; event-after-window failures; and gaze-quality exclusions/review states. Do not convert low-quality or incomplete Gazepoint trials into ordinary censoring.

Sensitivity plan

Pre-specify AOI geometry, detector/fixation rule, time origin, minimum fixation duration, quality threshold, Cox structure where scientifically relevant, AFT family, and review/exclusion alternatives. Rebuild the delegated eyeprocess survival table whenever the event or AOI definition changes.

Software provenance

Report both gp3tools and eyeprocess versions, plus the specialist model backends (survival and coxme) where relevant. Preserve the eyeprocess provenance columns returned by the adapter.

Reporting checklist

Report the target event/AOI, time origin, observation-window rule, gaze-quality criterion, event/censor/review counts, Cox structure, AFT family, effect estimate with 95% CI, diagnostics, sensitivity branches, and both package versions.