AbstractBackgroundThe early identification of patients at risk of dementia and timely medical intervention can be effective in preventing disease progression and mitigate burdens and costs of health insurance. Although the current diagnostic tests for preclinical and prodromal Alzheimer’s disease (AD) using neuropsychological evaluation and biomarkers can accurately predict the progression to AD dementia, the high cost, invasive nature, and limited availability restrict their use to wealthy clinics only. Therefore, we developed a model to predict cognitive impairment based on non‐invasive information, including eye movement (EM) data.MethodEye‐tracking data was extracted from 594 subjects, 428 cognitively normal (CN) controls, and 166 patients with mild cognitive impairment (MCI) while they performed Prosaccade/Antisaccade (PS/AS) and Go/No‐go tasks. EM behavior was compared by task after stratification by cognitive status and controlling for age, sex, and education. We used logistic regression to calculate odds ratios (OR) and 95% confidence intervals (CI) from the EM metrics.ResultLogistic regression models suggested increased OR of MCI from EM metrics, including wider latency variability, fixation duration variability, a higher number of errors committed, anticipations, and omissions with 95% CI of OR between 1.213‐1.621.ConclusionEM metric changes linked with MCI are associated with deficits in attentional and executive functions. The models built from non‐invasive data allow the discrimination of patients with MCI from cognitively normal adults. In summary, our results suggest that eye tracking, in combination with demographics and brief cognitive tests, might significantly improve the diagnostic prediction of MCI.