Clinical validation and machine learning optimization of MyCog: A self-administered cognitive screener for primary care settings.

Background: Primary care offers ideal opportunities for early detection of cognitive impairment, yet clinics face significant barriers to routine screening. MyCog is an electronic health record-integrated tablet application self-administered during a primary care visit designed to overcome barriers to screening. Methods: We compared MyCog performance between 65 adults age 65+ with diagnosed cognitive impairment and 80 cognitively normal adults. Five modeling approaches achieved consensus to select consistently discriminative variables for the final detection algorithm. Performance was primarily assessed via receiver operating characteristic area under the curve (AUC), sensitivity, specificity, and accuracy. Results: All models demonstrated strong diagnostic performance (AUC 0.817 to 0.873). Memory accuracy and executive function efficiency scores were consistently selected as predictors of impairment across models. The final logistic regression achieved AUC 0.890, with sensitivity 0.723 to 0.831, specificity 0.788 to 0.912, and accuracy 0.807 to 0.828 depending on threshold. Discussion: Findings suggest MyCog accurately detects cognitive impairment via a streamlined self-administered app that efficiently fits into primary care workflows.