Validity, diagnostic accuracy, and feasibility of trail making test in virtual reality for mild cognitive impairment

Background Immersive virtual reality (VR) adaptations of neuropsychological tests change how participants respond and must be validated in each clinical population. We evaluated the Trail Making Test in Virtual Reality (TMT-VR), a head-gaze version of the paper-and-pencil Trail Making Test (hereafter, TMT) that records completion times and wrong-target selections automatically, in older adults with and without mild cognitive impairment (MCI). Methods In this case–control evaluation, 84 adults aged 60–89 years (42 with clinically diagnosed MCI, 42 healthy comparison participants) completed the TMT-VR, the TMT and the Mini-Addenbrooke's Cognitive Examination (M-ACE). Planned analyses examined feasibility, convergence, group differences and diagnostic accuracy for all 26 TMT-VR scores against their TMT counterparts and the M-ACE, with demographic adjustment, multiplicity correction and sensitivity analyses. Results All participants completed both TMT-VR parts, and cybersickness ratings decreased over the session. Part B completion times correlated moderately across formats, more clearly in healthy participants. Participants with MCI were slower and made more wrong-target selections on TMT-VR Part B but not Part A. Part B completion-time discrimination did not differ detectably between formats. The Part B, total and B − A error scores and the Part B-specific derived scores discriminated MCI more strongly in the TMT-VR, with paired differences excluding zero for the error scores but none surviving Holm correction. The TMT's advantage lay in Part A completion time; its Part B difference depended more on education than the TMT-VR's. Education-matched and comparison-group sensitivity analyses showed the same direction. The M-ACE discriminated more accurately than any single trail-making score of either format. Conclusions In this initial case–control evaluation the TMT-VR was feasible and well tolerated, converged partially with the TMT and showed known-groups validity for MCI. Its Part B-specific and error-based scores showed higher point estimates of discriminative accuracy than the corresponding TMT scores, consistently across related scores, so it is a candidate trail-making format for neuropsychological batteries in MCI, albeit one that requires further examination. Residual confounding by education cannot be excluded, and the reference standard was asymmetric; test–retest reliability, reference values and incremental value beyond a global screen remain to be established in consecutive clinical referrals.