Reading the gene in the eye: Clinical discrimination and confidence-accuracy calibration in genetic frontotemporal dementia

BackgroundClinical evaluation in frontotemporal dementia (FTD) occurs under diagnostic uncertainty, particularly in stages when symptoms are subtle, heterogeneous, and may resemble primary psychiatric disorders. Less attention has been devoted to clinicians' evaluation of judgment reliability.ObjectiveTo characterize clinicians' decision-making and confidence calibration when evaluating individuals at risk for genetic FTD within the Genetic Frontotemporal Dementia Initiative (GENFI) cohort.MethodsTwo analyses were conducted. First, determinants of diagnostic confidence were examined in asymptomatic participants (N = 837) using univariate analysis and logistic regression including demographic, neuropsychological, and clinical factors. Second, clinicians' ability to predict disease, based on mutation status, was evaluated in a subset of asymptomatic individuals (N = 133) using signal detection theory. Discriminability (d'), response bias (c), and metacognitive performance (meta-d' and M-ratio) were estimated from confidence ratings.ResultsLower diagnostic confidence clustered around boundary neuropsychiatric presentations characterized by subtle symptoms, lower cognitive scores, older age, and features including psychiatric history and current depression or anxiety. Clinicians predicted mutation status above chance level (accuracy ≈ 0.71), with better identification of non-carriers than carriers. Metacognitive modelling revealed imperfect calibration between confidence and accuracy (d' ≈ 1.06; meta-d' ≈ 0.85; M-ratio ≈ 0.80), indicating that confidence carried information about reliability but did not fully reflect available evidence.ConclusionsClinical decision-making in genetic FTD shows moderate accuracy, with lower confidence at the interface between neurodegenerative and psychiatric symptoms. Improving clinical decisions requires not only diagnostic accuracy, but also awareness of when judgments are correct. Biomarkers and structured assessments may help identify cases requiring evaluation.