Insight discordance differentiates behavioural variant frontotemporal dementia from false positives.

Behavioural variant frontotemporal dementia (bvFTD) can present with behavioural symptoms similar to psychiatric mimics of the disease (sometimes termed 'phenocopies'). Despite this overlap, loss of insight is a cardinal feature of bvFTD unlike most psychiatric disorders. This study tested whether insight could be quantified to improve diagnostic accuracy in distinguishing these conditions, and examined neuroimaging correlates of impaired insight. In this prospective cohort study, 27 patients with frontotemporal lobar degeneration (FTLD) presenting as the bvFTD phenotype and 25 false-positives of bvFTD who ultimately had a psychiatric disorder underwent patient, informant, and clinician symptom ratings on a modified Frontal Behavioural Inventory (FBI). Discrepancy scores were calculated, diagnostic discrimination was evaluated by area under the curve (AUC) with leave-one-out cross-validation, alongside a neuroimaging analysis of network-level associations of insight in bvFTD. Patients with bvFTD endorsed fewer symptoms than false-positives (FBI 6.62 ± 8.80 vs 15.36 ± 8.42, P < 0.001), whereas clinician observations were markedly greater in bvFTD (27.67 ± 8.35 vs 5.64 ± 3.85, P < 0.001). Informant ratings did not differ between groups (25.15 ± 11.11 vs 21.32 ± 12.61, P = 0.14), even though bvFTD informants had more knowledge of the disease (FTD Knowledge Scale 16.20 ± 6.68 vs 9.41 ± 8.19, P = 0.003). A model incorporating patient-clinician and informant-clinician discrepancy showed excellent cross-validated discrimination (AUC 0.99, 95% confidence interval (CI) 0.97-1.00). A model independent of expert clinician ratings, the patient-informant discrepancy, showed fair cross-validated discrimination across the full cohort (AUC 0.71, 95% CI 0.56-0.85) and good discrimination in those who did not develop another FTLD syndrome on follow-up ('pure bvFTD'; AUC 0.78, 95% CI 0.63-0.94). In bvFTD, reduced insight was associated with salience and semantic appraisal network degeneration but not control frontotemporal regions. BvFTD, false-positives, and their informants demonstrated opposing patterns of symptom endorsement. Symptom under-endorsement (lack of insight) by patients relative to clinician and informant observation supports neurodegeneration, whereas informant and patient over-endorsement without clinician-observed disturbance suggests a false-positive. Unfamiliarity with bvFTD may contribute to over-reporting. Though the clinician-dependent findings may not translate to non-expert settings, the patient-informant model retained good discrimination. Formal quantification of discrepancy patterns in behaviour may reduce misdiagnosis in bvFTD. The neuroimaging findings support salience and semantic appraisal networks as the neuroanatomical basis for insight in bvFTD. Future studies should validate the FBI clinician version and examine factors underlying over-reporting by false-positives and their informants.