P1‐071: Intensity of dementia through latent variable modelling (I‐DeLV) in the AIBL cohort

For detection of AD the characterisation of clinical status of populations at single time points and disease progression from preclinical to clinical stages is desirable. No single instrument has been shown to sufficiently address these issues, thus combinations have been applied. As the scales of these tests often contain limitations (floor and ceiling effects) their combination introduces the potential for redundancy. Hence a latent-variable approach may provide a more rational and accurate analytic method for combining data from different clinical and cognitive tests to characterise both level of impairment and rates of decline in AD. Data from 936 participants in the Australian Imaging, Biomarkers and Lifestyle (AIBL) study were utilised. Two latent-variable models were specified using a structural equation model framework. The first, full, model accounted for the variance of ten cognitive tests whereas the second, reduced, model only considered MMSE, CDR sum-of-boxes and CVLT-II, both models were corrected for a variety of demographic information. The resulting scores were assessed to see how well the different clinical diagnosis groups were partitioned and whether or not differences were seen between groups of participants that transitioned (e.g. HC to MCI) or remained stable at eighteen and 36 month follow-up. The efficacy of separation was assessed using pairwise-t-tests (adjusted for multiple testing). For both models the fit and parameter estimates were significant. Resulting scores from the reduced model approximated those of the full model. The resulting scores from both models were significantly differentiated between clinical diagnostic groups (HC, MCI and AD) and between stable and transitioning groups. e.g. at eighteen months sensitivity and specificities of 79% and 81% for transition to MCI and 84% and 85% for transition to AD were achieved. The described models provide accurate predictions of dementia severity within the AIBL cohort. Resulting scores are shown to differentiate between the clinical and cognitive status of stable and transitioning groups, suggesting that this model may be useful in determining individuals at risk of developing dementia. Such a score may also prove useful in identifying suitable candidates for, or quantifying the efficacy of, diagnostic or therapeutic trials.