LAUSR

Objective: To construct and compare the dynamic prediction models of the risk of mild cognitive impairment (MCI) in the elderly based on six different cognitive function scales. Methods: Based on longitudinal data from the Alzheimer's Disease Neuroimaging Initiative from 2005 to 2020, Mini-mental state examination (MMSE), functional activities questionnaire (FAQ), Alzheimer's disease assessment scale-cognitive (ADAS-Cog) 11, ADAS-Cog13, ADAS delayed word recall (ADASQ4), and Rey auditory verbal learning test (RAVLT)_immediate were used as longitudinal cognitive function evaluation indicators to assess the longitudinal changes in cognitive function. The joint model was used to analyze association between indicators variation trajectory and survival outcome MCI, and construct the risk prediction model of MCI in the elderly, the linear mixed model was constructed the longitudinal sub-model which described the evolution of a repeated measure over time, a proportional hazards model was constructed the survival sub-model, and the two sub-models were connected through the correlation parameter (α). The areas under the receiver operator characteristic curve (AUC) were used to evaluate the predictive efficacy of the model in the follow-up period of (t, t+Δt). The starting point t was selected at the 30th, 42nd, and 54th month, and the Δt was selected as 15 and 21 months. Based on the prediction model, an example of the research object was selected for dynamic individual predictions of the risk of MCI. Results: Finally, 544 older adults (aged 60 years and above) with normal baseline cognitive status were included, of which 119 cases (21.9%) had MCI during the follow-up process were regarded as the case group, and 425 cases remained normal as the control group. The joint model suggests that the longitudinal trajectories of the six evaluation indicators are all related to the risk of MCI (P<0.001). The risk of MCI decreased by 32.3% (HR=0.677, 95%CI: 0.541-0.846) and 10.8% (HR=0.892, 95%CI: 0.865-0.919) for each one-point increase of MMSE and RAVLT_immediate longitudinal scores. The risk of MCI increased by 53.2% (HR=1.532, 95%CI: 1.393-1.686), 36.2% (HR=1.362, 95%CI: 1.268-1.462), 23.2% (HR=1.232, 95%CI: 1.181-1.285), and 85.1% (HR=1.851, 95%CI:1.629-2.104) for each one-point increase of FAQ, ADAS-Cog11, ADAS-Cog13, and ADASQ4 longitudinal scores. AUC results show that RAVLT_immediate (0.760 2) and ADASQ4 (0.755 8) have higher average prediction efficiency, followed by ADAS-Cog13 (0.743 7), ADAS-Cog11 (0.715 3), FAQ (0.700 8) and MMSE (0.629 5). ADASQ4 joint model was used to provide a dynamic individual prediction of the risk of MCI. The average probability of MCI after five years of follow-up and ten years of follow-up in the example individuals were 8% and 40%, respectively. Conclusions: The RAVLT_immediate and ADASQ4 scales, which are only for memory tests, have high accuracy in predicting the risk of MCI. Using the RAVLT_immediate and ADASQ4 scales as longitudinal cognitive function evaluation indicators to construct a joint model, the results can provide a basis for realizing MCI risk prediction for the elderly.