Dynamic Risk Prediction Of Mild Cognitive Impairment To Ad Conversion Based On Landmark Model

Objective:Our first aim of this article was to combine multiple longitudinal biomarkers to develop landmark models and compare in terms of their predictive performance,and then select the optimal model to provide estimate of the probability of patients with Mild Cognitive Impairment(MCI)to Alzheimer’s disease(AD).Our second aim was to provide an example for the other chronic disease researchers of how to develop a prediction model using landmark,illustrating the utility of this approach for marking the beat use of longitudinal and survival data.Methods:The data we used were from the Alzheimer’s Disease Neuroimaging Initiative(ADNI).We screened 312 patients with a 6-year follow-up period from 2004 to 2019.With AD as the outcome,the predictors were Mini Mental State Examination(MMSE),Alzheimer’s Disease Assessment Scale-Cognition Section(ADAS-13),Rey auditoy verbal learning test,functional activities questionaire(FAQ),hippocampal volume,APOE ε 4 genotype and demographic information(age,gender,and education years).The simple landmark model,IPL model and IPL * model were established by using longitudinal and survival data.The evaluation of the goodness of model fitting used Akaike Information Criterion(AIC)and Bayesian Information Criterion(BIC).We used C-index and Brier score to evaluate discriminative ability and prediction accuracy performance of the models.According to the model evaluation results,the optimal prediction model was selected to provide the dynamic probability prediction of the patient from MCI to AD.Results:Among the 312 selected patients,149 patients converted to AD during follow-up.The simple landmark model results showed that the cognitive longitudinal measurement factor: FAQ(b = 0.055,P = 0.0001),RAVLT immediate(b =-0.030,P = 0.013),and the hippocampal volume(b =-0.190,P = 0.027)significantly associated with patients from MCI to AD.The results of the ipl model showed that the cognitive longitudinal measurement factor: FAQ(b = 0.055,P = 0.0001),RAVLT immediate(b =-0.029,P = 0.014),and hippocampal volume(b =-0.189,P = 0.031)significantly associated with patients from MCI to AD.ipl * model results showed: ADAS-cog13 score(b = 0.031,P = 0.049),FAQ(b = 0.055,P = 0.0001),RAVLT immediate(b =-0.028,P = 0.015),hippocampus Volume(b =-0.190,P = 0.029)had an effect on the conversion of patients from MCI to AD.Each unit increase in the ADAS-cog13 score increased the risk of conversion to AD by 3.2%(HR: 1.032,95% CI: 1.000 ~ 1.065);each unit increase in the FAQ score increased the risk of conversion to AD 5.6%(HR: 1.056,95% CI: 1.027 ~ 1.086);each increase of RAVLT immediate score reduced the risk of conversion to AD by 2.9%(HR: 0.971,95% CI: 0.948 ~ 0.994);Each additional unit of volume reduced the risk of conversion to AD by 17.3%(HR: 0.827,95% CI: 0.679,0.981).The model’s goodness of fit evaluation results showed that the landmark model with the time smoothing term and the time cubic term function had better goodness of fit.The ipl * model had better discriminative ability.C-index were 0.894,0.807,and 0.768 for ipl * model,ipl model,and simple model,respectively.The Brier score range of the three models increased with time and its value close to 0.06,0.08,and 0.2 for ipl * model,ipl model,and simple model,respectively.The ipl * model had better prediction performance.ipl * model was used to provide individual prediction probabilities.The example patient was 78 years old,male,18 years of education,and carried the Apoe4 genotype.The prediction results showed that the patient’s probability of conversion from MCI to AD would be 31.9% after one year of follow-up,increasing to 42.0% after 3 years,and 87.3% after 5 years.Conclusion:In a longitudinal study of AD progression,cognitive longitudinal measurement indicators ADAS13,RAVLT immediate,FAQ,and imaging indicators of hippocampal volume were significant predictors for AD conversion.The increase of ADAS13 or FAQ scores and the decrease of RAVLT immediate scores or hippocampal volume will increase the risk of AD.Using landmark model to analyze the longitudinal measures and survival data in the AD progression can not only provide a reference and basis for the prevention guidance of the AD control window,but also provide methodological references for the other chronic disease researchers.