Dynamic Individual Prediction of Conversion from Mild Cognitive Impairment to Probable Alzheimer's Disease using Joint Modeling

Background We propose a joint model predicting the risk of conversion from MCI to AD that considers the association between biomarker evolution and disease progression. Methods We selected 814 MCI subjects (285 progressives, 529 stables) who had at least four follow-up MRI visits from the ADNI dataset. The values of Alzheimer’s Disease Assessment Scale-Cognitive (ADAS-Cog) were used as a surrogate of time. A mixed linear model was fitted for bilateral hippocampal volumes (HC) versus ADAS-Cog, education, age and sex and a Cox model for risk progression. The association between HC evolution and risk conversion was estimated by fitting a joint model. Results Our results show (1) significant association ( p < .0001, C.I.= [0.0864; 0.1217]) between bilateral HC and risk of conversion; (2) on average, the risk of progression increased as HC decreased; and (3) the individual prediction of the risk is dynamic, i.e., updated at each follow-up. The AUC of our model for the whole group increased to reach 0.789 at the last follow-up. Conclusions Applicable to AD and generalizable to other biomarkers and covariates, this joint methodology has a direct application in the clinical estimation of individual risk.