Diagnostic and prognostic multimodal prediction models in Alzheimer's disease: A scoping review

Background Multimodal prediction models for Alzheimer's disease (AD) are emerging as promising tools for improving detection and informing prognosis. Objective To summarize the predictive objectives, constituting predictors and algorithms, and performance of existing multimodal prediction models. Methods We performed a systematic literature search in Medline, Embase, and Web of Science up to January 15, 2024, to identify prediction models covering the full spectrum of AD, from the preclinical stage to subjective cognitive decline (SCD), mild cognitive impairment (MCI), and AD dementia. The predictors, algorithms, and model performance of prediction models were summarized narratively by their predictive objectives. The review protocol was registered with the Open Science Framework (osf.io/zkw6g). Results Predicting the future progression from MCI to AD dementia was the most common objective of prediction models for AD. The second most common objective was to classify AD stages (SCD versus MCI versus AD dementia), followed by detecting the presence of amyloid, tau, or neurodegeneration. More than half of the prediction models reported an area under the receiver operating characteristic curve exceeding 0.8 and an accuracy exceeding 70%. However, 66.7% of the prediction models were developed using data from the ADNI study, and only 10.1% of the models went through external validation. Conclusions Existing multimodal prediction models have mainly focused on the prediction of current or future AD stages and reported good performance. However, these models need to be validated using data other than the data used for model training before being considered for practical applications.