NeuropsychBrainAge: a biomarker for conversion from mild cognitive impairment to Alzheimer’s disease

Background BrainAge models based on neuroimaging data have shown good accuracy for diagnostic classification. However, they have replicability issues due to site and patient variability intrinsic to neuroimaging techniques. We aimed to develop a BrainAge model trained on neuropsychological tests to identify a biomarker to distinguish stable mild cognitive impairment (sMCI) from progressive mild cognitive impairment (pMCI) to Alzheimer’s disease (AD). Methods Using a linear regressor, a BrainAge model was trained on healthy controls (CN) based on neuropsychological tests. The model was applied to sMCI and pMCI subjects to obtain predicted ages. The BrainAge delta, the predicted age minus the chronological age, was used as a biomarker to distinguish between sMCI and pMCI. We compared the model to one trained on neuroimaging features. Findings The AUC of the ROC curve for differentiating sMCI from pMCI was 0.91. It greatly outperforms the model trained on neuroimaging features which only obtains an AUC of 0.681. The AUC achieved is at par with the State-of-the-Art BrainAge models that use Deep Learning. The BrainAge delta was correlated with the time to conversion, the time taken for a pMCI subject to convert to AD. Interpretation We suggest that the BrainAge delta trained only with neuropsychological tests is a good biomarker to distinguish between sMCI and pMCI. This opens up the possibility to study other neurological and psychiatric disorders using this technique but with different neuropsychological tests. Funding A full list of funding bodies that supported this study can be found in the Acknowledgments section. Research in Context Evidence before this study A major application of recent neuroimaging BrainAge models has been demonstrating its value in diagnostic classification. In spite of the good performance, most models based on neuroimaging data have limitations in real data as the distribution between sites can be different from training cohorts. They can also suffer from lack of specificity to a disease, for those based on BrainAge deltas trained on healthy controls or insufficient training data, for those trained to directly identify a specific disease. We develop a BrainAge model trained on neuropsychological tests used in Alzheimer’s disease research to identify a biomarker to distinguish sMCI from pMCI subjects. We propose a model that is trained on healthy controls for which there is more data to then reliably distinguish sMCI from pMCI subjects. Added value of this study This is the first study to use a BrainAge model based on neuropsychological test features to study Alzheimer’s disease. We suggest the NeuropsychBrainAge delta, which measure the difference between the model predicted age of the subject trained on healthy controls and the chronological age of the subject, as a biomarker of Alzheimer’s Disease. The NeuropsychBrainAge delta could differentiate between sMCI and pMCI. Moreover, we also show that the proposed biomarker is correlated with the time to conversion, the time taken for a pMCI subject to convert to Alzheimer’s Disease. Implications of all the available evidence Our approach could be used for the identification of patients with mild cognitive impairment at risk of developing Alzheimer’s disease. The NeuropsychBrainAge delta can also be used as a quantitative marker to measure disease severity due to its correlation with time to conversion. This study shows that using healthy controls for which there is more data but using features specific to a disease such as neuropsychological test can lead to reliable BrainAge models to identify specific neurological and psychiatric disorders.