Predicting conversion to Alzheimer’s disease in individuals with Mild Cognitive Impairment using clinically transferable features

Abstract

Patients with Mild Cognitive Impairment (MCI) have an increased risk of Alzheimer’s disease (AD). Early identification of underlying neurodegenerative processes is essential to provide treatment before the disease is well established in the brain. Here we used longitudinal data from the ADNI database to investigate prediction of a trajectory towards AD in a group of patients defined as MCI at a baseline examination. One group remained stable over time (sMCI, n = 357) and one converted to AD (cAD, n = 321). By running two independent classification methods within a machine learning framework, with cognitive function, hippocampal volume and genetic APOE status as features, we obtained a cross-validation classification accuracy of about 70%. This level of accuracy was confirmed across different classification methods and validation procedures. Moreover, the sets of misclassified subjects had a large overlap between the two models. Impaired memory function was consistently found to be one of the core symptoms of MCI patients on a trajectory towards AD. The prediction above chance level shown in the present study should inspire further work to develop tools that can aid clinicians in making prognostic decisions.

Personalized prognosis & treatment using an optimal predictor machine: An example study on conversion from Mild Cognitive Impairment to Alzheimer's Disease

“…• RAVLT-imm, RAVLT-del, and to a lesser degree RAVLT-rec are good predictors, both when used alone and when used jointly with all other predictors. Hippocampal Volume is a poorer predictor than any of the RAVLT when used alone, and likely also when used in combination with all others (contrast this with Rye et al, 2022). This last finding is also clear in Curtis's case: fig.…”

“…• The set of 12 predictors considered in the present work and in Rye et al (2022) can at most yield a prognostic accuracy of around 67.7% ± 0.7% over the full population, for any inference algorithm. This fact agrees with the (completely independent) findings in Rye et al (2022), where a maximal accuracy of 68.3% on a test dataset was found using an ensemble model. The present analysis also shows that the ensemble model managed to achieve the maximal accuracy possible with these predictors (but see § 4 for limitations of that model).…”

“…This longitudinal diagnostic label is used as the predictand variate cAD in our study; it categorizes each subject as either converting to Alzheimer The 12 predictor variates consist of the results from seven cognitive-test measures: a reading test (ANART), a word category fluency test (CFT), trail-making tests of executive function (TMTA, TMTB), the immediate-memory, delayed-recall and recognition-subsets of memory function (RAVLT-imm, RAVLT-del, RAVLT-rec); a geriatric depression scale (GDS); the presence of the APOE-e4 risk allele (Liu et al, 2013); a normalized measure of the sum of left and right hippocampal volume (HV); Age; Sex. Further details about these variates and their selection can be found in Rye et al (2022). The cognitive and GDS variates are integer-valued, hippocampal volume and Age are continuous, and APOE4 and Sex are binary.…”

“…The Ledley-Jaynes machine shows that the omission of any one of the 12 predictors, except RAVLT-del and possibly RAVLT-imm, does not lead to an appreciable decrease in accuracy (relative decrease of 0.3% or less) or in mutual information (relative decrease of less than 3%). This puts the prognostic-importance analysis of Rye et al (2022) into perspective. The exact quantification of these subtle differences is computationally quite expensive, and we did not carry it out further.…”

“…We showed its application in an example of prognosis and treatment of conversion from Mild Cognitive Impairment to Alzheimer's Disease for four different patients, where all three categories of differences listed above appeared. The patients were fictitious but the underlying learning database, originating from ADNI, is real, and was explored in a previous work (Rye et al, 2022).…”