The impact of AD drug treatments on event-related potentials as markers of disease conversion

Abstract

This paper investigates how commonly prescribed pharmacologic treatments for Alzheimer's disease (AD) affect Event-Related Potential (ERP) biomarkers as tools for predicting AD conversion in individuals with Mild Cognitive Impairment (MCI). We gathered baseline ERP data from two MCI groups (those taking AD medications and those not) and later determined which subjects developed AD (Convert->AD) and which subjects remained cognitively stable (Stable). We utilized a previously developed and validated multivariate system of ERP components to measure medication effects among these four subgroups. Discriminant analysis produced classification scores for each individual as a measure of similarity to each clinical group (Convert->AD, Stable), and we found a large significant main Group effect but no main AD Medications effect and no Group by Medications interaction. This suggested AD medications have negligible influence on this set of ERP components as weighted markers of disease progression. These results provide practical information to those using ERP measures as a biomarker to identify and track AD in individuals in a clinical or research setting.

Figure 1

Linear discriminant function. Component scores are generated by multiplying the vector of component scoring coefficients (measured with PCA) with the subject’s vector of ERP time points for task conditions (in this case, Number-Letter stimuli relevant or irrelevant to the task). It should be noted that the component waveforms depicted above are a vector of component loadings (shown for simplicity’s sake and mathematically similar to the scoring coefficients) with the metric restored by multiplying the loading at each time point by the standard deviation of the dataset at that corresponding time point [2, 22]. The voltage scale of the component and ERP waveforms is identical. The ERP component_conditions are shown in the order they were selected by the stepwise discriminant procedure. The ERP component scores are then multiplied by discriminant coefficients. These coefficients were derived by subtracting the classification coefficients developed for the Stable group from those developed for the Conversion group [34, 35]. After applying discriminant coefficients, the results are summed and then added to a constant (which is also a difference between the Conversion and Stable group constants) to produce a classification score for each subject. A list of the classification coefficients for each of the groups was presented in previous work [2].

Figure 2

ERP classification difference scores. The 30 MCI subjects (15 later Convert->AD, 15 remain Stable) are ordered by the absolute value of their difference scores (decreasing from left to right). The midline represents a difference score of 0 (not similar to either group). Subjects above this line were predicted to convert to AD. Subjects below this line were predicted to remain stable. The six subjects that were misclassified [2] fall on the “wrong side” of the midline for their clinical group (see x-axis 13–15). Subjects marked by an “X” were taking AD medications at the time of baseline ERP data collection.