JMIR Publications

ABSTRACT

Background:

Apolipoprotein E ε4 (APOE-ε4), the strongest genetic risk factor for late-onset Alzheimer’s disease (AD), is associated with early neuromotor vulnerability that may precede measurable cognitive decline. Because speech integrates fine neuromotor processes, acoustic analysis could offer a sensitive, noninvasive marker of preclinical effects.

Objective:

To determine if speech acoustics distinguish cognitively normal APOE-ε4 carriers from non-carriers, and to assess which speech tasks provide optimal classification performance.

Methods:

In this cross-sectional observational study, 80 cognitively normal adults aged 41–89 years (22 APOE-ε4 carriers and 58 noncarriers) completed sustained phonation, oral diadochokinetic (DDK) syllable-repetition, passage-reading, and spontaneous-speech tasks. Speech samples were collected remotely using participants’ personal devices and processed to extract 88 acoustic features from the extended Geneva Minimalistic Acoustic Parameter Set (eGeMAPS). Task-specific Random Forest classifiers were developed to distinguish APOE-ε4 carriers from noncarriers. A genetic algorithm (GA) was used to select informative features, and model performance was evaluated using leave-one-participant-out cross-validation (LOPOCV). Performance metrics included balanced accuracy, accuracy, precision, sensitivity, specificity, F1 score, and area under the receiver operating characteristic curve.

Results:

Spontaneous speech produced the strongest classification performance, with balanced accuracy of 0.84, accuracy of 0.89, sensitivity of 0.73, specificity of 0.95, F1 score of 0.78, and ROC-AUC of 0.75. Performance was lower for sustained phonation (F1=0.69), DDK syllable repetition (F1=0.70 for /ba/ and 0.64 for /pa/), and passage reading (F1=0.61). Combining recordings across tasks reduced performance (F1=0.53), indicating that task-specific acoustic patterns were more informative than pooled data.

Conclusions:

Automated analysis of task-specific acoustic speech features, particularly spontaneous speech, internally distinguished cognitively normal APOE-ε4 carriers from noncarriers. These findings support further validation of speech acoustics as a low-burden digital biomarker of preclinical AD risk and suggest that spontaneous-speech tasks may be especially well suited for future remote and longitudinal assessment.