Baseline predictors of response to a group-based speech and communication intervention in Parkinson's disease: A secondary analysis of a randomized controlled trial

BackgroundSpeech and voice symptoms are common in Parkinson's disease, yet predictors of response to behavioral speech interventions are unclear.ObjectivesTo identify predictors of responsiveness to a speech and communication group-intervention (HiCommunication) and contextualize findings against active controls.MethodsThis secondary analysis of a randomized controlled trial included intervention completers. Responders were defined for voice intensity (increase ≥2 decibels) and voice quality (Acoustic Voice Quality Index decrease ≥0.54). Nineteen baseline clinical, motor, cognitive, perceptual, and acoustic variables were entered into Random Forest classifiers. Primary models excluded the baseline value of the target domain; baseline-including variants were sensitivity analyses. Performance was evaluated using Cohen's kappa, precision, recall, and specificity; key predictors were examined using partial dependence.ResultsIn HiCommunication (n = 35), the primary voice intensity model showed moderate agreement (kappa = 0.57; precision = 0.79; recall = 0.90; specificity = 0.64). Higher baseline perceptual ratings of reduced loudness and overall speech deviation, as well as higher Acoustic Voice Quality Index values, were associated with a higher predicted probability of improvement in voice intensity, whereas higher levodopa equivalent daily dose was associated with a lower response probability; postural instability/gait difficulty or tremor-dominant motor phenotypes showed higher response probability than the indeterminate phenotype. Voice quality models were below chance without baseline Acoustic Voice Quality Index but reached kappa 0.38 when included. Active control models showed low performance.ConclusionsClinically accessible baseline measures predicted improvement in voice intensity following HiCommunication with moderate accuracy. Perceptual ratings may support expectation-setting and individualized planning of group-based speech intervention in Parkinson's disease, but findings require replication.ClinicalTrials.gov ID: NCT03213873, doi: 10.1177/1545968321999053.