Anomia, or difficulty naming common objects, is the most common, acquired impairment of language. Effective therapeutic interventions for anomia typically involve massed practice at high doses. This requires significant investment from patients and therapists. Aphasia researchers have increasingly looked to neurostimulation to accelerate these treatment effects, but the evidence behind this intervention is sparse and inconsistent. Here, we hypothesised that group-level neurostimulation effects might belie a more systematic structure at the individual level. We sought to test the hypothesis by attempting to predict the immediate (online), individual-level behavioural effects of anodal and sham neurostimulation in 36 chronic patients with anomia, performing naming and size judgement tasks. Using clinical, (pre-stimulation) behavioural and MRI data, as well as Partial Least Squares regression, we attempted to predict neurostimulation effects on accuracies and reaction times of both tasks. Model performance was assessed via cross-validation. Predictive performances were compared to that of a null model, which predicted the mean neurostimulation effects for all patients. Models derived from pre-stimulation data consistently outperformed the null model when predicting neurostimulation effects on both tasks' performance. Notably, we could predict behavioural declines just as well as improvements. In conclusion, inter-patient variation in online responses to neurostimulation is, to some extent, systematic and predictable. Since declines in performance were just as predictable as improvements, the behavioural effects of neurostimulation in patients with anomia are unlikely to be driven by placebo effects. However, the online effect of the intervention appears to be as likely to interfere with task performance as to improve it.
Keywords: aphasia; neurostimulation; prediction; stroke.