In this study, we examine the impact of domain adaptation and question-answer pooling on text-based aphasia prediction with standard and medically specialised BERT models for a German corpus. Modelling tasks comprise aphasia type classification as well as multitask regression of communicative, semantic, and syntactic skills. We found that domain adaptation before finetuning as well as question-answer pooling increased performance for the standard but not for the specialised models on all classification and regression tasks.