We included three trials (four interventions) involving 196 participants with stroke and co-morbid anxiety. One trial (described as a ‘pilot study‘) randomised 21 community-dwelling stroke survivors to four-week use of a relaxation CD or to wait list control. This trial assessed anxiety using the Hospital Anxiety and Depression Scale and reported a reduction in anxiety at three months among participants who had used the relaxation CD (mean (standard deviation (SD) 6.9 (± 4.9) and 11.0 (± 3.9)), Cohen’s d = 0.926, P value = 0.001; 19 participants analysed).
The second trial randomised 81 participants with co-morbid anxiety and depression to paroxetine, paroxetine plus psychotherapy, or standard care. Mean levels of anxiety severity scores based on the Hamilton Anxiety Scale (HAM-A) at follow-up were 5.4 (SD ± 1.7), 3.8 (SD ± 1.8), and 12.8 (SD ± 1.9), respectively (P value < 0.01).
The third trial randomised 94 stroke patients, also with co-morbid anxiety and depression, to receive buspirone hydrochloride or standard care. At follow-up, the mean levels of anxiety based on the HAM-A were 6.5 (SD ± 3.1) and 12.6 (SD ± 3.4) in the two groups, respectively, which represents a significant difference (P value < 0.01). Half of the participants receiving paroxetine experienced adverse events that included nausea, vomiting, or dizziness; however, only 14% of those receiving buspirone experienced nausea or palpitations. Trial authors provided no information about the duration of symptoms associated with adverse events. The trial of relaxation therapy reported no adverse events.
The quality of the evidence was very low. Each study included a small number of participants, particularly the study of relaxation therapy. Studies of pharmacological agents presented details too limited to allow judgement of selection, performance, and detection bias and lack of placebo treatment in control groups. Although the study of relaxation therapy had allocated participants to treatment using an adequate method of randomisation, study recruitment methods might have introduced bias, and drop-outs in the intervention group may have influenced results.