Sturm 1991

Methods	RCT, cross‐over group design Participants with left brain damage were allocated to 2 comparable groups with respect to age, sex and time post‐onset. Allocation of people to treatment group by random number table. There was a third training group with right brain damaged people. They were trained 'late' in the cross‐over design and their data were excluded from the current analysis Approach: restore attentional functions by computer training
Participants	Germany Total participants sample: 37; 0 people lost to follow‐up Treatment group: n = 13; mean age 51.5 ± 9.5 years; 69.2% males; 15 ± 9.7 weeks since onset; hemisphere of lesion: 13 left, 1 non‐stroke participant Control group: n = 14; mean age 49.6 ± 9.5 years; 71.4% males; 16.4 ± 9.2 weeks since onset; hemisphere of lesion: 14 left, 2 non‐stroke participants No inclusion or exclusion criteria specified, all participants had attentional deficits according to authors
Interventions	Treatment: 30 minutes' computer‐based training at clinic, 14 sessions spread over 3 consecutive weeks. Cognitrone and Wiener Determinationsgerat were used as training software Control: no treatment
Outcomes	Measured after intervention at 3 weeks, after cross‐over at 6 weeks and at 12 weeks' follow‐up Several measures, but no definition of primary outcome measure: Wiener Determinationsgerat (hits and false alarms, z‐scores) Cognitrone (hits and false alarms, z‐scores) Wiener Reaktionsgerat (reaction times for visual, auditory and choice subtests, z‐scores) Wiener Vigilanzgerat (hits and false alarms, z‐scores) d2 (hits minus errors, z‐scores) Non‐attentional tasks, such as Raven Standard Progressive Matrices and WAIS subtests
Notes	Cognitrone and Wiener Determinationsgerat were used for training, outcome measures based on these tasks were excluded for analysis Control group was more aphasic and scored significantly lower on WAIS, in the Intelligence structure task and the Ravens than the intervention group
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Use of random number table (personal communication of authors)
Allocation concealment (selection bias)	Unclear risk	No information
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	No information
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	No information
Incomplete outcome data (attrition bias) All outcomes	Low risk	No missing outcome data, but it is noteworthy that 90% of people with right brain damage initially agreed to participate in the study refused to participate when contacted later for starting the training
Selective reporting (reporting bias)	Low risk	No indication in article, but study protocol not available