Thirteen trials were included (896 participants, range 16 to 191). Six trials had adequate methods of randomisation and three of these also provided methods of allocation concealment. Eight trials reported losses to follow-up and six used intention-to-treat analysis. Five trials reported sample size calculations for their primary outcome. Ten trials reported attrition rates (range from 4% at 12 weeks to 32% at two years).
Timed get up and go tests (five trials): There was a reduction in timed get up and go with the intervention (WMD -1.39 seconds, 95% CI -2.59 to -0.19; three trials) but heterogeneity was very high (Ι²=96%). When the trial with outlying result was removed, this result was no longer statistically significant.
Timed walking tests (six trials): Three trials showed a significant increase in walking speed with the intervention, one an increase in walking distance and one a small but non-significant decrease in walking distance. The results of two trials that used a six-minute walk test were pooled but no significant difference was found and heterogeneity was high (Ι²=85%). Walking speed was pooled for four trials and showed a statistically significant improvement in speed (WMD 0.06 metres per second, 95% CI 0.01 to 0.1; Ι²=0%).
Balance and functional reach: Seven trials measured balance and one trial measured functional reach. Two trials that used the Berg balance scale were pooled and found a significant improvement for the intervention (WMD 3.4 points, 95% CI 1.08 to 5.72; Ι²=0%).
Flexibility and lower limb strength: Two out of three trials that reported flexibility found significant improvements with the intervention. Three out of four trials that reported lower limb strength found a significant improvement with the intervention.
Depression and quality of life: Four trials reported depression outcomes, all using a different measure. Three trials did not find any significant differences. One trial found a significant improvement for a subgroup of patients (58% of the total sample) at 24 months using the Cornell scale for depression in dementia. Two trials measured quality of life using a variety of questionnaires. One trial found significant differences in four out of seven measures with higher scores for comprehensive exercise compared to social conversation. The other trial found improvements in physical role function at three months.