Sensitivity and subgroup analyses
To explore the possible impact of risk of bias on the exercise effect, we removed from the analyses studies for which the risk of bias was unclear for at least three of the quality components considered,19 27 31 34 or for which the risk was at least unclear for one component and high for another.22 24 30 32. This barely changed the pooled estimate of the effect of exercise in any of the four injurious fall categories, but it greatly reduced the heterogeneity between studies included in the analysis of all injurious falls (rate ratio 0.69, 95% confidence interval 0.58 to 0.82, I2=26%, P=0.22, eight trials).
In the subgroup analysis of the effect of exercise in reducing injurious falls based on risk of falls at enrolment, no significant difference was found in pooled estimates between trials with participants selected for their higher risk of falling13 19 20 21 24 and trials with unselected participants,25 26 28 29 30 in any of the four categories.