The standardised mean differences were calculated check details by dividing the raw result by the standard deviation of the post-test score for the 14 trials for which this value was available. For the remaining three trials, the post-test standard deviation was estimated from the standard deviation of the change between initial and final assessment scores assuming a 0.6 correlation between pre and post scores. Meta-regression was also undertaken to assess whether there was a bigger effect on strength outcomes

of programs that specifically challenged strength. Meta-regression was not possible for any other outcomes due to the relatively small number of trials for those outcomes (ie, six) (Sterne et al 2001). The search strategy

identified 2198 studies (excluding duplicates). After screening, 23 eligible randomised trials were included in this review (Asikainen et al 2006, Bemben et al 2000, Bergstrom Metformin order et al 2007, Bravo et al 1996, de Jong et al 2006, Fu et al 2009, Garcia-Lopez et al 2007, Heinonen et al 1998, Janzen et al 2006, King et al 1991, Klentrou et al 2007, Levinger et al 2007, Lindheim et al 1994, Maiorana et al 2001, Mitchell et al 1998, Pereira et al 1998, Sallinen et al 2007, Shirazi et al 2007, Sillanpaa et al 2009, Singh et al 2009, Stefanick et al 1998, Teoman et al 2004, Uusi-Rasi et al 2003). Figure 1 presents the flow of studies through the review. The 23 included trials

involved a total of 2550 participants. Table 1 summarises the features of the included trials. Table 2 presents the characteristics of participants, interventions, and adherence to the intervention. Quality: Three trials performed concealed allocation ( de Jong et al 2006, Fu et al 2009, King et al 1991) and two trials used blinded assessment of outcomes ( Fu et al 2009, only Uusi-Rasi et al 2003). This information is also presented in Table 2. Participants: The majority of trials recruited postmenopausal women. The mean age of participants in the included studies ranged from 41 to 60 years of age. Intervention: Most trials included a strength component, followed by a combination of the strength and endurance components. Three trials included a combination of all three physical activity components (ie, strength, balance, and endurance). Included trials were heterogeneous regarding the total prescribed physical activity hours and adherence. Outcome measures: Lower limb strength was measured in 13 trials, endurance was measured in 7 trials, and balance in 6 trials. No studies reported effects of physical activity on falls soon after receiving the intervention program. One study reported longer-term (15 year) effects of physical activity on falls. We were able to pool data from 17 of the included trials in the meta-analyses. The data used in the meta-analyses are shown in Table 3.