Ten studies were included (n=2,981 participants and 2,549 included in analyses; this represented a 14.5% drop-out rate). Five studies evaluated multifactorial interventions and the other five evaluated exercise-only interventions.
The funnel plot showed a degree of asymmetry, which suggested publication bias. The authors stated that the Fail-safe N value was high, which suggested that too many unpublished non significant studies were required to make the findings non significant (Fail-safe N value not reported).
All interventions combined were associated with a statistically significant reduction in the number of falls compare to controls (RR 0.67, 95% CI 0.52 to 0.85). Significant heterogeneity was found (p<0.001).
Multifactorial interventions were associated with a statistically significant reduction in the number of falls compare to controls (RR 0.90, 95% CI 0.82 to 1.00) and were statistically homogeneous. Exercise-only interventions were also associated with a statistically significant reduction in the number of falls compare to controls (RR 0.45, 95% CI 0.29 to 0.71). Significant heterogeneity was found (p<0.001). Data were homogeneous after removal of one study. Exercise–only interventions were significantly more effective at reducing falls than multifactorial interventions.
Moderator analysis showed that interventions that lasted less than 12 months were effective (p<0.001), but those that lasted more than 12 months did not appear to be effective (p=0.10). All five exercise-only interventions plus one multifactorial intervention lasted less than 12 months. All interventions that lasted more than 12 months were multifactorial.
Meta-regression showed that smaller studies had larger effect sizes and interventions that targeted older people (80 years or over) were less effective than interventions that targeted younger people (around 70 years). Sample size accounted for 22% of the variability among studies (p<0.001).