Table 3

Results of meta-regression exploring the impact of trial-level characteristics on the effect of exercise on falls in general community-dwelling older populations

Variable tested in meta-regression analyses (number of trials with this characteristic for dichotomous variables)Coefficient (95% CI), p, % heterogeneity explained
Study design
 PEDro score*, number/101.01 (0.93 to 1.08), 0.88, −5%
 Participants randomised, number1.00 (1.00 to 1.00), 0.37, −2%
 Weeks of follow-up, number1.00 (1.00 to 1.00), 0.07, 18%
Sample characteristic
 Average age >75 years (33)0.95 (0.80 to 1.12), 0.52, −3%
 Control group fall rate, available for 65 comparisons0.99 (0.90 to 1.07), 0.73, −9%
 Unselected population (37)1.02 (0.87 to 1.21), 0.79, −7%
Programme characteristic
 Inclusion of moderate or high-intensity strength training† (28)0.97 (0.82 to 1.15), 0.73, −4%
 Inclusion of high-intensity strength training (10)1.23 (0.96 to 1.57), 0.11, 9%
 Inclusion of moderate or high challenge balance training‡ (47)0.85 (0.71 to 1.00), 0.06, 19%
 Inclusion of high challenge balance training (31)0.85 (0.73 to 1.00), 0.04, 28%
 Inclusion of walking training or practice (29)1.01 (0.86 to 1.20), 0.87, −6%
 2+ hours per week of exercise (40)0.98 (0.83 to 1.16), 0.83, −6%
 3+ hours per week of exercise (20)0.77 (0.65 to 0.91), 0.003, 61%
 Better adherence to exercise§ (52)0.95 (0.79 to 1.13), 0.54, −1%
  • *Coded using PEDro rating scale.14

  • †Moderate intensity (40–60% of the 1-repetition maximum (RM) that is, a weight so heavy that it can only be lifted once) or high intensity (>60% 1RM).

  • ‡Moderately challenging=two of the following criteria or highly challenging=all three criteria: movement of the centre of mass, narrowing of the base of support and minimising upper limb support.

  • §Greater than or equal to 75% participants attended 50% or more sessions and/or >50% attendance rate.

  • Note: a meta-regression coefficient <1 indicates a greater impact of exercise on falls in trials with that characteristic; a negative number for percentage of heterogeneity explained reflects no heterogeneity explained; number of trials with a particular characteristic indicated for dichotomous meta-regression variables only; statistically significant comparisons shown in italics.