Article type
Year
Abstract
Introduction/Objective: To compare the use of the original metric versus standardized difference effect sizes for calculating primary outcomes from studies on the effects of aerobic exercise on resting systolic blood pressure in adult women.
Method: Studies were retrieved from computer searches (MEDLINE, Current Contents) as well as from cross-referencing the bibliographies of retrieved studies and review articles. Inclusion criteria for studies were as follows: (1) randomized controlled trials, (2) aerobic exercise as the only intervention, (3) changes in resting systolic blood pressure assessed in adult women ages 18 and older, and (5) published in English-language journals between January 1992 and January 1998. Since all trials were parallel trials, the original metric effect size (net changes in resting systolic blood pressure, measured in mmHg) was calculated as the difference (exercise minus control) between changes (initial minus final) in these mean values. The corresponding standardized difference was calculated by subtracting the difference (exercise minus control) between changes (initial minus final) in these mean values divided by the pooled standard deviation of exercise and control group standard deviations. All standardized differences were corrected for small sample bias according to procedures developed by Hedges and Olkin (1985). For both these approaches, pooled differences were calculated by assigning parametric weights equal to the inverse of the total variance. Heterogeneity (Q) was examined using procedures developed by Hedges and Olkin (1985).
Results: Twelve primary outcomes met the criteria for inclusion. Using a fixed-effects model, decreases in resting systolic blood pressure (8 +- SD) averaged -2.50 +- 3.66 mmHg (95% confidence interval = -4.59 to -0.39). From a clinical standpoint, this is considered a "small" effect The standardized difference approach yielded an average effect (d) of 8 +- SD= -0.20 +- 0.22 (95% confidence interval = -0.34 to -0.06). Using Cohen's (1988) categories for classifying average effect sizes (<0.41 = small, 0.43 to 0.70 = moderate, >0.70 = large), this is also considered a "small" effect. No significant heterogeneity was observed for original metric or standardized difference results (original metric, pooled = 3.83, p = 0.97; standardized difference, Q pooled = 5.09, p = 0.93).
Discussion: The use of either the original metric or the standardized difference will yield similar results. However, if all studies included in a meta-analysis report the same metric, then it is recommended that the original metric be preserved when calculating effect sizes. For clinicians, use of original metric effect sizes will be more clinically meaningful and will enhance interpretation of results for a wider range of readers
Method: Studies were retrieved from computer searches (MEDLINE, Current Contents) as well as from cross-referencing the bibliographies of retrieved studies and review articles. Inclusion criteria for studies were as follows: (1) randomized controlled trials, (2) aerobic exercise as the only intervention, (3) changes in resting systolic blood pressure assessed in adult women ages 18 and older, and (5) published in English-language journals between January 1992 and January 1998. Since all trials were parallel trials, the original metric effect size (net changes in resting systolic blood pressure, measured in mmHg) was calculated as the difference (exercise minus control) between changes (initial minus final) in these mean values. The corresponding standardized difference was calculated by subtracting the difference (exercise minus control) between changes (initial minus final) in these mean values divided by the pooled standard deviation of exercise and control group standard deviations. All standardized differences were corrected for small sample bias according to procedures developed by Hedges and Olkin (1985). For both these approaches, pooled differences were calculated by assigning parametric weights equal to the inverse of the total variance. Heterogeneity (Q) was examined using procedures developed by Hedges and Olkin (1985).
Results: Twelve primary outcomes met the criteria for inclusion. Using a fixed-effects model, decreases in resting systolic blood pressure (8 +- SD) averaged -2.50 +- 3.66 mmHg (95% confidence interval = -4.59 to -0.39). From a clinical standpoint, this is considered a "small" effect The standardized difference approach yielded an average effect (d) of 8 +- SD= -0.20 +- 0.22 (95% confidence interval = -0.34 to -0.06). Using Cohen's (1988) categories for classifying average effect sizes (<0.41 = small, 0.43 to 0.70 = moderate, >0.70 = large), this is also considered a "small" effect. No significant heterogeneity was observed for original metric or standardized difference results (original metric, pooled = 3.83, p = 0.97; standardized difference, Q pooled = 5.09, p = 0.93).
Discussion: The use of either the original metric or the standardized difference will yield similar results. However, if all studies included in a meta-analysis report the same metric, then it is recommended that the original metric be preserved when calculating effect sizes. For clinicians, use of original metric effect sizes will be more clinically meaningful and will enhance interpretation of results for a wider range of readers