Data Synthesis and Analysis
We conducted random-effects meta-analyses to estimate
the effect size of counseling on all intermediate health
outcomes and behavioral outcomes. We combined all trials
with a given outcome and conducted separate analyses for
each of the 3 intervention targets (physical activity, healthful
diet, and combined) and, if applicable, for the specific
dietary message (sodium reduction, focus on fruits and
vegetables only, or general low-fat or heart-healthy dietary
counseling). Analyses were stratified by estimated intervention
intensity (low [30 minutes], medium [between 31
minutes and 6 hours of contact], or high [6 hours of
contact]). Trials were also categorized by population risk as
being unselected or selected only on the basis of age; selected
for suboptimal behavior (such as sedentary behavior
or poor dietary intake); or selected for individual or population
risk factors for increased incidence of cardiovascular
disease (such as mildly elevated diastolic blood pressure or
fasting glucose or serum lipid levels, obesity, or poverty or
poor access to health care).
We assessed the presence of statistical heterogeneity
among the studies by using standard chi-square tests and
estimated the magnitude of heterogeneity by using the I2
statistic (23). Tests of publication bias to determine
whether the distribution of the effect sizes was symmetric
with respect to the precision measure were performed by
using funnel plots and the Egger linear regression method
(24, 25). Meta-regressions were performed on the basis of
the random-effects models to examine the effect of 4 a
priori variables of heterogeneity (intervention intensity, intervention
target, study population risk, and recruitment
method [volunteer vs. study-identified]) on effect size. To
interpret effect sizes of standardized mean differences, we
used the Cohen d statistic, in which an effect size of 0.2 to
0.3 generally represents a small effect; 0.5, a moderate effect;
and 0.8, a large effect (26).