I would like to meta-analyse raw mean differences (MDs) in systolic blood pressures (BPs) according to soda intake (i.e. drinkers vs. non-drinkers). The problem is that there are important differences between drinkers and non-drinkers in each study (e.g. age, bmi, gender) which blur the association between soda intake and BP. I would like to account for these differences when calculating a pooled MD in BP between drinkers and non-drinkers.
Below is the analysis (using metafor
):
### set-up data
study=c(1,2,3,4,5,6)
md_BP=c(3,4,4.5,6,7,10) ### raw mean differences in BP
se_md_BP=c(0.81,0.88,1.38,2.57,1.39,1.95) ### standard error of the raw mean differences in BP
md_age=c(-4.2,-3,2,0,2,3) ### raw mean differences in age
md_bmi=c(-4,-1.5,0,2,3,4) ### raw mean differences in bmi
d_gender=c(29.9,0,0,0,0,8) ### differences in % males (i.e. drinkers%-non-drinkers%)
df=data.frame(study,se_md_BP,md_BP,md_age,md_bmi,d_gender)
model1<-rma(yi=md_BP,sei=se_md_BP,data = df,mods = ~ d_gender+md_age+md_bmi)
summary(model1)
Results:
Mixed-Effects Model (k = 6; tau^2 estimator: REML)
logLik deviance AIC BIC AICc
-3.3755 6.7511 16.7511 10.2168 76.7511
tau^2 (estimated amount of residual heterogeneity): 0 (SE = 2.9702)
tau (square root of estimated tau^2 value): 0
I^2 (residual heterogeneity / unaccounted variability): 0.00%
H^2 (unaccounted variability / sampling variability): 1.00
R^2 (amount of heterogeneity accounted for): 100.00%
Test for Residual Heterogeneity:
QE(df = 2) = 0.6828, p-val = 0.7108
Test of Moderators (coefficient(s) 2,3,4):
QM(df = 3) = 14.5999, p-val = 0.0022
Model Results:
estimate se zval pval ci.lb ci.ub
intrcpt 4.8431 0.7021 6.8977 <.0001 3.4670 6.2193 ***
md_gender 0.0532 0.0505 1.0535 0.2921 -0.0458 0.1521
md_age -0.1818 0.4471 -0.4066 0.6843 -1.0580 0.6944
md_bmi 1.0360 0.5219 1.9853 0.0471 0.0132 2.0588 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Is the analysis above a methodologically sound approach (e.g. with mean differences [for age and bmi] and % difference [for gender] as moderators) to solving the stated problem?
Also, are the following statements regarding the results correct?
After accounting (correct word?) for differences in age, gender and bmi between the two groups, soda drinkers had significantly higher BP than non-drinkers (MD=4.84; 95% CI 3.47-6.22). (Note that these confidence intervals are more certain than when the moderators are not included in the analysis - I would be grateful for reassurance that this is fine.)
The test for residual heterogeneity was not significant QE(df = 2) = 0.6828, p-val = 0.7108. I2 was 0.00%.
There combined influence of the moderators (age, gender and bmi) on BP was statistically significant, according to the omnibus test (p-val = 0.0022).
Thank you!