Quality Assessment
Quality assessment was conducted independently using the AMSTAR 2 tool
(11). Studies were categorised as having high, moderate, low and
critically low confidence in the results based on the number of
‘critical domains’. Critical domains related to each review containing:
an explicit statement that the methods were established a priori within
a protocol; if a satisfactory technique for assessing the risk of bias
was conducted and sufficiently discussed; if the meta-analysis used
appropriate methods; and if publication bias (small study bias) was
conducted.
Data analysis and
synthesis
The random-effects meta-analysis model was used to statistically combine
the measure of effects for those outcomes that were reported by more
than one study, stratified by the three level of exposure (ACEIs/ARBs,
ACEIs, ARBs). We conducted several sub-group analyses based on numerous
variables including: whether the reported measure of effects was crude
or adjusted, the study was peer-reviewed or not, and the study’s
methodological quality as per the quality assessment. Furthermore, to
assess the impact of ACEIs/ARBs among patients with hypertension (the
most common indication for ACEIs/ARBs), we also conducted sub-group
analysis based on whether the studies had included either patients with
hypertension only or at least had hypertension as one of the
comorbidities versus those studies which did not recorded the
hypertension status of their study population. The combined pooled
estimates were presented as odds ratios and 95%CI and graphically as
forest plots. I2 statistic (12) was used to assess
heterogeneity between the studies with I2 of 0%
indicating lack of heterogeneity, whereas 25%, 50%, and 75%
indicating low, moderate and high heterogeneity, respectively (12).
Publication bias was assessed using funnel plots and Egger’s asymmetry
test (13) only for those outcomes where >10 studies were
included in the analysis as recommended by Cochrane guidelines (14).
Furthermore, we evaluated the influence of individual reviews on the
summary pooled estimate for each outcome by conducting influential
analyses (15) whereby the pooled meta-analysis estimates for each
outcome were computed by omitting one study at a time. Data were
analysed using STATA 12.
Role of the Funding
Source
None