Several study limitations of this meta-analysis should be considered. First there was some methodological heterogeneity between the studies, including the individual objectives, study populations, risk factors for development and progression of CVD, heterogeneity on the definitions of the outcomes of interest, in particular for HF and MACE, dosages of pioglitazone, types of control, treatment, and follow-up duration. Overall, differences across the individual studies were observed for medical history and a number of cardiovascular risk factors, including BMI, HbA1c, smoking status, presence of hypertension, systolic blood pressure, LDL-cholesterol, total cholesterol and medication use, as can be seen in Additional file
1: Table S4. These differences should be taken into account when interpreting the results of this meta-analysis. For example, cardiovascular risk management between the two major trials of this meta-analysis—PROactive and IRIS—are quite different. In the IRIS trial more strict treatment of risk factors was evident with 70% of participants on statins compared to 40% in the PROactive study. Furthermore, there is difference of ~ 10 mmHg in systolic blood pressure in favor of the IRIS trial (Additional file
1: Table S4). Nevertheless, both trials indicate a lower risk for recurrent cardiovascular events by pioglitazone [
2,
10]. In addition, heterogeneity was observed for duration of follow-up and treatment. Most of the studies with a follow-up duration shorter than 12 months are however small studies with a subsequent low number of events [
29,
31,
33,
34]. Consequently, the weight of these studies in the meta-analyses is quite small and therefore they do not have a large effect on the overall results of the meta-analyses, as can been in Additional file
1: Figures S2–S6, after stratifying the included studies on follow-up of 12 months or longer. Moreover, the study from Suryadevara et al. from 2012 was somewhat different compared to the other included studies. Although the participants from this study differ on some aspects from the other included studies, for example 100% of the participants are hypertensive and all of them took statins, it is not likely that these differences affect the results of the meta-analyses and the interaction between pioglitazone and the outcomes of interest, since the weight of this study on the meta-analyses is quite small [
31]. Also, a number of studies included participants with recent percutaneous coronary intervention (PCI) [
29,
30,
32‐
34]. Since the studies with participants that underwent recent PCI are limited in number of participants and events, separate analyses on CVD outcomes in participants that underwent recent PCI were not performed. Although it would be interesting for future research, to study whether patients on pioglitazone with recent PCI have different CVD outcomes compared to patients on pioglitazone with coronary artery disease (CAD) without recent PCI. In some of the included studies treatment duration versus follow-up duration was not reported [
8,
29,
34]. We assumed that treatment duration was equal to follow-up duration when studies only mentioned follow-up duration. Furthermore, publication bias could not entirely be excluded, since less than ten studies were included in the individual analysis. The methodological quality for the included studies was variable and the risk of bias among the included studies should be taken into account when interpreting the results of this meta-analysis.