In this study, we systematically searched RCTs reporting mortality and cardiovascular events in patients with T2DM randomised to a specific glucose-lowering strategy or to a specific drug, to quantify the rates of these outcomes in control arms and describe their trends. As RCTs included participants from mid-1960 to 2015, it was possible to quantify outcome trends across 5 decades. We found no important changes over the observed years in most of the relevant diabetes-related outcomes, including death from cardiovascular causes, myocardial infarction, or death from any cause; 3-P MACE, a combined cardiovascular endpoint commonly used in RCTs (particularly recent ones), showed an increasing trend which, however, was less evident when accounting for participants’ baseline characteristics across RCTs. Notable exceptions were the declining trends for the individual outcome total stroke, nonfatal stroke, and nonfatal myocardial infarction, and a possible reduction for total myocardial infarction.
Recent decades have been characterised by significant improvements in the diagnosis and treatment of cardiovascular disease risk factors. As a result, declining trends of major cardiovascular disease have been repeatedly reported in observational studies from several countries, both in the general population and in people with T2DM  (Table S9). As cardiovascular diseases represent the main cause of death in patients with T2DM, such reduction also translates in a lower mortality risk , albeit with wide variation in absolute mortality rates across different countries. The reasons behind such heterogeneity in mortality rates are likely related to clinical (including access to healthcare; screening, early detection and management of T2DM and its complications; proactive ongoing management of hyperglycaemia and other risk factors; patient education and self-management; and prevalent comorbidities), biological/genetic, and socioeconomic differences. Along with the multifaceted syndemic interplay between these elements , differences in the processes of measuring (data quality, exposure definitions and assessment, outcome ascertainment) and synthesising (study design and analysis) information could also have contributed. Such heterogeneity in mortality rates was also observed in RCTs included in our analysis; however, in this situation it is more likely attributable to clinical differences rather than study design and analysis.
Variations in rates of single and combined cardiovascular outcomes comparing observational studies and RCTs are more difficult to interpret than mortality. Differences in the definitions and ascertainment of outcomes are well recognised in observational studies (i.e., physician vs self-reported T2DM or cardiovascular outcome), particularly for fatal events, where there are spatiotemporal differences in the definition and reporting of the underlying cause of death [18, 19]. In an attempt to limit heterogeneous comparisons, efforts have been made to standardise definitions of cardiovascular outcomes and their composites in RCTs, thus making geographical and temporal comparisons more reliable. With this in mind, our results indicate a nonsignificant 30% increased risk of major adverse cardiovascular events every 5 years, accounting for differences in demographic and clinical characteristics of RCTs’ participants. These results are possibly linked to rising trends of cardiovascular mortality seen in the analysis of RCTs reporting 3-P MACE, while the contribution of nonfatal myocardial infarction and nonfatal stroke to this trend is uncertain. In fact, there are only seven studies with stratified data for these two outcomes among the RCTs reporting 3-P MACE.
When including all available RCTs, however, we found declining trends for both nonfatal myocardial infarction and stroke. The divergent trends between cardiovascular death and nonfatal cardiovascular events have several possible explanations. More intensive glucose control in recent years (change in glycemic targets), coupled with an increased prevalence of diabetes in multimorbid elderly patients, may have resulted in increasing rates of hypoglycaemia which has been associated with a higher risk of cardiovascular death in post-hoc analysis of RCTs, observational, and experimental studies [20,21,22,23]. There is also a possibility that other cardiovascular phenotypes are increasingly contributing to the risk of cardiovascular death. The reduction of cardiovascular death attributable to fatal atherothrombosis due to a widespread use of statin and aspirin, along with the increased risk of heart failure associated with aging , could have changed the phenotype “responsible” for the majority of cardiovascular complications and cardiovascular deaths in patients with T2DM, with a shift from myocardial infarction and stroke to chronic heart failure [25,26,27]. The recent suggestion to include heart failure in CVOTs as a pre-specified component of MACE would help in reducing the misclassification of outcomes and clarify whether and how changes in the cardiovascular death phenotype explain the contrasting trends between fatal and nonfatal events observed in this analysis . Further insights will also be provided by several ongoing CVOTs which included only T2DM with heart failure or were specifically designed to assess the risk of heart failure [14, 24].
Notably, in the sensitivity analysis excluding HEART 2D, TOSCA and DEVOTE, there was an inversion of trends with rising rates for nonfatal myocardial infarction and nonfatal stroke. These findings are likely related to the very low rates for both outcomes reported in TOSCA (3 per 1000 person-years for nonfatal myocardial infarction and 2.5 for nonfatal stroke) when compared to those observed in other RCTs. The reasons for such a striking difference are unknown although, as pointed out by the investigators of this study, they could be attributable to the ubiquitous use of statins, anti-hypertensive and antiplatelet agents .
To our knowledge, this is the first study to assess trends of key diabetes-related outcomes including all CVOTs studies conducted after the 2008 FDA guidance on CVOTs. This resulted in a much larger number of studies compared to previous systematic investigations and therefore in a substantial increase in the statistical reliability of the findings [11,12,13]. We also extracted data simultaneously on several outcomes and baseline characteristics of included participants, to give as clear a picture as possible of cardiovascular complications adjusted for potential confounders associated with outcomes’ rates. The study has also some limitations. We had no access to patient-level data which would have allowed a more detailed assessment of the contribution of confounders (including cardioprotective drugs, such as β-receptor antagonists, ACE-inhibitors, aldosterone antagonists, statins, and anti-hypertensive treatments) on trends and of a possible presence of ecological (aggregation) bias . Moreover, we were not able to extract information across all studies for other potential study-level confounders; however, we adjusted for key covariates strongly related to the risk of cardiovascular disease and death, namely age, sex, duration of diabetes and, more importantly, prevalent cardiovascular disease . Among other possible cardiovascular diseases at baseline, we selected myocardial infarction because it was the only confounder reported in all studies. The adjustment for prevalent myocardial infarction lessens the impact on the estimates of the different baseline risk of outcomes, particularly when comparing RCTs published after vs before 2008.
In contrast to observational data, in this study there was no evidence from RCTs of reducing rates of all-cause and cardiovascular mortality in patients with T2DM. For both RCTs and observational studies, more homogenous definitions of exposure and outcomes, the inclusion of heart failure among pre-specified endpoints, and an easier access to individual participant data will help quantify the differences between experimental and real-world evidence and further elucidate the reasons behind such divergences. Moreover, as prediction models for cardiovascular disease and mortality risk are instrumental in defining treatment strategies, targets, and clinical guidelines, health care decisions should consider that models’ performance could be highly influenced by the nature of the data, as the absolute risk of events is highly heterogeneous comparing RCTs and “real-world” patients.