Globalization Does not Lead to Homogenization in Randomized Controlled Trials: Impact on Sample Size Estimation

This article refers to ‘Geographic differences in patients with acute myocardial infarction in the PARADISE-MI trial’ by J.H. Butt et al., published in this issue on pages 1228–1242. The Prospective Angiotensin Receptor–Neprilysin Inhibitor (ARNI) versus Angiotensin-Converting Enzyme (ACE) Inhibitor Trial to Determine Superiority in Reducing Heart Failure Events after Myocardial Infarction (PARADISE-MI) was an international, randomized, double-blind, active-controlled (ramipril) trial recruiting 5661 patients with acute myocardial infarction (MI) complicated by a reduced left ventricular ejection fraction, pulmonary congestion, or both.1 PARADISE-MI enrolled patients in 41 countries on six continents, becoming a ‘paradigm’ for global trials, like his big brother.2 In this issue of the Journal, we can find a post hoc study of this trial evaluating geographical differences in patient characteristics, management of MI, clinical outcomes, and treatment effect.3 By focusing on regional heterogeneity, the study paradoxically underscores that globalization does not lead to uniformity in randomized controlled trials (RCTs), but to heterogeneity in many dimensions. Representativeness means that a subset of participants (sample size) accurately reflects the characteristics of a larger group (full population).4 Mas-Llado et al.5 have recently evaluated all RCTs cited in the American College of Cardiology/American Heart Association and European Society of Cardiology clinical practice guidelines for the management of acute coronary syndrome, concluding that most of these RCTs had recruited patients from North America and Europe (52.6% and 73.2%, respectively), whereas other continents were largely underrepresented. There were only 25.3%, 20.4%, 19.9%, and 12.8% studies recruiting patients from Asia, Latin-America, Australia, and Africa, respectively.5 This recent paper elegantly show that, before global trials became popular, the majority of RCTs were conducted in Western Europe and Northern America, and consequently most of the recommendations provided by clinical practice guidelines are based on trials conducted in a limited range of geographic locations and subgroup of patients.6 Randomized controlled trials have become increasingly globalized in modern times.7 Globalization has improved representativeness, ensured timely recruitments, and allowed the evaluation of the treatment effect across different subsets of patients.8 Nevertheless, global trialists face new challenges, most of them in the form of heterogeneity.9 Heterogeneity in the study population, in clinical outcomes, and in treatment effect. The key question is whether heterogeneity, which might a priori jeopardize the internal validity of RCTs, can be used to the trialist advantage to streamline the study design of future trials. Global trials are biased towards academic institutions with a primary interest in research, which in turn tend to care for patients that might be different from non-academic centres.5 Participating sites in trials have been shown to practice more evidence-based medicine than non-participating sites. Similarly, trials tend to exclude very high-risk patients, and the degree of exclusion may vary by region.5 These disparities might be translated into an exacerbated within- and between-region heterogeneity in demographics and comorbidities, as well as in background pharmacological and interventional therapies among the patients enrolled. Let us put the focus on the background therapy. In the post hoc study of the PARADISE-MI published in this issue,3 the rates of prescription of evidence-based secondary prevention were high (statin 94.9%, beta-blocker 85.3%, ACE inhibitor or angiotensin receptor blocker [ARB] 78.4%, mineralocorticoid receptor antagonist [MRA] 41.3%). Some might argue that these data are similar to some US registry data (US National Cardiovascular Data Registry, 2019) showing a percentage of patients with statins at discharge of 98.9%, beta-blockers 97.4%, and ACE inhibitor or ARB in those with left ventricular systolic dysfunction 92.2%.10 However, others might argue that global registries using high-risk patients have shown lower prescription rates,11 similar to those registries not solely including high-risk MI patients.12, 13 Beyond overall prescription rates, in this post hoc study of PARADISE-MI there were substantial geographic differences in the use of secondary preventive treatments.3 In South Asia, there was a low rate of use of ACE inhibitor/ARB (34.8%) and beta-blockers (65.5%), in line with previous observational reports evaluating the same region.12, 13 MRA use was (paradoxically) lowest in North America (22%) and highest (53%) in Eastern Europe/Russia. This observed heterogeneity in background evidence-based therapies leads to two main questions. Can the use of quality indicators14 reduce the ‘translational gap’, improve the suboptimal implementation of evidence-based care, and subsequently decrease regional disparities? Should differences in study populations have an impact on the study design of future RCTs if consistently observed across RCTs? In addition to disparities in patient characteristics, there was some heterogeneity in the event rate of the primary outcome across regions of the PARADISE-MI trial (e.g. 4.6 primary events per 100 person-year in South Asia, and 9.2 primary events per 100 person-year in Latin America).3 South Asia had a high cardiovascular death rate, whilst it had a low rate of incident heart failure, potentially explained by a competing risk issue.4 In contrast, Latin America had a high hospitalization rate, but it also had a high mortality rate, in accordance with previous reports.12, 13 We might try to explain these discrepancies in outcomes using both individual-level (e.g. comorbidities, background therapy), and system-level arguments (e.g. different health care systems, bed availability). We might also want to discuss about composite outcomes and the weight of each individual component, and about whether death is unambiguous but a rare outcome, whilst worsening heart failure is more frequent, but also health care-dependent and connected with subsequent mortality,15 but the key question is whether we can learn from geographic disparities in clinical outcomes to better design future RCTs. Can sample size estimation be optimized according to the expected event rate in each region? Randomized controlled trials are powered to provide evidence of an overall treatment effect, and are generally underpowered to yield reliable regional estimates on treatment efficacy. As with other subgroup analyses, caution should be exercised when interpreting signs of geographic heterogeneity in treatment effect, unless the evidence of heterogeneity is overwhelming (e.g. interaction p < 0.001).16 The PLATO trial of ticagrelor in acute coronary syndrome, and the MERIT-HF trial of metoprolol in heart failure were controversial in this regard, since both showed overall superiority of the new treatment, which was not so evident in US patients.16 Whilst subgroup analyses are generally seen as exploratory and hypothesis-generating, the question here is not whether such geographic discrepancies can be of sufficient magnitude to unveil real differences in treatment efficacy, but whether we can use these data, whether statistically significant or not, to streamline the sample size estimation. If a region consistently recruits higher risk patients, then, even if the relative benefits of a new treatment are geographically consistent, the absolute benefit would be greater in such region. Differences in the expected number of events in the control group between regions have an impact on the sample size estimation. Most geographic regions in previous studies have been selected based on their arbitrary world-map location, but might not be well supported by racial, social, health care, or medical practice grounds.8 It has been shown that the country income (per capita) and income inequality, as assessed by the Gini coefficient, have an impact on prognosis in patients with MI.17 Patients from low-income countries or with higher income inequality had higher mortality rates compared to patients from high-income and low inequality areas,8 and sex differences tend to be exacerbated in this context.17 Whether the impact of income inequality and income per capita on event rates have a subsequent impact on treatment effects in cardiovascular RCTs is yet to be determined. Hence, there is a last question before we move on to the implications of all these layers of heterogeneity for the study design of future trials. Should we categorize based on health care and social provision instead of map location, would we find larger or smaller differences between ‘regions’? Are we grouping the regions in the most comparable way? In an era where event rates in cardiovascular trials are falling progressively, global trials are needed to obtain larger sample sizes, as well as to improve representativeness and generalizability.5 Whilst geographic differences in patient characteristics, outcomes and treatment effect can be perceived as new challenges for the study design and interpretation of future trials,18 an optimistic trialist might use all this new useful information to ‘personalize’ the estimation of sample size according to the regions involved. Between-country disparities in comorbidities, background therapies, and baseline risk, as well as in health care systems and level of attainment to evidence-based management might be used to streamline the sample size calculation. Because of these differences, both the event rate in the control group and the treatment effect may vary across geographic locations, and therefore the number of patients needed to show efficacy might vary from country to country. Whilst the focus would still be on showing overall efficacy, small differences between countries might be considered among the parameters for sample size estimation. An expected number of events might be calculated for each country based on the baseline risk of the control group, and the expected treatment effect. The overall sample size (aimed to test the primary hypothesis in the overall population) should result from adding up the expected number of events from each country. Of note, there will be needed to estimate and/or cap the number of participants recruited in each country. Yet, imaginative and flexible solutions might be applied to monitor and adapt the needed sample size according to the number of participants recruited in each country. Conflict of interest: none declared.