Portal de Periódicos da CAPES

A tentative interpretation of the TOPCAT trial based on randomized evidence from the brain natriuretic peptide stratum analysis

2016; Elsevier BV; Volume: 18; Issue: 12 Linguagem: Inglês

10.1002/ejhf.621

1879-0844

Nicolas Girerd, João Pedro Ferreira, Patrick Rossignol, Faı̈ez Zannad,

Cardiovascular Function and Risk Factors

Heart failure with preserved ejection fraction (HFpEF) affects a large proportion of patients with the clinical syndrome of heart failure (HF).1-3 These patients share impaired quality of life and poor prognosis with their HF counterparts with reduced ejection fraction (HRrEF).4, 5 Despite the burden of this condition, no treatment tested in a clinical outcome trial has convincingly been shown to improve outcomes in HFpEF.6-8 Mineralocorticoid receptor antagonists reduce hospitalizations and mortality across the HFrEF spectrum.9, 10 The Treatment of Preserved Cardiac Function Heart Failure with an Aldosterone Antagonist (TOPCAT) trial tested spironolactone vs. placebo in patients with HFpEF.11 TOPCAT was a phase III, multicentre, international, randomized, double-blind, placebo-controlled trial. Patients were eligible if they were ≥50 years of age, exhibited at least one sign and one symptom of HF, and had an LVEF of ≥45%. In addition, eligible patients had either a history of hospitalization for worsening HF within the previous 12 months or an elevated natriuretic peptide level within 60 days before randomization (BNP ≥100 pg/mL or NT-proBNP ≥360 pg/mL). A total of 3445 patients were randomized in six countries [USA, Argentina, Brazil, and Canada ('the Americas') as well as Georgia and Russia ('Eastern Europe')].1 Overall, the TOPCAT trial did not achieve a significant reduction in the primary composite outcome of cardiovascular death, aborted cardiac arrest, or hospitalization for management of HF.11 At times, significant interactions (statistically significant differences) between treatment groups are found among different subgroups within a multicentre trial.12 Several post-hoc analyses of the TOPCAT trial have been performed. For example, the interactions for location of randomization and LVEF spectrum with spironolactone treatment effect have been published and commented upon.13-16 However, these are post-hoc, non-randomized analyses, and should be regarded as hypothesis generating, as hereafter discussed. Briefly, TOPCAT subanalyses identified a large variation in treatment effect on cardiovascular mortality between different geographical areas [hazard ratio (HR) 0.74, 95% confidence interval (CI) 0.57–0.97 in 'the Americas' vs. HR 1.31, 95% CI 0.91–1.90 in 'Eastern Europe', P for interaction = 0.012].12, 14 However, cardiovascular mortality was a secondary endpoint and no significant interaction was found regarding the primary composite outcome of HF hospitalization and cardiovascular mortality (HR 0.82, 95% CI 0.69–0.98 in 'the Americas' vs. HR 1.10, 95% CI 0.79–1.51 in 'Eastern Europe', P for interaction = 0.12). We reviewed geographic differences in HF trials—including TOPCAT—in a recent issue of the European Journal of Heart Failure.13 In addition, another subanalysis suggested that the potential efficacy of spironolactone was greatest at the lower end of the LVEF spectrum (P for interaction for continuous LVEF = 0.046).15 In this analysis, stronger benefits of spironolactone were possibly present at the lower end of the LVEF spectrum with respect to the primary composite endpoint (LVEF <50%: HR 0.72, 95% CI 0.50–1.05). These results suggest a gradient in treatment efficacy according to LVEF. Specifically, the treatment effect in patients with intermediate or mid-range LVEF (i.e. EF between 40% and 50%)17, 18 appeared closer to that observed in patients with LVEF <35–40%. This is of crucial importance given the current large gaps of evidence regarding the treatment of this 'middle child in heart failure'.18 Importantly, in the absence of specific randomized evidence regarding spironolactone effect in patients with mid-range LVEF, it is difficult currently to recommend its use in this population. Unfortunately, as patients with mid-range LVEF usually represent <20% of HF patients, specific trials addressing this question are unlikely. It is beyond the aim of this 'Viewpoint' to speculate as to the reasons behind the 'bias' in patients enrolled in Russia/Georgia or treatment effect differences across the LVEF spectrum.12, 16 However, the very unfortunate result is that our HFpEF patients are deprived of what could have been an evidence-based life-saving therapy.14 In an attempt to overcome these limitations, Pfeffer and Braunwald19 recommend the following: 'based on the findings in TOPCAT in North and South America and in the absence of other more definitive data, it now appears reasonable to treat patients with HFpEF resembling those enrolled in North and South America with spironolactone to improve outcomes'. We would like at this juncture to offer another tentative interpretation of the findings of TOPCAT based on the BNP stratum. Indeed, no matter how striking are the interactions described to date based on post-hoc subgroup analyses, from a methodological perspective, interactions between pre-specified randomized strata have a higher value. Indeed, in stratified trials, such as TOPCAT, randomization is performed within each stratum. As a consequence, the results derived within a given stratum are 'true' randomized evidence as already emphasized by Armitage and Gehan20 and Kernan et al. 'with the knowledge that patients within a subgroup are presumed similar in all regards except treatment, investigators can draw valid inferences about treatment effectiveness within subgroups'.21 Moreover, in clinical trials, stratification is by nature always pre-specified. It is based on the strong association of the strata criterion with the primary endpoint of the trial. The virtues of randomization make the results of strata analyses immune to bias which inevitably hampers all analyses made on subgroups defined from non-randomized baseline characteristics. These methodological aspects necessarily imply that the analysis of TOPCAT according to its strata is by nature a greater level of evidence than the usual subgroup analyses. In TOPCAT, trial randomization was stratified according to whether patients were enrolled hierarchically on the basis of the first criterion (designated the hospitalization stratum) or the second criterion (designated the BNP stratum). The second criterion was considered only for those who did not meet the first criterion.11 No significant differences between treatment and placebo groups were observed when randomization was performed on the basis of hospitalization.11 Although history of HF hospitalization is a very strong predictor of outcome, variation in the healthcare systems and subjectivity for deciding when to admit a patient for worsening HF makes it a source of heterogeneity, which was admittedly stretched to the limits in that, in Russia and Georgia TOPCAT patients, the primary event rate was a fifth of that of the Americas' patients, hardly superior to that of non-HF patients. In contrast, a significant beneficial treatment effect was identified in the BNP stratum, with a significant interaction between treatment effect and randomization strata (P = 0.01; Figure 1). In the BNP stratum (981 patients: 490 spironolactone/491 placebo), spironolactone was associated with lower rates of the primary composite outcome events [78 (15.9%) events in the spironolactone group vs. 116 (23.6%) in the placebo group; HR 0.65, 95% CI 0.49–0.87, P = 0.003]11 (Figure 1). Of note, there were 38 fewer primary endpoint events in the spironolactone vs. the placebo group in this BNP stratum, which is greater than the 31 fewer primary endpoint events in the active vs. the placebo group in the entire study. This may be due to the 'risk homogenization' provided by natriuretic peptides13. Stratification is usually performed to ensure that strong predictors are balanced between randomization groups. However, stratification is also the only situation in which the benefit of randomization is maintained in subgroups since randomization is performed at a stratum level. As a result, in TOPCAT, stratification allows the evaluation of spironolactone effects in HFpEF patients with elevated natriuretic peptides without recent HF hospitalization in a randomized fashion. Hence, these data support that in patients randomized on the basis of elevated BNP, spironolactone treatment was effective in reducing the primary outcome while showing a trend for significance in reducing cardiovascular mortality. Natriuretic peptides correlate with symptomatic LV diastolic dysfunction22 and are less susceptible to bias as compared with history of hospitalization criteria, prone to great heterogeneity across healthcare systems and variation in geographic area.13 The 'homogenized' disease definition and severity provided by BNP may be key to the observed positive spironolactone effect in this stratum, potentially eliminating subjects who were less likely to suffer from HFpEF, as discussed elsewhere.13 Several limitations could prevent this result from having an immediate impact on clinical guidelines. The first one is multiple testing. The P-values reported in the TOPCAT trial for interaction (P = 0.01) and effect in the BNP stratum (P < 0.001) are lower than the threshold derived from the Bonferroni method for two (and even three) statistical tests. However, a priori, clinical trials intend to perform only one statistical test, not several (e.g. one for interaction and one for treatment effect). As stated above, stratification is currently used as a balancing tool. Based on this case study, we might recommend that a small part of the alpha risk could be devoted a priori to interaction between strata testing prior to treatment effect quantification. If a 0.01 alpha had been devoted to initial interaction testing between strata, ∼0.04 alpha would have been left for treatment effect testing. Using this strategy, TOPCAT is likely to be a positive trial in patients with HFpEF and elevated BNP. One further limitation is that, for good reasons, guideline authors and regulatory agencies are reluctant to accept evidence based on a subgroup analysis based only on a segment of the trial population, even if pre-specified. However, as stated above, this is the consequence of the subgroup analysis being weakened by non-randomization bias, whereas stratification preserves randomization benefit. However, both non-randomized subgroup analyses and randomized strata analyses are open to potential lack of power. In the present instance, the trial was not powered a priori to identify treatment effect within the BNP stratum, which is a significant limitation. In addition, patients were only entered in the BNP stratum if they had not been hospitalized; hence these results are not generalizable to patients who have been hospitalized for HF in the last 12 months. However, it is likely that, at least in certain geographic areas with a low threshold for HF admission, patients with a recent history of HF hospitalization would benefit equally if they also satisfy the BNP criteria. Nevertheless, even if with an admittedly moderate level of evidence, we would like to weigh in the possibility that spironolactone is beneficial in HFpEF patients (and has an acceptable safety profile), where no alternative therapy exists. We offer our personal view that such patients should be given spironolactone if presenting with the TOPCAT patient characteristics and high BNP levels. Of note, another TOPCAT subanalysis, published as an abstract to date,23 identified that patients in the lower BNP tertile derived more benefits from spironolactone, despite having lower event rates (HR 0.3, 95% CI 0.2–0.6 in tertile 1 vs. ∼0.9 in tertiles 2 and 3, P for interaction = 0.02). Such an interaction is intriguing and is reminiscent of similar findings in the I-PRESERVE trial with irbesartan.24 Anand et al., authors of both reports, speculated that the benefit of spironolactone or irbesartan, respectively, in lower risk HFpEF patients may indicate effects of the drugs on early, but not late higher risk stages of the disease. As recognized by the authors, a major limitation of this type of post-hoc secondary analysis of data is that the findings may be spurious and need to be confirmed in other samples. The observation that the lower the BNP the better is the benefit from treatment is intriguing and is yet to be explained. As a matter of fact, these findings were not confirmed in a similar analysis of data from another ARB HFpEF trial (CHARM-Preserved), as stated by Kao et al.22 However, irrespective of how challenging these findings may be, these do not contradict the observation that the patient population enrolled on the basis of a specific BNP entry criterion is more likely to have a homogeneous risk, which makes any real effect of the study drug statistically more likely to be significant, since it is less diluted by the heterogeneity of patient severity that is observed in the 'history of HF hospitalization' stratum in TOPCAT. Spironolactone, as compared with placebo, reduced the composite outcome of hospitalization for HF or cardiovascular mortality among patients with HFpEF and elevated BNP. Given the stratified nature of TOPCAT, this result is randomized evidence. In the absence of alternative evidence-based treatment in patients with HFpEF, we concur with the viewpoint of Pfeffer and Braunwald19 that 'HFpEF is often a disabling and life-shortening condition. Other than the administration of diuretics for fluid accumulation and the management of hypertension (if present), there is little to offer these patients. Based on the findings in TOPCAT in North and South America and in the absence of other more definitive data, it now appears reasonable to treat patients with HFpEF resembling those enrolled in North and South America with spironolactone to improve outcomes. This drug is generic, inexpensive, and generally well tolerated, although periodic monitoring of electrolytes and creatinine must be conducted to detect the occasional development of hyperkalaemia and renal dysfunction'. We would like to add the further possibility that clinicians should also consider spironolactone in HFpEF patients when BNP is elevated. Conflict of interest: none declared.