Cardiovascular Safety of Roxadustat in CKD Anemia

Considerable excitement has been building in nephrology. A new paradigm for the treatment of the anemia associated with advanced CKD has arisen from Nobel Prize-winning science on the hypoxia-inducible factor (HIF) pathway, an ancient biologic system designed to protect the organism from damage in acute or chronic hypoxia (1). A considerable unmet need in the treatment of CKD anemia had been addressed with the approval of recombinant epoetin alfa in 1989. However, although treatment with epoetin alfa and, subsequently, other erythropoiesis-stimulating agents (ESAs) increased hemoglobin concentrations and reduced transfusion needs, evidence that treating anemia would reduce cardiovascular events has remained elusive. Even worse, several pivotal trials in persons with kidney disease or failure and anemia generated concerning cardiovascular safety signals from more aggressive anemia treatment using ESAs that led to boxed warnings in ESA labels.

Thus, it comes as no surprise that the community was excited by the prospect of a new treatment paradigm that promised to treat CKD anemia in a more physiologic way through manipulation of the HIF pathway by prolyl-hydroxylase inhibitors (PHIs) (1). Leading the pack of innovators was a small biotech company, Fibrogen, whose HIF-PHI roxadustat was first to enter the all-important stage of phase 3 trials. Other HIF-PHIs, including Akebia’s vadadustat and GlaxoSmithKline’s daprodustat, were not far behind. However, in light of the cardiovascular safety concerns with ESAs, regulators in the United States and European Union were not willing to approve these novel treatments on the basis of anemia treatment efficacy data alone; they mandated that comprehensive trial programs be designed and executed to prospectively evaluate the cardiovascular safety of these new drugs.

The first program to report results from these prespecified cardiovascular safety analyses was the roxadustat program, for which trial-level efficacy and pooled safety analyses were presented at the 2019 Kidney Week of the American Society of Nephrology separately for dialysis-dependent and for non–dialysis-dependent patients with CKD. This issue of CJASN contains the long-anticipated formal report on the pooled analyses of the pivotal phase 3 roxadustat trials in non–dialysis-dependent CKD (2).

In this pooled analysis, data were used from three randomized, double-blinded, placebo-controlled trials of generally similar design in which eligible patients with CKD (eGFR<60 ml/min per 1.73 m²) and anemia (hemoglobin ≤10 g/dl) initiated treatment with thrice-weekly per os roxadustat or matching placebo to achieve hemoglobin response assessed as difference in mean hemoglobin for the US Food and Drug Administration (FDA) and, for the European Medicines Agency (EMA), the proportion of patients with an achieved hemoglobin of ≤11 g/dl (and at least 1-g/dl increase from baseline) or an increase of at least 2 g/dl if the baseline was below 8 g/dl (3). Although each of the individual studies was separately powered to demonstrate erythropoietic efficacy and all three have previously been reported to have met these efficacy criteria (4–6), data from all three trials were pooled for the evaluation of cardiovascular safety. The primary end point was an expanded major adverse cardiovascular event (MACE) composite consisting of time to the earliest of nonfatal myocardial infarction, nonfatal stroke, or mortality from any cause. A key secondary end point was MACE complemented by any occurrence of unstable angina or hospitalized heart failure (MACE+). End points were independently adjudicated in an exposure-blinded fashion. Individual trials were analyzed for these outcomes, and then, a summary estimate (hazard ratio [HR]) was to be generated using meta-analysis. These analyses were designed and powered to show noninferiority relative to a stated noninferiority margin at HR of 1.3 (2).

Among 4277 randomized patients (2391 randomized to roxadustat and 1886 randomized to placebo) with CKD (mean eGFR, 20 ml/min per 1.73 m²) and anemia (mean hemoglobin concentration 9.1 g/dl), 830 MACE events and 1010 MACE+ events were observed, with all-cause mortality contributing the majority of MACE events (74%); in total, there were 138 adjudicated myocardial infarctions and 92 strokes, and 701 subjects died. The pooled analyses of the primary and key secondary end points yielded an HR of 1.10 (95% confidence interval, 0.96 to 1.27) for MACE and an HR of 1.07 (95% confidence interval, 0.94 to 1.21) for MACE+. Thus, assuming the stated noninferiority margin at HR of 1.3 (2), roxadustat demonstrated noninferiority in cardiovascular safety as assessed through the incidence of expanded MACE and MACE+ relative to placebo. However, although seemingly a reasonable statement and conclusion, the interpretation of these pooled cardiovascular safety findings is not straightforward and requires deeper consideration.

The pooled analysis for cardiovascular safety used a noninferiority analytic framework, a design insufficiently understood by many. In brief, such a design does not aim to demonstrate that Treatment A is better than (i.e., superior to) Treatment B (or placebo) but rather, to demonstrate that Treatment A is not unacceptably worse than Treatment B relative to the outcome of interest. Now, the questions are what defines “not unacceptably worse” and what constitutes an acceptable increase in risk. If the answer was “no increase is acceptable,” then we would be back to a superiority design, and only interventions that significantly reduced the risk of the end point would be considered for adoption. However, such a criterion can be prohibitive in many situations and could stifle innovation, especially if established treatments are available to treat the condition of interest and a placebo comparison may be impractical or unethical. The decision to compare roxadustat with placebo and use a noninferiority framework to assess cardiovascular safety is a questionable choice (FDA guidance specifically calls for active treatment comparison for noninferiority trials [7]) and contrasts with other innovators developing HIF-PHI drugs, whose corresponding phase 3 programs all use(d) active ESA treatment for comparison. Maybe the roxadustat sponsors believed that they had a shot at demonstrating cardiovascular superiority and hedged their bet by using a legitimate “demonstrate noninferiority first, then test for superiority approach.” Unfortunately, their thinking has not been codified in an a priori design paper.

Now, having established an acceptable noninferiority margin (and herein, we are not arguing for or against 1.3 being an appropriate threshold) (regulatory guidance is in refs. 7 and 8), the choice of what components constitute a composite end point is particularly important. In the case of a superiority trial, a sponsor is motivated to include only component end points that can reasonably be expected to be affected by the treatment of interest. Including individual events that are unlikely to be affected by the intervention would dilute the trial; these end points would be expected to occur in equal rates in both treatment groups and reduce power to demonstrate superiority, if present, for the other potentially treatment-affected end points in the composite. This would yield effect estimates that are closer to the null value of no association, which is a conservative bias, and reduce the chances of rejecting the null hypothesis of no effect, unless sample size is (considerably) increased to account for the noise induced by the nuisance component end point(s). By contrast, when using a noninferiority framework, inclusion of events unlikely to be affected by the intervention induces a nonconservative bias. One would be increasingly likely to infer noninferiority as the proportion of end points that are not plausibly affected by the treatment increases. Although deaths adjudicated as cardiovascular would plausibly complement nonfatal myocardial infarctions and strokes (the traditional definition of MACE), especially as it relates to the treatment of anemia and the intervention via the HIF pathway, using all-cause mortality includes noncardiovascular deaths that cannot be plausibly affected by this intervention of interest. Although not detailed in this report, another HIF-PHI program reported that 41% of deaths were from cardiovascular causes (9). Thus, about 60% of deaths can be assumed to have been equally distributed between the roxadustat and placebo groups, merely by design, and served as a main driver toward accepting the noninferiority assumption tested in this trial.

Finally, a similar argument can be made for the use of an intention-to-treat analysis, especially in a setting of high treatment discontinuation. Dependent on how quickly a putative treatment effect on the study outcome weans off, many events will once again occur equally between treatment groups, some long after treatment was discontinued, contributing events that would spuriously lead to an acceptance of cardiovascular noninferiority.

Thus, although much of the upcoming discussion may center on what noninferiority margin to use, an HR of 1.30 as stated in the report by Provenzano et al. (2), 1.25 as stated for the vadadustat program (9), or any other number, it is more important to recognize that this number and finding of cardiovascular noninferiority are not rooted in strong design, valid assumptions, and data. Use of noncardiovascular deaths and of unexposed person time in an intention-to-treat analytic approach combined with high rates of treatment discontinuation make noninferiority an almost self-fulfilling prophecy. As stated eloquently in a guidance document by the FDA, “[i]n a superiority trial, sloppiness can lead to study failure. In contrast, poor quality in a [noninferiority] trial can sometimes lead to an apparent finding of non-inferiority that is incorrect. Therefore, particular attention to quality is critical when planning and conducting [a noninferiority] study. Adjustment for poor quality after the fact is usually not possible” (7). Strikingly, the FDA has agreed to design features in the roxadustat program (and others) that seem to contradict their own guidance. Solid evidence on safety from appropriately designed trials satisfying key quality criteria is required to fully understand the true potential of excess risk. Only then will consumers—patients, prescribers, and payors—be able to fully understand the trade-off when weighing these risks relative to the potential benefits of new treatments. Fig leaves are sometimes used in visual arts to conceal what might be tempting, beautiful, or sometimes ugly. Unless the leaf gets plucked away, we will not know the truth that is behind it.

Disclosures

W.C. Winkelmayer reports having served as a consultant and receiving honoraria from Akebia/Otsuka, AstraZeneca, Bayer, Janssen, Merck, Reata, Relypsa, and Vifor FMC Renal Pharma during the past 36 months; serving as cochair of Kidney Disease Improving Global Outcomes; serving as an associate editor of the Journal of the American Medical Association; and serving on the editorial boards of American Journal of Kidney Diseases, American Journal of Nephrology, CJASN, and Seminars in Dialysis. The remaining author has nothing to disclose.

Funding

C.P. Walther is supported by National Institutes of Health grant K23 DK122131.