Abstract Background Hypoxia-inducible factor prolyl hydroxylase inhibitors (HIF-PHIs) represent a potential new therapeutic option to treat anaemia of chronic kidney disease (CKD). Purpose In the recent ASCEND-ND trial (NCT02876835)1, conducted in patients with anaemia of CKD not requiring dialysis, the HIF-PHI daprodustat was non-inferior to darbepoetin for cardiovascular (CV) events in the primary intention-to-treat analysis (HR 1.03; 95% CI 0.89, 1.19); however, a prespecified on-treatment analysis raised concerns about a higher risk of CV events associated with daprodustat compared to darbepoetin (HR 1.40; 95% CI 1.17, 1.68). Our analysis explored potential reasons to explain the differences between on-treatment and intention-to-treat analyses in ASCEND-ND. Methods Overall, 3872 patients were randomised to receive either oral daprodustat (daily) or darbepoetin alfa given weekly, every 2 weeks, or every 4 weeks in an open-label fashion (last dosing frequency for darbepoetin was 7% weekly, 15% every 2 weeks, and 78% every 4 weeks). Patients were followed for development of the composite CV outcome (all-cause death, first non-fatal myocardial infarction, or first non-fatal stroke). The prespecified on-treatment approach included CV events up to 28 days following the last non-zero dose date of randomised therapy (Figure 1). Post-hoc analyses used Cox regression models to assess the impact of different follow-up periods (indexed to last non-zero dose date, treatment stop (discontinuation) date, and dosing intervals) on the treatment effect estimate. Results Different definitions of “on-treatment” using alternative censoring approaches resulted in hazard ratios for the CV composite outcome for daprodustat vs. darbepoetin that ranged from 1.06 (95% CI 0.89, 1.27) censored at treatment stop date; 1.09 (95% CI 0.89, 1.33) censored at last non-zero dose date + dosing interval; 1.54 (95% CI 1.20, 1.97) censored at the last non-zero dose date (Figure 2). As a result of the differential dosing interval, the gap between the last non-zero dose date and CV event date was 15 [1 to 134] days in the daprodustat arm, and 35 [13 to 134] days prior in the darbepoetin arm. This resulted in identical patients (i.e., identical treatment stop dates and event dates) being more likely to count as “on-treatment” in the daprodustat arm (Figure 1). This artefactual difference was not observed in analyses that were indexed to the treatment stop date, nor in analyses that accounted for the different dosing intervals for darbepoetin. Conclusion In the ASCEND-ND trial, different dosing frequencies introduced longer intervals between the last non-zero dose date and date of CV outcomes in the darbepoetin arm, compared with daprodustat. This artefact led to inappropriate undercounting of CV events in the darbepoetin arm. Accounting for the differential intervals resulted in neutral effect estimates, which were consistent with those observed with intention-to-treat approaches. Funding Acknowledgement Type of funding sources: Private company. Main funding source(s): ASCEND-ND was funded by GlaxoSmithKline
Read full abstract