Prioritised endpoints for device-based hypertension trials: the win ratio methodology

Abstract

Aims: Multiple endpoints with varying clinical relevance are available to establish the efficacy of device-based treatments. Given the variance among blood pressure measures and medication changes in hypertension trials, we performed a win ratio analysis of outcomes in a sham-controlled, randomised trial of renal denervation (RDN) in patients with uncontrolled hypertension despite commonly prescribed antihypertensive medications. We propose a novel prioritised endpoint framework for determining the treatment benefit of RDN compared with sham control.

Methods and results: We analysed the SPYRAL HTN-ON MED pilot study data using a prioritised hierarchical endpoint comprised of 24-hour mean ambulatory systolic blood pressure (SBP), office SBP, and medication burden. A generalised pairwise comparisons methodology (win ratio) was extended to examine this endpoint. Clinically relevant thresholds of 5 and 10 mmHg were used for comparisons of ambulatory and office SBP, respectively, and therefore to define treatment "winners" and "losers". For a total number of 1,596 unmatched pairs, the RDN subject was the winner in 1,050 pairs, the RDN subject was the loser in 378 pairs, and 168 pairs were tied. The win ratio in favour of RDN was 2.78 (95% confidence interval [CI]: 1.58 to 5.48; p<0.001) and corresponding net benefit statistic was 0.42 (95% CI: 0.20 to 0.63). Sensitivity analyses performed with differing blood pressure thresholds and according to drug adherence testing demonstrated consistent results.

Conclusions: The win ratio method addresses prior limitations by enabling inclusion of more patient-oriented results while prioritising those endpoints considered most clinically important. Applying these methods to the SPYRAL HTN-ON MED pilot study (ClinicalTrials.gov Identifier: NCT02439775), RDN was determined to be superior regarding a hierarchical endpoint and a "winner" compared with sham control patients.

Figure 1

Conceptual illustration of the prioritised endpoints methodology involving four hypothetical subjects – two randomised to renal denervation (R1 and R2) and two randomised to sham control (S1 and S2). A) The generalised prioritised endpoints method requires each RDN subject to be compared to each sham control subject, resulting in 2×2=4 pairs. B) Hypothetical outcomes (change from baseline 24-hr mean ASBP; ΔASBP) for each of the four subjects. C) Endpoint comparison of each subject (RDN vs sham control) and the resulting classification of the pair based on a threshold of 5 mmHg. D) For pairs that result in a tie, compare patients using the next specified endpoint, in this case change from baseline mean OSBP; ΔOSBP. ASBP: ambulatory systolic blood pressure; OSBP: office systolic blood pressure; RDN: renal denervation; NR: number of RDN patients; NS: number of sham control patients

Figure 2

Win ratio analysis of the hierarchical endpoint of ΔASBP at six months (with threshold of 5 mmHg), ΔOSBP at six months (with threshold of 10 mmHg), and ΔIndex at six months (with threshold of one unit) in the SPYRAL HTN-ON MED pilot study. Every patient in the RDN arm is compared with every patient in the sham control arm. For each pair, it is determined over the six-month follow-up, first whether the RDN randomised patient “wins” or “loses” on ASBP; then, if “tied” on ASBP, whether the RDN randomised patient “wins” or “loses” on OSBP; and then, if “tied” on OSBP, whether the RDN randomised patient “wins” or “loses” on medication index. ASBP: ambulatory systolic blood pressure; OSBP: office systolic blood pressure; RDN: renal denervation