Win ratio aligns with and enhances interpretation of REPRIEVE’s primary findings

Abstract

We performed a win ratio (WR) analysis of REPRIEVE, a randomized trial of pitavastatin vs placebo for CVD in people with HIV, to assess robustness to event severity. WR aligns with primary findings, and provides insight into pitavastatin protections: persistent advantage over placebo on CV Death, Stroke, and Myocardial Infarction.

Background

Cardiovascular disease (CVD) is increased two-fold among people living with HIV (PWH).[1] The Randomized Trial to Prevent Vascular Events in HIV (REPRIEVE) assessed whether statin therapy prevents major adverse cardiovascular events (MACE) among PWH at low-to-moderate traditional CVD risk.[2] REPRIEVE demonstrated a 36% reduction in the hazard of MACE among participants randomized to pitavastatin vs placebo using a time-to-first-event (TTFE) analysis,[3] and treatment guidelines have now changed accordingly.[4] However, there are limitations to TTFE analyses in CVD trials, as the first event experienced may not be the most severe or most clinically relevant event.[5] For example, myocardial infarction (MI) and stroke may precede death in high-risk populations, leading TTFE HRs to largely ignore death.

The win ratio (WR) incorporates severity and is a popular alternative effect measure in CVD trials.[6] The WR compares each treatment participant to each control participant, declaring the participant with the less severe outcome the “winner.” Estimating the WR requires dividing treatment wins by control wins, with WR>1 suggesting treatment benefit. The reciprocal of the WR is the hazard ratio (WR=1/HR) when considering a single time-to-event outcome that satisfies proportional hazards.[7] The WR measures how much better the treatment arm performs relative to the control arm, and the HR measures how much worse (i.e., HR>1 suggests treatment harm). Unlike the HR, the WR applies to multiple events with ordered severity and win tabulation provides interpretable insights. Based on the Finkelstein-Schoenfeld test,[8] the WR’s introduction in 2012[6] has prompted several re-analyses of CVD trials,[9] and has been featured as a primary analysis in others.[10]

REPRIEVE is an interdisciplinary trial of relevance to infectious disease and cardiology practitioners. Here, we estimate the WR in REPRIEVE to assess alignment with prior findings and demonstrate the insights WR analysis reveals when multiple events differ in severity. Because more severe events tend to accumulate with longer follow-up,[11] we investigate the impact of varied trial follow-up by re-estimating the WR after censoring at each follow-up year.

Methods

Trial Design

REPRIEVE randomized 7769 PWH in 12 countries to receive pitavastatin or placebo (1:1) between March 2015 and July 2019[2]. The trial was stopped early in March 2023 for efficacy,[3] with final visits between April and August 2023. PWH with low-to-moderate risk of CVD, receiving stable antiretroviral therapy, with CD4 count > 100 cells/mm³; and eGFR ≥ 60 mL/min/1.73m² were included. Prior occurrence of CVD events was exclusionary.

Events

MACE is a composite endpoint comprising multiple, independently adjudicated CVD events.[2] Analysis used a consensus-determined hierarchy, by decreasing severity: 1: CV Death (including undetermined causes) > 2: Stroke > 3: MI > 4: Other CV Event (transient ischemic attack, unstable angina, or peripheral arterial ischemia) > 5: CV Procedure (revascularization of a coronary, carotid, or peripheral artery). Tiers 4 and 5 were reversed in sensitivity analyses. A secondary outcome included only Hard MACE events, 1: Confirmed CV Death (excluding undetermined causes) > 2: Stroke > 3: MI. Participants were censored upon non-CVD death or loss to follow-up.

Win Ratio

Estimating the WR requires constructing all possible pairs of pitavastatin and placebo participants, then comparing time from randomization to event in order of severity within each pair until a winner is declared or events are exhausted. For a stratified WR, pairs in the same strata are constructed.

Pitavastatin wins an event if the event/censoring time is greater than the placebo event time. Declaring a winner is impossible if either participant is censored before an event occurs; the pair is unevaluable, and comparison continues to the next event. The pair is tied if there is no winner for any event, or all events are unevaluable. See Supplementary Material for explanatory flowchart.

Example:

A placebo participant experiences stroke at 6 months and CV death at 2 years of follow-up. A pitavastatin participant is censored at 1 year. The pair is unevaluable on CV death, because the pitavastatin participant may die after censoring but before the placebo participant. Evaluation proceeds to stroke. Placebo participant experiences stroke at 6 months and pitavastatin participant is stroke free: pitavastatin wins. If both are stroke free, the pair is unevaluable and evaluation proceeds to MI, and so on.

Dividing pitavastatin wins by placebo wins estimates the WR. Variance and confidence interval estimation is more complex and derives from large sample U-statistic theory, but software solutions exist (see Supplementary Material).[12] Estimation is performed after censoring at each follow-up year, i.e., assuming the trial ended after 1 year, 2 years, etc., up to 8 years of follow-up in the present analysis using SAS Version 9.4M5 on a Linux platform .

Results

Participants

All 7,769 REPRIEVE participants are included, with a median follow-up of 5.6 years. The median age was 50 years, 31% (n=2,419) were female at birth, and 65% (n=5,065) were non-white (41% Black, 15% Asian, 9% Other). Median LDL was 108 mg/dl and median 10-year ASCVD was 4.5%. Median CD4 count was 621 cells/uL and 98% (n=5,997) of captured viral loads were <400 copies/mL.

Events

Among 3,888 pitavastatin participants, 123 MACE and 61 Hard MACE events occurred: 14 CV deaths, 21 undetermined deaths, 21 strokes, 26 MIs, 19 other CV events, and 22 CV procedures.

Among 3,881 placebo participants, 202 MACE and 110 Hard MACE events occurred: 20 CV deaths, 32 undetermined deaths, 42 strokes, 48 MIs, 18 other CV events, and 42 CV procedures.

Win Ratio

The WR for MACE is presented for each follow-up year in Figure 1, with wins by event and arm for 15,089,328 pairs. Win tabulation appears for MACE and Hard MACE in Supplementary Tables 1 and 2. Results were unchanged in analyses stratified by natal sex and enrollment region (Supplementary Tables 3 and 4), and with MACE tiers 4 and 5 reversed (Supplementary Table 5).

Figure 1. Wins by Event and Arm for MACE, Censoring at Each Follow-up Year.

3,888 Pitavastatin and 3,881 Placebo participants were compared, corresponding to 15,089,328 total pairs. Win Ratio and 95% Confidence Interval displayed on the x-axis. Bar labels show proportion of all wins contributed by event and arm. *CV Death includes death of undetermined cause.

Alt text: A shaded and segmented bargraph presenting the win distribution across follow-up years in the REPRIEVE trial. Bars are presented separately for the pitavastatin and placebo arms, and segments are shaded according to event severity. Within each bar segment, the proportion of total wins is presented. Beneath each year, the win ratio is presented with its 95% confidence interval.

In the first follow-up year, the two arms contributed equally to wins on CV death (11% each) and CV procedures (4% each). Pitavastatin contributed twice as many wins on stroke (15% vs 7%) and had a smaller advantage on MI (17% vs 13%), with no advantage on other less severe events. The MACE WR was 1.18 (95% CI, 0.66 to 2.11) and Hard MACE WR was 1.59 (95% CI, 0.74 to 3.43), both favoring pitavastatin. During years 2 through 5, MACE WRs fluctuated between 1.43 and 1.58, and Hard MACE WRs between 1.51 and 1.93.

About 65% of participants ended follow-up by year 6, after which WRs did not change. Ninety-five percent of pairs resulted in a tie, and 3.2% a win. The overall or year 8 MACE WR was 1.55 (95% CI, 1.20 to 1.99) and its reciprocal 1/1.55=0.645 (95% CI, 0.50 to 0.83), similar to REPRIEVE’s TTFE HR=0.64 (95% CI, 0.48 to 0.84). The WR suggests the probability of having the better CVD outcome (“winning”) is 55% greater for pitavastatin vs placebo. The year 8 Hard MACE WR was 1.65 (95% CI, 1.20 to 2.27).

Discussions

In REPRIEVE, WR analysis, which incorporates a hierarchy of event severity, revealed effects similar to prior TTFE analysis,[2] while providing additional insights through win tabulation. These analyses help explain how pitavastatin protects against MACE. Its advantage over placebo on stroke and MI was observed at year 1 of follow-up, though initially nonsignificant due to few events, and persisted through all follow-up years. Pitavastatin’s advantage on death did not appear immediately but grew over time. REPRIEVE’s primary findings are robust to event severity, and WR analysis suggests that pitavastatin protects against MACE in PWH mainly due to an early and increasing advantage over placebo on CV death, stroke, and MI.

In part, alignment of the MACE WR and TTFE HR is because few REPRIEVE participants experienced multiple events (<1% of participants: 27 pitavastatin, 46 placebo), a result of eligibility criteria. Thus, the TTFE analysis was effectively a time-to-only-event analysis. Declaring a winner requires one or both of the participants to experience the event, leading to similar event contributions between both methods.

WRs changed with longer follow-up, but the advantage of pitavastatin on Hard MACE persisted. Wins for CV death, stroke and MI, drove the overall trial results in favor of pitavastatin. Pitavastatin demonstrated a less clear advantage for less severe CVD events. These differences explain why the Hard MACE WR was greater than the MACE WR during the trial. Differences across event hierarchy may be due to lack of specificity or ascertainment for less severe endpoints. These results emphasize the importance of examining the WR over time and selecting events and hierarchies carefully. While the WRs ability to consider multiple events may increase power, inclusion of events less affected by treatment can also dilute effects.

The WR statistic has gained traction in CVD trials, extending the Finkelstein-Schoenfeld test derived to simultaneously quantify treatment effects for death and CD4 in early HIV trials. Use of the WR enables improved assessment of outcomes, increasing our understanding of the effects of pitavastatin across a hierarchy of event severity within the composite MACE endpoints in REPRIEVE. Our data demonstrate how the WR analysis can yield new insights and provide a useful analytic approach for large trials with hierarchical endpoints.