Effect of intensive blood pressure and blood glucose control on cardiovascular outcomes driven by reductions in cardiovascular death and nephropathy: win ratio analysis of ADVANCE trial

Abstract

Aims

This study aims to re-analyse the ADVANCE (Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation) trial using the win ratio approach to better understand the relative contributions of individual outcomes in a composite endpoint.

Methods and results

We applied an unmatched win ratio analysis to the 11 140 participants of the ADVANCE trial, comparing intervention and control groups for blood pressure and glucose control. Outcomes were prioritized hierarchically by severity: cardiovascular death, non-fatal stroke, non-fatal myocardial infarction, nephropathy, and retinopathy. The ADVANCE trial is registered with ClinicalTrials.gov, number NCT00145925. For blood pressure lowering, the perindopril-indapamide group had a win ratio of 1.11 [95% confidence interval (CI) 1.01–1.22, P = 0.028] compared to placebo and a net benefit of 1.5% (95% CI 0.2%–2.8%) for a number needed to treat (NNT) of 68 patients. For intensive glucose control, the win ratio was 1.10 (95% CI 1.01–1.20, P = 0.027) compared to standard glucose control with a net benefit of 1.6% (95% CI 0.2%–3.0%) for a NNT of 63 patients. When both interventions were combined, the win ratio increased to 1.19 (95% CI 1.04–1.35, P = 0.010) and the NNT decreased to 41 patients. Cardiovascular death and nephropathy were the main contributors to the observed benefits.

Conclusion

The win ratio approach confirms the robustness of the original ADVANCE findings while providing a more detailed understanding of the relative importance of individual outcomes. This method enhances the interpretation and communication of composite endpoint analyses.

Graphical Abstract

Graphical Abstract.

Introduction

The ADVANCE [Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release (MR) Controlled Evaluation] trial was a landmark study that investigated the effects of blood pressure lowering and intensive glucose control in patients with type 2 diabetes. This large-scale, randomized controlled trial enrolled 11 140 participants from 20 countries, aiming to reduce the incidence of major macrovascular and microvascular events through a combination of perindopril-indapamide for blood pressure lowering and gliclazide MR for glucose control.^1,2

The primary outcome of the ADVANCE trial was a composite of macrovascular outcomes including cardiovascular death, stroke, and myocardial infarction (i.e. MACE) and microvascular outcomes comprising new or worsening nephropathy and retinopathy. The original analysis consisted of a survival analysis of time to first composite event using a Cox proportional hazard model. The initial results of the ADVANCE trial demonstrated significant benefits in reducing cardiovascular events and mortality among patients with type 2 diabetes. In particular, the trial showed that routine blood pressure lowering reduced the risk of major vascular events by 9% [hazard ratio of 0.91, 95% confidence interval (CI) 0.83–1.00] and all-cause mortality by 14%,¹ while intensive glucose control led to a 10% relative risk reduction (hazard ratio of 0.90, 95% CI 0.82–0.98) in major macrovascular and microvascular events.²

One limitation of the ‘time-to-first-event’ analysis performed on the ADVANCE data is that each of the five components of the primary composite outcome is given equal importance. Under that approach, the effect of the intervention is driven by whichever event occurs first, regardless of its relative importance and ignoring later events of potentially higher importance (e.g. cardiovascular death occurring after worsening retinopathy).

General Pairwise Comparison methods, including the win ratio approach,³ are becoming increasingly popular for the analysis of composite outcomes (e.g. MACE) and have been suggested as an alternative method for the analysis of cardiovascular trials.⁴ As opposed to an analysis of time-to-first event which ignores the importance of the first event, the win ratio approach classifies outcomes by order of importance, first assessing the most important component and only assessing less important components if a ‘win’ cannot be determined on the most important ones. Due to this hierarchical approach, less important components do not mask more important components when they occur earlier.

General Pairwise Comparison methods were initially introduced by Buyse in 2010.⁵ The win ratio approach has gained popularity in cardiovascular outcome trials since its introduction by Pocock in 2012,⁴ and major cardiovascular and renal trials are now adopting the win ratio approach as their primary analysis including the EMPULSE and PEERLESS trials.^6,7

In this manuscript, we present the results of a confirmatory re-analysis of the ADVANCE trial data using the win ratio approach. This analysis aims to provide deeper insights into the benefits of blood pressure lowering and glucose control in patients with type 2 diabetes by providing a more detailed assessment of the contribution of each component. With this re-analysis, we also aim to inform researchers who wish to use the win ratio approach for the analysis of future trials.

Methods

Study design and participants

The ADVANCE trial was a multicentre, randomized controlled trial conducted across 20 countries. A total of 11 140 participants with type 2 diabetes were enrolled between 2001 and 2003 and followed for a median duration of 4.3 years and 5 years for the blood pressure and blood glucose intervention, respectively. Eligible participants were aged 55 years or older, with a history of major macrovascular or microvascular disease, or at least one other risk factor for vascular disease. ADVANCE was a 2 × 2 factorial trial with participants randomly assigned to both a blood pressure lowering intervention (perindopril and indapamide vs. placebo) and a glucose control intervention (gliclazide MR-based intensive glucose control regimen vs. standard glucose control).

Data collection

Baseline data were collected on demographic characteristics, medical history, and cardiovascular risk factors. Follow-up visits were conducted every 6 months to monitor blood pressure, HbA1c levels, and other relevant clinical parameters. Adverse events and outcomes were recorded throughout the study period.

Outcome measures

The primary outcome was a composite of major macrovascular events (non-fatal myocardial infarction, non-fatal stroke, or cardiovascular death) and major microvascular events (new or worsening nephropathy or retinopathy). Secondary outcomes included all-cause mortality and individual components of the primary outcome.

Statistical analysis

We used an unmatched win ratio approach⁴ where every participant from the intervention arm (perindopril-indapamide for the BP comparison and intensive glucose monitoring for the GL comparison) was paired to every participant from the corresponding control arm. We also paired patients who were randomized to both active interventions (perindopril-indapamide + intensive glucose monitoring) with those randomized to both control arms (placebo + standard glucose monitoring). Participants in each pair were compared during a shared follow-up time defined as the minimum of their follow-up times. Participants who died of a non-cardiovascular cause were censored at the time of death. As a sensitivity analysis, we considered all deaths as the first level. A pair was classified as a ‘win’ if the control participant experienced the outcome first during the shared follow-up time and as ‘loss’ if the intervention participant experienced the event first. Participants were first compared according to cardiovascular death, then sequentially according to the other non-fatal events. The stepwise approach is described in Supplementary material online, Table S1.

We calculated three complementary statistics: the win ratio, the net benefit, and the win odds.⁸ The win ratio was computed as the proportion of winner pairs divided by the proportion of loser pairs (ties are excluded from the calculation) as follows:

$$Winratio=\frac{\text{\%}ofwinners}{\text{\%}oflosers}$$

The net benefit uses the same quantities but is calculated as the absolute difference instead of as a ratio using the following formula:

$$Netbenefit=\text{\%}ofwinners-\text{\%}ofloosers$$

Following recent recommendations for the handling of ties,⁹ we also computed the win odds by counting half the ties to winners and half as losers. The win odds was thus calculated as

$$Winodds=\frac{\text{\%}ofwinners+halfthe\text{\%}ofties}{\text{\%}oflosers+halfthe\text{\%}ofties}$$

These three statistics are derived from the same percentages and all test the same hypothesis, i.e. that the chance of winning is the same in both the intervention and control groups. The net benefit provides an absolute measure of effect which represents the probability of having a better outcome when assigned to the experimental arm minus the probability of having a better outcome when assigned to the control arm, i.e. the absolute improvement in the chance of having a better outcome. The number needed to treat (NNT) was derived as the inverse of the net benefit.

Confidence intervals around the win statistics were computed using the delta method to approximate the variance as described by Dong et al.^8,10

We also performed a cumulative analysis to assess how the win ratio evolves as each component is added one-by-one according to the pre-specified hierarchy, i.e. starting with CV death and finishing with retinopathy.

All analyses were performed using R version 4.3.1 with RStudio version 2024.09.1¹¹ and the WINS package.¹²

Data and resource availability

The data that support the findings of this study are available from the corresponding author upon reasonable request. The R code used for the win ratio analysis is available from the WINS package (version 1.4.2).

Results

All 11 140 participants randomized to the ADVANCE trial were included in our analyses. For the blood pressure intervention, 5569 were allocated to perindopril-indapamide and 5571 were allocated to placebo. For the glucose control intervention, 5571 were allocated to intensive glucose control and 5569 were allocated to standard glucose control. Using an unmatched win ratio approach, we paired every patient in the intervention arm to every patient in the control arm, resulting in 31 024 899 pairs of participants for each intervention.

Primary composite outcome

The win ratio re-analysis of the primary composite outcome for the blood pressure lowering intervention showed a significant benefit for the perindopril-indapamide group compared to placebo. Overall, participants from the perindopril-indapamide group won 14.8% of the time, while those allocated to placebo won 13.4% of the time with 71.8% of comparisons resulting in ties. This resulted in a win ratio of 1.11 (95% CI 1.01–1.22, P = 0.028) and a net benefit of 1.5% (95% CI 0.2%–2.8%). This corresponds to a NNT of 68 patients (95% CI 36–644). The biggest contributors to the overall net difference were CV deaths and nephropathy with net differences of 0.8% and 0.6%, respectively (Figure 1).

Figure 1.

Win ratio decision tree showing the percentage of wins and ties by group at each step of the hierarchy for the blood pressure lowering intervention. For each additional component, the ties from the previous step are reclassified as wins or ties. The rightmost column shows the percentage of win difference (Perindopril-Indapamide wins − Placebo wins).

For the glucose control intervention, 17.1% of participants allocated to the intensive arm won against 15.5% in the control arm (67.4% of ties) resulting in a win ratio of 1.10 (95% CI 1.01–1.20, P = 0.027) and a net difference of 1.6% (95% CI 0.2%–3.0%) (Figure 2). This corresponds to NNT of 63 patients (95% CI 33–574). The nephropathy outcome was the largest contributor to the overall intervention effect, with a net difference of 0.8%, representing approximately half of the overall net difference. This was followed by cardiovascular death and retinopathy with net differences of 0.5% and 0.4%, respectively.

Figure 2.

Win ratio decision tree showing the percentage of wins and ties by group at each step of the hierarchy for the blood glucose control intervention. For each additional component, the ties from the previous step are reclassified as wins or ties. The rightmost column shows the percentage of win difference (Intensive control wins − Standard control wins).

Cumulative analysis

The cumulative analysis shows how the win ratio evolves as each component is added one by one (Figures 3 and 4). For the blood pressure lowering intervention, CV death was the main contributor to the overall intervention effect explaining 52% of the overall net difference. With CV death alone, the win ratio significantly favoured the intervention (win ratio 1.22, 95% CI 1.02–1.47). The win ratio decreased after adding non-fatal strokes and non-fatal MIs with a CI spanning unity but increased again after adding the nephropathy outcome, the second largest contributor (win ratio 1.15, 95% CI 1.03–1.28). After adding the retinopathy outcome, the last one in the hierarchy, the win ratio was slightly attenuated but still favoured the intervention (win ratio 1.11, 95% CI 1.01–1.22) (Figure 3). For the glucose control intervention, CV death, the first outcome in the hierarchy, contributed to around 30% of the overall net difference. On its own, it led to a win ratio of 1.11 (95% CI 0.94–1.32). The CI only excluded unity after adding new or worsening retinopathy, the last component in the hierarchy, which contributed to around 24% of the total net difference (Figure 4).

Figure 3.

Cumulative win ratio analysis across five clinical outcomes for the blood pressure lowering intervention. Horizontal bars show the percentage of wins by group, along with the percentage of ties (marked with an asterisk). A forest plot displays the win ratios with 95% confidence intervals.

Figure 4.

Cumulative win ratio analysis across five clinical outcomes for the blood glucose control intervention. Horizontal bars show the percentage of wins by group, along with the percentage of ties (marked with an asterisk). A forest plot displays the win ratios with 95% confidence intervals.

Joint effect of blood pressure lowering and intensive glucose interventions

Two thousand and seven hundred eighty-three patients were allocated to both the perindopril-indapamide and intensive glucose control interventions. We compared them to the 2783 patients who were randomized to placebo and standard glucose control to assess the combined effect of both interventions. This resulted in 7 745 089 pairs analysed using the win ratio approach. Overall, patients allocated to the double intervention arm won 15.8% of the time while those allocated to the double control arm won 13.3% of the time (70.9% of ties). This results in a win ratio of 1.19 (95% CI 1.04–1.35, P-value = 0.01) and corresponds to an overall net difference of 2.5% in favour of the intervention, or a NNT of 41 patients (95% CI 23–174). The net difference was mainly driven by CV deaths and nephropathy outcomes with net differences of 1.0% and 1.2%, respectively (Figure 5). The cumulative analysis (Figure 6) shows that CV deaths alone, the first outcome in the hierarchy, results in a win ratio of 1.30 (95% CI 1.01–1.69). The win ratio decreases after adding stroke but increases again once nephropathy outcomes are added to the hierarchy (win ratio 1.21 95% CI 1.04–1.41) and remains consistent after adding the retinopathy outcome.

Figure 5.

Win ratio decision tree showing the percentage of wins and ties by group at each step of the hierarchy for both interventions combined. For each additional component, the ties from the previous step are reclassified as wins or ties. The rightmost column shows the percentage of win difference (Active wins − Control wins).

Figure 6.

Cumulative win ratio analysis across five clinical outcomes for both interventions combined. Horizontal bars show the percentage of wins by group, along with the percentage of ties (marked with an asterisk). A forest plot displays the win ratios with 95% confidence intervals.

Sensitivity analysis using all-cause mortality

When using all-cause mortality instead of cardiovascular mortality as the first level in the hierarchy, results remained largely consistent across all comparisons (see Supplementary material online, Tables S2–S4). The only notable difference was the increase in overall net benefit for the blood pressure intervention which increased from 1.5% to 1.9% with all-cause death and nephropathy contributing to approximately two-thirds and one-third of the overall effect, respectively (see Supplementary material online, Table S1).

Discussion

The aim of this study was to explore whether a re-analysis of the ADVANCE trial using the win ratio approach would confirm the original findings based on time-to-first event analyses and provide additional insights. The win ratio analysis showed significant reductions in the composite of major macrovascular and microvascular outcomes with both the blood pressure lowering and intensive glucose control interventions, thus confirming the robustness of the original analysis. It also demonstrated that the effect of blood pressure lowering was mostly driven by reductions in cardiovascular deaths and new or worsening nephropathy while the effect of glucose control was mostly driven by new or worsening nephropathy.

Participants allocated to the perindopril-indapamide combination were more likely to have a better outcome 14.8% of the time while those allocated to placebo were more likely to have a better outcome 13.4% of the time, a net benefit of 1.5%. Participants allocated to the intensive blood glucose intervention had a 17.1% chance of having a better outcome than those allocated to standard glucose control, while control participants won 15.5% of the time, a net benefit of 1.6%. Participants allocated to both the perindopril-indapamide and intensive blood glucose interventions had a net benefit of 2.5% when compared to patients allocated to both control interventions.

Using the win ratio approach, we were able to separately estimate the contribution of each outcome to the overall intervention effect. For the blood pressure intervention, we showed that the main contributor was cardiovascular death with 52% of the overall intervention effect, followed by nephropathy (43% of the overall effect). For the blood glucose intervention, the main contributor was new or worsening nephropathy with 51% of the overall effect, followed by CV deaths and new or worsening retinopathy (30% and 24% of the overall effect, respectively). When both interventions are combined, CV deaths and nephropathy contribute to ∼90% of the overall effect. The net benefit of the combined intervention (2.5%) confirms the fact that the effects of blood pressure lowering (1.5%) and blood glucose control (1.6%) are largely independent and cumulative. The combined benefit of both interventions corresponds to a NNT of only 41 patients (68 and 63 respectively for the blood pressure and blood glucose interventions when considered individually).

The lack of clear effects on non-fatal myocardial infarction or non-fatal stroke is largely consistent with the original ADVANCE analysis which found no effect of blood pressure lowering on major cerebrovascular events¹ and no effect of glucose control on either major cerebrovascular or major coronary events.² Analyses of non-fatal events alone are however potentially misleading and must be interpreted with caution as they exclude potential effects on fatal events, i.e. cardiovascular deaths. Under the win ratio approach, the analysis of non-fatal myocardial infarction/stroke is conditional on the analysis of cardiovascular death. That is, the analysis of stroke and MI is only performed in pairs of participants where no decision could be made on cardiovascular deaths.

Overall, our findings confirm the strong benefits of blood pressure lowering on both macrovascular and microvascular outcomes including cardiovascular mortality. This is in line with guidelines that recommend that most patients with diabetes and hypertension should be treated, with ACE inhibitors or ARB recommended as first-line therapy.^13,14 Our analysis also supports the benefits of glucose-lowering medications on microvascular outcomes and, to a lesser extent, cardiovascular deaths; however, these apparent macrovascular benefits did not convert into long-term benefits as shown by the extended follow-up of ADVANCE patients (ADVANCE-ON study).¹⁵ Since the completion of ADVANCE, glucagon-like peptide 1 receptor agonists (GLP-1 RA) and sodium–glucose cotransporter 2 inhibitors (SGLT2i) have shown strong cardiovascular benefits in patients with diabetes and have increasingly become part of standard care while sulfonylureas are no longer recommended as a glucose-lowering strategy.¹⁶ It is also worth noting that overall background care has evolved including the use of statins. At the time of the ADVANCE trial, only 28% of participants were on statins at baseline while in the recent SOUL trial,¹⁷ that proportion was around 90%. Our findings should therefore be interpreted in the context of the recent evidence and recommendations in relation to blood pressure or glucose control in people with diabetes.

A limitation of the win ratio approach is that the results, including the relative contribution of each outcome, depend on their order in the hierarchy. When in doubt about the relative importance of each outcome and the most appropriate hierarchy, sensitivity analyses are therefore recommended. Given the clear severity grading across our five outcomes, we feel confident about our pre-specified hierarchy which was approved by all clinical authors. One might consider flipping the order between strokes and MIs or consider them together as one level; however, given these were the two smallest contributors, these changes are unlikely to impact the results. Another important consideration is the handling of ‘competing risks’ and, in particular, non-cardiovascular deaths. In accordance with the original ADVANCE analysis, our primary analysis only considers cardiovascular deaths as the first level, effectively censoring the analysis in case of deaths from other causes. To assess the robustness of the results, we ran sensitivity analyses using all-cause deaths as the first level in the hierarchy. The analysis resulted in more decisions, i.e. fewer ties, but very consistent win ratios of 1.12, 1.08, and 1.18 for the blood pressure, blood glucose, and combined interventions, respectively. The main difference was seen for the blood pressure comparison which achieved an overall net benefit of 1.9% and with all-cause death being the main contributor (63%) of this overall effect.

The win ratio approach is becoming increasingly popular for the analysis of composite outcomes in cardiovascular trials. Several trials have also recently been re-analysed using win ratio approaches¹⁸ introducing new visually compelling methods of reporting the results which we have used, and improved upon, in our paper. Other trials are now using a win ratio approach as their primary analysis.^6,18 The ability to prioritize outcomes based on their relative importance allows one to consider additional, less important outcomes, thus potentially increasing statistical power. Simulations have also shown that even when considering the same mix of outcomes, the win ratio approach can increase power compared to a standard ‘time-to-first event’ analysis when treatment effects differ across components or when the analysis is dominated by earlier, less important events.⁴ A limitation to a wider adoption of win ratio approaches is the difficulties in estimating the sample size. Sample size calculators typically require one to specify the expected win ratio and proportion of ties; however, at the design stage, these two quantities are largely unknown due to a lack of existing data and familiarity with the win ratio approach.

Despite the inclusion of five outcomes in the win ratio hierarchy, the analyses ended up with many indecisive comparisons with approximately two-thirds of ties. The high proportion of ties limits the ability to make a decision and thus statistical power. Some have argued that in the presence of ties, the win ratio tends to overestimate the effect of the intervention and have instead suggested reporting the win odds^8,9 which allocates half of the ties to the intervention arm and half to the control. For both the blood pressure and blood glucose control comparisons, the win odds were 1.03, indeed much closer to unity than the win ratio. A strategy to reduce the number of ties and improve statistical power consists in adding further outcomes to the composite, taking into account the relative importance of each additional component.

In conclusion, the re-analysis of the ADVANCE trial using a win ratio approach confirms the robustness of the original findings while accounting for the relative importance of each component. The concordant findings between the survival and win ratio analyses imply that the tested interventions both delay the composite outcome over time and yield a net hierarchical benefit when prioritizing more serious outcomes. This analysis also provides detailed insights about the relative contribution of each component, showing that cardiovascular deaths and nephropathy contributed to over 90% of the effect of blood pressure lowering while 75% of the effect of glucose control was explained by microvascular outcomes.

Although ADVANCE was conducted nearly two decades ago, this re-analysis remains clinically informative as it clarifies which specific components of the composite outcome drove the observed benefits—information that was not accessible from the original time-to-first-event analyses. We encourage other completed trials to perform a re-analysis and future trials to consider the win ratio approach. Reporting win ratio analyses will not only help future trials with their design and sample size calculations, it also has the potential to facilitate the reporting of trial results to the community with more intuitive metrics.

Supplementary material

Supplementary material is available at European Heart Journal—Cardiovascular Pharmacotherapy online.

Author contributions

A.P., L.B., and J.C. conceived the study. S.B. and L.B. performed the analyses and wrote the first draft of the manuscript. J.C., M.W., K.H., S.H., G.M., N.P., and A.P. reviewed and edited the manuscript. All authors approved the final version of the manuscript. S.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. During the course of preparing this work, the author (L.B.) used Microsoft Copilot for the purpose of creating an initial draft of the manuscript. Following the use of this tool, L.B. formally reviewed the entire content for its accuracy and edited it as necessary. The author(s) take full responsibility for all the content of this publication.

Funding

The ADVANCE trial was supported by grants from Servier and the National Health and Medical Research Council of Australia (211086 and 358395).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Ethical approval

The ADVANCE trial was conducted in accordance with the Declaration of Helsinki and was approved by the ethics committees at all participating centres. All participants provided written informed consent prior to enrolment.

Pre-registered clinical trial number

The pre-registered clinical trial number is NCT00145925.