Long-term nitrate use in chronic coronary artery disease: Need for a randomized controlled trial☆☆☆★★★

See related article on page 577.

Nitrates have been used for more than a century for the treatment of angina pectoris. During the last 2 decades, their use has also been expanded to include the treatment of unstable angina and congestive heart failure. They are among the most often used drugs in coronary artery disease (CAD). This widespread acceptance and use of nitrates in CAD is mainly based on data from the large volume of animal and mechanistic human studies,1, 2, 3, 4, 5 which have clearly shown how nitrates can be beneficial in myocardial ischemia and, by extrapolation, should be beneficial in reducing clinical events in individuals with CAD. Whether or not the available data support this assumption of a beneficial clinical effect can be examined by a review of the evidence from clinical trials on the effects of nitrates tested under the various clinical situations of acute myocardial infarction, stable and unstable angina, and congestive heart failure.

The evidence on nitrates in acute myocardial infarction is derived from 2 separate groups of randomized, controlled trials. Data from 10 intravenous and 5 oral trials of nitrate in acute myocardial infarction, carried out before the widespread use of β-blockers, aspirin, and thrombolytics, suggest that the routine use of nitrates is beneficial in acute myocardial infarction.6 Data from 3 recent large trials on a total of more than 80,000 patients, however, have been neutral and therefore less promising.7, 8, 9 Indeed, each of these trials, which individually has adequate power to detect a beneficial effect, if it exists, failed to show any benefit. On the basis of these data, it can only be concluded that the data do not support the routine use of nitrates after acute myocardial infarction.

The treatment of unstable angina has been marked by the paucity of randomized data on the nitrates. Only one randomized controlled trial of a nitrate in unstable angina has been reported, with inconclusive results.10 A review of 9 uncontrolled studies with intravenous nitroglycerin in 280 patients with unstable angina suggests that the treatment is effective in controlling pain but there are no data on other clinical outcomes.11 Nitrates have been shown to be beneficial in relieving the pain of chronic stable angina12 and in improving the exercise time to ischemia and/or angina in exercise studies.13, 14 In contrast to the large volume of data from animal laboratory and short-term clinical studies with surrogate end points such as exercise time, there are no randomized clinical trial data on long-term clinical outcomes when nitrates are used routinely in patients with CAD, after acute myocardial infarction, or in unstable or chronic stable angina. The V-HeFT I trial reported that a combination of hydralazine and a nitrate is beneficial long term in congestive heart failure.15

The finding by Nakamura et al16 that the use of long-acting nitrates in chronic CAD may be associated with an increased risk of cardiac death is therefore provocative but important in setting the stage for a critical re-examination of one of the oldest forms of CAD and angina therapy. In this study, the authors analyzed the relation between long-acting nitrates and mortality outcome in patients with CAD treated during the chronic phase of the disease. Two databases were examined, one from an observational cohort study on patients who were enrolled after the acute phase of myocardial infarction or unstable angina, the Multicenter Study of Myocardial Ischemia (MSMI), and the other on survivors from acute myocardial infarction in a long-term clinical trial of a calcium channel blocker, the Multicenter Diltiazem Postinfarction Trial (MDPIT). Cox proportional hazards regression analyses on both databases individually showed that patients who were receiving nitrates at baseline had a significantly increased mortality risk compared with those not taking nitrates. This increased mortality risk also appeared to be present when the data were re-analyzed by the propensity score method.

As the authors have rightly emphasized, the findings of this study should be regarded as “hypothesis generating,” given the uncontrolled nature of the data. Intuitively, clinicians can think of several reasons to explain why patients taking nitrates in these databases may have a worse outcome. Chief among these is the view that the use of nitrates serves as a surrogate for one or more variables that may be truly associated with a poor prognosis and that have not been adequately captured and accounted for in the databases and analyses. Patients taking nitrates would more likely be those who have angina, and it is possible and likely that these patients would tend to carry a greater CAD burden, be less stable, or both. No information is available on the subsequent treatment of the patients in the studies, not even whether they continued taking the nitrates during follow-up. The nitrates could have been used for patients in congestive heart failure or in those who are symptomatic but deemed unsuitable or at unacceptably high risk for revascularization therapy. These patients are at high mortality risk. Furthermore, the main findings of the MSMI, which was published previously,17 showed that ST-segment depression in the resting electrocardiogram was significantly associated with risk of subsequent cardiac events and that other noninvasive tests of silent or symptomatic myocardial ischemia tended to be associated, but not significantly, with increased risk for subsequent recurrent cardiac events; these findings were not included in this analysis. It is possible that nitrate use may have been highly correlated with these positive findings.

These concerns about the finding by Nakamura et al16 should be examined in the context of the methods used in the study and the robustness of the data. The authors have appropriately used 2 powerful statistical methods, the Cox proportional hazards regression model and the propensity score method, as their analytic tools. The findings, however, could be affected by the ability to control adequately for potential confounders, and further by the need to impute for a considerable amount of missing data. In the MSMI, data on 1042 patients (106 more patients than the previously reported study17) who were followed up for an average of 26 months were analyzed. The total of 25 deaths during follow-up was surprisingly few, and it is possible that several fewer deaths in the nitrate group and a few more in the non-nitrate group would drastically affect the eventual outcome. Furthermore, fitting a Cox regression model with a total of only 25 cardiac deaths and 15 independent variables may lead to an unstable estimation of the relative importance of the explanatory variables, including nitrate use. For logistic regression analysis, which often yields similar results as those obtained with a failure time model, a rule of thumb is to have approximately 5 events (deaths) at each level of the more important explanatory variables. Thus the conclusion from the MSMI database alone should not be regarded as reliable.

The original MDPIT enrolled 2466 patients.18 Of these, 1779 patients who survived the first 30 days of the trial and who had well-documented concurrent medications were included in the analyses for this study. There were 125 cardiac deaths during the average follow-up of about 24 months, and the data appear more robust. A simple χ2 test showed that patients taking nitrates had a worse outcome than those not taking nitrates, and Cox regression analysis showed that nitrates independently contributed to cardiac death with a hazard ratio of 1.61 (P = .019), strengthening the conclusion from the MSMI database analysis.

To minimize the impact of treatment selection bias inherent in observational studies, the authors then estimated a “propensity score” as a means of matching individuals with similar observable characteristics.19, 20 This score is the probability of a patient receiving nitrate treatment based on a number of factors that may influence the decision to use or not use this treatment. The point of this type of matching is that the comparison of 2 individuals with the same observable characteristics, one of whom was treated and one of whom was not, is like comparing these 2 individuals in a randomized experiment. In this paper, 5 subclasses of individuals were formed on the basis of quintiles of the estimated propensity scores for the 2 individual databases. The effect of nitrate use was then estimated within each subclass.

Matching on propensity score will lead to unbiased estimates of treatment effect, provided determination of the propensity score was based on all potential confounders.20 In a series of simulations, Drake showed that when an important confounder is omitted, either in the estimation of the propensity score or in model-based approach, severely biased results can occur.21 The bias in the estimate of the odds ratio ranged from 7% to approximately 70%. Nakamura et al performed a sensitivity analysis to estimate the effect such omitted or “unobservable” confounders might have on the observed relation between nitrate use and mortality rate and concluded that the omitted confounder would have to increase the risk of cardiac death 3- to 6-fold to eliminate the observed finding. However, on the basis of the results in the paper by Drake,21 the potential bias in the Nakamura study may equally account for the observed “positive” finding.

The results of these analyses also depend on data robustness. In the MSMI, the quintile of propensity has resulted in highly uneven number of patients in each group. Because of the small number of deaths, the absence of this event in one quintile in each of the 2 treatment groups did not allow the calculation of hazard ratio in 2 of the 5 quintiles. The small number of events in the study also means that conclusions from this analysis can change drastically if just a couple of deaths were to fall into other quintiles. In this setting, the data do not warrant more than 2 or 3 subclasses. Although event rates from MDPIT were higher in each of the quintiles, the hazard ratio, although tending to be higher in the nitrate group in each quintile, was less striking.

Despite these shortcomings, Nakamura et al16 have made an admirable effort in attempting to draw some useful conclusions from their databases. Although their study does not conclusively show that the use of long-acting nitrates in chronic CAD, perhaps on a long-term basis, can lead to an increased mortality risk, the lack of randomized trial data to support or refute such practice is troublesome. Although the pharmacologic and physiologic mechanisms of benefit of nitrates in CAD appear well known, the mechanisms for potential harm can only be speculated.16 It is also unknown whether this potential harm might be from all or any particular long-acting nitrate preparations. We echo their call for a randomized trial to answer this important question.