In the GUSTO trial, 41021 patients with acute myocardial infarction, presenting with ST-segment elevation within 6 hours of symptom onset, were randomly allocated to four thrombolytic strategies: streptokinase with subcutaneous heparin, streptokinase with intravenous heparin, accelerated-dose tissue plasminogen activator (tPA) with intravenous heparin, and the combination of streptokinase and tPA with intravenous heparin. Complete details about the thrombolytic agent and heparin dose regimen in each treatment arm, as well as about other adjunctive therapy, are contained in the recent GUSTO report [1]. For the primary end point of 30-day mortality, the results are given in Table 1. No significant difference between the two streptokinase strategies was noted (P = 0.731). Accelerated-dose tPA led to a 14% (95% CI, 5.9% to 21.3%) decrease in mortality compared with the two streptokinase strategies combined (7.3% compared with 6.3%, P = 0.001); a mortality benefit was also noted when tPA was compared with each streptokinase strategy separately (tPA compared with streptokinase with intravenous heparin, P = 0.003; tPA compared with streptokinase with subcutaneous heparin, P = 0.009). No significant difference in mortality was noted between the thrombolytic arm involving both streptokinase and tPA and the strategies involving streptokinase monotherapy (P = 0.352); however, a marginally significant difference was noted in 30-day mortality between accelerated-dose tPA and combination therapy with streptokinase and tPA (P = 0.04). As reported by the GUSTO investigators [1], an excess was noted of approximately 2 hemorrhagic strokes per 1000 patients treated with accelerated-dose tPA compared with streptokinase, but an overall statistically significant benefit was noted for accelerated-dose tPA compared with each of the streptokinase strategies for the combined end points of mortality plus any nonfatal stroke, mortality plus nonfatal hemorrhagic stroke, and mortality plus nonfatal disabling stroke.

Consistency of Results Across Subgroups

Ridker and colleagues [4] claim that the benefit of accelerated-dose tPA observed in GUSTO is limited to selected patient subgroups. Major subgroups specifically mentioned in their critique, where they contend the data show no treatment benefit, included patients treated 4 or more hours after symptom onset and patients treated outside the United States. The issue of which thrombolytic agent to use in certain patient subgroups (patients older than 75 years, patients with small inferior infarctions, and patients treated 4 or more hours after symptom onset) was also raised in the editorial that accompanied the GUSTO report [5]. Established principles have been articulated in multiple publications about the perils of subgroup analyses and the dangers of misinterpreting subgroup results [6-9]. Regardless of whether a trial shows an overall treatment benefit, careful analysis and considerable caution are required before drawing conclusions that a treatment benefit exists in one subgroup but not in another. Concern for the perils of subgroup analyses is reflected by the cautionary warning from Peto and colleagues [6] that "marked heterogeneity of the treatment comparison in different strata can arise by chance more easily than would intuitively be expected ..." The difficulties of interpretation that arise in subgroup analyses have been effectively highlighted by the International Study of Infarct Survival (ISIS-2) investigators [7] using the example of subgroups defined on the basis of astrological birth signs.

Of particular concern is the approach of comparing treatments within selected subgroups (particularly subgroups with considerably fewer patients than the overall trial), of finding that the P value does not achieve conventional significance, and of concluding that no treatment benefit exists for the patients in those particular subgroups. A study is generally never designed with sufficient patients in all major subgroups of interest to provide adequate power for statistically detecting clinically important treatment differences when treatments are compared within the individual subgroups. In the landmark Gruppo Italiano per lo Studio della Streptochinasi nell'Infarto Miocardico (GISSI-1) trial [10], although clear evidence was given that, overall, streptokinase achieved a statistically significant decrease in mortality compared with placebo, this effect was not statistically significant if one examined, for example, just the elderly (older than 65 years), or women, or those treated after 6 hours. This observation does not imply, however, that the treatment is of no benefit to patients in those particular subgroups. Given the number of patients in the GUSTO trial treated 4 or more hours after symptom onset, the power to detect a 1% mortality difference in that subgroup is less than 35%. Rather than statistically comparing treatments within multiple individual subgroups, it is more relevant to examine the direction of the treatment differences and the consistency of the estimate of treatment benefit across subgroups. This evaluation can be done quantitatively by carefully assessing whether interactions exist between treatment and specific patient strata (for example, by statistically assessing whether the relative benefit of a treatment varies across specific characteristics of the patients). Judgments about heterogeneity in treatment effects across specific subgroups should not be made by simply testing whether treatment differences achieve conventional significance within each individual subgroup.

Ridker and colleagues [4] provide misleading results when they claim that "for the more than 9000 patients in GUSTO-1 who were treated 4 or more hours after symptom onset, mortality was nonsignificantly lower among patients receiving streptokinase with subcutaneous heparin compared with those receiving accelerated-dose tPA with intravenous heparin". As shown in the published report from the GUSTO investigators [1], nearly 8000 (85%) of the 9400 patients treated 4 or more hours after symptom onset were treated within 6 hours. For patients treated 4 to 6 hours after symptom onset, the mortality rate was nonsignificantly lower in the accelerated-dose tPA arm than in the streptokinase arms (8.9% compared with 9.3%). Although this difference is smaller than the overall mortality benefit observed in the trial and does not achieve statistical significance if compared within this stratum (P = 0.256), it is not correct to conclude there is no benefit of tPA for patients treated 4 to 6 hours after symptom onset and to claim that the mortality is lower among patients treated with streptokinase. Less benefit may exist for tPA among patients enrolled within this time interval, and the data do suggest that the benefit diminishes as the time to treatment increases. However, less benefit in these smaller subgroups does not imply no benefit. It is only in the small number of patients treated after 6 hours that an indication exists of a possible lower mortality rate in streptokinase-treated patients. The mortality numbers we have quoted above for streptokinase patients treated 4 hours or more after symptom onset were taken from the published GUSTO report [1] and are the results in the pooled group of streptokinase patients. The discussion by Ridker and colleagues, however, refers to patients treated with streptokinase and subcutaneous heparin. Because the mortality in patients treated 4 hours or more from symptom onset with streptokinase and subcutaneous heparin was actually slightly higher than in the pooled streptokinase group, the argument above also applies if one considers tPA relative to streptokinase with subcutaneous heparin.

A further point made by Ridker and associates [4] and others [3] is that the patients in GUSTO were treated quite early and thus perhaps the study findings would not be generalizable. The median time to randomization in GUSTO was 2.0 hours, and the median time to treatment was 2.8 hours. Additional perspective on those figures can be obtained from comparative patients in Gruppo Italiano per lo Studio della Sopravvivenza nell'Infarto Miocardico (GISSI-2) and its International Study Group extension, the only other large-scale comparative trial of thrombolytic agents where patients were enrolled within a 6-hour time window [11, 12]. Those investigators reported time to randomization (rather than time to treatment) as shown in Table 2.

The GUSTO trial actually enrolled a lower percentage of its patients within the first hour than did GISSI-2, and the proportion of patients enrolled within 3 hours of symptom onset was similar in the studies. The issue of generalizability because of the relatively short time to treatment thus applies equally to both studies. However, given the current state of our knowledge of reperfusion therapy in acute myocardial infarction, no disagreement should exist that the goal of therapy is to treat patients early. In fact, widespread recognition of this need has stimulated national initiatives to study factors that impede rapid access to thrombolysis. In GUSTO, patients were enrolled from 1081 hospitals in 15 different countries, and most of those hospitals were community based. In North America, less than half of GUSTO sites had previously participated in a clinical trial involving patients with acute myocardial infarction. Hospitals participating in GUSTO, at least in the higher enrolling countries, were a representative cross-section of typical health care settings in which patients with acute myocardial infarctions are treated. With the increasing emphasis on treating patients earlier after symptom onset, and with education and a heightened awareness about decreasing the time to treatment, the results of studies such as GUSTO and GISSI-2 are even more relevant.

Ridker and colleagues claim that the benefit of accelerated-dose tPA observed in GUSTO was restricted to patients treated within the United States. In GUSTO, as in any large international trial, some degree of country-to-country variation exists in clinical event rates, and such variation is expected among geographically defined subgroups. Indeed, observations about intercountry variation have been reported from the International TPA and Streptokinase Mortality Trial [13]. Ridker and colleagues correctly point out that the observed mortality difference between patients receiving accelerated-dose tPA and those receiving streptokinase was larger and more statistically significant among U.S. patients (17% decrease; 1.2% absolute difference) than among non-U.S. patients (7% decrease; 0.5% absolute difference). A larger treatment difference was observed in the United States despite the fact that mortality among streptokinase-treated patients was lower in the United States than elsewhere (7.1% compared with 7.6%). To aid in judging whether the U.S. compared with non-U.S. variation in the benefit of accelerated-dose tPA compared with streptokinase represents real or random fluctuation, we used logistic regression analysis and examined the data for a treatment by country interaction. The test for whether tPA had a different effect compared with streptokinase in U.S. compared with non-U.S. patients yielded a P value of 0.304. Thus, although the variation observed in different health care settings is interesting and deserves thoughtful reflection, the U.S. compared with non-U.S. variation observed in GUSTO about the benefit of tPA compared with streptokinase falls within the range of what one could expect from the play of chance.

Other analyses done to assess the degree of heterogeneity of the treatment benefit across subgroups other than those defined by country and time to treatment showed a striking consistency of the treatment effects across subgroups defined by age, sex, infarction location, risk factors (hypertension, diabetes, history of smoking, and hypercholesterolemia), baseline medical history (previous myocardial infarction, coronary artery bypass surgery, or percutaneous transluminal coronary angioplasty), and presenting characteristics (blood pressure, heart rate, and Killip class). Certainly it is true that within particular subgroups (especially those with relatively small numbers of patients), treatment differences may not be statistically significant using conventional criteria for significance. However, as clearly advocated by the ISIS investigators [7] and emphasized by other clinicians involved in the clinical trials [9], where there is little evidence of any real heterogeneity (as occurred in GUSTO), more weight should be given to the overall results of the trial than to the apparent effect observed within a particular subgroup.

For clinicians faced with the responsibility of applying the results of these trials to individual patients, interpreting subgroup analyses can be a particularly vexing challenge. Individual patients obviously reflect a combination of clinical factors that influence prognosis. A given patient may fall into a low-risk subgroup on the basis of one factor or a high-risk subgroup on the basis of another factor, and have varying degrees of risk conferred by virtue of other characteristics. Along with a complete assessment of the randomized treatment comparisons in these trials, carefully developed statistical models that take into account multiple patient characteristics (which we are currently developing with the GUSTO data) can be usefully applied to estimate the risk of an event for an individual patient.

Unblinded Design

Specific criticism from Ridker and colleagues concerned the "open-label" design of GUSTO, suggesting that this aspect of the study casts doubt on the validity of observed treatment benefits and that any such benefits may simply be the result of bias introduced by the absence of double-blinding. We agree totally with the scientific desirability of a double-blind design whenever feasible. The inclusion of the combination thrombolytic agent arm in GUSTO, through which important unanswered questions could be addressed about the safety and efficacy of aggressive pharmacologic therapy, made blinding more difficult because of the requirement for dual intravenous infusion lines. This situation would have necessitated insertion of an extra intravenous line in more than 30 000 patients for the sole purpose of blinding. Apart from the operational and economic practicalities of doing a blinded trial, however, the GUSTO Steering Committee (with its collective clinical trials experience) was convinced that by using a study design that focused on the objective end point of mortality (and by using appropriate training of site personnel, careful data collection, and monitoring of the data), a valid open-label trial could be done. Although not mentioned by Ridker and colleagues [4], one of the two other large-scale trials discussed in their editorial (namely the GISSI-2 International trial [11, 12]) was also an unblinded study. Yet, as they point out, the major results of GISSI-2 and of ISIS-3, which was a double-blind trial [14], were essentially superimposable. The GISSI-1 trial also had an open design, yet its results have been highly acclaimed and widely accepted [10]. Further, the results of the unblinded GISSI-1 trial were replicated by the blinded ISIS-2 trial.

In considering the validity of the open-label design of GUSTO, it is important to keep in mind several points: 1) Patients enrolled in GUSTO were randomly allocated to the treatment arms, thus eliminating bias in the selection of thrombolytic therapy; 2) treatments were compared according to intention to treat; 3) compliance with the assigned thrombolytic therapy occurred in 98% of patients in each treatment arm; and 4) with a primary end point that involved the most objective assessment possible (namely, whether the patient was dead or alive at 30 days), this information could be (and was) collected in an unbiased and impartial way.

For the important secondary outcomes involving stroke incidence, data (consisting of computerized axial tomography of the head, magnetic resonance imaging, or autopsy in 94% of all patients where a focal neurologic dysfunction was noted) were independently examined by a stroke review committee, the members of which were fully blinded to treatment. Additionally, end point information (indeed, all patient information collected in the trial) was obtained and processed independently of the sponsors [15]. Furthermore, on-site monitoring of all submitted data, including comparisons and cross-checking with source documentation in the hospital medical records, was done in a randomly selected sample of over 10% of the patients. Every hospital participating in the trial had a sample of their patients monitored. In a trial where double-blinding was not practical without adding enormously to the complexity and expense of the study, it is difficult to conceive of any greater safeguards to ensure the quality and completeness of the data than were built into the conduct of GUSTO.

As further support to the credibility of the GUSTO mortality results, the GUSTO angiographic substudy (which conclusively showed a higher patency rate of the infarction-related artery at 90 minutes with accelerated-dose tPA) provides a logical pathophysiologic explanation for the effect of accelerated-dose tPA in decreasing mortality [16].

In connection with the issue of an open-label design, Ridker and colleagues suggest that the clinical benefits observed in GUSTO may be attributable to the fact that a higher proportion of patients treated with accelerated-dose tPA had coronary artery bypass surgery than did patients treated with streptokinase (9.0% compared with 8.3%). However, additional analyses showed that the difference in surgical procedures cannot be the explanation for the tPA benefit. Among GUSTO patients who did not have coronary artery bypass surgery during their hospitalization, 30-day mortality was 6.5% in patients treated with accelerated-dose tPA and 7.6% in patients treated with streptokinase, a highly statistically significant difference (P < 0.001). Among the small group of patients who did have bypass surgery during hospitalization, the observed 30-day mortality was similar for accelerated-dose tPA and streptokinase (4.0% compared with 4.1%, respectively). In regression analyses of these data, accelerated-dose tPA was still beneficial (P = 0.001) after adjusting for the effects of bypass surgery and for numerous other baseline prognostic factors. Interpretation of these data about the influence of bypass surgery on mortality must be made cautiously because bypass surgery is not a baseline variable and its occurrence may be influenced by patient selection or other clinical events after randomization. Nevertheless, the treatment benefit cannot be attributed to the slightly higher rate of bypass surgery among tPA-treated patients.

Compliance Issues

Criticism has been leveled at GUSTO for the degree of heparin noncompliance that occurred in the streptokinase plus subcutaneous heparin treatment arm [2, 3]. It was clearly pointed out in the GUSTO report that 36% of patients randomized in this group received intravenous heparin at some point during their hospitalization and that this problem occurred more often in the United States than in other countries [1]. However, in most of the instances where patients in this arm were given intravenous heparin, it was medically indicated. This was consistent with the design of the trial, which allowed for clinically indicated use of heparin. The most common indications for more aggressive heparin were symptoms of recurrent ischemia or the use of coronary angiography or coronary angioplasty or both, with the latter occurring more frequently in the United States. The use of intravenous heparin in conjunction with angiographic procedures or as treatment of postinfarction unstable angina was considered acceptable clinical practice within the study protocol. If one examines results from the 14 countries other than the United States that enrolled patients in GUSTO (assuming those countries would have a "purer" result with subcutaneous heparin), the streptokinase plus subcutaneous heparin arm had the lowest 30-day mortality among the four treatment groups in only one country and that particular country enrolled considerably fewer than 1000 patients. In the eight non-U.S. countries that enrolled more than 1000 patients in the study, the arm with the lowest 30-day mortality in every case was either accelerated-dose tPA or the combination of streptokinase and tPA. This degree of consistency adds further credence to the overall conclusions of GUSTO and suggests neither that the results are simply an artifact because of noncompliance with the subcutaneous heparin regimen nor that they are limited to only U.S.-treated patients.

Magnitude of the Treatment Differences

Another area of concern is with the magnitude of the treatment benefit. The clinical relevance of the mortality difference the study was designed to detect (a 15% decrease in mortality or an absolute difference of 1%, whichever was larger) has been openly questioned. Critics argue that a difference requiring 41000 patients to show statistical significance cannot be important. Our response is that the treatment differences GUSTO was designed to detect are entirely consistent with the design goals of ISIS-3 and the GISSI-2 International trials. Further, a 15% relative decrease or a 1% absolute decrease in mortality in a trial designed to compare several active therapies (in contrast to active drug compared with placebo) is indeed an important advance in myocardial infarction therapy. In the era of placebo-controlled trials of thrombolytic agents, 26 lives per 1000 patients treated were saved with the use of thrombolytic therapy for patients treated within the first 6 hours of symptom onset [17]. Nine to 11 additional lives saved per 1000 patients treated—nearly 40% of the previously established standard—is a vitally important public health benefit, even though the cost of this further mortality decrease is 1 disabling stroke per 1000 patients treated. Although a decrease in mortality risk from approximately 7% to approximately 6% is small in absolute terms, it is highly clinically significant to the hundreds of patients (and to the families of those patients) whose lives would be saved if the therapeutic strategy was appropriately used in the large number of persons throughout the world who annually have myocardial infarction.

Concerning the statistical significance of treatment differences, it is noteworthy that even at the first major prespecified interim efficacy analysis, when key outcome data were prepared for confidential review by the GUSTO Data and Safety Monitoring Board (which occurred when slightly more than 10 000 patients had been enrolled), the difference between accelerated-dose tPA and the better of the two streptokinase arms was statistically significant if judged only by the conventional 0.05 level criterion. The trial was continued, however, because, among other reasons, the relatively stringent O'Brien-Fleming group-sequential stopping boundaries prespecified in the interim analysis plan had not been crossed [18, 19]. The apparent benefit of tPA appeared relatively early and was maintained at subsequent interim analyses throughout the trial.

Cost-Effectiveness

The cost-effectiveness of therapy using accelerated-dose tPA is obviously an important issue in the context of debate about the magnitude of the treatment benefit. We share the concern expressed by others about the cost implications of widespread use of tPA. We join others in desiring to see the drug made available at an even more cost-effective level. We caution readers to be wary, however, of some of the relatively simplistic cost calculations [20] that neither account for what happens to patients beyond the 30 days reported in the initial GUSTO publication nor place the cost-effectiveness data in the context of other accepted, and likely more expensive, therapies such as bone marrow transplantation, treatment of hypercholesterolemia, and renal dialysis. Fortunately, ongoing follow-up of GUSTO patients continues so that the longer-term consequences of early therapeutic benefits can be appropriately assessed and factored into cost-effectiveness calculations. Analyses in progress from the economic substudy of the GUSTO investigative group will shed important additional light on these issues. Were it not for its cost-related implications, we doubt that the critique of the trial would be so vociferous [2-4, 20].

Combining GUSTO Data with Other Trials

The commentary by Ridker and colleagues contains their synthesis of data from the three large thrombolytic trials (GISSI-2 International, ISIS-3, and GUSTO-1) considered "both individually and together" [4]. (Although they propose several conclusions "drawn from the randomization of more than 100 000 patients in these three large-scale trials" [4], their conclusions are presumably not based on analyses that statistically combined the results from all three studies. For several reasons, we caution readers against deriving inferences about treatment differences based on a simple combination of the data from all three studies. First, a biological reason exists for considering GUSTO separately from the other trials. In the tPA monotherapy arm, GUSTO used accelerated-dose tPA with intravenous heparin. This represents a considerably different tPA treatment strategy than was used in the other trials. The dose of tPA was also weight-adjusted, differing from the GISSI-2 International trial. Second, a statistical basis exists for considering GUSTO separately. The mortality odds ratios for tPA compared with streptokinase in each of the three studies are shown in Table 3.

Comparing treatment effects among the three trials (for example, using logistic regression), an interaction [P = 0.012] exists between the effect of treatment and study. The effect of tPA compared with streptokinase is statistically different in GUSTO compared with that observed in the other studies. Combining data from the three studies for the purpose of judging the benefit of tPA compared with streptokinase is thus problematic. Important conclusions that can be derived from these three studies, however, are 1) the GISSI-2 International and ISIS-3 trials provide clear evidence that standard-dose tPA given with subcutaneous heparin has no mortality advantage compared with streptokinase and 2) GUSTO provides clear evidence that accelerated-dose tPA given with intravenous heparin does decrease mortality. There is neither a biological nor a statistical basis for combining dissimilar studies and clouding the clarity of these conclusions. We agree with Ridker and colleagues that early treatment is important, and considerations of which thrombolytic agent to give should not delay the initiation of treatment.

One (Not Two) Accelerated-Dose Tissue Plasminogen Activator Arms

The statement by Ridker and colleagues that the GUSTO trial evaluated two "accelerated" tPA regimens should be amplified, and their use of the terminology "accelerated tPA" in conjunction with both arms that involved tPA should be clarified. In the combination therapy arm in GUSTO (streptokinase and tPA), the dose of each thrombolytic agent was different (lower) than the amount administered in the individual monotherapy arms. In the "accelerated tPA" arm (the tPA monotherapy arm), the intravenous thrombolytic agent infusion was given over a period of 1.5 hours, with two thirds of the dose given in the first 30 minutes. This infusion is the so-called front-loaded, or accelerated-dose, regimen studied by Neuhaus and colleagues [21, 22]. In contrast to the accelerated-dose tPA arm, patients randomly assigned to the combination of tPA and streptokinase received less tPA for a given patient weight, and the administration schedule did not involve the same rapid degree of front-loading. For example, a typical patient weighing 79 kg (the median weight of patients in GUSTO) received 65 mg of tPA during the first 30 minutes in the accelerated-dose tPA arm compared with 43 mg of tPA in the combination therapy arm (although patients receiving combination therapy also received 500 000 international units of streptokinase in the first 30 minutes). Although the tPA dose in the combination therapy arm (1 mg/kg up to 90 mg given over 1 hour) was a more accelerated dose than the 3- to 4-hour dosing used in the previous large comparative trials, we emphasize that in GUSTO, the tPA dosing used in the combination arm was considerably less accelerated than the regimen referred to as the "accelerated tPA" arm in the GUSTO report.

Final Comment

The design and execution of this large, multinational, cooperative trial of thrombolytic strategies involved the collective experience and wisdom of more than 30 experts in the field who composed the Steering Committee [15]. We sought to achieve the optimal balance and autonomy of the trial and of patients involved in the trial, beginning with its architectural blueprints and extending through the completion of the analysis phase (> 3.5 years). Our findings of an important stepwise survival benefit of accelerated-dose tPA compared with previously established thrombolytic therapy indicate that the speed and permanence of coronary thrombolysis are directly related to preservation of left ventricular function and survival in patients with acute myocardial infarction. This insight should lead to clinical trials of other compounds that will open arteries faster than accelerated-dose tPA (which had only 54% grade III flow at 90 minutes using the Thrombolysis in Myocardial Infarction [TIMI] study criteria) and that decrease rates of reocclusion. However, as large-scale, randomized trials shed new light on ways to improve medical therapy, appropriate interpretation of the results will enhance the potential for clinical and scientific understanding. Careful review of the GUSTO trial data, with many of the key issues clarified herein, will help to forge meaningful progress in our contemporary approach to myocardial reperfusion.