Patient selection bias in primary percutaneous coronary intervention trials: a critical issue

Abstract

The ‘frequentist’ approach in randomized study designs and the inclusion of a large number of patients (high statistical inference) have frequently produced inconclusive, or conflicting, or negative results despite the high potential for a strong impact on outcome of the study drug, or device, or strategy. In many studies, the adoption of rigid criteria for randomization resulted in the exclusion of many patients at high risk of adverse events and prevented the correct assessment of the efficacy of the study treatment. This review will focus on two issues that are still matter of debate due to the incapacity of concluded randomized trials to provide an unequivocal answer: the routine use of an emergency revascularization strategy in cardiogenic shock complicating acute myocardial infarction (AMI) and the routine use of abciximab in the setting of primary percutaneous coronary intervention (PCI) for AMI.

Evidence based medicine is supported by the results of randomized trials. Blinded or unblinded patient randomization to a drug, or device, or therapeutic strategy should avoid any bias in patient selection and provide the way to a correct assessment of the efficacy of the study treatment. Randomized studies have allowed undeniable and extraordinary advancements in the treatment of diseases. However, in the setting of acute coronary syndromes, many studies based on the same strong hypothesis have produced negative or conflicting results and delayed the routine use of an effective treatment or strategy for a long time. Clear instances of inappropriate delays in the correct assessment of the superiority of a new treatment or therapeutic strategy are primary coronary angioplasty when compared with fibrinolysis, primary infarct artery stenting when compared with balloon angioplasty, a strategy of emergency coronary revascularization when compared with an initial conservative approach in patients with cardiogenic shock (CS) complicating acute myocardial infarction (AMI), an invasive strategy when compared with medical treatment for patients with non-ST-segment elevation acute coronary syndromes, and the routine use of IIb/IIIa inhibitors in the setting of primary percutaneous coronary intervention (PCI) for AMI. Some of these cases are definitively closed, whereas others are still open. The superiority of primary PCI over fibrinolysis is a closed case, and currently it is recommended by the American and European guidelines. However, the first primary PCI was performed >20 years ago and more than 20 randomized studies were performed before the definitive establishment of its superiority when compared with fibrinolysis, but only just a review of 23 randomized studies has been published.¹ Limitations in study designs and definition of enrolment criteria resulting in highly selected study populations may, at least in part, explain the negative or unexpected results of many randomized trials and suggest that randomization ‘per se’ cannot avoid a paradoxical patient bias phenomenon. Many concluded randomized trials of PCI in AMI share a bias in patient selection resulting in the exclusion from randomization of elderly patients and high-risk patients, and this bias occurred also in studies that did not use an upper age limit for randomization. This type of bias have important clinical implications because age is a strong predictor of death in patients with AMI, and with ageing population in western countries, elderly patients represent one-third of all patients admitted with a diagnosis of ST-segment elevation AMI. Moreover, the exclusion or the poor representation of elderly patients in many trials make difficult or impossible the assessment of the efficacy of a treatment in this high-risk population and simultaneously may mask the benefit of a treatment in a population at low risk and with a low expected mortality rate. This review will focus on two issues that are still matter of debate due to the incapacity of concluded randomized trials to provide an unequivocal answer: the routine use of an emergency revascularization strategy in CS complicating AMI and the routine use of abciximab in the setting of primary PCI for AMI.

Early revascularization in CS complicating AMI

After the results of the Should We Emergently Revascularize Occluded Coronaries for Cardiogenic Shock (SHOCK) registry,² which showed a strong bias in patient selection for emergengy revascularization excluding sicker patients, in 1993 the SHOCK investigators began a randomized trial based on a sample size of 302 patients with CS, where an immediate coronary revascularization strategy (PCI or surgery) was compared with a more conservative approach that included fibrinolytic treatment and aortic balloon pumping (initial medical stabilization arm). The enrolment ended in 1998, 6 years after the start of the study. The primary endpoint of the study was a 20% reduction in 30-day mortality in the emergency revascularization arm when compared with the initial medical stabilization arm. The results of this trial were negative because the reduction in 30-day mortality was inferior to 20% (46.7 vs. 56.0%; difference between the groups, 9.3%; 95% CI for the difference, 20.5–1.9%; P=0.11).³ The 30-day mortality rate was 45.3% among the 75 patients who underwent PCI and 42.1% among those who underwent coronary surgery. A positive interaction between early revascularization strategy and 30-day mortality was found for age <75 years, men, a time from myocardial infarction onset to randomization <6 h, and a previous myocardial infarction. In contrast, randomization to emergency revascularization was associated with an increased mortality in elderly and women. At 6 months, the overall mortality from all causes was lower in the emergency revascularization group than in the initial medical stabilization group (50.3 vs. 63.1%, P=0.027), but 6-month mortality was only a secondary endpoint of the study. The results of this study, which is the only concluded randomized trial comparing an emergency revascularization strategy with an initial medical stabilization strategy in patients with CS, were considered for the ACC/AHA Guideline Task Force that recommended an emergency revascularization strategy only in patients <75 years.⁴

Analysis of several variables, including patient selection bias, may help to explain the results of this study. The reasons for the very high mortality rate in the early revascularization group, significantly higher than the 25–30% rate reported by single high-volume centres and the 35% rate by the GUSTO investigators,⁵ may have several explanations. First, the success rate of coronary angioplasty remained very low until the last 2 years of enrolment, only 58% of patients achieving a TIMI grade 3 flow after the procedure. This poor result may be explained, in part, by the limited use of stents in the first 4 years of enrolment, whereas in 1997 and 1998 with the use of stents in 76% of treated cases, a TIMI grade 3 flow was achieved in 68% of patients. As expected, successful PCI resulted in decreased mortality (38%) when compared with the mortality rate in patients with an unsuccessful procedure (79%). Secondly, the delay from CS diagnosis to treatment was abnormally long: the median time from myocardial infarction onset to shock diagnosis was 5 h, whereas the median time from myocardial infarction onset to randomization was 11 h. This delay in treatment may, in part, explain the relatively poor outcome of the patients of the SHOCK trial. Among patients randomized to emergency revascularization, only 25% were randomized <6 h, and subgroup analysis suggested a positive interaction between revascularization and 30-day mortality in this subgroup of patients. This figure is consistent with the results of previous studies showing that the benefit of revascularization in shock patients is time dependent.^6–8 Thus, one may infer that a higher PCI success rate achievable with stent-supported angioplasty, as well as a shorter delay from diagnosis of shock to treatment, could have been associated with a better outcome.

Finally, a patient selection bias is evident. The study design did not include an upper age limit, but patients selected for randomization were younger than non-randomized patients (mean age of randomized patients 65.8±10.4 vs. 68.5±12.1 years of non-randomized patients, P<0.001), while the incidence of patients >75 years old in the SHOCK Registry cohort was twofold higher than that of the randomized trial cohort (32.4%).^9,10 After the publication of the results of the randomized study, the same investigators have produced other studies based on the registry patient cohort that have disavowed some conclusions of the randomized study, suggesting a strong benefit of emergency revascularization also in women and carefully selected elderly patients.^11,12

Abciximab as adjunctive treatment in patients undergoing primary PCI for AMI

Five concluded randomized studies comparing abciximab with placebo in patients undergoing primary PCI for AMI^13–16,21 have produced different and conflicting results with a broad spectrum of possibilities: no benefit of the drug in patients receiving infarct artery stenting or benefit of the drug only in patients undergoing conventional balloon angioplasty with provisional stenting, benefit limited to the early phase and mainly driven by the decrease in the need for urgent target vessel revascularization (TVR), benefit in terms of decreased mortality, reinfarction, and target vessel revascularization in the early phase but not maintained at 6 months, long-term benefit in the composite of death, reinfarction, and urgent target vessel revascularization, improved early and late outcome including long-term survival.

The protective effect of abciximab against the clinical events related to the failure of the target vessel could be clearly demonstrated in the Reopro and Primary PTCA Organization and Randomized Trial (RAPPORT) that compared coronary angioplasty alone and coronary angioplasty plus abciximab in 483 patients with AMI.¹³ The RAPPORT trial showed a better early outcome of patients randomized to abciximab, with a >70% relative reduction in the composite of death, myocardial infarction, and TVR at 1 month when compared with placebo (2.8 vs. 10.6%, P<0.006), and this benefit was driven by the reduction of ischaemic events related to early target vessel failure. The benefit was no more evident at 6 months, with similar event-free survival curves of the two groups (28.1 vs. 28.2%). When non-urgent TVR procedures were excluded, the incidence of the composite of death, myocardial infarction, and urgent TVR at 6 months was 17.8% in the placebo group and 11.6% in the abciximab group (P=0.05). These results may be explained by the very high incidence at the 6-month follow-up of TVR due to late restenosis that is not affected by abciximab treatment and confirm the overwhelming impact on outcome of a component of the composite endpoint that is unrelated to the study treatment. In this trial, the use of stents was strongly discouraged, and only 70 patients had unplanned coronary stenting. This characteristic of the study design resulted in the easy demonstration of a reduction of ischaemic events in abciximab treated patients due to the very high incidence of early target vessel failure after balloon angioplasty, and at the same time, in the impossibility to identify potential mechanisms of benefit other than the protection from ischaemic events related to target vessel failure.

It is important to highlight that the impact on survival of early target vessel failure after balloon angioplasty is modest. Only a minority of deaths after successful PCI for AMI is due to target vessel failure, and the relationship between death and target vessel failure may emerge only in patients with large area at risk, or severe ventricular dysfunction or CS, thus, just in high-risk patients who were not fully represented or excluded from enrolment in concluded randomized trials.

In the Intracoronary Stenting and Antithrombotic Regimen (ISAR)-2 trial, 200 patients with AMI were randomized to infarct artery stenting alone and 201 to stenting plus abciximab.¹⁴ The study was based on the hypothesis of a positive effect of abciximab on late in-stent restenosis, and the sample size of the study was defined according to a significant reduction of late loss at 6 months in patients randomized to abciximab. At 1 month, the incidence of the composite of death, non-fatal reinfarction, and TVR was lower in the abciximab group when compared with the stent alone group (5.0 and 10.5%, respectively, P=0.038). However, at 1 month, there were no differences between the two groups in the incidence of the individual components of the composite clinical endpoint, including mortality (2.0% in the abciximab group and 4.5% in the stent alone group, P=0.16). Despite a relative reduction in mortality >50%, this difference was not significant due to the small sample population. The 1-month benefit was not maintained at the 6-month follow-up, and this figure may be explained by the high incidence of late TVR in both groups and the lack of effect of abciximab on late in-stent restenosis. Late TVR was the largely dominant component of the composite endpoint and overcame the other two hard endpoints of death and non-fatal reinfarction preventing a significant difference in outcome between groups.

In the Abciximab before Direct angioplasty and stenting in Myocardial Infarction Regarding Acute and Long term follow-up (ADMIRAL) trial, 149 patients with AMI were randomized to abciximab plus infarct artery stenting and 151 to stenting alone.¹⁵ The 1-month mortality rate was 3.4% in the abciximab group and 6.6% in the stent alone group. Again, despite a relative reduction in mortality of nearly 50%, this difference did not reach significance (P=0.19), and this was due to the small patient cohort. The population enrolled in this study had a more ‘realistic’ risk profile when compared with those of previous randomized trials of primary coronary angioplasty in AMI. The primary endpoint of the study was the composite of death, reinfarction, or urgent target vessel revascularization at 30 days. Urgent target vessel revascularization was defined as repeat PCI performed within 24 h after a new ischaemic episode. The difference in the primary endpoint rates (6.0% abciximab group and 14.6% placebo group, P=0.01) was driven by the difference in the incidence of urgent target revascularization (1.3% abciximab group and 6.6% in the placebo group, P=0.01). The differences in the other two components of the endpoint, death and reinfarction, both favoured abciximab but did not reach significance. Different from the ISAR-2 trial, the non-inclusion of late non-urgent TVR in the endpoint allowed the benefit of abciximab treatment maintained at the 6-month follow-up: the primary endpoint rate resulted halved in the abciximab group when compared with the placebo group (7.4 vs. 15.9%, respectively, P=0.02) with a remarkable relative risk value of 0.46 (95% CI 0.22–0.93). The benefit was superior in those patients who received early abciximab administration (RR=0.12, 95% CI 0.01–0.98) when compared with patients randomized in the coronary care unit or in the catheterization laboratory. The ADMIRAL investigators explained this finding with a higher incidence of a pre-procedural TIMI grade 3 flow in patients randomized to abciximab (16.8 vs. 5.4%, P=0.01), suggesting an effect of abciximab on the early reopening of the target vessel and increased myocardial salvage.

The largest randomized trial comparing primary coronary angioplasty and primary coronary stenting with and without abciximab administration is the Controlled Abciximab and Device Investigation to Lower Late Angioplasty Complications (CADILLAC) trial.¹⁶ This trial included 2082 patients who were randomized to four arms: conventional angioplasty and provisional stenting with and without adjunct abciximab treatment and primary infarct artery stenting with and without adjunct abciximab treatment. The study was primarily designed to test the effect of routine infarct artery stenting on outcome, despite four concluded randomized studies had already shown infarct artery stenting to be highly superior to conventional coronary balloon angioplasty.^17–20 The primary endpoint of the study was the 6-month composite of death, reinfarction, disabling stroke, and ischaemic TVR. The study confirmed the superiority of routine stenting over conventional angioplasty with provisional stenting: the primary endpoint rate was 20% in the angioplasty group, 16.5% in the angioplasty plus abciximab group, 11.5% in the stent group, and 10.2% in the stent plus abciximab group. In the two balloon angioplasty arms, there was a benefit of abciximab treatment at 1-month follow-up. The rate of the composite of death, reinfarction, TVR, and stroke was 4.8% in the abciximab arm and 8.8% in the placebo arm (P<0.02), but this benefit was not maintained at 6 months due to the overwhelming effect of late restenosis and subsequent TVR (primary endpoint rate 20% in the placebo arm and 16.5% in the abciximab arm). The comparison between the two groups of stent alone (516 patients) and stent plus abciximab (529 patients) did not show any clinical benefit of abciximab in patients undergoing routine infarct artery stent implantation. The incidence of the primary endpoint was 10.2% in the abciximab treated patients and 11.5% in the stent alone group, and no differences were revealed for each individual component of the composite endpoint. In the overall population, the randomization to abciximab was associated with a significant reduction in 1-month occurrence of death, reinfarction, TVR, or stroke (4.6 vs. 7.0%, relative risk reduction 0.65, 95% CI 0.46–0.93, P=0.01), but at 12 months, abciximab was no longer associated with significant difference in the composite endpoint (16.9 vs. 18.4%, P=0.29). The main criticism raised from the results of this trial is the low risk of the enrolled population and, as a consequence, the impossibility to translate the trial results to the ‘real world’ AMI patients. The 6-month mortality rate was very low for all four arms (4.5% in the angioplasty alone arm, 2.5% in the angioplasty plus abciximab arm, 3.0% in the stent alone arm, and 4.2% in the stent plus abciximab arm), as was the median age of the four groups (59, 60, 60, and 59 years, respectively). Furthermore, patients with CS were excluded, and the baseline left ventricular function of the enrolled patients was nearly normal (median value >56% in all groups). Finally, only in a minority of patients, the left anterior descending artery was the infarct vessel (38.7%). All these figures make the enrolled population poorly representative of the real world of AMI. Moreover, exclusion criteria included high-risk coronary anatomy such as ostial target lesion, left main disease, calcification, tortuosity, a major branch involved in the target lesion, small target vessel, and all conditions where it could be presumed a difficult placement of a stent and a high risk of early target vessel failure due to thrombosis. In fact, 599 of patients with AMI eligible for the study were excluded from enrolment (22% of all eligible patients). Most patients excluded from randomization had a high-risk anatomy and had a worse baseline risk profile when compared with enrolled patients (Turco MA et al., 52nd ACC Session, Chicago, 2003). Patients included in the registry were older (median age 69 vs. 59 years, P=0.002) and had a greater incidence of previous myocardial infarction (20 vs. 14%, P=0.001), previous coronary surgery (12 vs. 2%, P<0.0001), and more severe left ventricular dysfunction (median left ventricular ejection fraction 45%). The differences in baseline characteristics resulted in a poorer but more realistic outcome in the registry patient cohort when compared with the outcome of randomized patients (in-hospital mortality 4.0 vs. 1.5%, P=0.001; in-hospital repeat PCI 14 vs. 2.5%, P<0.001).

The Abciximab and Carbostent Evaluation (ACE) trial assigned 400 patients with ST-segment elevation AMI to undergo infarct artery stenting alone or stenting plus abciximab.²¹ The primary endpoint of this study was a composite of death, reinfarction, TVR, and stroke at 1 month. Key secondary endpoints of the study were effectiveness of reperfusion as assessed by early ST-segment resolution and infarct size as assessed by 1-month ^99mTc-sestamibi scintigraphy.

The incidence of the primary endpoint was lower in the abciximab group when compared with that in the stent alone group (4.5 and 10.5%, respectively, P=0.023), and randomization to abciximab was independently related to the risk of the primary endpoint (OR 0.41, 95% CI 0.17–0.97, P=0.041). Early ST-segment resolution was more frequent in the abciximab group (85 vs. 68%, P<0.001). Infarct size, as assessed by 1-month ^99mTc-sestamibi scintigraphy, revealed smaller infarcts in the abciximab group (12.50% of left ventricular size vs. 16.60%).

At 1 year, the cardiovascular mortality rate was 5.1% in the abciximab group and 10.7% in the stenting alone group (P=0.040), with a relative difference in mortality of 47.6%.²² The two Kaplan–Meier survival curves were nearly identical in the first month and progressively diverged during the following 11 months as a consequence of the increased mortality in the stent alone group.¹¹ All, except one, cardiovascular deaths were cardiac (a fatal non-haemorrhagic stroke occurred in a patient of the stent alone group). In the abciximab group, seven out of 10 deaths occurred within 1 month of randomization, whereas in the stenting alone group, the majority of the deaths occurred after 1 month from randomization (12 out of 21 deaths). Heart failure accounted for the majority of deaths in both groups (five deaths in the abciximab group and 14 in the stenting alone group were due to heart failure). The 1-month difference in reinfarction rates (0.5% in the abciximab group and 4.5% in the stenting alone group) was maintained at 1 year (1% in the abciximab group and 6% in the stenting alone group, P=0.006). As expected, there were no differences between groups in TVR rates (16.7% in the abciximab group and 17.8% in the stenting alone group) at 1-year follow-up.

Several unique features of this trial should be highlighted. In this study, the mean age was higher when compared with previous studies (median age 63 years in the stenting alone group and 64 years in the abciximab group), more than one-third of patients were over 70 years, 66% were at ‘not low-risk’ according to the TIMI criteria, the more frequent location of the AMI was anterior, diffuse disease or multiple lesions within the infarct artery resulted in multiple stent implantation in 25% of patients, patients with CS on admission were included, as well as patients with coronary anatomy at high risk. The study design resulted in the enrolment of a population whose characteristics are consistent with those of large survey studies on AMI.^23,24 Thus, the study sample can be considered representative of the entire population of patients with ST-segment elevation AMI.

The ACE trial confirmed that abciximab provides a benefit related to the decrease in post-PCI early coronary complications such as myocardial infarction or the need for urgent TVR in the early phase of AMI, and this effect may be easily and directly related to the glycoprotein IIb/IIIa receptor blockade. However, the most intriguing figure of the study is the demonstration of a significant decrease in long-term mortality due to heart failure in patients randomized to abciximab. As shown by the secondary endpoints of the study, abciximab provides a better myocardial reperfusion with the potential for reduction of microvessel disruption due to embolization and of reperfusion injury. The improved myocardial salvage in the early phase may translate into improved survival in the long-term follow-up.

The pathophysiologic endpoints related to the effectiveness of reperfusion (more frequent early ST-segment resolution) and myocardial salvage (smaller infarct size) suggest a benefit of abciximab beyond the effects on the epicardial vessel and are consistent with the results of a mechanistic randomized trial from the Munich group based on a sample of 200 patients treated by stenting and randomized to abciximab or placebo.²⁵

This study showed that abciximab may prevent or reduce microvascular damage after reperfusion and improve coronary flow, coronary reserve, and left ventricular function after reperfusion, as assessed by intracoronary Doppler measurements and left ventricular angiography. Moreover, at 1 month, the incidence of the composite of death, myocardial infarction, and TVR was lower in the abciximab group when compared with the placebo group: two major adverse events occurred in the abciximab group and nine in the control group (OR 0.20, 95% CI 0.04–0.94, P=0.031).

The effects on microvessel network may be related to the prevention or reduction of platelet microvascular plugging by inhibition of glycoprotein IIb/IIIa receptors, as well as neutrophil activation by inhibition of alfaMbeta2 receptors and Bbeta3 receptors that are present on granulocytes and monocytes.^26,27 The inhibitory effects of abciximab also include one of the most important platelet mediators, the CD40 ligand, which is principally expressed in monocytes, macrophages, and endothelial cells, and plays a major role in the unleashing of inflammation and production of interleukins and chemochines.²⁸

After the conclusion of the ACE trial, a preliminary analysis of the pooled data from RAPPORT, ISAR-2, ADMIRAL, CADILLAC, and ACE trials was performed by Topol et al.²⁹ The analysis includes 3666 patients (1843 patients randomized to abciximab and 1823 to placebo). This analysis shows a robust benefit of abciximab at 1-month follow-up with a reduction of >45% in the composite of death, myocardial infarctions, and TVR (4.8 vs. 8.8%), of 33% in the composite of death and myocardial infarction (3.2 vs. 4.8%), and of 26% in mortality (2.3 vs. 3.1%). The benefit of abciximab in terms of significant decrease in 1-month and long-term mortality in patients undergoing PCI for AMI was demonstrated by a more recent and complete meta-analysis: 1-month mortality 2.4 vs. 3.4% (P=0.047), 6–12-month mortality 4.4 vs. 6.2% (P=0.01).³⁰

Similar to other effective therapies, in which benefit parallels the risk of the patient, the potential benefits of abciximab on the effectiveness of mechanical reperfusion may be revealed more easily in high-risk AMI patients, because it may be difficult or even impossible to show the benefit of a strong treatment such as abciximab in low-risk patients. This is a crucial issue for a correct interpretation of the results of concluded trials and assessment of the efficacy of abciximab in patients undergoing routine infarct artery stent implantation for AMI.

Conclusions

The frequentist approach in randomized study designs and the inclusion of a large number of patients (high statistical inference) have frequently produced inconclusive, or conflicting, or negative results despite the high potential for a strong impact on outcome of the study drug, or device, or strategy. In many studies, the adoption of rigid criteria for randomization resulted in the exclusion of many patients at high risk of adverse events and prevented the correct assessment of the efficacy of the study treatment. Other confounding variables, which may explain the conflicting or inconclusive results of studies assessing the same treatment, may be related to other major design limitations, such as the formulation of an erroneous hypothesis for the study endpoint, or the use of composite endpoints including adverse events that are not related to the study treatment, and at the same time are so far the more frequent events and the decisive factor of patient outcome. New studies should use a ‘bayesian’ approach instead of a frequentist approach and should be based on ‘realistic’ or even high-risk populations.

Conflict of interest: none declared.

References