Abstract
Background:
Dual antiplatelet therapy is the standard of care for acute coronary syndrome, but uncertainty exists regarding the optimal regimen for patients in North America. We sought to compare the effectiveness and safety of acetylsalicylic acid (ASA) and ticagrelor or clopidogrel in patients with acute coronary syndrome from a single tertiary academic centre in Montréal, Canada.
Methods:
We conducted a pragmatic, open-label, time-clustered (bimonthly between October 2018 and March 2021), randomized controlled trial. The primary effectiveness end point was a composite of all-cause mortality, nonfatal myocardial infarction, or ischemic stroke. The primary safety end point was hospital admissions for bleeding. We ascertained 12-month outcomes from the Quebec universal electronic health databases. We designed and analyzed the study within a Bayesian paradigm to supplement existing knowledge. The primary analysis was a Bayesian logistic regression model with an informed focused prior from previously randomly assigned North American patients. Robustness was evaluated with vague and other prespecified informative priors, spanning reasonable pre-existing beliefs. We defined clinically important benefits and harms as risk reductions exceeding a 10% difference.
Results:
We randomly assigned 1005 patients with acute coronary syndrome to ticagrelor (n = 450) or clopidogrel (n = 555). Major acute cardiovascular events occurred in 50 (11.1%) patients assigned to ticagrelor and 64 (11.5%) assigned to clopidogrel (relative risk [RR] 0.95, 95% credible interval 0.67–1.35, with a vague prior). The primary analysis with an informed focused prior resulted in probabilities of a clinically meaningful ticagrelor benefit (RR < 0.9), equivalence (0.9 ≤ RR ≤ 1.1) or harm (RR ≥ 1.1) of 2%, 41%, and 57%, respectively. For the safety end point, there was no consistent signal of benefit or harm with ticagrelor. Sensitivity analyses with a range of prior beliefs gave generally consistent results.
Interpretation:
Whether we analyzed this trial with a vague or a range of reasonable informed priors, we found no strong evidence for the superiority of ticagrelor over clopidogrel in North American patients. Current guidelines favouring ticagrelor over clopidogrel might take this new evidence into future consideration.
Trial registration:
Dual antiplatelet therapy (DAPT), the combination of a P2Y12 receptor inhibitor and acetylsalicylic acid (ASA), is the contemporary treatment strategy for the prevention of secondary ischemia-related events in patients with acute coronary syndrome. According to clinical guidelines1–3 the P2Y12 inhibitor ticagrelor is favoured over clopidogrel. The strength of this endorsement is based on the Platelet Inhibition and Patient Outcomes (PLATO) randomized controlled trial (RCT)4 performed in 862 study centres across 43 countries that reported a 16% reduction (hazard ratio 0.84, 95% confidence interval [CI] 0.77–0.92; p < 0.001) in major acute cardiovascular events (MACE) with ticagrelor, without any excess in major bleeding events.
The lack of reproducibility in medical research is a major concern; 5 the credibility of a single study, even a multicentre one, may therefore be limited. Thus, replication studies are important. Although PLATO included many centres, the researchers incorrectly assumed homogeneity across regions. The estimates of effectiveness in PLATO varied across the study regions, as the 1814 patients in North America (NA) showed a statistically nonsignificant increase in MACE (relative risk [RR]NA 1.25, 95% CI 0.93–1.67)4,6 with ticagrelor and a statistically significant deviation from the overall PLATO pooled result (p = 0.045). A Bayesian hierarchical (random effects) model accounting for the regional variability provides a compromise between the pooled and individual geographical rates by a structured borrowing of information. Under this model, the PLATO ticagrelor risk estimate for patients in North America moved toward the PLATO global mean (RRNA = 1.13 [95% CI 0.75–1.47]), but clinically important uncertainty about ticagrelor benefit or harm in these patients remains.7 People may have different prior beliefs about which data to consider or which statistical model best describes the underlying data-generating mechanism, but if these different beliefs produce different inferences, then the data cannot be considered robust. This provided the justification for the RCT we designed to resolve the uncertainties about the comparative effectiveness of DAPT in a North American population with acute coronary syndrome.
We sought to compare the effectiveness and safety of ASA and ticagrelor or clopidogrel in patients with acute coronary syndrome from a single tertiary academic centre in Montréal, Canada.
Methods
Trial design and randomization
The Ticagrelor Compared to Clopidogrel in Acute Coronary Syndromes trial (TC4) (NCT04057300, https://clinicaltrials.gov/study/NCT04057300) was a pragmatic, open-label, active-control, time-clustered RCT designed to assess the effectiveness and safety of ticagrelor versus clopidogrel as DAPT therapy in a population with acute coronary syndrome undergoing percutaneous coronary intervention in a single tertiary academic centre in Montréal, between October 2018 and March 2021. Randomization followed a time-cluster randomized design.8 In summary, all patients with acute coronary syndrome arriving to the McGill University Health Centre (MUHC) during a 2-month cluster received the scheduled DAPT for that period. We determined drug assignment for the first exposure period by a random number with simple alternation for subsequent 2-month drug assignment periods. The COVID-19 pandemic severely hampered recruitment efforts in 2020, particularly disrupting ticagrelor recruitment clusters. Therefore, we added an extra 2-month ticagrelor recruitment period before the study end. Given the large number of potential health care physicians (from the emergency department, cardiac catheterization suite, cardiology consultation, and coronary care unit) involved with acute coronary syndrome care, this time-clustering randomization simplified administrative hurdles and made the trial feasible. Given the acute care population studied, we did not expect the unblinded exposure assignment to bias the results.
Trial oversight
The trial was investigator initiated, sponsored by a public funder (Canadian Institutes of Health Research [reference no. PJT-156344]). The trial was funded in 2018, and the protocol is available online (https://www.brophyj.com/upload/TC4protocol.pdf). We followed the Consolidated Standards of Reporting Trials reporting checklist. All statistical code is freely available online (https://github.com/brophyj/TC4-code), but ethical restrictions from the Quebec provincial health authorities prohibit the sharing of individual patient–level outcome data.
Patients
We approached consecutive patients with acute coronary syndrome admitted to the MUHC emergency department, cardiology, or intensive care services between Oct. 1, 2018, and Mar. 31, 2021, and undergoing cardiac angiography, to participate. We achieved this by examining the cardiac catheterization roster daily. Specifically, patients with ST- and non–ST-elevation myocardial infarction (MI) with positive biomarkers and patients with a diagnosis of unstable angina were eligible for study inclusion. A research nurse obtained informed consent after confirmation from the treating physician that DAPT was the appropriate treatment. Exclusion criteria included lack of patient-specific clinical equipoise by the attending physician, patient refusal, prior intolerance to either drug, recent (< 30 d) previous hospital admission for acute coronary syndrome, and out-of-province residency.
Exposure
We randomly assigned patients to receive ticagrelor or clopidogrel depending on the date of their hospital admission for acute coronary syndrome. In accordance with guidelines1–3 and previous trials,4 we advised physicians to prescribe a 180-mg loading dose followed by a 90-mg twice-daily dose of ticagrelor, or a 300-mg loading dose followed by a daily dose of 75 mg of clopidogrel to their patients. Both therapies were to be accompanied by a 325-mg loading dose and 81-mg daily dose of ASA, with patients encouraged to take their medications for 12 months following hospital discharge. Beyond the choice of DAPT regimen, all other treatment decisions were made by treating physicians, independently of this study.
Outcome
Similar to the PLATO trial,4 the primary effectiveness measure was a 3-point MACE outcome — a composite of all-cause mortality, nonfatal MI (International Statistical Classification of Diseases and Related Health Problems 10th Revision [ICD-10] codes I21.X, I22.X, I23.X, or I25.2), or ischemic stroke (ICD-10 codes H34.1, I63.X, I64.X, or I67.X) — within 12 months of the index hospital admission for acute coronary syndrome. The primary safety outcome was a composite of gastrointestinal bleeds (ICD-10 code K92.X) or hemorrhagic strokes (ICD-10 codes I60.X, I61.X, or I62.X) requiring hospital admission. The secondary outcomes of interest were the individual MACE and safety outcomes. We extracted the outcomes electronically from Quebec medico-administrative hospital databases using ICD-10 codes and death certificates from hospital and provincial health records, which have been previously validated.9–11 We performed outcome data extraction blinded to treatment assignment.
Bayesian statistical design
In addition to specifying type I and type II error rates, frequentist sample size methods depend on point estimates and standard deviations of the outcomes in the intervention and control groups, which are typically not accurately known at the study design stage. Bayesian methods offer several advantages, including the following: input uncertainties are represented by prior densities rather than point estimates, with uncertainty resolved through the progressive accumulation of new evidence; this incorporation of existing prior knowledge with new data follows the laws of probability via Bayes’ theorem; the resulting posterior distributions allow multiple direct probability statements to be formulated about clinically meaningful benefits, harm, or practical equivalency; and statistical penalties are not required for Bayesian sequential data analyses because posterior probabilities computed at the moment of stopping the trial are perfectly calibrated.12
Given the existence of previous, but inconclusive, evidence regarding the choice of DAPT in North American patients with acute coronary syndrome,4,6 we designed this Bayesian study to leverage the best available prior evidence with new data from the current study, to allow the refinement of both efficacy and safety estimates with a smaller sample compared with standard non-Bayesian designs. Using the PLATO4,6 regional North American cohort as an informed focused prior (RRNA 1.25, 95% CI 0.93–1.67) and assuming the same PLATO event rates in North America, we calculated that a study with 1000 patients would result in a 96% posterior probability of an increased ticagrelor risk and an 83% probability that the difference exceeded an absolute 1% difference in MACE. We believed these projected changes would be clinically important, and that adding data from 1000 newly randomly assigned patients to the existing 1800 randomly assigned North American patients from PLATO would provide a clinically meaningful addition to the evidence base.
Bayesian priors
Because our sample size was calculated by incorporating existing prior evidence, we used prespecified Bayesian methods updating prior beliefs with the current data in our analyses. The resulting posterior distribution is a weighted mean of the prior and new data. The probabilities for clinically meaningful harm, benefit, or equivalence are directly proportional to the area under the posterior probability density function.
Bayesian analyses with both vague (enabling current study data to dominate) and informative priors are useful in most studies but especially those with modest sample sizes. Priors have historically been considered a weakness of Bayesian analyses, but this can be mitigated by acknowledging that clinical beliefs vary13 and considering a community of priors, which provides a test of robustness.
The priors we considered were a “vague” prior represented by a Student t distribution around the null effect (RR = 1.0), with 3 degrees of freedom and a standard deviation of 5, and 4 informative priors. The first informative prior was a “focused” prior (as noted in our 2018 funding protocol [https://www.brophyj.com/upload/TC4protocol.pdf], this prior was prespecified as our primary Bayesian analysis) (defined relative to our study population) from the MACE and bleeding estimates of the North American PLATO population (RRMACE = 1.25, 95% CI 0.93–1.67; RRBleed = 1.05, 95% CI 0.76–1.45). The second was an “enthusiastic” prior (defined relative to previous ticagrelor benefit) using the overall PLATO MACE (RRMACE = 0.84, 95% CI 0.77–0.92) and bleeding estimates (RRBleed = 1.04, 95% CI 0.95–1.13). The third was a “modified enthusiastic” prior that recognizes the geographic heterogeneity of MACE results within PLATO and uses a hierarchical model with an increased standard deviation (RRMACE = 0.84, 95% CI 0.57–1.11; geographic safety bleeding data are not available).7 The fourth was a “summary” prior13 elicited from a Bayesian network meta-analysis of 17 previously published RCTs involving 57 814 participants14 (RRMACE = 0.95, 95% credible interval 0.81–1.14; RRBleed = 1.07, 95% credible interval 0.99–1.17). In addition to these population-level treatment-effect priors, we used noninformative priors for the nuisance parameters, a Student t prior, with 3 degrees of freedom (Student_t [3, 0, 2.5]) for the standard deviation (SD) and a Lewandowski–Kurowicka–Joe uniform distribution (η = 1) for the correlation structure of the cluster levels.
Statistical analysis
We summarized the baseline characteristics of study participants using means and standard deviations for continuous variables and proportions for categorical groups. Analyses examined the time to the first occurrence of a previously specified outcome with censoring at 1 year after randomization. As there were no differences between survival and logistic analyses,15 we report results from the logistic models. We also fitted both fixed-effect (pooled sample) and hierarchical models16 that account for the time-clustered randomization.
Bayesian analyses provide posterior distributions for each parameter, thereby providing estimates not only of their mean but also of their variability, commonly expressed as credible, or probability, intervals. Consequently, we were able to evaluate not only the probability of one treatment being better than the other, but also the probability that it exceeded the smallest change in an outcome that is considered worthwhile, often referred to as the minimal clinically important difference (MCID).17 While these thresholds are arbitrary, they can be used to enhance the appreciation of the posterior treatment distributions. For example, we report MCID probability statements based on a ± 10% change in the RRs as well as the probability for the range of practical equivalence (ROPE), where the posterior RRs fall between 0.9 and 1.11. The full posterior distributions permit the evaluation of differing clinical thresholds.
We performed Bayesian statistical inference using the brms R package,18 a front-end wrapper for the Stan probabilistic programming language,19 which samples posterior distributions with the No-U-Turn Sampler,20 an extension of the Hamiltonian Monte Carlo algorithm. We used 3 Hamiltonian Monte Carlo chains, each a minimum of 10 000 iterations with a 5000 burn-in period to produce a total of 15 000 posterior samples. We monitored all samples to ensure their convergence. We conducted model comparisons using the leave-one-out cross-validation, to evaluate model fit.20 The leave-one-out cross-validation evaluation suggested that the pooled model, when using a vague prior, had a better overall fit to the data than the hierarchical model.15 Consequently, unless otherwise specified, results refer to this pooled model.
We performed all analyses within the RStudio integrated developmental environment (RStudio Team, Integrated Development for R, PBC, http://wwwrstudiocom/) using the R programming language (https://www.r-project.org/) and followed Bayesian analysis reporting guidelines.21,22
Ethics approval
The trial was approved by the MUHC Research Ethics Board.
Results
Between Oct. 1, 2018, and Mar. 31, 2021, 1005 patients with acute coronary syndrome were randomly assigned and analyzed (Figure 1) from 13 (7 clopidogrel and 6 ticagrelor) 2-month cluster periods and 1 (ticagrelor) 4-month cluster period. Follow-up, using the provincial electronic health care databases, ended in April 2022 (Figure 2); no patients were lost to follow-up. The baseline characteristics of the clopidogrel (n = 555) and ticagrelor (n = 450) DAPT groups were generally well balanced (Table 1), with a mean age of 67 years and 75% males. All patients had a 12-month follow-up unless a terminating outcome occurred before this time (Figure 3). The number of patients who declined enrolment in the study was proportional to the numbers recruited to each group (244 [56%] in clopidogrel months and 192 [44%] ticagrelor months).
Figure 1:
Recruitment timeline for randomization of group assignments from October 2018 to March 2021. See Related Content tab for accessible version.
Figure 2:
Flow chart of the study participants. See Related Content tab for accessible version.
Table 1:
Baseline characteristics of the study population, by dual antiplatelet therapy
| Characteristic | No. (%) of patients* | |
|---|---|---|
| Clopidogrel n = 555 |
Ticagrelor n = 450 |
|
| Age, yr, mean ± SD | 67.5 ± 10.9 | 65.2 ± 11.32 |
| Sex, male | 420 (75.7) | 338 (75.1) |
| Height, cm, mean ± SD | 170.6 ± 9.5 | 171.0 ± 9.3 |
| Weight, kg, mean ± SD | 83.1 ± 22.0 | 83.3 ± 17.8 |
| Smoking status | ||
| Never | 192 (34.6) | 181 (40.2) |
| Former, daily | 224 (40.4) | 157 (34.9) |
| Current, daily | 134 (24.1) | 110 (24.4) |
| Race | ||
| White | 451 (81.3) | 376 (83.6) |
| Other | 104 (18.7) | 74 (16.4) |
| Previous DAPT | ||
| None | 409 (73.7) | 341 (75.8) |
| Clopidogrel | 137 (24.7) | 88 (19.6) |
| Ticagrelor | 6 (1.1) | 17 (3.8) |
| Prasugrel | 0 (0.0) | 1 (0.2) |
| ACS diagnosis | ||
| STEMI | 116 (20.9) | 94 (20.9) |
| NSTEMI | 210 (37.8) | 207 (46.0) |
| Unstable angina | 228 (41.1) | 148 (32.9) |
| Hypertension | 387 (69.7) | 300 (66.7) |
| SBP, mean ± SD | 140.6 ± 22.2 | 140.0 ± 22.6 |
| DBP, mean ± SD | 79.7 ± 13.7 | 80.4 ± 15.0 |
| Heart rate, mean ± SD | 72.9 ± 15.4 | 72.4 ± 15.1 |
| Dyslipidemia | 376 (67.7) | 301 (66.9) |
| Diabetes | 185 (33.3) | 139 (30.9) |
| Type 2 | 168 (90.8) | 130 (93.5) |
| Previous MI | 159 (28.6) | 120 (26.7) |
| Previous PCI | 144 (25.9) | 114 (25.3) |
| Congestive heart failure | 32 (5.8) | 15 (3.3) |
| Previous CABG | 77 (13.9) | 32 (7.1) |
| Previous stroke | 27 (4.9) | 14 (3.1) |
| History of PAD | 5 (0.9) | 2 (0.4) |
| Serum creatinine, median (IQR) | 83.0 (71.0–97.0) | 83.0 (71.0–97.0) |
| COPD | 97 (17.5) | 64 (14.2) |
| Troponin, median (IQR) | 205 (16–2440) | 416 (25–2521) |
| Missing | 144 (25.9) | 85 (18.9) |
Note: ACS = acute coronary syndrome, CABG = coronary artery bypass grafting, COPD = chronic obstructive pulmonary disease, DAPT = dual antiplatelet therapy, DBP = diastolic blood pressure, IQR = interquartile range, MI = myocardial infarction, NSTEMI = non–ST elevation myocardial infarction, PAD = peripheral arterial disease, PCI = percutaneous coronary intervention, SBP = systolic blood pressure, SD = standard deviation, STEMI = ST-segment elevation myocardial infarction.
Unless stated otherwise.
Figure 3:
Kaplan–Meier survival curves for the clopidogrel and ticagrelor arms from the Ticagrelor Compared to Clopidogrel in Acute Coronary Syndromes trial (TC4) showing no difference in survival outcomes for major adverse cardiac events between the 2 treatments. The shaded areas represent 95% confidence intervals.
Effectiveness outcomes
The TC4 MACE effectiveness outcome occurred at similar rates in both exposure groups (clopidogrel, 11.5%; ticagrelor, 11.1%) and the RR with a vague prior was 0.95 (95% credible interval 0.67–1.35) (Table 2). When using a vague prior, this equated to a 37% probability that ticagrelor is responsible for a clinically meaningful reduction (RR < 0.9) in MACE, a 21% chance of a clinically meaningful increase (RR > 1.1), and a 42% probability of being within the ROPE, when compared with clopidogrel (Table 2).
Table 2:
Effectiveness outcomes in the Ticagrelor Compared to Clopidogrel in Acute Coronary Syndromes trial (TC4) (raw TC4 data, prior and posterior distributions, and probabilities for clinically meaningful benefits, equivalence, and clinically meaningful harms)
| Outcome | TC4; no. (%) | Prior distributions, mean (95% CI) | Posterior distributions | Posterior probabilities | |||
|---|---|---|---|---|---|---|---|
| Clopidogrel n = 555 |
Ticagrelor n = 450 |
Relative risk (95% credible interval) | MCID benefit (PrHR < 0.9) | ROPE (PrHR[0.9, 1.1]) | MCID harm (PrHR > 1.1) | ||
| MACE | 64 (11.5) | 50 (11.1) | Vague 0.0 (−15.9 to 15.9) |
0.95 (0.67 to 1.35) | 0.37 | 0.42 | 0.21 |
| Focused* 1.23 (1.01 to 1.45) |
1.12 (0.91 to 1.39) | 0.02 | 0.41 | 0.57 | |||
| Enthusiastic 0.84 (0.76 to 0.92) |
0.85 (0.78 to 0.93) | 0.89 | 0.11 | 0.00 | |||
| Modified enthusiastic 0.84 (0.57 to 1.11) |
1.03 (0.73 to 1.44) | 0.22 | 0.35 | 0.43 | |||
| Summary† 0.97 (0.72 to 1.22) |
0.99 (0.69 to 1.38) | 0.30 | 0.44 | 0.26 | |||
| All-cause mortality | 15 (2.7) | 7 (1.6) | Vague 0.0 (−15.9 to 15.9) |
0.60 (0.24 to 1.39) | 0.83 | 0.09 | 0.08 |
| Focused 1.24 (0.61 to 1.87) |
1.05 (0.69 to 1.58) | 0.24 | 0.35 | 0.41 | |||
| Enthusiastic 0.79 (0.50 to 1.08) |
0.75 (0.54 to 1.06) | 0.78 | 0.11 | 0.11 | |||
| Summary 0.86 (0.60 to 1.11) |
0.67 (0.27 to 1.52) | 0.76 | 0.11 | 0.13 | |||
| MI | 46 (8.3) | 38 (8.4) | Vague 0.0 (−15.9 to 15.9) |
1.02 (0.67 to 1.54) | 0.29 | 0.37 | 0.35 |
| Focused‡ 1.24 (0.93 to 1.55) |
0.99 (0.61 to 1.60) | 0.35 | 0.32 | 0.33 | |||
| Enthusiastic 0.83 (0.73 to 0.93) |
0.90 (0.81 to 1.01) | 0.48 | 0.52 | 0.0 | |||
| Summary 0.96 (0.74 to 1.18) |
1.05 (0.70 to 1.54) | 0.23 | 0.36 | 0.41 | |||
| Stroke | 3 (0.5) | 5 (1.1) | Vague 0.0 (−15.9 to 15.9) |
1.99 (0.52 to 7.88) | 0.12 | 0.07 | 0.81 |
| Focused‡ 1.73 (−1.93, 5.40) |
1.93 (0.75 to 5.07) | 0.06 | 0.07 | 0.87 | |||
| Enthusiastic 1.17 (0.82 to 1.52) |
1.20 (0.93 to 1.55) | 0.01 | 0.24 | 0.75 | |||
| Summary 1.02 (0.71 to 1.33) |
1.11 (0.94 to 1.32) | 0.01 | 0.43 | 0.56 | |||
| Bleeding | 28 (5.0) | 20 (4.4) | Vague 0.0 (−15.9 to 15.9) |
0.89 (0.52 to 1.52) | 0.51 | 0.26 | 0.23 |
| Focused 1.05 (0.60 to 1.50) |
1.03 (0.78 to 1.36) | 0.17 | 0.51 | 0.32 | |||
| Enthusiastic 1.04 (0.94 to 1.14) |
1.03 (0.95 to 1.12) | 0.00 | 0.93 | 0.07 | |||
| Summary 1.16 (0.85 to 1.47) |
1.07 (0.99 to 1.16) | 0.00 | 0.74 | 0.26 | |||
Note: CI = confidence interval, MACE = major adverse cardiac events, MCID = minimally clinically important difference, MI = myocardial infarction, PLATO = Platelet Inhibition and Patient Outcomes, ROPE = region of practical equivalence.
The “focused” prior considers only the PLATO participants in North America (Canada and the United States).
The “summary” prior considers all randomized controlled trials from a network meta-analysis of ticagrelor versus clopidogrel.
The “focused” prior for MI and stroke are based on PLATO US data alone as separate Canadian data could not be found.
However, our prespecified primary Bayesian analysis was with a “focused” informative prior, based on the patients in North America previously randomly assigned in PLATO. This resulted in a posterior mean MACE RR of 1.12 (95% credible interval 0.91–1.39) with the probability of clinically meaningful ticagrelor benefit of 2%, probability for harm of 57%, and being within the ROPE of 41% (Table 2). These results are also displayed graphically in Figure 4, which shows that the posterior distribution is a weighted mean of the prior and new data from this trial.
Figure 4:
Graphical display showing the posterior distribution as a weighted mean of the prior and new data from this trial. The probabilities for clinically meaningful harm, benefit, or equivalence are directly proportional to the area under the posterior probability density. Note: PLATO = Platelet Inhibition and Patient Outcomes trial, prior = focused prior limited to PLATO patients in North America, RR = relative risk, TC4 = Ticagrelor Compared to Clopidogrel in Acute Coronary Syndromes trial.
Posterior MACE probabilities employing the “enthusiastic” prior of all PLATO data shifted the posterior distribution to more favourable ticagrelor inferences, with an 89% probability of a clinically meaningful improvement. Using all the PLATO data, but accounting for the observed regional heterogeneity as a “modified enthusiastic” prior, the probability of a ticagrelor clinical benefit was only 22%, with a twofold higher probability (43%) of clinical harm.
Posterior MACE probabilities with the “summary” prior, which was not restricted to the single PLATO study, but rather included all high-quality prior RCT evidence, gave less encouraging findings for ticagrelor benefits (Table 2); the MACE posterior distribution mean was RR 0.99 (95% credible interval 0.69–1.38) with a 30% probability of clinically meaningful ticagrelor benefit, but also a 26% probability of clinically meaningful ticagrelor harm relative to clopidogrel and a 44% probability of clinical equivalency.
Safety outcomes
With a vague prior, major bleeding events requiring hospital admission did not substantially differ across the TC4 treatment groups (clopidogrel, 5.0%; ticagrelor, 4.4%; RR 0.89 [95% credible interval 0.52–1.52]). This translated into a 51% probability for a clinically meaningful reduction in bleeding (RR < 0.9) with ticagrelor, a 26% probability for clinical equivalence (0.9 ≤ RR ≤ 1.1), and a 23% probability for an increase (RR > 1.1) in bleeding. As expected, conclusive safety inferences from this study alone were not possible. However, the integration of bleeding events from this study with the other priors (Table 2) showed only a small chance of a clinically meaningful reduction in bleeding with ticagrelor (0%–17%), a moderate-to-high probability of clinical equivalence (51%–93%), and a modest probability of a clinically meaningful increase in bleeding rates (7%–32%).
Secondary outcomes
There were fewer all-cause deaths identified in the ticagrelor group (1.6% v. 2.7%) than the clopidogrel arm (RR 0.60, 95% credible interval 0.24–1.39). There was no difference for MI events between patients treated with ticagrelor and clopidogrel (8.4% v. 8.3%; RR 1.02, 95% credible interval 0.67–1.55). Finally, we observed a similar number of ischemic strokes in the 2 groups (clopidogrel, n = 3 v. ticagrelor, n = 5) (RR 1.99, 95% credible interval 0.52–7.88). The power of this trial to make meaningful inferences for these secondary outcomes was limited. However, the primary analysis with the “focused” informative North American PLATO prior showed that the probability of clinically meaningful harm with ticagrelor exceeded its probability of clinical benefit for all the individual MACE components.
Interpretation
The 1005 patients who were enrolled in this pragmatic, time-clustered randomized trial represent a 60% increase in the number of North American patients randomly assigned to ticagrelor or clopidogrel DAPT regimens for acute coronary syndrome.4 We did not design our trial to detect meaningful differences in outcomes solely on its own merits, but we powered it to assist in resolving existing uncertainties about the choice of DAPT in North American patients.4 On its own (with a vague prior), this trial revealed similar probabilities for a clinically important composite MACE benefit (37%), equivalence (42%), or harm (21%) with ticagrelor compared with clopidogrel. However, our primary prespecified analysis using an informative “focused” prior, based on PLATO North American data, found that there was only a very small 2% probability of a clinically meaningful MACE benefit with ticagrelor compared with clopidogrel, and moderate probabilities of clinical equivalency (41%) or harm (57%) (Table 2 and Figure 3).
To account for other possible prior beliefs, we have presented the MACE results in combination with an “enthusiastic” prior based on the overall PLATO results,4 and a systematic “summary” prior, based on a Bayesian network meta-analysis.14 Using the “enthusiastic” prior, the probability of a clinically meaningful ticagrelor benefit was 89%. However, this prior minimizes uncertainty by ignoring the observed geographic variation in the PLATO results. Therefore, using a wide spectrum of reasonable priors, the probability of a clinically important ticagrelor MACE benefit does not exceed 30% and may be as low as 2% in patients in North America. The MACE probability of clinical equivalence was approximately 40% and the probability of clinically important harm between 20% and 50%. The probability of a meaningful clinical benefit associated with ticagrelor was seen only when we used the “enthusiastic” prior of PLATO data alone (with a statistical model that underestimates uncertainty by neglecting between-country variations) and ignoring the results of the 8 other prior comparative RCTs (n = 8016).14 Using the systematic “summary” prior, the probability of a clinically meaningful ticagrelor MACE benefit (RR < 0.9) was 30%, roughly equal to the probability (26%) of a meaningful increase in harm (RR > 1.1). However, the “summary” prior is also affected by an overestimation of the benefit of ticagrelor, because PLATO data account for approximately 50% of the “summary” prior sample size.
Regarding the individual MACE components, our primary analysis with a “focused” prior showed little support for ticagrelor superiority. There was a 24% probability of a clinically meaningful decrease, a 35% probability of practical equivalence, and a 41% probability of a clinical meaningful increase in all-cause mortality. Similarly for MI and stroke, there were no clear signals of ticagrelor benefit with the “focused” prior, as the probability of clinically meaningful harm often exceeded that for benefit. As predicted at the planning stage, data analyses with a vague prior lacked the power to draw any meaningful inferences (Table 2). Analyses with the “enthusiastic” priors did suggest moderate probabilities for ticagrelor reducing all-cause mortality and MI. However, these “enthusiastic” priors are at risk of bias, and any ticagrelor benefits are attenuated with the less biased “summary” prior. Although the number of patients with stroke was small, all analyses independent of the chosen prior found the posterior probabilities of a clinically meaningful increase in stroke with ticagrelor to exceed 50%.
Unsurprisingly, our study on its own was underpowered to reliably assess clinically meaningful safety outcomes. With a vague prior, the probability of major bleeding events requiring hospital admission being reduced by a clinical meaningful benefit was 51% for ticagrelor compared with clopidogrel. There was also a 26% probability of equivalence and a 23% probability of a clinically meaningful increase in bleeding with ticagrelor. When using informative priors, there was at least moderate probability (50%–93%) of clinical equivalence compared with a smaller probability for an increase in harm with ticagrelor (25%).
The findings from our TC4 trial, when taken alone or integrated with a reasonable spectrum of prior beliefs, do not fully align with recent North American guidance on acute coronary syndrome,2,3 which recommends ticagrelor over clopidogrel based on the PLATO trial, a multinational study dominated by non–North American centres. The TC4 trial efficacy results fall between the overall PLATO and its North American subgroup results and are compatible with the prior using the Bayesian network meta-analysis efficacy summary. Our results are also consistent with the recent ALPHEUS trial23 of 1910 patients in France and Czechia randomly assigned ticagrelor or clopidogrel, which found no difference in their primary outcome, a composite of percutaneous coronary intervention–related outcomes at 48 hours (odds ratio 0.97, 95% CI 0.80–1.17; p = 0.75). Owing to the short duration of this trial, it is not included in the Bayesian network meta-analysis, which examines longer-term outcomes.
The heterogeneity between the overall PLATO results and the PLATO North American subgroup has been attributed to the play of chance,24 a post hoc hypothesis that the higher ASA doses in North America had a negative interaction with ticagrelor24 or irregularities in the management of the PLATO trial.25–28 There are several reasons that the high-dose ASA hypothesis is unlikely. First is the possibility of a type 1 error, because this putative interaction was observed after interaction testing involving 37 factors. 24 Second, in a large meta-analysis of antithrombotic trials29 comparing high- and low-dose ASA (n = 10 RCTs, 6767 participants) or ASA alone versus ASA combined with other antiplatelet medications (n = 27 RCTs, 34 452 participants), there was no evidence for any impact of dose on outcomes. However, the greater use of revascularization procedures in North America is another potential explanation for the North American subgroup findings in PLATO because revascularization may reduce the long-term benefit of ticagrelor’s more potent antiplatelet activity.
Regardless, acute coronary syndrome RCT data comparing ticagrelor and clopidogrel are lacking in North America since the original 2009 PLATO publication.4 Baseline covariates were well balanced across the DAPT treatment arms, and the pragmatic nature of this trial allowed for the enrolment of a more clinically representative population than in a typical RCT. This may explain why the average age of the TC4 patients was older than that of the PLATO study participants (about 67 yr v. about 62 yr). We also recruited only patients with acute coronary syndrome undergoing cardiac catheterization, which is typical of the North American approach to patients with acute coronary syndrome, compared with only 64% in PLATO. By leveraging the provincial electronic health care databases, and focusing on clinically validated outcomes, we lost no patients during the 1-year follow-up period for identifying clinical outcomes. The Bayesian analytical approach eliminated null hypothesis significance testing and p values,5 allowed incorporation of a wide range of prior beliefs, and permitted formulation of direct probability statements regarding the effectiveness and safety of the 2 treatments, as well as their interpretation with respect to clinically meaningful effect sizes. Finally, large comparative phase 3 or phase 4 cardiovascular RCTs can cost approximately $33 000 per patient randomly assigned,30 while our study cost approximately $300 per patient randomly assigned.
Limitation
A limitation of the TC4 trial was its reliance on an intention-to-treat analysis, ignoring any possible benefits from per-protocol analyses.7 Unfortunately, we could not conduct per-protocol and as-treated analyses, because prescription follow-up data were only available in the administrative databases for patients older than 65 years. Although this single-centre study could be seen as having limited external generalizability, it provided excellent internal validity with extensive local data quality control and no missing data. It could be argued that this single-centre study offers better contemporary external generalizability to North American centres than PLATO, with its results from 43 different countries whose health care systems may or may not be comparable to North America’s. The open-label design raises the possibility of a differential exposure to post-randomization cointerventions, but we are unaware of physicians behaving differently according to whether a patient received clopidogrel or ticagrelor. Empirical evidence supporting this opinion is the absence of any difference in statin use between the 2 groups at 12 months (Bayes factor = 14 in favour of the null hypothesis of no difference between the groups).
Conclusion
This pragmatic RCT has added a substantial quantity of North American evidence to the DAPT acute coronary syndrome literature. As expected, the stand-alone TC4 results did not show conclusive evidence for the superiority of ticagrelor over clopidogrel for either the primary effectiveness or safety outcomes. After the incorporation of a range of clinically relevant priors selected from the published literature, the results still do not show, in any convincing manner, the superiority of ticagrelor over clopidogrel for DAPT treatment. In fact, the probability of ticagrelor harm is approximately equal to or exceeds the probability of its benefit. Given these observations of a low probability for a benefit of ticagrelor, as well as increased costs and the inconvenience of twice-daily dosing, we suggest that current clinical acute coronary syndrome guidelines regarding the choice of DAPT should be re-evaluated, especially in North American patients.
Footnotes
Competing interests: None declared.
This article has been peer reviewed.
Contributors: James Brophy contributed to the conception of the study. Stephen Kutcher, Nandini Dendukuri, and James Brophy contributed to the study design. Stephen Kutcher, Lyne Nadeau, Sonny Dandona, and James Brophy contributed to the data acquisition. Stephen Kutcher, Lyne Nadeau, and James Brophy contributed to the data analysis. Stephen Kutcher, Nandini Dendukuri, Sonny Dandona, and James Brophy contributed to the interpretation. Stephen Kutcher and James Brophy contributed to drafting the article. All authors gave final approval of the version to be published and agreed to be accountable for all aspects of the work.
Funding: This work was funded by the Canadian Institute of Health Research (reference no. PJT-156344). The funder had no role in the study’s design, conduct, or reporting.
Data sharing: All statistical code is freely available on request, but ethical restrictions from the Quebec provincial health authorities prohibit the sharing of individual patient–level outcome data.
References
- 1.Collet J-P, Thiele H, Barbato E, et al.; ESC Scientific Document Group. 2020 ESC guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J 2021;42: 1289–367. [DOI] [PubMed] [Google Scholar]
- 2.Levine GN, Bates ER, Bittl JA, et al. 2016 ACC/AHA guideline focused update on duration of dual antiplatelet therapy in patients with coronary artery disease. J Am Coll Cardiol 2016;68:1082–115. [DOI] [PubMed] [Google Scholar]
- 3.Mehta SR, Bainey KR, Cantor WJ, et al. Members of the Secondary Panel. 2018 Canadian Cardiovascular Society/Canadian Association of Interventional Cardiology focused update of the guidelines for the use of antiplatelet therapy. Can J Cardiol 2018;34:214–33. [DOI] [PubMed] [Google Scholar]
- 4.Wallentin L, Becker RC, Budaj A, et al. Ticagrelor versus clopidogrel in patients with acute coronary syndromes. N Engl J Med 2009;361:1045–57. [DOI] [PubMed] [Google Scholar]
- 5.Baker M. 1,500 scientists lift the lid on reproducibility. Nature 2016;533:452–4. [DOI] [PubMed] [Google Scholar]
- 6.Center for Drug Evaluation and Research. Complete Response Review Addendum Sponsor Safety Reporting Submissions: NDA 22–433 and IND 65,808 SD 632 Drug: ticagrelor (BrilintaTM). Available: https://www.accessdata.fda.gov/drugsatfda_docs/nda/2011/022433orig1s000medr.pdf (accessed 2025 Mar. 12).
- 7.Brophy JM. Bayesian analyses of cardiovascular trials—bringing added value to the table. Can J Cardiol 2021;37:1415–27. [DOI] [PubMed] [Google Scholar]
- 8.Arnup SJ, McKenzie JE, Hemming K, et al. Understanding the cluster randomised crossover design: a graphical illustration of the components of variation and a sample size tutorial. Trials 2017;18:381. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Lambert L, Blais C, Hamel D, et al. Evaluation of care and surveillance of cardiovascular disease: Can we trust medico-administrative hospital data? Can J Cardiol 2012;28:162–8. [DOI] [PubMed] [Google Scholar]
- 10.Tamblyn R, Lavoie G, Petrella L, et al. The use of prescription claims databases in pharmacoepidemiological research: the accuracy and comprehensiveness of the prescription claims database in Québec. J Clin Epidemiol 1995;48: 999–1009. [DOI] [PubMed] [Google Scholar]
- 11.Wilchesky M, Tamblyn RM, Huang A. Validation of diagnostic codes within medical services claims. J Clin Epidemiol 2004;57:131–41. [DOI] [PubMed] [Google Scholar]
- 12.Harrell F. Continuous learning from data: no multiplicities from computing and using Bayesian posterior probabilities as often as desired. Statistical Thinking 2017. Available: https://www.fharrell.com/post/bayes-seq/ (accessed 2025 Mar. 12).
- 13.Spiegelhalter DJ, Freedman LS, Parmar MKB. Bayesian approaches to randomized trials. J R Stat Soc Ser A Stat Soc 1994;157:357–416. [Google Scholar]
- 14.Kutcher S, Flatman L, Haber R, et al. The efficacy and safety of the prasugrel, ticagrelor, and clopidogrel dual antiplatelet therapies following an acute coronary syndrome: a systematic review and Bayesian network meta-analysis. medRxiv 2023. Aug. 16. doi: 10.1101/2023.08.12.23294021. [DOI] [Google Scholar]
- 15.Kutcher S. Investigating the effectiveness and safety of dual-antiplatelet therapies in acute coronary syndrome patients [thesis]. Montréal: McGill University Library; 2024:127–33. Available: https://escholarship.mcgill.ca/concern/theses/3t945x28c (accessed 2025 Mar. 11). [Google Scholar]
- 16.Cox DR. Regression models and life-tables. J R Stat Soc Series B Stat Methodol 1972;34:187–202. [Google Scholar]
- 17.Bloom DA, Kaplan DJ, Mojica E, et al. The minimal clinically important difference: a review of clinical significance. Am J Sports Med 2023;51:520–4. [DOI] [PubMed] [Google Scholar]
- 18.Bürkner P-C. brms: an R Package for Bayesian multilevel models using Stan. J Stat Softw 2017;80:1–28. [Google Scholar]
- 19.Stan modeling language users guide and reference manual. Stan Development Team. Available: https://mc-stan.org (accessed 2025 Mar. 11). [Google Scholar]
- 20.Vehtari A, Gelman A, Gabry J. Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Stat Comput 2017;27:1413–32. [Google Scholar]
- 21.Kruschke JK. Bayesian analysis reporting guidelines. Nat Hum Behav 2021;5:1282–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.van de Schoot R, Depaoli S, King R, et al. Bayesian statistics and modelling. Nat Rev Methods Primers 2021;1:1–26. [Google Scholar]
- 23.Silvain J, Lattuca B, Beygui F, et al.; ALPHEUS investigators. Ticagrelor versus clopidogrel in elective percutaneous coronary intervention (ALPHEUS): a randomised, open-label, phase 3b trial. Lancet 2020;396:1737–44. [DOI] [PubMed] [Google Scholar]
- 24.Mahaffey KW, Wojdyla DM, Carroll K, et al.; PLATO Investigators. Ticagrelor compared with clopidogrel by geographic region in the Platelet Inhibition and Patient Outcomes (PLATO) trial. Circulation 2011;124:544–54. [DOI] [PubMed] [Google Scholar]
- 25.DiNicolantonio JJ, Can MM, Serebruany VL. Lost in follow-up rates in TRACER, ATLAS ACS 2, TRITON and TRA 2P trials: challenging PLATO mortality rates. Int J Cardiol 2013;164:255–8. [DOI] [PubMed] [Google Scholar]
- 26.DiNicolantonio JJ, Tomek A. Misrepresentation of vital status follow-up: challenging the integrity of the PLATO trial and the claimed mortality benefit of ticagrelor versus clopidogrel. Int J Cardiol 2013;169:145–6. [DOI] [PubMed] [Google Scholar]
- 27.DiNicolantonio JJ, Tomek A. Inactivations, deletions, non-adjudications, and downgrades of clinical endpoints on ticagrelor: serious concerns over the reliability of the PLATO trial. Int J Cardiol 2013;168:4076–80. [DOI] [PubMed] [Google Scholar]
- 28.Doshi P. Doubts over landmark heart drug trial: ticagrelor PLATO study. BMJ 2024;387:q2550. [DOI] [PubMed] [Google Scholar]
- 29.Antithrombotic Trialists’ Collaboration. Collaborative meta-analysis of randomised trials of antiplatelet therapy for prevention of death, myocardial infarction, and stroke in high risk patients. BMJ 2002;324:71–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Sertkaya A, Beleche T, Jessup A, et al. Costs of drug development and research and development intensity in the US, 2000–2018. JAMA Netw Open 2024; 7: e2415445. [DOI] [PMC free article] [PubMed] [Google Scholar]