Author + information
- Elliott M. Antman, MD∗ ()
- Cardiovascular Division, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts
- ↵∗Address for correspondence:
Dr. Elliott M. Antman, TIMI Study Group, 350 Longwood Avenue, Boston, Massachusetts 02115.
Drugs prescribed for noncardiovascular conditions may interact with those prescribed for cardiovascular conditions and vice versa. For example, drug A may alter the exposure or physiological effect of drug B. Regulatory authorities emphasize that studies of drug-drug interactions should be planned prospectively (1,2). An example would be measurements of drug exposure in the presence of a potentially interacting agent and the creation of a population-based pharmacokinetic model. Cardiologists are also familiar with a factorial design in which multiple drugs are tested through independent randomization in a single clinical trial. Factorial designs provide information on the effect of drug A alone and when coadministered with drug B, a situation often encountered in clinical practice.
In large clinical trials it is not always practical or even possible to randomize subjects prospectively to multiple drugs. Typical approaches to evaluating potential drug interactions include stratifying randomization on the basis of the presence or absence of a previously prescribed drug of interest and/or reporting a subgroup analysis on the basis of concomitant use of a drug of interest. Stratification of randomization distributes subjects taking the concomitantly prescribed drug with roughly equal balance across the test arms. Subgroup analyses should be prospectively defined but must be interpreted cautiously because of limited statistical power (potential for type II error) and the risk for a false-positive observation because of multiple comparisons (type I error).
Armed with this background, we can now approach the aspirin analysis from the PRECISION (Prospective Randomized Evaluation of Celecoxib Integrated Safety Versus Ibuprofen or Naproxen) trial by Reed et al. (3) in this issue of the Journal. PRECISION was a phase IV, double-blind study of patients with osteoarthritis or rheumatoid arthritis who required nonsteroidal anti-inflammatory drug (NSAID) treatment and either had established cardiovascular disease (CVD) or were considered at high risk for CVD (4). The population enrolled had a lower rate of the primary endpoint (Antiplatelet Trialists’ Collaboration [APTC]–defined major adverse cardiovascular events [MACE]: cardiovascular death, nonfatal myocardial infarction, or nonfatal stroke) of about 1% per year compared with the pretrial estimate of 2% per year. Because 90% of the subjects had osteoarthritis and there were regulatory restrictions on the maximum celecoxib dose, a comparatively low dose of celecoxib (200 mg) was tested against naproxen and ibuprofen, which were prescribed at about 85% of the maximum allowed by the protocol (5). The naproxen and ibuprofen regimens were associated with significantly lower rates of discontinuation for insufficient clinical response but, not unexpectedly, given the comparatively higher NSAID dose regimens, were also associated with significantly higher rates of adverse events (5). PRECISION was a difficult trial to conduct, with 69% of subjects discontinuing study drug during the course of the trial, an unknown number of crossovers to a nonstudy NSAID, and 27% of subjects lost to follow-up (4).
The aspirin analysis is one that clinicians were waiting for given the previously reported interaction of aspirin with naproxen and ibuprofen. Specifically, the acetylation of cyclooxygenase (COX)–1 by aspirin (a beneficial effect of aspirin) is blocked if naproxen or ibuprofen access a critical channel in COX-1 before aspirin (6). The ability of naproxen and ibuprofen to inhibit aspirin’s acetylation of COX-1 is dose dependent and is more prominent at lower rather than higher doses of aspirin (7). It is a general recommendation that patients take aspirin prior to an NSAID so as to achieve the protective effects of aspirin.
In PRECISION, subjects underwent stratified randomization on the basis of a history of aspirin use (as well as type of arthritis and geographic region). The protocol specified that the aspirin dose be ≤325 mg/day, but the exact doses are not reported (3,4). About 45% of subjects reported a history of aspirin use at the time of enrollment; that information was used to assign them to the aspirin or no-aspirin subgroup. As per Figure S5 in the Appendix to the main PRECISION trial, there was no significant interaction between the aspirin subgroups for the primary APTC-defined MACE endpoint of the trial: celecoxib 200 mg/day was noninferior to naproxen and ibuprofen in the doses used in the trial (4).
Recognizing that aspirin stratification at enrollment was a reflection of a subject’s prior history, the investigators of the present study performed several maneuvers. Missing baseline data (1.5% of subjects) were filled in by imputation using a series of simulation procedures (8). Adjustment for baseline differences (potential confounders) was accomplished using a logistic regression–based propensity score for receiving aspirin. Each subject was assigned a score on the basis of the propensity analysis, which was adjusted further by the inverse of the probability of being treated with aspirin (9). The effect of these sophisticated manipulations was to convert the observational data regarding the use of aspirin at enrollment into a pseudo-randomized dataset that adjusted for differences in baseline characteristics; of course, there cannot be any correction for unmeasured confounders (9). Instead of using the APTC-defined MACE primary endpoint of the main PRECISION trial, the investigators created a new post hoc 9-component composite safety endpoint that included cardiovascular, gastrointestinal (GI), and renal components (3). This increased the number of endpoints for analysis from 607 using the APTC-defined MACE endpoint in the main trial to 1,275 using the composite safety endpoint of the present aspirin analysis (3). The main PRECISION analyses (and the original aspirin subgroup analysis reported) were performed in the intention-to-treat cohort, while the present aspirin analyses were performed in the modified intention-to-treat cohort that the investigators report was confirmed to be taking study drug during follow-up (details not provided), referred to as the “on-treatment” cohort (3,4).
Overall this aspirin analysis used 5 endpoints: the composite safety endpoint, extended MACE, APTC-defined MACE, GI events, and renal events. There were 4 sets of comparisons: aspirin versus no aspirin among all patients, aspirin versus no aspirin with each NSAID analyzed separately, each NSAID separately in patients not on aspirin, and each NSAID separately in patients on aspirin. Inspection of the array of Kaplan-Meier plots and forest plots shows oscillation of the findings depending on the endpoints used and the groups being compared. Many of the confidence intervals overlap, and the interaction tests for the aspirin subgroups are either nonsignificant or of borderline significance. In general, the findings replicate those of the main PRECISION trial: celecoxib 200 mg/day appears similar to comparatively higher doses of naproxen or ibuprofen with respect to cardiovascular events but appears to be associated with fewer GI or renal events, all at the cost of being less effective for controlling arthritis pain (5). Subjects with a history of aspirin use at enrollment had less protection against GI and renal events with celecoxib 200 mg/day compared with those without a history of aspirin use at enrollment who were allocated to celecoxib (3).
Despite the sophisticated statistical approaches used in this aspirin analysis, many questions remain. The assignment of “aspirin status” was based on a history of aspirin use at enrollment. Enrollment aspirin status was used as a surrogate for aspirin actually taken concomitant with the NSAID study drugs. We are not given critical information about the relative timing of aspirin versus the NSAID and the actual dose of aspirin used during the trial (compared with that reported at the time of enrollment). New information appears in this report: 304 patients discontinued aspirin after randomization, and 964 who were not on aspirin at enrollment started it during the study (3). The relative timing of these aspirin crossovers and cross-ins with respect to events is not reported. We are told that the exclusion of these patients had no impact on the primary composite safety endpoint, but no information is provided about the impact on any of the other endpoints (3).
The investigators acknowledge that the present aspirin analysis was pre-specified in the trial protocol but was not designed to detect an interaction between study NSAIDs and aspirin (3). Also, the investigators report that the statistical methodology for the aspirin analysis was developed post hoc. These factors, the absence of correction for multiple analyses, insufficient details on the actual use of aspirin during the trial period, and the limitations of the main PRECISION trial make it difficult to provide any definitive recommendations to direct clinical practice. It seems best to minimize the use of NSAIDs in general and especially in patients with CVD (10,11). If they must be used, the drug with the safest profile in the lowest dose for the shortest period of time remains the best advice for practice (10). A meta-analysis of the vascular and GI effects seen in trials of NSAIDs suggests that naproxen 1,000 mg/day was associated with fewer events than celecoxib usually tested in doses ≥400 mg/day (Online Appendix materials in the Coxib and Traditional NSAID Trialists’ meta-analysis ). It is not clear how celecoxib 200 mg/day compares with lower doses of naproxen or ibuprofen than were used in PRECISION—and were needed to more effectively control pain than celecoxib 200 mg/day (5). Aspirin probably ameliorates some of the apparent relative benefit of low-dose celecoxib (200 mg/day) with respect to GI and renal events. No information from PRECISION or this aspirin analysis informs the clinical community about 400 mg or 800 mg/day of celecoxib, the dose range in which previous reports raised concerns about cardiovascular safety (Online Figure 15 in the Coxib and Traditional NSAID Trialists’ meta-analysis) (12). A lesson learned from the entire COX-2 story is that there is an urgent need for new effective and safe analgesic agents that would also allow patients with CVD to take aspirin with less concern about a drug interaction (13,14). Finally, despite an enlarging statistical toolkit to enhance subgroup analyses, there is no substitute for a well-designed prospective collection of exposure-outcome data when a drug-drug interaction is investigated in a clinical trial (1,2).
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Dr. Antman has reported that he has no relationships relevant to the contents of this paper to disclose.
- 2018 American College of Cardiology Foundation
- ↵European Medicines Agency. Guideline on the investigation of drug interactions. Available at: http://www.ema.europa.eu/docs/en_GB/document_library/Scientific_guideline/2012/07/WC500129606.pdf. Accessed February 14, 2018.
- ↵Center for Drug Evaluation and Research. Clinical drug interaction studies—study design, data analysis, and clinical implications; guidance for industry. Available at: https://www.fda.gov/downloads/drugs/guidances/ucm292362.pdf. Accessed February 14, 2018.
- Reed G.W.,
- Abdallah M.S.,
- Shao M.,
- et al.
- Nissen S.E.,
- Yeomans N.D.,
- Solomon D.H.,
- et al.
- Antman E.M.
- Li X.,
- Fries S.,
- Li R.,
- et al.
- Antman E.M.,
- Bennett J.S.,
- Daugherty A.,
- Furberg C.,
- Roberts H.,
- Taubert K.A.
- Bally M.,
- Dendukuri N.,
- Rich B.,
- et al.
- Grosser T.,
- Theken K.N.,
- FitzGerald G.A.
- Institute of Medicine