Why does it seem like no patient in the hospital can get NSAIDs? If they don’t have acute kidney injury, they have heart failure and they’re going to retain more fluid, or they have no platelets and they’re going to bleed, or man seriously there are like 30 contraindications to NSAIDs. So hey, I’ve prescribed oxycodone for headache when Tylenol doesn’t work, but it felt pretty bad (for me, anyway). It makes me wonder — just how real is the basis for our fears about NSAIDs?
With regard to the idea that NSAIDs increase risk of myocardial infarction (MI) at least, Michèle Bally and colleagues have undertaken a systematic review and meta-analysis to try to better understand the degree to which NSAID use, even for a short time, can increase the risk of myocardial infarction. So read on, and think about whether the moment in which you prescribe a statin for primary prevention ought also to be the moment in which you counsel the patient to avoid NSAIDs indefinitely.
If You’re Only Going to Read One Paragraph
In an individual patient data meta-analysis of observational studies of NSAID use (N = 444,763), taking any dose of NSAIDs for as long as a week was associated with a higher risk of MI. Odds ratios were 1.48 (95% CI 1.00 to 2.26) for ibuprofen, 1.50 (1.06 to 2.04) for diclofenac, 1.53 (1.07 to 2.33) for naproxen, and 1.58 (1.07 to 2.17) for rofecoxib, with only celecoxib not crossing the threshold for significance at 1.24 (0.91 to 1.82). There was a dose-dependent increase in risk, and risk seemed to peak with exposure for one month and plateau thereafter, though the increase was observable with as little as 7 days of therapy. This is a great reason to think hard about whether NSAIDs are really a good option for your patients already at risk for MI, to counsel patients on minimizing their use, and, maybe, to consider reaching for celecoxib a little earlier than you currently do.
So this is our first big meta-analysis on IM HEAT! I have been dreading the day. I’m going to take a very, very high altitude view of this, because it is incredibly easy to get bogged down in statistical methods related to meta-analysis. But let’s at least talk briefly about two important components: study selection and patient-level vs study-level analysis.
First, study selection. There are always more papers on a given topic than it is feasible to include, so knowing how authors pared it down is important when assessing risk of bias. Oh God you’re bored already, aren’t you? This is the part of the conversation where my date used to fake a family emergency. Hang in there, I promise this will be fast and relatively painless (same thing I used to promise on a date). The authors searched standard medical databases for observational studies of NSAIDs that included MI as an outcome, and further searched bibliographies of prior meta-analyses on the topic. Only studies that documented MI rather than a composite cardiovascular outcome were included. Because the authors wanted to describe the temporal relationship between NSAID use and incident MI, they selected only studies that allowed for time-to-event analysis (i.e. that had captured such data). The 82 studies were pared down to just four on the basis of the above criteria, which still resulted in a whopping 444,763 patients for analysis.
Second, let’s talk about patient-level analysis. If you want to understand the benefits of patient-level versus study-level meta-analysis, you probably want someone else to explain it to you. But, briefly: if you’re looking at subgroups in studies (i.e. patients with low dose NSAID use in the 7 days prior to an MI), working at study-level data (i.e. taking the reported mean and standard deviation and weighting according to the number of patients) misses a significant amount of heterogeneity between patients in a given subgroup. Working with patient-level data is harder (first of all you have to get the data, then you have to fit it into your structure for analysis), but because it captures the differences between individual patients, it can be more accurate in revealing a subgroup effect. The paper I linked to uses both approaches in a real-world example and notes that they result in different statistical conclusions. The Bally paper uses the harder, better approach.
So back to it: for patients who had an MI, the investigators took the date of the event as the index date and used calendar time-matched dates for the non-MI comparators. They broke patients out by recency of NSAID use and high vs low dose usage. The dosing breakdown for low- vs high-dose is reasonable for each NSAID considered (e.g. < 750mg of naproxen daily is low dose, more is high dose).
Bally et al., BMJ 2017;357:j1909
NSAID exposure in the prior 365 days was the primary covariate. They adjusted for a multitude of factors related to MI risk including age, sex, diabetes/HTN/hyperlipidemia, renal failure, CHF, aspirin or clopidogrel use, known CAD, and several others. They employed Bayesian analysis to examine the difference in MI risk for patients on each of the NSAID classes in each of the use categories above.
A Bit of Context
Remember Vioxx (rofecoxib)? Selective COX-2 inhibitor, great market penetration, kind of felt like a magic bullet for osteoarthritis pain? Well I know I talk about the FDA a lot, but if it weren’t for post-marketing surveillance and mandated safety studies, we might never have gotten the VIGOR trial and its associated 4-fold increase in acute MI for patients on rofecoxib compared to naproxen (0.4 percent vs. 0.1 percent). The authors cheekily tried to suggest that this was due to naproxen’s “cardioprotective effect,” which may actually exist, but would have to be about three times as effective as aspirin’s to account for the difference (it isn’t).
No, rofecoxib produced between 88,000 and 140,000 excess cases of serious heart disease because (we think) COX inhibition of prostaglandin synthesis results in a selective shunting of arachidonic acid to thromboxanes (oh yeah, we’re doing biochem) rather than prostaglandins. Thromboxanes increase platelet aggregation, which a very bad thing when it happens in your coronary arteries. That this risk is similar between COX-2 inhibitors and traditional NSAIDs was recently proven by the PRECISION trial, which found that celecoxib is noninferior to ibuprofen or naproxen (WHERE IS YOUR CARDIOPROTECTIVE EFFECT NOW?) with regard to cardiovascular safety, and predictably has a better GI safety profile. However, there was no comparison to placebo in this trial.
So we all agree in principle that NSAIDs increase risk of myocardial infarction. But ask any cardiologist and you will learn that MI, despite being quite common, is an endpoint that requires a large sample size to properly assess changes in risk. Despite a few studies and not a few smaller meta-analyses in the past, questions remain incompletely answered about the effects of NSAIDs on MI risk.
Of the 446,763 individuals included, most were male, ~10% had diabetes, and ~25% had known coronary artery disease. Data for several such confounders was not collected by the individual trials comprising the meta-analysis. Only one trial collected data on aspirin and clopidogrel use, in which 23% of patients were on ASA and 1.7% on clopidogrel.
The odds ratios of increased MI risk for each NSAID with any dose for as little as 1-7 days were significantly elevated for all drugs — 1.48 (95% CI 1.00 to 2.26) for ibuprofen, 1.50 (1.06 to 2.04) for diclofenac, 1.53 (1.07 to 2.33) for naproxen, and 1.58 (1.07 to 2.17) for rofecoxib), except celecoxib (1.24 (0.91 to 1.82)). Higher doses generated a larger observed risk, though it’s worth noting that there was a relative paucity of data for patients taking lower doses (after all, look at the doses — who’s prescribing less than 750 mg of naproxen a day?). The largest increase in risk happened within the first 7 days, though it did increase slightly with 8-30 days of use. There was no further increase with use > 30 days. Notably, rofecoxib had odds ratios significantly higher than any other NSAID, so Vioxx really does seem to be more generative of MI than other drugs in its class.
Granted, this is a meta-analysis, and the included studies are nonrandomized. However, this is likely the best data we are going to get — the idea that we could recruit enough patients to an NSAID vs placebo trial, or get approval for that trial given the known association of NSAID use with MI (as expressed here) is fairly ridiculous. So the study has its limits, but these are not limits we are likely to be able to overcome. For its part, it did adjust appropriately for confounders where possible, and it is the highest-powered analysis yet to detect this effect.
While I am suitably terrified by the idea that NSAIDs increase MI risk in the first week of their use, I don’t know how this changes my practice for chronic pain patients. If the options are NSAIDs or opioids for chronic pain (assuming patients are already on maximum dose acetaminophen), it’s not at all clear that this MI risk is significant enough to overcome the risk of abuse and dependence for opioids. However, it does make me more likely to talk to my patients about the potential harms of NSAID use, especially those who have already had an MI in the past. It’s also a reminder to lean hard on your adjunctive therapies for chronic pain, including treating depression, helping them get to sleep, and using agents that may help with any neuropathic components of pain. And for acute pain in patients at elevated risk of MI? Maybe that ankle sprain will do better with oxycodone than naproxen after all, as long as you can get the opioids off once the pain gets better. Which, you know, what could go wrong?
Read the original study here.
Coming Up on IM HEAT
So in getting feedback from some colleagues, it seems I’m not the only one struggling to keep up with our current pace of updates. Apparently it’s almost as difficult to read three essays per week as it is to write them. So, we’re going to make a change: from here out, IM HEAT gets an entry by me every Monday, and a guest post every Thursday! Same great updates, easier to digest schedule. And the podcast isn’t dead — only resting. We’re going to try something week after next. Obviously get excited.
Think I’m going to talk about cystic fibrosis and insurance on Monday. Specifically, we’ll explore why if you do have a pre-existing condition, lacking insurance coverage may be hazardous to your health. Which is to say you might die. Don’t believe me? Some guys wrote about it, and you can read about it on IM HEAT!