Cardiovascular safety of non-steroidal anti-inflammatory drugs: network meta-analysis

This is a meta-analysis of clinical trials of NSAIDs, against each
other or against placebo, using network analysis methodology to enable
direct and indirect comparisons. Clinical trials were included only if
they contributed at least 100 person-years of follow-up per study arm. The
indications for use of NSAIDs were varied, including prevention of
Alzheimer's disease, prevention of cancer degeneration of adenomatous
polyps, osteoarthritis and rheumatoid arthritis. Outcomes are multiple:
myocardial infarction, stroke, cardiovascular death, death from any cause,
APTC composite outcome. The results are variable, showing globally that
the risk of these vascular outcomes is increased with most of these drugs
to varying degrees, variable also according to the outcome looked at, with
little consistency. Using the composite outcome, only ibuprofen (2.26,
1.11 to 4.89), rofecoxib (1.44, 1.00 to 1.99), and lumiracoxib (2.04, 1.13
to 4.24) seemed to be worse than placebo. Comparing the drugs two by two,
again the results are variegated. For instance diclofenac is worse than
naproxen and celecoxib for stroke, but not for other events, and the same
is true for most drugs, that will be better than others or worse depending
on the outcome.

My reservations and comments:

- the number of outcomes are extremely low: myocardial infarction was 554
cases in 117218 person-years, i.e., one in 216 patient-years (0.46 per
hundred patient-years), death one in 173 patient-years (0.6%), stroke
0.3%, etc, which seem low in this patient population of mostly older
patients with OA. Especially, for instance there were only 2 trials
including ibuprofen vs celecoxib or lumiracoxib, with very few events, and
these served for extrapolations for comparisons with other drugs. The same
is true for other drugs: there were 9 trials with diclofenac, dominated by
the MEDAL study including about 20000 patients per group treated with
diclofenac vs etoricoxib, i.e. about ten times more than any other study.
Of course all these are taken into account in the analysis, but this makes
me wary of drawing too firm conclusions.

- the indications for the drugs are often unusual, with about a third of
the studies concerned with prevention of Alzheimer's disease, and another
third in the prevention of colorectal polyps or other cancer, which are
not the usual indications for, or the usual patient population of NSAIDs.
I am not certain risks in these populations extrapolate to NSAIDs users in
real life.

- the use of the drugs is also quite unusual compared to real-life usage
patters, in that the doses used are high (2400 mg for ibuprofen, 100-150
mg for diclofenac), and duration of continuous use if very long (up to 3
years for the prevention of Alzheimer or the adenomatous polyps studies).
Mean duration of studies was 117 weeks for Alzheimer, 171 weeks for
polyps, 45 weeks for OA, and 64 weeks for RA. Common dispensation for OA
is around 100 DDD per year (14 weeks), for RA continuous (360 DDD per
year). Since in this meta-analysis OA represents about 72% of patients and
62% of patient-years contributed to the analysis, how this discrepancy
with real-life usage patterns can impact on the representativity or
transposability of results is difficult to apprehend.

My major issue with all these trials, especially those against
placebo is that one arm is treated with powerful analgesics at analgesic
or anti-inflammatory doses on a daily basis over long periods of time,
whereas the other arm receives placebo. This could mean that the untreated
or placebo-treated patients may be more prone to episodes of common pain,
flu or injury, for which they would quite naturally seek relief. Obviously
the situation would be much worse in OA or RA patients who have an
actively painful disorder. Thankfully most OA and RA studies are not
placebo controlled, so the problem would concern mostly the Alzheimer and
polyp prevention studies. In these patients, what would be or was the pain
relief used? Was it another NSAID, was it paracetamol (which is not an
effective painkiller, as the authors point out), or was it aspirin? If the
placebo controls used more aspirin for common pain relief (including
because they were told not to use NSAIDs), then that might modify the risk
in the placebo group rather than in the treatment group. This reasoning
might also apply to low-dose (200 mg) celecoxib, which is not an effective
painkiller below 400 - 800 mg.

Finally comparisons with placebo are important to demonstrate formal
risk in the absolute. However the fact is that OA and even more so RA
patients cannot go without pain relief, and this pain relief will be
mostly NSAIDs, as the authors point out. The study shows that there is no
drug that is consistently better or worse than all the others, so that in
a situation of choice between drugs, the prescriber still has little
information to go on. These studies were all designed to demonstrate
efficacy in the various primary endpoints, with as little as possible
interference from concomitant diseases or risk factors, not safety.
Consequently, most of the patients included in the clinical trials were
probably free from heart disease or major cardiovascular risk factors: the
overall death rate is 0.6 per hundred patient-years, which is lower than
the population average, if one expects these patients to be generally
above the age of 60: in France, one of the countries with the lowest death
rates, yearly death rate is 0.8% for the 60-64, 1.1% for 65-69, 2.2% for
70-79, 6.6% for 80-89 and 21.3% above 90. There is therefore no
information on any possible interaction with patients at real-life risk
from these clinical trials, and the real risk estimates can only come from
clinical trials in representative populations (i.e. non-selected
populations), or from appropriate pharmacoepidemiological studies.
It is not certain how the risk of NSAIDs-related thrombotic events is
modified by or modifies other cardiovascular risk factors, or if it is
simply additive. If not, then the recommendation to take into account the
patient's cardiovascular risk when choosing NSAIDs is meaningless, since
the risk is essentially the same for all drugs at reasonably the same
dose. The present recommendation to start with paracetamol (considering
the patient's nutritional status and liver function), then move to NSAIDs
if the analgesic effect of paracetamol is insufficient appears reasonable.
In patients with cardiovascular disease or cardiovascular risk factors,
these should be taken care of, obviously, and independently from the
eventual use of NSAIDs, as part of normal medical practice.

On the other hand no study has shown an added cardiovascular risk for
NSAIDs when used at the low doses and short durations of OTC dosages and
use, or rather several studies have positively shown the absence of
relevant risk (point estimate of risk at or near 1) at the lower end of
the dose-response curve, which is the usual OTC dose of ibuprofen or
diclofenac. Since these also have short half-lives, the exposure to low-
dose drug is short and the risk should be even lower. Maybe that could be
a relevant recommendations: in elderly patients (or for that matter in any
patient), use the lowest effective analgesic dose for the shortest
possible time. But isn't that already part of good medical practices?