NSAIDs (Non-Steroidal Anti-Inflammatory Drugs)

Median follow-up was 56 days, far too short to pick up the harms that build with long-term use, and the review says outright that it cannot make firm statements about safety. The reported adverse-event ratio of 1.04 almost certainly understates real-world harm for the same reason. Only English, German and Dutch language trials were eligible. No individual drug in the class outperformed another.

Diclofenac at 150 mg a day, the most effective oral option in the network, also carried a raised risk of people dropping out because of side effects, so the ranking rewards a preparation more people stopped taking. Only English-language trials with at least 100 patients per group were eligible, which leaves the smaller ones out. This covers knee and hip osteoarthritis, not pain in general. Several authors declare consulting relationships with industry.

The review rates most of its own evidence as low quality, chiefly because the trials reported their methods poorly, and the studies disagreed with each other substantially on the main pain comparison. Fifty-nine per cent of the trials were commercially funded and a further 31% did not say who paid. No one drug in the class could be shown safer or more effective than another, because most head-to-head comparisons rested on very few small trials.

Most of the included trials were at high risk of bias or raised concerns, and all of them ran for a year or less, so nothing here bears on years of combined use. The result for ulcer complications, as opposed to ulcers found by endoscopy, had an interval crossing no effect, from 0.10 to 1.07. Comparisons against other stomach medicines rest on single small studies. The trials were published between 1996 and 2014.

Most of the underlying data come from wisdom-tooth extraction in otherwise healthy adults, a standardised research model rather than everyday pain. These are single doses measured over four to six hours and say nothing about taking something repeatedly. The overview ran no statistical test between the drugs, so these numbers sit side by side and were never compared. Some drug and dose combinations were judged unreliable because of susceptibility to publication bias.

The best diclofenac figure rests on two studies, and some of the ketoprofen data come from 1980s trials with loosely defined outcomes. Around 5,900 participants' worth of registered but never fully reported efficacy data was identified and could not be obtained, which is the direction publication bias usually runs. Only 220 people across all 61 trials were given a tablet for comparison, so the comparison with swallowing one is thin.

About half the people using the carrier gel alone also counted as a success, roughly twice the usual response to a dummy pill, so part of what people feel comes from the vehicle and the act of rubbing rather than from the drug. Every trial in the topical review was in osteoarthritis, almost all of it knee osteoarthritis. Up to 6,000 participants' worth of completed but unpublished data was unavailable to the reviewers. Trials with fewer than 100 patients per group were excluded from the network.

These are relative risks, and the review does not supply an absolute one, which turns heavily on age and ulcer history. What counted as a serious complication varied between the pooled studies. The high-dose randomised data come from prescription-level regimens given to people with arthritis, not from the occasional tablet. Risk did not appear to keep climbing past the first month.

An online self-reported diary, so what people recorded may not be what they swallowed, and volunteers who sign up online are not a random sample. Going over the limit was worked out from the recorded times and doses, never checked against an independent record. The study counted behaviour and not harm, so it cannot say how many of those days caused anyone a problem. It ran in the United States, where pack sizes and single-pill doses differ from other markets.

Opioids showed a similar association, at 1.69, with no proposed bone-healing mechanism at all, which points at more painful or more severe fractures attracting stronger prescriptions rather than at the drugs themselves. Prescriptions filled stand in for drugs actually swallowed, and anything bought over the counter is invisible, which weakens the non-selective comparison specifically. Nonunion was rare, following 0.9% of episodes, so these odds sit on small absolute numbers.

These come from observational studies whose definitions of a serious complication and methods of recording who took what varied, which the authors flag themselves. The intervals are wide for the less-studied drugs, running from 1.07 to 17.81 for diflunisal. No absolute risk is supplied, and it turns heavily on age and ulcer history. Several of the drugs ranked are not in common use in most markets.

Not one of the pooled studies reported how often acute kidney injury happens without the drug, so there is no baseline here to turn these odds into a personal number, and the authors say so outright. Definitions of kidney injury differed between studies. The estimate for people with existing chronic kidney disease rests on five studies whose own figures ranged from 1.12 to 5.25. Prescription-based records miss anything bought over the counter.

Both reviews agree any benefit is small, and the Cochrane authors call the size probably not clinically relevant. Low confidence means the evidence cannot settle the question, not that the drugs were shown not to work. Only 7 of the 32 trials in the Cochrane review were judged at low risk of bias, almost half were industry-funded, and most were never registered. Neither review followed anyone long enough to speak to repeated use.

Two reviews of acute low back pain agree which way anti-inflammatories point; they disagree on whether trials this small, mostly unregistered and often industry-funded can say so.

Nearly seven in ten participants stopped taking the study drug and more than a quarter stopped follow-up altogether, which blunts every comparison the trial makes. It was funded by celecoxib's manufacturer. A non-inferiority design can show one drug is not clearly worse than another; it cannot show that any of them is safe. Everyone enrolled had arthritis and raised cardiovascular risk.

Whether celecoxib is as heart-safe as ibuprofen and naproxen is unsettled. Prescribing data rank it riskier; the one large trial found it no worse but dosed the drugs unequally, and no study has compared them at matched doses.