Intended for healthcare professionals

Education And Debate

Outcomes studies of drug induced ulcer complications: do we need them and how should they be done?

BMJ 2000; 321 doi: (Published 29 July 2000) Cite this as: BMJ 2000;321:291
  1. C J Hawkey (cj.hawkey{at}, professor
  1. Division of Gastroenterology, University Hospital, Queen's Medical Centre, Nottingham NG7 2UH
  • Accepted 4 May 2000

The hazards of non-steroidal anti-inflammatory drugs are considerable, and they result in the death of up to 2000 patients a year in the United Kingdom.1 2 Protective strategies that use the prostaglandin analogue misoprostol37 or proton pump inhibitors8 9 have been developed by using endoscopic evaluation. These strategies have been shown to reduce the incidence of clinically important events such as ulcer bleeding.10 11 The structure of primary care in the United Kingdom lends itself to such studies

Non-steroidal anti-inflammatories exert their therapeutic and toxic effects mainly by inhibiting the cyclo-oxygenase enzyme, which is involved in prostaglandin synthesis. There are two isoforms of cyclo-oxygenase: the constitutive cyclo-oxygenase-1 is principally responsible for mucosal protection and the inducible cyclo-oxygenase COX-2 principally contributes to disease.12 13 The recent discovery of these two forms has stimulated development of selective inhibitors of cyclo-oxygenase-2,14 of which rofecoxib and celecoxib are available for clinical use.15 16 Unlike standard non-steroidal anti-inflammatories, these selective drugs do not affect gastric mucosal prostaglandins17 and do not injure the stomach at clinical doses.18 19

Summary points

Long term use of non-steroidal anti-inflammatory drugs can cause stomach ulcers

Selective cyclo-oxygenase-2 inhibitors reduce the risk of ulcers

Most studies have been based on endoscopic findings rather than clinical outcomes

Outcome studies need to be large, and problems exist in evaluating clinical outcomes

Future studies need to have improved outcome definitions or could be conducted in primary care groups

Requirement for outcomes studies

Regulatory authorities have been reluctant to remove warnings about ulcer complications from patient labels even though endoscopic studies show no increase in ulceration with cyclo-oxygenase-2 inhibitors compared with placebo2023 and data from phase IIb and III trials show a reduction in ulcers compared with the rates with non-selective non-steroidal anti-inflammatory drugs.23 24 The authorities have not shown a similar reluctance to accept endoscopy as a surrogate for clinical outcomes in ulcers induced by Helicobacter pylori, and their decision is ironic in view of the fact that the authors of the misoprostol ulcer complications outcomes and safety assessment (MUCOSA) trial concluded that they had validated endoscopic assessment.10 Recent preliminary results of prospective studies on celecoxib and rofecoxib show that celecoxib reduced clinical outcomes only in patients not taking aspirin whereas rofecoxib reduced ulcers in the whole study group (Merck press release, March 27 2000; L Sima, American College of Physicians satellite symposium, Philadelphia, April 2000). Rofecoxib seemed to be associated with an increase in cardiovascular events (which disagrees with data reported from phase IIb and III studies), but no such association was reported for celecoxib. These inconsistencies are likely to fuel the appetite for outcomes studies.

Problems with outcomes studies

Thus outcomes studies, in which the effect of a treatment on clinically important events is measured directly, seem here to stay. The problem is that the superficial appeal of studying an end point of direct interest may be more than counterbalanced by the problems of doing such studies. Firstly, outcomes studies need to be very large to achieve even marginal power. The MUCOSA study enrolled 8843 patients in 664 centres and followed them for six months. It identified 42 definite events (33 ulcers) in patients taking placebo with non-steroidal anti-inflammatory drugs versus 25 events (16 ulcers) in those taking active misoprostol with non-steroidal anti-inflammatories (P=0.049).10 The phase IIb and III data analysed for rofecoxib and celecoxib concerned 2155 and 1763 patient years respectively, 23 24 and the recently concluded prospective studies have each enrolled around 8000 patients.

A more important limitation relates to the imprecision of end points reported in such studies. In an outcomes study the patients present unexpectedly, often to someone who is not a direct participant in the study. Adjudication committees have to decide retrospectively, on the basis of non-systematic evidence, whether a true clinically important event occurred. To make matters worse, there is no universal consensus about what a true clinically important event is. Consequently considerable uncertainty exists about the comparability of end points in different studies. The end point most commonly studied is perforations, ulcers, or bleeds. This tends to sound more serious than it really is. Only a minority of perforations, ulcers, or bleeds are due to complications from non-steroidal anti-inflammatory drugs, the rest being uncomplicated ulcers discovered because of endoscopy for dyspepsia or soft end points such as a haem positive stool, depending on the design of the study. Thus, the end points that outcomes studies try to measure could end up no more serious than those that are the target of endoscopic studies.

Could the design and conduct of outcome studies be improved?

Given the substantial investment needed to do such studies, it is important to ask how the yield and fidelity of information from them could be improved. Firstly, since ulcer complications are the prime cause of concern with non-steroidal anti-inflammatory drugs, these should be an end point and preferably the primary one. It is likely that this would sharpen the difference between treatments by reducing the number of irrelevant events that are recorded. In both the MUCOSA study of misoprostol10 and the short term meloxicam large scale international study safety assessment (MELISSA)25 there were no differences between treatments for softer end points such as haem positive stools or melaena without endoscopic evidence of an ulcer. Nevertheless, a doubling or trebling of patient numbers would be needed for studies to show a reduction in ulcer complications compared with the rates with non-selective non-steroidal anti-inflammatories. Moreover, it may become clear that existing cyclo-oxygenase-2 inhibitors are sufficiently safe to make a non-selective comparator arm unethical. Trials of new drugs would then need to be larger still to show equivalence to cyclo-oxygenase-2 inhibitors, although only of a comparable size to outcomes studies in other therapeutic areas.26 27 However, if such increases in size were achieved by including more (and thus more varied) centres, concerns about end point precision would grow.

Despite these problems, it is possible to conduct effective and better controlled outcomes studies at reduced cost in a country like the United Kingdom, where a small number of large district general hospitals predictably serve a well defined population. Conventionally, this would be an efficient way of conducting a standard randomised study. End point precision could be improved substantially: presentation to hospital with ulcer complications could be recorded according to a consistent, protocol driven approach using well defined end points, and patients could be assessed by a limited number of gastroenterologists trained to carry out formal trial assessments.

A more radical approach would also be possible. Trial drugs could be made available to all the patients in some primary care groups. Rates of ulcer complications in these groups could be compared with rates in matched groups that continued to use existing drugs. Although decisions to take the test drug among patients in the trial primary care groups, as well as the specific selection of drugs in the control groups, may be affected by risk factors such as patients' history, these factors are sufficiently well defined for their influence to be quantified and allowed for. Moreover, use of non-steroidal anti-inflammatory drugs is the only identifiable factor that discriminates between the variable rates of presentation with ulcer complications in different general practices.28

In addition, by studying drugs in a real life setting and allowing an epidemiological approach to analysis (including comparison with patients not taking non-steroidal anti-inflammatories) the proposed strategy could support more radical conclusions. It might be possible to show not only that cyclo-oxygenase-2 inhibitors cause fewer ulcer complications than non-selective non-steroidal anti-inflammatories but also that they cause none compared with non-use. Equally, it might emerge that previous ulceration associated with non-steroidal anti-inflammatory drugs continues to leave some patients at risk even when they stop or change their drug.

Endoscopic image of stomach ulcer



  • Competing interests CJH has received research funding or honoraria from Astra, Boehringer Ingelheim, Glaxo, Merck, NicOx, Novartis, Parke Davis, Searle, SmithKline Beecham, and Wyeth Lederle.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.