Skip navigation
Link to Back issues listing | Back Issue Listing with content Index | Subject Index

Real world use of COXIBs

Studies
Canada
UK
Patient characteristics
Canada
UK
Results
Canada
UK
Comment
A minimum requirement?

Understandably makers of new products get very excited about them, and expect us to be as excited about the results of their clinical trials as they themselves are. If we were, we would be hopping around like frogs in a jam jar. Older and wiser heads ask about real world results, about what happens to the next ten treated rather than the NNT. What we want is some evidence that in the real world the new interventions work as well as they do in clinical trials. An important academic point this, too, because if the results were very different, we may have to re-think the way we do the trials.

For coxibs, persuasive evidence exists that they reduce gastrointestinal complications associated with older NSAIDs, without any loss of efficacy. The clinical trials showing this are large, and we have guidelines for treatment indicating that patients at higher risk of gastrointestinal problems with NSAIDs should be given coxibs. Now we have two real world observational studies [1,2] from Canada and the UK that appear to confirm the results of the randomised trials.

Studies

Canada

This was a population based retrospective study using databases with 1.3 million people aged 66 years or older in Ontario. It covered a period of one year from April 2000, the date on which coxibs became available for prescription. There is free access for prescriptions to all elderly people in Ontario.

People taking coxibs, non-selective NSAIDs and a random sample of 100,000 people were chosen (a general elderly population, not matched by age, or sex). The outcome chosen was admission to hospital for upper gastrointestinal haemorrhage using comprehensive linked databases. Users were defined as those who were given at least two successive prescriptions and who received enough drug for at least 30 days of observation.

UK

This study used the UK general practice research database of about 700 practices between 1987 and January 2001. A single cohort of 1.4 million users of NSAIDs or coxibs, with periods of exposure and non-exposure, were identified for every user. The outcome was gastrointestinal haemorrhage, melaena, or haematemesis.

Patient characteristics

Canada

Patients had an average age of 75 years. Users of NSAIDs and coxibs were different (Table 1) from controls.

UK

Patients had an average age of 62 years. Users of NSAIDs and coxibs were different (Table 1).

Table 1: Characteristics of patients included in studies from Canada and the UK

Canada
Controls
NSAID
Diclofenac + misoprostol
Coxib
Number
100,000
5,391
5,087
33,491
Hospital admission in last year
12
19
18
20
Use of GPA within 180 days
17
25
25
42
Use of opioid analgesics within 180 days
11
26
26
31
Prior GI procedure
18
20
21
32
Prior upper GI haemorrhage
1
1
1
3
Aspirin use
12
19
19
18
Anticoagulants
7
5
5
10
PPI
6
8
8
22
UK
NSAID
Meloxicam
Coxib
Number
1,380,000 for all drugs combined
Use of GPA within 180 days
8
19
23
Any history of dyspepsia
6
13
17
History of gastroduodenal events
12
26
35
Anticoagulants
0.3
0.4
0.7
History of H2A use
15
35
42
History of PPI use
5
24
36

Results

Canada

Users of coxibs outnumbered users of NSAIDs by 6 to 1 in numbers of users, and 10 to 1 in exposure. The number of admissions for gastrointestinal haemorrhage, patient years of exposure, and events per 1000 patient years are in Table 2. There were 187 events in total, nut only 17 (9%) were in patients using NSAIDs. Coxibs had about half the rate of admissions for gastrointestinal haemorrhage than for NSAIDs or diclofenac plus misoprostol.

Table 2: Events, exposure, and event rates for upper gastrointestinal bleeding

Canada
Controls
NSAID
Diclofenac + misoprostol
Coxib
Admissions for upper gastrointestinal haemorrhage
82
17
13
75
Exposure (thousand patient years)
38
1.4
1.4
14.7
Number per 1000 person years
2.2
12.6
9.6
5.1
UK
Not exposed
NSAID
Meloxicam
Coxib
Gastrointestinal haemorrhage, melaena, or haemtemesis
5611
2875
36
4
Exposure (thousand patient years)
5816
628
7.1
1.6
Number per 1000 person years
1.0
4.6
5.1
2.6

For coxib users compared with NSAID users the relative risk was 0.4 (0.2 to 0.7). For diclofenac plus misoprostol users compared with NSAID users the relative risk was 0.8 (0.4 to 1.6).

UK

Users of NSAIDs had 400 times more exposure than those of coxibs. The number of cases of gastrointestinal haemorrhage, melaena, or haemetemesis, patient years of exposure, and events per 1000 patient years are in Table 2. There were 8,526 events, of which 36 occurred with meloxicam and four with coxibs.

For coxib users compared with NSAID users the relative risk was 0.4 (0.1 to 0.97) after adjustment for covariates. For meloxicam users compared with NSAID users the relative risk was 0.8 (0.6 to 1.2).

Comment

These two examples provide some fertile territory for thinking about observational studies. Both studies were well done, used standard methods to capture lots of information, were population based, and were large, drawing on total populations of about 1.3 million people in each case. Both studies show pretty much the same result, confirming a lower rate of upper gastrointestinal bleeding with coxibs than with NSAIDs in a high-risk population.

While it is tempting to accept the results, at some point we have to compare and contrast the studies, and look at the numbers. The Canadian study covered the first year of coxib introduction. Of the 44,000 older people taking coxibs or NSAIDs, 76% were using coxibs, and as a result the number of events occurring in the much smaller proportion of people taking NSAIDs was low, at 17. The UK study covered a longer period of about 13 years, and 99% of the information concerned use of NSAIDs, with only a quarter of one percent of the information on coxibs, with only four actual events.

The Canadian study was good at measuring event rates with coxibs but not NSAIDs, and the UK study was good at measuring event rates with NSAIDs but not coxibs. It is tempting to put the two studies together, but different outcomes, patients, and time span make this problematical. The difference in rates of events in people not exposed (Table 2) is very considerable.

A minimum requirement?

Bandolier frequently gets cross at observational studies that seem to be important, but, on closer examination, have a very small number of events. It is possible to generate statistical significance with a small number of events. For instance if events occur in 3 of 30 people in one group but 10 of 30 in another, we have a result which is conventionally statistically significant. If we have one more event in the first group or one fewer in the latter, the result is not statistically significant.

We need some rules to help us here, because it isn’t just the number of events that is important. The number of observations also contributes both to the statistical calculation and the weight we put on any result. Is there a minimum number of events that should be allowable in observational studies? Bandolier would love pointers from numerate readers.

We should all keep reminding ourselves that observational studies, powerful though they may be, tell us only about associations they find in the context of what they have examined. They do not prove cause and effect. And however well they are done, there is likely to be some elasticity in what is going on. In these particular studies, neither could absolutely guarantee that all events were captured, or that those captured were the events we want. So is four, or 17, the lower number of events in the studies, a number we can trust?

That's the problem. Just as we have clever people working on reporting guidelines for randomised trials and systematic reviews to help peer-reviewers, editors and readers, we need help in establishing guidelines for observational studies. Please put on the list the number of events that need to be observed for us to be sure that we can trust the result, and relieve our concern about findings being due just to the random play of chance.

References:

  1. M Mamdani et al. Observational study of upper gastrointestinal haemorrhage in elderly patients given selective cyclo-oxygenase-2 inhibitors or conventional non-steroidal anti-inflammatory drugs. BMJ 2002 325: 624-629.
  2. TM MacDonald et al. Channelling bias and the incidence of gastrointestinal haemorrhage in users of meloxicam, coxibs and older, non-specific non-steroidal anti-inflammatory drugs. Gut 2003 52: 1265-1270.

previous or next story