Use of single and combined antithrombotic therapy and risk of serious upper gastrointestinal bleeding: population based case-control study

Setting

The data for this study came from three different sources—Odense University pharmacoepidemiological database, the Funen County patient administrative system, and the Danish central person register.

In brief, information on reimbursed drug dispensing in Funen County (population 470 000) has been recorded in the pharmacoepidemiological database since 1990. Each prescription record includes a person identifier; the date of dispensing; and the brand, quantity, and form of the drug. The substances and quantities are registered according to the World Health Organization's anatomical-therapeutic-chemical system and defined daily doses methods.12 The indication for treatment and the dosing instruction are not recorded. Drugs not reimbursed, and therefore not recorded in the database, are over the counter drugs and some non-reimbursed prescription drugs, mainly oral contraceptives, sedatives, hypnotics, and some antibiotics.13

Data on patient contacts came from the Funen County patient administrative system. All Funen County residents have had their secondary care contacts registered since 1977. As inpatient care is provided almost exclusively by the national health services, this data source in effect allows true population based epidemiological studies. Inpatient data are available from 1977 and outpatient data from 1989. Diagnoses are encoded by the ICD-8 (international classification of diseases, 8th revision) until January 1994 and ICD-10 thereafter.

We used the central person register to extract the controls and to ensure that all cases and controls were Funen County residents on their index dates and during the previous 365 days.

Sample size calculation

By using our prescription database, we estimated the cumulative exposure to the aspirin-clopidogrel regimen throughout the study period at 700 person years with an average age of 70 years. With an age specific bleeding incidence of 4 per 1000 person years and an odds ratio of 5, this would yield about 14 exposed cases. This would allow us to estimate the odds ratio with a confidence interval of −45% to 70% around the point estimate.

Cases and controls

We defined cases by fulfilment of three criteria: admission with peptic ulcer or gastritis as the main diagnosis to one of the county's hospitals during 1 January 2000 to 31 December 2004; significant bleeding defined by melaena, a subnormal haemoglobin, or the need for transfusions; and a potential bleeding source in the stomach or duodenum identified by endoscopy or surgery. We excluded gastric varices.

To retrieve all relevant cases, we manually reviewed discharge summaries of all 4449 admissions with a main diagnosis of peptic ulcer (complicated or not) or gastritis (ICD-10 code K25-9) within the study period. We were blinded to the exposure status of the subjects. For each case, we considered only the first episode within the study period. Eventually, 1443 validated cases could be included.

We sampled controls, 40 for each case, by using a risk set sample technique.14 In brief, we selected 40 random controls from among people who, on the admission date of the corresponding case, had one year's residency within the county and who matched the case with respect to sex and age (within 10 years). We assigned the controls an index date identical to the admission date of the corresponding case. Cases were eligible to be selected as control subjects until their first admission with upper gastrointestinal bleeding. The odds ratios derived by such sampling are unbiased estimates of the incidence rate ratio.14

Analysis

The antithrombotic drugs included in this study were aspirin, clopidogrel, dipyridamole, and vitamin K antagonists. We defined current use as the redeeming of a prescription within the previous 90 days. We classified people whose latest prescription was redeemed 91-180 days before the index date as recent users and those whose latest prescription was redeemed more than 180 days before the index date as past users. We used the same exposure criteria for all drugs in the analysis. Unless otherwise stated, we based analyses on current exposure, and the reference was person time currently non-exposed to all four classes of antithrombotic drugs. When analysing for recent use we excluded current users of any antithrombotic drugs, and when analysing for past use we excluded both current and recent users.

As potential confounders we included current use of non-steroidal anti-inflammatory drugs (including selective cyclo-oxygenase-2 inhibitors), selective serotonin reuptake inhibitors, antiulcer drugs, systemic corticosteroids, or nitrate vasodilators and a past diagnosis of peptic ulcer or upper gastrointestinal bleeding (more than one month before the index date), diabetes, ischaemic heart disease, or alcohol related diagnoses or use of disulfiram. Each of these showed a statistically significant association with upper gastrointestinal bleeding in univariate analyses of people non-exposed to antithrombotic drugs. We controlled for confounders by conditional logistic regression modelling. Results are presented with 95% confidence intervals, whenever relevant.

In order to express the risk caused by use of antithrombotic treatment in absolute terms, we used the “the number of patients needed to be treated for one additional patient to be harmed” (NNTH) principle.15 We used the adjusted odds ratios (OR) and the incidence of upper gastrointestinal bleeding in the background population unexposed to any antithrombotic drug and aged 50 or above. For the period 2000 to 2004, the cumulative person time unexposed to antithrombotic drugs for people aged 50 or more was 762 000 person years. We found 960 cases aged 50 or above who were unexposed to all four antithrombotic drugs, yielding an unexposed event rate ER_unexp of 1.26 per 1000 person years in this age group. NNTH is then calculated simply as NNTH = 1/(ER_unexpx(OR −1)).15

This estimate can be construed as the number of person years' treatment needed to produce one excess case, if the drug is given to “average” citizens aged over 50 years. We introduced the 50 year limit because otherwise the low risk of bleeding and low rate of use of antithrombotic drugs among young people would dilute the NNTH towards unrealistically high values. Confidence intervals for NNTH are presented as suggested by Altman.16

To describe the shift in use of antithrombotic drugs, we extracted all 830 000 prescriptions for antithrombotic drugs for the entire background population for the period 1995 to 2004. We used the same exposure criteria as when we assigned exposure status to cases and controls. That is, we assigned each prescription a 90 day exposure window and charted all users with respect to periods of single and combined use within each calendar year. Finally, we summed the periods of exposure for each antithrombotic regimen within each year during the period.

Results

In all, 380 (26.3%) of the 1443 cases were exposed to at least one of the four categories of antithrombotic drugs. Table 1 shows the clinical characteristics of cases and controls. Cases had more chronic disease and were more likely to use ulcerogenic drugs such as non-steroidal anti-inflammatory drugs, as well as antithrombotic drugs, gastroprotective drugs, and selective serotonin reuptake inhibitors. In all, 127 (8.8%) cases died within 30 days of their index date.

Table 2 shows the results of the univariate and multivariate analyses. We found consistently higher odds ratios for combinations than for the single drug regimens. Too few cases were exposed to three or more antithrombotic agents to allow a meaningful estimate of the odds ratios. Furthermore, we found only one case using a vitamin K antagonist and clopidogrel and no cases exposed to either clopidogrel and dipyridamole or vitamin K antagonist and dipyridamole.

All estimates for past use were consistently lower than for current use, and confidence intervals crossed the null value, except the estimate for vitamin K antagonists (1.3 to 2.4). Results for recent use showed intermediate values between current use and past use for aspirin and vitamin K antagonists (data not shown).

The estimated NNTHs ranged between one bleeding episode per 124 treatment years for the clopidogrel-aspirin combination to one bleeding episode per 1040 treatment years for vitamin K antagonist monotherapy (table 3). For clopidogrel, the confidence interval of the odds ratio crossed the null value, so the upper limit of the confidence interval for the NNTH was infinite.

The 1 shows trends in use of combined antithrombotic drugs. All regimens showed a dramatic increase over the period 1995-2004. The total exposure to combined regimens was 758 person years in 2000 and 3978 person years in 2004, an increase of 425% over four years.

Use of combined antithrombotic regimens in outpatient settings. Data from Funen County, Denmark, 1995-2004. Note logarithmic y axis. VKA=vitamin K antagonist

• Download figure
• Open in new tab
• Download powerpoint

Discussion

The main findings of our study are that combinations of antithrombotic drugs are increasingly prevalent and that they are associated with high rates of bleeding. The apparent effect reflects a true synergism—that is, the effect of combined treatment is more than a simple addition of the effects of individual drugs. Some of the combinations involving antiplatelet drugs or vitamin K antagonists have not been assessed formally in randomised trials, and safety data are very scarce.17 18 With, for example, one excess case per 184 treatment years for the vitamin K antagonist-aspirin regimen, clinicians should consider the expected benefit carefully before prescribing this combination.

The finding of a null value odds ratio for clopidogrel monotherapy is interesting. Evidence on gastrointestinal effects of clopidogrel monotherapy is scarce and mostly indirect. Evidence exists to show that clopidogrel causes little gastric mucosal inflammation,19 that clopidogrel monotherapy is safer than aspirin monotherapy,3 and that safety is better with aspirin plus esomeprazole than with clopidogrel monotherapy.20 In a large case-control study, Lanas et al found an odds ratio of 2.8 after adjustment for use of aspirin.21 Our own study suggests that clopidogrel by itself carries little if any risk of upper gastrointestinal bleeding, but when it is given with aspirin the risk increases beyond the effect of aspirin given alone. To our knowledge, this is not contradicted by data from the large clinical trials.

Another interesting finding is the moderately elevated odds ratio for dipyridamole monotherapy. Again, safety data from the large clinical trials are scarce, essentially limited to the second European stroke prevention study.4 The number of serious bleeding events reported for dipyridamole did not differ from placebo, but figures were small and bleeding sources were not stated.

One of the strengths of our study is a thorough retrieval and ascertainment of cases, all of which we reviewed manually. The validity of the prescription data is also well established.13 As the data are used for reimbursement between retail pharmacies and the county, pharmacies have a strong economic incentive to be comprehensive and accurate. Low dose and high dose aspirin and 200 mg ibuprofen tablets were available over the counter throughout the study period. By comparing the quantity registered in our database with the gross volume sales,22 we can estimate that the amount sold without prescription was on average 1.5% for low dose aspirin, 12% for non-aspirin non-steroidal anti-inflammatory drugs, and 93% for high dose aspirin.

One limitation of our study is that we did not have data on certain risk factors for upper gastrointestinal bleeding that might be potential confounders, such as smoking, high alcohol consumption, or use of over the counter non-steroidal anti-inflammatory drugs. Although an alcohol related marker, a diagnosis of chronic obstructive pulmonary disease, and use of non-steroidal anti-inflammatory drugs were risk factors for upper gastrointestinal bleeding by themselves (odds ratios 6.0, 3.2, and 4.9), their inclusion or exclusion in the multivariate models did not change our main estimates (data not shown). We find it unlikely that strong residual confounding by smoking, alcohol use, or over the counter non-steroidal anti-inflammatory drugs would be present.

Finally, we need to consider the possibility of selection bias, mainly that patients taking antithrombotic treatment would be more readily admitted when presenting with symptoms suggestive of upper gastrointestinal bleeding.23 We have focused only on cases of severe bleeding that would be admitted irrespective of their drug exposure. Both our incidence and our case fatality are within the typical ranges for this patient group.24

Our study presents data on only the adverse effects of antithrombotic treatment. The decision to treat a patient with an aggressive antithrombotic regimen ultimately relies on the balancing of benefits and risks. Although we included a sizeable number of cases, we had too few to identify subgroups with particularly high or low risk. Future studies with higher power might aid in individualising treatment.