Association of upper gastrointestinal toxicity of non-steroidal anti-inflammatory drugs with continued exposure: cohort studyBMJ 1997; 315 doi: https://doi.org/10.1136/bmj.315.7119.1333 (Published 22 November 1997) Cite this as: BMJ 1997;315:1333
- T M MacDonald, reader in clinical pharmacologya,
- S V Morant, statisticianb,
- G C Robinson, clinical project managerb,
- M J Shield, medical directorb,
- M M McGilchrist, senior computer programmera,
- F E Murray, consultant gastroenterologista,
- D G McDevitt, professor of clinical pharmacologya
- a Medicines Monitoring Unit, Department of Clinical Pharmacology, University of Dundee, Ninewells Hospital and Medical School, Dundee DD1 9SY
- b Searle, PO Box 53, High Wycombe, Buckinghamshire HP12 4HL
- Correspondence to: Dr MacDonald
- Accepted 8 July 1997
Objectives: To determine the profile of risk of upper gastrointestinal toxicity during continuous treatment with, and after cessation of, non-steroidal anti-inflammatory drugs.
Design: Cohort study with a prospectively constructed, population based, record linkage database containing details of exposure to all community dispensed non-steroidal anti-inflammatory drugs and also all admissions to hospital for upper gastrointestinal diagnoses.
Setting: The population of Tayside, Scotland.
Subjects: 52 293 subjects aged 50 and over who received one or more non-steroidal anti-inflammatory between 1 January 1989 and 31 December 1991 and 73 792 subjects who did not receive one during the same period (controls).
Main outcome measures: Admission to hospital for upper gastrointestinal bleeding and perforation, and admission for other upper gastrointestinal diagnoses.
Results: About 2% of the non-steroidal anti-inflammatory cohort were admitted with an upper gastrointestinal event during the study period compared with 1.4% of controls. The risk of admission for upper gastrointestinal haemorrhage and perforation was constant during continuous non-steroidal anti-inflammatory exposure and carried over after the end of exposure. The results were similar for admissions for all upper gastrointestinal events.
Conclusion: This study provides evidence that non-steroidal anti-inflammatory toxicity persists with continuous exposure. There seems to be carryover toxicity after the end of prescribing. These findings have implications for the management of patients requiring non-steroidal anti-inflammatory drugs.
The risk of upper gastrointestinal toxicity associated with non-steroidal anti-inflammatory drugs is constant with continuous exposure
Gastrointestinal toxicity continues for some time after treatment stops
Such toxicity is common in older patients and patients with a history of upper gastrointestinal disease
Non-steroidal anti-inflammatory drugs should be avoided when possible; when they are used the lowest effective dose of the least toxic drug should be used for the shortest period possible
Many studies have established the association between use of non-steroidal anti-inflammatory drugs and admission to hospital for upper gastrointestinal haemorrhage and perforation and other upper gastrointestinal events.1 2 3 4 5 6 7 8 9 10 11 12 13 Data on the associations between duration of exposure to non-steroidal anti-inflammatory drugs and these events are limited. Previous studies have suggested that upper gastrointestinal toxicity shortly after exposure does not persist,5 9 11 13 14either because adaptation to the drugs takes place15 16 or because there is a depletion of susceptible patients among long term users.17 We carried out a population based cohort study in patients aged 50 years or over to test the hypothesis that upper gastrointestinal toxicity is constant during exposure to non-steroidal antiinflammatory drugs.
Subjects and methods
The study was carried out with the record linkage database of the Tayside Medicines Monitoring Unit (MEMO).18 Since 1989 all dispensed prescriptions for non-steroidal anti-inflammatory drugs other than aspirin for the population of Tayside who are registered with a Tayside general practitioner have been entered on a database. Each record includes a community health number which is unique to each patient and from which their date of birth and sex can be derived.
All discharges from Tayside hospitals are coded with up to six diagnostic codes from International Classification of Diseases, ninth revision (ICD-9)19 and up to four operation or procedure codes from the Office of Population Census and Surveys (fourth revision).20 These discharge diagnoses are entered on the Scottish Morbidity Record 1 (SMR1) database, which also includes each patient's community health number. Thus, dispensed prescriptions and admission to hospital can be temporally linked.
Study population and study cohorts
The study population consisted of all subjects resident in Tayside and registered with a Tayside general practitioner between 1 January 1989 and 31 December 1991. The non-steroidal anti-inflammatory cohort consisted of all subjects in the study population aged 50 or over on l January 1989 who redeemed one or more prescriptions for a non-steroidal anti-inflammatory during the study period, and the control cohort consisted of all other subjects aged 50 or over.
Events with a broad range of upper gastrointestinal diagnostic codes (ICD-9 280, 530.1-536.9, 578.0-578.9, and 787.1) were extracted from the SMR1 database for the study cohorts. The diagnoses were verified from original case records by medically qualified staff blinded to non-steroidal anti-inflammatory exposure and, in equivocal cases, were further reviewed by two consultant physicians, one of whom was a gastroenterologist.21
The highest ranking of the following diagnoses (in descending rank order) was ascribed to each event: (a) oesophageal varices; (b) gastric, duodenal, stomal, or small bowel ulcer; (c) oesophageal ulcer; (d) gastric or duodenal erosions; (e) gastritis or duodenitis; (f) gastric or duodenal erosions, or both; (g) oesophagitis; (h) other diagnoses—for instance, the Mallory Weiss syndrome; (i) investigated but no cause identified (for upper gastrointestinal haemorrhage only); and (j) a haemorrhage recorded in the case records but not investigated or an SMR1 diagnostic code indicating an upper gastrointestinal event in patients for whom the case records could not be located.
Two categories of diagnoses were used for analysis: complicated upper gastrointestinal events (upper gastrointestinal haemorrhage or upper gastrointestinal perforation) and any upper gastrointestinal event (any of diagnoses (a) to (j)).
Definition of drug exposure
There was no evidence that a substantial proportion of patients received regular repeat prescriptions for non-steroidal anti-inflammatories at intervals of other than 4 weeks. Sixty per cent of repeat prescriptions were filled within 45 days of the pervious prescription, and these were taken to imply continuous treatment. An interval between prescriptions of more than 45 days was taken to imply a new course of treatment. Exposure to a non-steroidal anti-inflammatory drug was assumed to last for 45 days after the final prescription in a continuous course of treatment.
Definition of risk factors
The potential risk factors examined were: age, sex, history of previous admission for an upper gastrointestinal event, current exposure to non-steroidal anti-inflammatory drug, type of drug, dose, duration of continuous exposure and non-exposure to non-steroidal anti-inflammatory drug, and the use of a drug for ulcer healing at any time during the study period.
All admissions for complicated or any upper gastrointestinal events in the nine years before 1 January 1989 were regarded as historical events.
Drug doses were categorised as low, medium, or high according to their data sheets. Ulcer healing drugs were those defined in section 1.3 of the British National Formulary.22
All courses of treatment were divided into the first, second, fourth to sixth, or subsequent months of exposure. The total exposure time and the total number of events in each of these periods were calculated. Non-exposure between courses of non-steroidal anti-inflammatory treatment were divided into similar consecutive periods. In the non-steroidal anti-inflammatory cohort a distinction was made between the period before first exposure and subsequent periods of non-exposure.
Each patient's first admission for any upper gastrointestinal event in the study period was defined as the outcome event. All subsequent events and exposure were censored. All uncensored patient days of exposure and non-exposure to non-steroidal anti-inflammatory drugs were classified by the above risk factors and event rates for every combination of these factors were calculated. Events rates in the control cohort were also classified by the relevant risk factors.
Formal analyses were carried out by using a Poisson model for the number of events. The total number of days at risk to each combination of factors (in log units) was used as an offset variable.23 Separate analyses were carried out for complicated and any events, and the results are presented as relative risks with 95% confidence intervals.
Exposure to non-steroidal anti-inflammatory drugs
During the study period 52 293 subjects received 297 594 prescriptions for non-steroidal anti-inflammatories (Table 1). Over half the subjects in the study received only one or two prescriptions, and only 7.9% received 20 or more prescriptions. The average was 5.7 prescriptions per subject over 3 years. The control group consisted of 73 792 subjects.
In the non-steroidal anti-inflammatory cohort there were 1034 events of which 474 were complicated (Table 2). The unadjusted event rates (Table 3) were substantially different in subjects with and without a history of admission for any upper gastrointestinal event, and a highly significant (P<0.001) interaction existed between history of admission and exposure to non-steroidal anti-inflammatories which could not be accounted for by the other risk factors in the statistical model. Subsequent analyses were therefore confined to subjects without a history of admission.
In the control cohort there were 1005 events, of which 510 were complicated. The unadjusted event rates in the control cohort were higher for those with a history of admission and for those who took ulcer healing drugs. Event rates in the control cohort were higher than those in the non-steroidal anti-inflammatory cohort before those patients' first exposure to the drugs.
Age and use of ulcer healing drugs
In the non-steroidal anti-inflammatory cohort the risk of both any and complicated events increased with age and was higher among patients who used an ulcer healing drug during the study period (Table 4). Similar relative risks for these factors were observed in the control cohort (1.63, 2.93, 4.78, and 5.05 in the 60-69, 70-79, 80-89, and ≥90 age bands, respectively, relative to the 50–59 age band and 2.10 for users of ulcer healing drugs for complicated events).
Duration of continuous exposure
The risk of a complicated event rose between five and 10-fold during exposure relative to the pre-exposure periods in the same subjects. There was no evidence of a trend during periods of continuous exposure (Table 4), 1. The risk of any upper gastrointestinal event rose between threefold and fivefold and also did not show a trend with continuous exposure.
Duration of continuous non-exposure
After the discontinuation of treatment the risks of both complicated and any events remained higher than those observed in the periods before exposure in the same subjects. The risk decreased with increasing time since last exposure, but it remained higher than the baseline risk after a year of non-exposure (Table 4), 1.
Dose and type of non-steroidal anti-inflammatory
There was evidence that both complicated and any events varied between drugs (P=0.005 and P<0.001, respectively, for the overall tests of significance). The risks during exposure to each drug relative to ibuprofen, the non-steroidal anti-inflammatory with the greatest exposure, are shown in Table 5.
There was evidence of a greater risk of complicated events on medium and high doses of non-steroidal anti-inflammatory drugs relative to low doses.
The major finding in this study is that the increased risks of admission to hospital for both complicated and any upper gastrointestinal event associated with exposure to non-steroidal anti-inflammatory drugs are constant during continuous exposure. Furthermore, some excess risks seem to persist for at least a year after last exposure.
Our findings differ from those of case-control studies in which outcome events determine the duration of a course of treatment and which then use that duration as an explanatory variable in statistical analyses.5 9 11 13 14 This can induce apparent relations which are arithmetic artefacts.
We found higher risks in the control cohort than in the period before exposure in the non-steroidal anti-inflammatory cohort. This suggests that general practitioners are taking account of risk factors not available in this analysis when prescribing non-steroidal anti-inflammatory drugs and that consequently the drug cohort is a lower risk population than the control group. The non-steroidal anti-inflammatory group have therefore been used as their own controls in our analysis.
We found that the risk of complicated and any events remained after exposure ceased. This may be due to subjects making their prescription last longer than intended. The effect of intermittent self treatment with leftover drug combined with the use of non-prescription use could account for the increased risks observed one year after last apparent exposure.
Our results agree closely with previous observations of the relative risks of complicated upper gastrointestinal events among different non-steroidal anti-inflammatory drugs and that the risk increases with age and dose.12 13 Event rates were higher among subjects who used drugs for ulcer healing during the study period, perhaps because their use is a marker of previous or symptomatic upper gastrointestinal diseases.
A history of admission for an upper gastrointestinal event was a major risk factor. Risks attributable to non-steroidal anti-inflammatory exposure in patients with such a history were similar to those in patients with no history, but the relative risks were smaller because of the substantially higher baseline risk before exposure (Table 3).
In this study, data on prescribed aspirin and over the counter drugs were unavailable, and the indications for the prescribed non-steroidal anti-inflammatory drugs were not known. We could not control for the confounding effects of smoking, alcohol consumption, and the use of anticoagulants or oral steroids, and we did not know which subjects had a history of upper gastrointestinal disease as outpatients. Any of these factors may confound the results of this study.
This study, however, is a cohort rather than a case-control study, with complete follow up over the study period. It used a population representing all socioeconomic groups and with universal healthcare coverage. Dispensed prescribing data was used thus eliminating primary non-compliance as a source of bias.24 The wide range of diagnostic codes searched ensured a high probability that all admissions for the events of interest were captured, and primary hospital case records were used to validate the diagnoses.
The conclusions were robust to two sensitivity analyses. One confirmed that using a 45 day exposure period after a prescription for a non-steroidal anti-inflammatory drug was not a critical assumption. The second confirmed that excluding subjects not resident in the Tayside area for the entire study period had no material effect on the analyses.
Our findings have important implications for non-steroidal anti-inflammatory treatment. Such treatment seems to increase the risks of upper gastrointestinal events by similar proportions in all age groups over 50 years, but the attributable risk in elderly patients is substantially greater because their baseline risk is higher. The risk varies with type and dose, persists with continuing treatment, and seems also to carry over after the end of treatment. Patients at high risk should ideally avoid non-steroidal anti-inflammatory drugs altogether, but should such treatment be necessary, drugs known to carry least risk should be used and consideration given to the use of measures to reduce toxicity.
The Medicines Monitoring Unit is supported by the Medicines Control Agency.
Funding: The study was supported by a research grant from Searle.
Conflict of interest: Searle, which funded the research, manufactures non-steroidal anti-inflammatory drugs.