Risk factors for gastrointestinal bleeding : a hospital-based case-control study

a Hospital Pharmacy, Kantonsspital Winterthur/University Children’s Hospital Zurich, Switzerland b Department of Pharmaceutical Sciences, University of Basel, Switzerland c Clinical Pharmacology & Toxicology, University Hospital Basel, Switzerland d Department of Internal Medicine, Kantonsspital Winterthur, Switzerland e Basel Pharmacoepidemiology Unit, Clinical Pharmacology & Toxicology, University Hospital Basel, Switzerland

lands the most frequent ADR-related diagnosis of hospital admissions was bleeding (8.6%), and the drugs most commonly associated with ADRrelated hospitalizations were anticoagulants (17.8%) [2].A Swiss study retrospectively analyzed all hospital admissions during one year and found that about 4% of them were directly related to ADRs.Analyzed by affected organ system, the most frequent ADRs were gastrointestinal complications (33%) caused by platelet aggregation inhibitors, NSAIDs, oral anticoagulants or digoxin.21% of all ADRs were due to drug-drug interactions (DDIs), whereof the combinations of NSAIDs and oral anticoagulants as well as the combination of platelet aggregation inhibitors and corticosteroids were most frequently observed [3].Various former studies focused on the interaction between NSAIDs and oral anticoagulants as risk factor for GI bleeding.The short term risk for upper GI bleeding was six times higher (relative risk 5.8, 95% confidence interval (95% CI) 2. [3][4][5][6][7][8][9][10][11][12][13][14] when anticoagulated patients were also exposed to NSAIDs compared with use of anticoagulants alone [4].According to Battistella et al., 0.3% of anticoagulated patients (≥66 years) were hospitalized with upper GI bleeding per year, and the concomitant intake of NSAIDs was a risk factor for GI bleeding [5].However, NSAIDs also bear a risk for GI bleeding without concomitant anticoagulant therapy.
An observational cohort study showed that the relative risk of upper GI bleeding for elderly users (≥66 years) of non-selective NSAIDs was 4.0 (95% CI 2.3-8.5)[6].
The aim of the present hospital-based casecontrol study was to investigate risk factors for a first-time GI bleeding leading to hospitalization with a special emphasis on the role of drugs and DDIs.

706
Risk factors for gastrointestinal bleeding: a hospital-based case-control study

Study population and data source
The study has been reviewed and accepted by the local Ethics Committee.This retrospective hospitalbased case-control study was conducted at the Kantonsspital Winterthur, a 500-bed teaching hospital providing primary and secondary care to a population of approxi-mately 200 000 inhabitants.Between January and December 2005, patients admitted to the Department of Medicine were eligible to be included in the study.
Information on drugs prescribed at hospital admission (according to the anatomical therapeutical chemical (ATC) classification), demographic information (age and sex), admission date and length of hospital stay, main and additional diagnoses (according to the international classification of diseases, 10 th revision (ICD-10) classification), history of non-bleeding GI ulcer, body mass index (BMI) and international normalized ratio (INR) value were obtained from the electronic patient records.

Controls
Controls were patients who were admitted to the Department of Medicine for diseases other than GI bleedings.Among all such potential control patients without current or previous GI bleeding, we identified at random two controls per case, matched on age (±1 year), sex and calendar time of hospital admission (±1 month).

Exposure definition
Patients were defined as current users of a drug of interest when, according to the medical history, they were using the drug at the time of the hospital admission.Glucocorticoids included use of betamethasone, cortisone, hydrocortisone, prednisolone, or prednisone, nonsteroidal antiinflammatory drugs (NSAIDs) included acemetacin, celecoxib, diclofenac, etodolac, ibuprofen, indometacin, mefenamic acid, or meloxicam, oral antico-agulants included acenocoumarol or phenprocoumon, selective serotonin reuptake inhibitors (SSRIs) included citalopram, fluoxetine, paroxetine, or sertraline, and thrombocyte aggregation inhibitors included clopidogrel, dipyridamole, high dose aspirin, or low dose aspirin.

Analysis of DDIs
Prescriptions at hospital admission were screened for DDIs potentially causing GI bleeding.As a result of our previous evaluation study of frequently used drug interaction screening programs [7], Pharmavista [8] was chosen to check prescriptions for DDIs.

Statistical analysis
We conducted a matched analysis (conditional logistic regression model) using the software program SAS, version 8.02 (SAS Institute, Inc, Cary, NC).Relative risk estimates (odds ratios (ORs)) are presented with 95% CIs.P-values less than 0.05 were considered statistically significant.
In a second step, we investigated the role of DDIs and explored whether concomitant use of NSAIDs, glucocorticoids, oral anticoagulants, thrombocyte aggregation inhibitors or SSRIs affected the risk of GI bleeding.

707
S W I S S M E D W K LY 2 0 0 7 ; 1 3 7 : 7 0 5 -7 The final multivariate models included use of glucocorticoids, NSAIDs, oral anticoagulants, SSRIs, thrombocyte aggregation inhibitors, PPIs and hypertensive diseases.We evaluated DDIs in separate models in which we classified patients into mutually exclusive groups of non-users of drug A and B, users of drug A only, users of drug B only, or users of a combination of A and B, and we adjusted these models for all other drugs not involved in a particular DDI of interest.

Characteristics of the patients and dropouts
During the study period January to December 2005, the Kantonsspital Winterthur registered 19 385 admissions, of which 24% (4713) were allocated to the Department of Medicine, wherefrom 1.9% (90) due to GI bleeding as the main diagnosis.16 cases were excluded (15 patients showed evidence for previous GI bleedings, one patient lacked sufficient clinical information).The detailed main diagnoses of the 74 cases and further characteristics of the cases and of the matched controls are displayed in table 1.During hospitalization, 4 (5.4%)cases and 13 (8.8%)controls died.

Multivariate regression models
In the multivariate model, adjusted for the drugs listed above and for hypertensive diseases, use of NSAIDs (adjusted OR 8.6, 95% CI 3.1-23) and use of thrombocyte aggregation inhibitors (adjusted OR 2.2, 95% CI 1.1-4.6)yielded statistically significantly increased risks for GI bleeding (table 2).Furthermore, SSRI use was also associated with an increased bleeding risk, (adjusted OR 3.3, 95% CI 0.9-12).Use of oral anticoagulant drugs alone in the therapeutic INR range was not associated with bleedings (adjusted OR 0.9, 95% CI 0.4-2.3).However, high INR values ≥4 were associated with an increased bleeding risk (adjusted OR 13, 95% CI 1.2-150).
According to the multivariate DDI models, use of NSAIDs alone, use of glucocorticoids alone, or concomitant use NSAIDs and glucocorticoids, as compared to non-use of both NSAIDs and glucocorticoids, yielded adjusted ORs of 8.3 (95% CI 3.0-23), 1.4 (95% CI 0.3-7.2) and 20 (95% CI 1.6-257), respectively.Furthermore, there were 8 cases and 0 controls who concomitantly used NSAIDs and oral anticoagulants.Due to the zero cell, we could not assess an adjusted OR in a multivariate model, but we calculated a crude OR under the assumption that one (instead of 0) control patient used both an NSAID and oral anticoagulation at the time of the hospitalization, which yielded a crude OR of 20.The adjusted relative risk estimates of developing a GI bleeding were also increased for concurrent use of thrombocyte aggregation inhibitors and oral anticoagulants as well as of thrombocyte aggregation inhibitors and SSRIs (without statistical significance), as compared to single use of these drugs.The detailed results from DDI models are displayed in table 2.

Discussion
Almost 2% of all admissions to the Department of Medicine at the Kantonsspital Winterthur were due to GI bleeding.First-time GI bleeding was registered in slightly more male than female patients (54% vs 46%).The number of cases increased with age, more than two thirds of all patients admitted with first-time GI bleeding (69%) were at least 70 years old.
Our study suggests that the risk for GI bleeding under treatment with oral anticoagulants alone was not elevated (adjusted OR 0.9, 95% CI 0.4-2.3),if the INR did not exceed 4, and if patients were not exposed to other risk factors.However, an INR value ≥4 was associated with an increased GI bleeding risk (adjusted OR 13, 95% CI 1.2-150).This finding is in line with a recent Norwegian study reporting that 74% of patients treated with warfarin had, according to the authors, INR values above the therapeutic range at the time of GI bleeding [9].According to a meta-analysis [10], the OR for major bleeds for INR 3 to 4 compared with INR 2 to 3 was 2.3 (95% CI 0.5-10) and did not reach statistical significance.However, the OR for INR >4 compared with the INR 2 to 3 reference group was highly significant (OR 33, 95% CI 9.1-121).Various studies showed that the safety management and monitoring of an oral anticoagulant therapy is a difficult challenge for both patients and physicians.In such studies, the INR values were beyond the therapeutic range in 41 to 57% of the observation period [11][12][13].
Patients treated with NSAIDs showed a 9-fold risk (adjusted OR 8.6, 95% CI 3.1-23) for hospitalization due to GI bleeding compared to patients without NSAID treatment.The results of two recent cohort studies showed a 3.6-fold and a 5.5-fold higher risk for current NSAID users of developing upper GI bleeding [14,15].According to another large case-control study the ORs ranged from 1.4 for aceclofenac to 25 for ketorolac, suggesting substantial differences between individual NSAIDs [16].The annual incidence of NSAID-associated GI bleeding was also estimated in prospective outcome studies.Upper GI bleeding occurred in 3 to 4.5% of patients ingesting NSAIDs per year, and serious bleeding episodes due to bleeding of large blood vessel and/or gastric or intestinal perforation in approximately 1.5% [17].
In our study, there was a suggestion that patients with combined use of NSAIDs and glucocorticoids had a higher GI bleeding risk (OR 20, 95% CI 1.6-257) than patients treated with NSAIDs alone (OR 8.3, 95% CI 3.0-23).Similar results were published by Hallas et al. [14] (increase in risk from 5.5 for patients using NSAIDs alone to 10 for patients using NSAIDs and glucocorticoids), by Mellemkjaer et al. [15] (increase in risk from 3.6 to 7.4), by Piper et al. [18] (increase in risk from 1.1 to 4.4) and by Weil et al. [19] (increase in risk from 3.8 to 9.0).The combination of NSAIDs with oral anticoagulants is also associated with a higher risk of GI bleeding than use of NSAIDs alone.In the study of Mellemkjaer et al. [15], the risk for GI bleeding increased from 3.6 in NSAIDs users to 11.5 for combined use of anticoagulants and NSAIDs.In a cohort study among NSAIDs users (≥65 years), the risk for hospitalization due to a bleeding ulcer was 13fold increased (95% CI 6.3-26) for combined use of anticoagulants and NSAIDs, and 4.0 (95% CI 3.4-4.8)for use of NSAIDs only [20].In our study, eight patients were exposed to both NSAID and oral anticoagulants, but none in the control group, precluding the adjusted calculation of an OR, but the OR is approximately 20.
We also analyzed concurrent illnesses such as obesity, diabetes mellitus, hypertensive diseases, disorders of lipoprotein metabolism and history of non-bleeding GI ulcer as risk factors for GI bleeding (table 1).Unadjusted conditional regression analyses yielded significant ORs for patients with hypertensive diseases (OR 2.4, 95% CI 1.0-5.4).However, after adjusting for confounders, the risk estimate decreased (adjusted OR 2.2, 95% CI 0.8-6.0),which is in line with the conclusion of the authors of a recent review article who stated that hypertension may not be an independent risk factor for anticoagulant-related bleeding, when other risk factors were controlled for [21].On the other hand, the presence of co-morbidities in patients with a GI bleeding is associated with an increased mortality [22].

Limitations
Any epidemiologic studies may be subject to limitations such as confounding factors.Our data were retrieved from electronic medical records, with missing values for certain laboratory parameters such as INR, particularly in control patients.In addition, possible diagnosis misclassifi-cation and incomplete patient records cannot be fully excluded.Recent studies showed discrepancies of up to 40 or 50% of patients' medication when medical records and patient-reported use of drugs were compared [23][24][25].For non-prescription NSAIDs, disagreement was found in 74% of patients' medication [25], and a systematic review reported that up to 61% of patients had at least one omission error in prescription medication histories [26].Therefore bias due to possible underreporting of NSAID use can not be excluded.
No statement about the GI bleeding risk for individual NSAIDs was possible in our study due to the small number of cases.A meta-analysis suggested that ibuprofen, followed by diclofenac bear the lowest risk for GI bleeding.Azapropazone, tolmetin, ketoprofen, and piroxicam ranked highest for risk where indometacin, naproxen, sulindac, and aspirin occupied intermediate positions [27].According to a case-control study, ketorolac was associated with the highest risk followed by piroxicam, indomethacin, ketoprofen and naproxen.Lower risks were found for aceclofenac, ibuprofen, nimesulide and diclofenac [16].
In our study, doses and durations of exposure to the drugs were not taken into consideration.In previous studies, increasing doses were shown to be a risk factor for upper GI bleeding [16] especially for ibuprofen and naproxen [15].
Finally, the number of cases (n = 74) and matched controls (n = 148) was rather small.Bias due to sparse matched sets can affect the magnitude of OR estimates, even when there is no confounding, selection bias or measurement error [28].
In conclusion, the results of this small hospital-based case-control analysis suggest that firsttime GI bleeding is associated with high INR values, but not necessarily with oral anticoagulation alone if other risk factors such as DDIs are not present.Oral anticoagulants and NSAIDs as well as glucocorticoids and NSAIDs are frequently prescribed concomitantly in daily practice, and our results emphasize the problems related to concomitant use of these drugs.
AbbreviationsADR adverse drug reaction ATC anatomical therapeutical chemical BMI body mass index CI confidence interval DDI drug-drug interaction ICD-10 international classification of diseases, 10 th revision

Table 2
10 • w w w .s m w .ch * adjusted for glucocorticoids, NSAIDs, oral anticoagulants, SSRIs, TAIs, PPIs and hypertensive diseases Strategies for reducing GI bleedings include close monitoring of INR values, careful dose adjustment and prescription of drugs with a known and low potential for DDIs.