Risk of Hospitalization for Upper Gastrointestinal Tract Bleeding Associated With Ketorolac, Other Nonsteroidal Anti-inflammatory Drugs, Calcium Antagonists, and Other Antihypertensive Drugs

Background  Nonsteroidal anti-inflammatory drugs (NSAIDs) cause substantial morbidity and mortality from upper gastrointestinal tract disease. Ketorolac tromethamine has been singled out as an NSAID with a distinct gastrotoxicity profile. Calcium channel blockers, a class of antihypertensive drugs, have also been found to increase the risk of gastrointestinal tract bleeding.

Methods  We identified 1505 patients hospitalized because of upper gastrointestinal tract bleeding and/or perforation, and we randomly sampled 20000 controls in the source population.

Results  The adjusted relative risk (RR) for upper gastrointestinal tract bleeding and/or perforation in NSAID users compared with nonusers was 4.4 (95% confidence interval [CI], 3.7-5.3). The risk increased with higher daily doses. Ketorolac presented the highest risk (RR, 24.7; 95% CI, 9.6-63.5) and piroxicam ranked second (RR, 9.5; 95% CI, 6.5-13.8). Ketorolac was 5 times more gastrotoxic than all other NSAIDs (RR, 5.5; 95% CI, 2.1-14.4). The excess risk with ketorolac was observed with both oral and intramuscular administration and was already present during the first week of therapy. Among the various antihypertensive drug classes, β-blockers were associated with the lowest relative risk (RR, 1.0; 95% CI, 0.7-1.4), and current use of calcium channel blockers with the highest (RR, 1.7; 95% CI, 1.3-2.1). The association with calcium channel blockers declined when adjusting for various markers of comorbidity (RR, 1.4; 95% CI, 1.1-1.8). Past use of calcium channel blockers was also associated with an increased risk (RR, 1.5; 95% CI, 1.3-1.8).

Conclusions  The excess risk of major upper gastrointestinal tract complications associated with outpatient use of ketorolac suggests an unfavorable risk-benefit assessment compared with other NSAIDs. More data are required to reduce the uncertainty about the apparent small increased risk of upper gastrointestinal tract bleeding in patients using calcium channel blockers.

APPROXIMATELY 30 million patients worldwide take nonsteroidal anti-inflammatory drugs (NSAIDs) every day. In addition to their anti-inflammatory and analgesic effects, NSAIDs have unwanted effects on the upper gastrointestinal tract, mainly by inhibition of the enzyme cyclooxygenase (COX).¹ The gastroduodenal injury ranges from dyspepsia to fatal upper gastrointestinal tract bleeding (UGIB) and perforation. Many epidemiologic studies have shown that NSAIDs increase the risk of peptic ulcer bleeding 3- to 5-fold.^2,3 Ketorolac tromethamine is an NSAID that has attracted the attention of a number of regulatory bodies after the reporting of fatal cases of gastrointestinal tract bleeding.⁴ A recent meta-analysis assessed the relative risk (RR) of UGIB associated with individual NSAIDs⁵; however, the effect of ketorolac could not be analyzed because of lack of information in the original studies.

Calcium antagonists are 1 of the therapeutic groups most frequently prescribed in hypertension. Little evidence exists for an association between use of calcium channel blockers and UGIB, in contrast with NSAIDs. However, a randomized study was stopped prematurely because of the occurrence of more deaths caused by major bleeding among patients taking nimodipine than among the placebo group.⁶ A cohort study found that, compared with β-blockers, use of calcium antagonists was associated with an increased risk of UGIB.⁷

We carried out this study to provide quantitative estimates of the RR of hospitalization for UGIB associated with the use of individual NSAIDs, including ketorolac, by means of the automated health care data files from the region of Friuli-Venezia Giulia (F-VG), Italy. In the same population, we also assessed the risk of UGIB among users of antihypertensive agents, including calcium channel blockers.

The region of F-VG maintains complete and accurate computerized records of hospitalizations since 1985 and outpatient drug prescriptions since 1991 for the 1.2 million inhabitants of the region. The Hospital Services Database collects data on all admissions to public and private hospitals of F-VG. This database contains information on personal demographics of enrollees, admitting and discharge date, vital status, 1 admitting diagnosis, and a maximum of 4 discharge diagnoses (International Classification of Diseases, Ninth Revision [ICD-9])⁸ and surgical procedures. The Outpatient Prescription Database keeps a record of each prescription dispensed to F-VG residents and covered by the National Health Service. The information includes the drug name, date of prescription, strength, and number of refills. The Patient Identification Database keeps a record for each resident in F-VG. It includes a unique personal identifier (present in all computerized files that allows linking of the various databases), name, sex, date of birth, address, and general practitioner, as well as special medical conditions, such as diabetes, neoplasia, and glaucoma. This enrollment file is updated daily for address and general practitioner changes, births, deaths, and the arrival and departure of residents from the region.

The source population included permanent residents, aged 25 to 89 years, of the province of Udine, Italy, from January 1, 1991, until June 30, 1995 (361383 persons), of Gorizia, Italy, from July 1, 1991, until June 30, 1995 (98397 persons), of Pordenone, Italy, from March 1, 1992, until June 30, 1995 (183304 persons), and of Trieste, Italy, from March 1, 1993, until June 30, 1995 (193775 persons). Persons with a hospitalization for cancer, coagulopathy, alcohol abuse, esophageal varices, Mallory-Weiss syndrome, chronic liver disease, or regional enteritis before the beginning of the study period were excluded. Women with a pregnancy during the study period were also excluded. A total of 780463 persons formed the study cohort.

All members of the study cohort were followed up until the earliest occurrence of a UGIB hospitalization (primary discharge diagnosis of bleeding or perforated peptic ulcer [ICD-9 codes 531-534], hematemesis [578.0], melena [578.1], or gastrointestinal tract hemorrhage [578.9]), a hospitalization with 1 of the exclusion conditions (as defined above), death, emigration from the region, or end of the study period. The following computer-based exclusion criteria were used: presence of a code of an exclusion condition in second, third, or fourth discharge diagnoses during the UGIB hospitalization, or in a hospitalization within 3 months after the UGIB hospitalization. Persons with an admission to the hospital in the month preceding the hospitalization for UGIB were also excluded. This left a total of 1779 potential cases.

A case of UGIB was defined as an episode of hemorrhage or perforation located in the stomach and/or duodenum, or a bleeding or perforated peptic ulcer (site not reported) and no exclusion conditions. The date of admission to the hospital was used as the index date. We obtained from the hospitals copies of the complete clinical chart corresponding to the UGIB hospitalization for a stratified random sample of 196 potential cases. A positive endoscopic examination present in the hospital records was requested for a patient to be a case. Ninety percent of patients with ICD-9 codes 531 and 532 were confirmed cases, as were close to 60% of patients with ICD-9 codes 533, 534, and 578. On the basis of the predictive values of specific ICD-9 codes in this first validation process, we reviewed hospital charts for all remaining patients with ICD-9 codes 533, 534, and 578 as the primary discharge diagnosis (n=465). Personal identifiers were masked before manual review of records. After the review, 274 patients were classified as noncases, either for lack of UGIB or for presence of an exclusion condition. Finally, 1505 patients constituted our group of UGIB cases.

A group of 20000 persons was randomly sampled from the study cohort, with a random date included in the follow-up period of that individual taken as the index date. All cases' eligibility criteria were applied to the control series.

A person was defined as a current NSAID user when the supply of the most recent NSAID prescription lasted until the index date. Other NSAID exposure definitions employed were recent NSAID user, when the end of the current use period was between 1 and 30 days before the index date; past NSAID user, when the end of the current use period was between 31 and 60 days before the index date; distant past NSAID user, when the end of the current use period was between 61 and 150 days before the index date; and nonuser, when there was no NSAID use in the 150 days before the index date.

Among current users of NSAIDs, 3 mutually exclusive categories were created: current single users, who had not received any other individual NSAID within 2 months before the date of current NSAID prescription; current switcher users, who had a prescription for another NSAID within 2 months before the current prescription date, but with no overlap; and current multiple users, who had a prescription for another NSAID within 2 months before the current prescription date, but with overlap. The risk of individual NSAIDs was examined only among current single users. More than 40 individual NSAIDs were available on the Italian market. Aspirin was also sold as an over-the-counter drug. A few other NSAIDs (mainly ibuprofen) were also available over the counter, representing less than 5% of total use of nonaspirin NSAIDs.

The presence of a dose-response relationship was investigated among current single users by means of 2 dose categories, as in a previous study: low through medium and high daily dose.⁹ Duration of use was assessed in current users, defined as the number of months of continuous NSAID therapy. Route of administration was also evaluated among current single users as oral, parenteral, and suppository formulations. A separate variable was created to record use of NSAID gel preparation.

A person was defined as a current antihypertensive user when the last antihypertensive prescription before the index date was filled during the previous month. Other antihypertensive exposure definitions employed were recent antihypertensive user, when the last antihypertensive prescription before the index date was filled 31 to 60 days before that date; past antihypertensive user, when the last antihypertensive prescription before the index date was filled 61 to 365 days before that date; distant past antihypertensive user, when the last antihypertensive prescription before the index date was filled between 1 and 2 years before that date; and nonuser, when there was no antihypertensive prescription in the last 2 years before the index date. Six therapeutic classes of antihypertensive drugs were identified: diuretics, β-blockers, calcium channel blockers, angiotensin-converting enzyme inhibitors, α-antagonists, and central agonists. Combination antihypertensive drugs were assigned to each respective class of the active ingredients.

Several other potential risk factors for UGIB were identified through the computerized records. These included previous hospitalization for UGIB (before the beginning of the study period) and hospitalization for other diseases, as well as filling of prescriptions for several categories of drugs used as morbidity markers (β-agonists, insulin, oral hypoglycemic drugs, cardiovascular drugs, and other drugs) or as independent risk factors (corticosteroids and anticoagulants).

We used SPSS to run unconditional logistic regression and compute the RRs and 95% confidence intervals (CIs) of UGIB associated with NSAID and antihypertensive use, adjusting for risk factors mentioned above.¹⁰ Nonusers of the respective drug class were used as the reference group. Interactions were examined by means of stratified analyses and use of a common reference group.

Among the 1505 cases of UGIB, 623 had a gastric site of bleeding or perforation, 805 had a duodenal site, and 77 had multiple sites or a diagnosis of peptic ulcer with no specific site. Overall, the case fatality rate was 4.6%. A perforated lesion was found in 223 patients, with a case fatality rate of 9%.

The adjusted RR for current NSAID use was 4.4 (Table 1). Among current users, patients exposed concomitantly to 2 or more individual NSAIDs in the 2 months preceding the UGIB presented an RR of 7.8 (95% CI, 5.6-11.0). The risk rapidly decreased as time since last NSAID prescription increased. Persons with their most recent NSAID use more than a month before the index date presented a risk of UGIB similar to that in nonusers. No difference in risk associated with NSAID use was observed between gastric and duodenal cases (Table 2). No significantly different estimates of RR were found for bleeding, perforation, or fatality.

Table 1 presents estimates of risk for several risk factors. Age was a major predictor of the risk of UGIB. Men had higher risks than women. The adjusted RRs for current use of corticosteroids, antiplatelet drugs, anticoagulants, and aspirin were 1.4 (95% CI, 0.7-2.8), 2.0 (95% CI, 1.4-2.7), 2.2 (95% CI, 1.4-3.4), and 2.3 (95% CI, 1.7-3.2), respectively. Adjustment for exposure to all these drugs did not alter the estimate of RR associated with NSAID use. The crude RR for exposure to NSAID gel preparations was 3.3 (95% CI, 2.4-4.6). However, after controlling for systemic NSAID use, the risk disappeared completely (1.0; 95% CI, 0.7-1.5).

A previous hospitalization for complicated peptic ulcer was the single most important determinant of UGIB (19.7; 95% CI, 13.9-28.1), and use of acid-suppressing drugs was an independent marker of risk (Table 1). Ulcer antecedents were also a strong modifier of NSAID effect on the risk of UGIB, with the RR for current NSAID use in persons without ulcer antecedents being greater than in those with ulcer antecedents (Table 3). We used a common reference group to explore the interaction between NSAIDs and ulcer antecedents: persons without ulcer antecedents not exposed to NSAIDs. The adjusted RR for those with both risk factors was 12.5 (95% CI, 9.5-16.4). To study the independent effect of NSAID use, we restricted the remaining analyses to the subset of persons without ulcer antecedents.

The risk among individual NSAIDs showed marked differences (Table 4). All individual NSAIDs except 2 had RRs smaller than 6 compared with nonuse. Piroxicam was associated with an RR of 9.5, and ketorolac was the NSAID with the highest RR (24.7; 95% CI, 9.6-63.5). The RR associated with use of piroxicam vs all other NSAIDs (piroxicam not included) was 2.6 (95% CI, 1.6-4.0), and the RR with use of ketorolac vs all other NSAIDs (ketorolac not included) was 5.5 (95% CI, 2.1-14.4). The estimate of RR with oral formulation of ketorolac (19.9; 95% CI, 4.2-93.0) was smaller than that with intramuscular formulation (28.3; 95% CI, 8.7-92.0). The RR in the first week of therapy was 19.1 (95% CI, 4.6-79.2), and the RR was 30.3 (95% CI, 8.5-107.5) for longer durations. The indication for ketorolac treatment was requested from attending physicians. Information was not available in 1 patient and 1 control subject. In 9 patients and 6 control subjects, the indication was either osteoarthritis or chronic pain. Only 1 patient and 1 control subject were prescribed ketorolac for acute pain. Table 5 indicates that NSAID users taking high daily doses had a risk 2 times greater than that in persons using low through medium doses. Among current users, the RR was rather constant during the first year of therapy, suggesting increasing cumulative risk with longer treatment duration. Suppository and oral form of NSAIDs presented similar risk, while the intramuscular form was associated with the highest risk (Table 5). However, when we analyzed intramuscular preparations excluding piroxicam and ketorolac preparations, the risk of intramuscular administration was no different from that of either oral or suppository forms. The RR for parenteral piroxicam was 24.9 (95% CI, 8.9-69.9), and the corresponding risk for diclofenac sodium was 4.8 (95% CI, 1.3-17.9). Table 6 presents estimates of RR of UGIB associated with individual NSAIDs by daily dose among individual NSAIDs with 5 or more cases exposed overall. Ketorolac was the only NSAID with an RR greater than 10 when used at low through medium daily dose (20.0; 95% CI, 4.3-93.6).

One hundred seventy-one episodes of UGIB occurred among single users of antihypertensive drugs. These patients had a significantly increased risk of UGIB (1.7; 95% CI, 1.4-2.1) relative to nonusers (Table 7). The extent of risk was comparable in the initial months of antihypertensive therapy than after more than 1 year of treatment. Among the various antihypertensive drug classes, users of calcium-channel blockers presented the highest risk of UGIB (Table 8). Past users of calcium antagonists still had an elevated RR compared with nonusers (1.5; 95% CI, 1.3-1.8). The RR associated with use of digoxin (marker of heart failure) was 1.4 (95% CI, 1.1-1.7). Previous hospitalization for angina or myocardial infarction was not associated with the risk of UGIB. When we analyzed the subset of single users in each antihypertensive drug class (exposed to only 1 antihypertensive drug within 2 months before the index date), results were similar to the ones among all current users (data not shown). Adjusting for anticoagulant treatment did not change the results. However, when we introduced markers of comorbidity in the model, such as number of previous hospitalizations and history of cerebrovascular disease, diabetes, and use of aspirin and antiplatelet drugs, the RR associated with calcium channel blockers decreased to 1.4 (95% CI, 1.1-1.8). Estimates of RR within individual calcium channel blockers were all overlapping, ranging from 1.3 (95% CI, 0.5-3.2) for nimodipine to 3.0 (95% CI, 1.6-5.7) for lacidipine. Relative to users of β-blockers, patients exposed to calcium channel blockers had an RR of 1.7 (95% CI, 0.9-3.2).

The goal of this study was to examine the risk of UGIB associated with a therapeutic group known to present such risk (NSAIDs) and to compare it with the risk with another therapeutic group recently suspected to be gastrotoxic (calcium channel blockers) in the same general population. We found estimates of RR of UGIB between 4 and 5 associated with the use of NSAIDs, which are in line with the literature.^2,3 Our findings suggest a higher risk of severe gastrointestinal tract toxic reactions among ketorolac users than among users of other NSAIDs. This increased risk in users of ketorolac was already apparent in the first week after starting therapy, was present at both low through medium daily dose and high dose, and was seen with both the oral and the intramuscular preparations. The annual background incidence rate of hospitalizations for UGIB was 1 per 1000 persons in our study population, an estimate identical to the one observed in a general population in Canada.¹¹ The corresponding incidence rate among ketorolac users was 20 per 1000 person-years. In a recent meta-analysis on the risk of gastrointestinal tract complications associated with individual NSAIDs,⁵ apazone was found to have the greatest RR compared with ibuprofen (9.2; 95% CI, 4.0-21.0); ketorolac was not studied in any of the articles reviewed. In our study, the RR of ketorolac compared with ibuprofen was 11.7 (95% CI, 2.5-54.0). The information on the indication for prescribing ketorolac did not provide any evidence of higher susceptibility to gastrointestinal tract complications in these recipients. A large hospital-based epidemiological study comparing parenteral ketorolac and opiates found a slightly higher incidence of gastrointestinal tract and operation-site bleeding among patients receiving ketorolac.¹² The authors estimated the risk of clinically serious gastrointestinal tract bleeding to be between 1% and 2%. Both the study population and the studied drugs were noticeably different relative to the current study, preventing a direct comparison of results. One other published study found a higher risk of ulcer and erosions with ketorolac than with other NSAIDs.¹³ Our study is the first population-based study that compares the risk of UGIB between outpatient use of ketorolac and other NSAIDs.

Antihypertensive drug use was associated with a slightly elevated risk of UGIB in our study, and this was mainly observed in users of calcium channel blockers, with an RR close to 2. Lately, a randomized double-blind controlled trial^6,14 that was designed to determine whether nimodipine (a calcium channel blocker) would reduce the incidence of neurologic deficits during the postoperative period among patients undergoing cardiac valve replacement was stopped prematurely because of the occurrence of more deaths among patients randomly assigned to nimodipine than among patients in the placebo group. The underlying cause of death was major bleeding in 6 of 9 deaths, 5 in the nimodipine group and 1 in the placebo group. Another recent study showed that use of calcium antagonists increased the risk of hemorrhage in the elderly compared with β-blockers. Their estimate of risk for calcium channel blockers was similar to ours, whereas their risk for NSAIDs (RR, 1.5) was surprisingly lower than the one observed in our study and in most published studies.^2,3,5 In our study, the risk associated with calcium channel blockers was reduced when we adjusted for markers of comorbidity such as number of previous hospitalizations, chronic medical conditions, and use of digoxin. In addition, we saw that patients exposed to calcium channel blockers more than a year before the episode of gastrointestinal tract bleeding still presented a slightly elevated risk. This observed risk among past users does not support the presence of an acute effect associated with calcium antagonists. Therefore, it is unclear whether use of calcium channel blockers or characteristics of the persons exposed to these drugs, not included in our analysis, are the determinant of this small increased risk.

The estimates of risk obtained in this study were adjusted for age and history of peptic ulceration, the 2 most important risk factors for UGIB. Further adjustment for use of several groups of drugs and various comorbidities did not change the risk associated with NSAIDs, but gradually reduced the one associated with antihypertensive drugs and especially the one with calcium channel blockers. As in all observational studies, there is some concern for residual confounding caused by unknown or inaccurately measured risk factors. Selection bias is also a possibility that could partially explain our findings. In the case of ketorolac, we contacted the prescribing physicians and excluded with confidence the presence of confounding by indication. With respect to the slightly elevated risk of calcium antagonists, several findings, such as a reduction of risk when adjusting for comorbidity status, no decrease in risk with time elapsing since stopping calcium channel blockers use, and a similar risk of UGIB in users of digoxin, all point to some uncontrolled confounding. As for misclassification of exposure, drug exposure in this administrative database is recorded prospectively at the time a person fills the prescription in the pharmacy. It is known that compliance with drug regimens is never complete, and this could introduce some minor underestimation of the observed associations.

In conclusion, our data confirm an overall RR associated with use of NSAIDs between 4 and 5. The risk with ketorolac was distinctly higher than with the remaining NSAIDs. This suggests an unfavorable risk-benefit assessment for outpatient use of ketorolac compared with other NSAIDs. These data also support use of the lowest effective doses of any NSAID to reduce the burden of serious gastrointestinal tract complications. In light of the NSAID-associated gastrointestinal risk, more data are needed to elucidate whether the RR of less than 2 observed in users of calcium antagonists is real, or whether other causal pathways mediated through external risk factors would account for this small association.

Accepted for publication April 23, 1997.

This research was supported in part by Laboratorios Dr Esteve, Barcelona, Spain.

We thank Ana Ruigómez, MD, PhD, for her multiple comments on earlier versions of the manuscript.

Reprints: Chiara Cattaruzzi, PhD, Azienda Ospedaliera "S. Maria della Misericordia" Farmacia, Udine 3100, Italy (e-mail: lagarcia@ceife.es).