P. Wells, A. Holbrook, N. Crowther
Nov 1, 1994
Citations
7
Influential Citations
512
Citations
Quality indicators
Journal
Annals of Internal Medicine
Abstract
Warfarin is the most widely used oral anticoagulant drug in North America, in part because of its relatively predictable onset and duration of action and its excellent bioavailability [1-3]. Warfarin, a racemic compound with a more potent S enantiomer and a less potent R enantiomer, achieves its anticoagulant effect by inhibiting the activation of vitamin Kdependent clotting factors. Oral anticoagulants are effective in the prevention and treatment of deep venous thrombosis and also in the prevention of thromboembolic events in patients with atrial fibrillation [4-8], prosthetic heart valves [9-12], and indwelling central venous catheters [13], as well as in patients who have had myocardial infarction [14-20]. Despite excellent evidence [21] for its clinical value, warfarin may be underused because of the inconvenience of monitoring and the concern about potential complications, primarily bleeding [22-24]. The risk for complications may be increased when concomitant drug therapy is required [25, 26]. The potential for warfarin to interact with other drugs, resulting in changes in its anticoagulant effect, is widely recognized among health professionals and informed patients. Many published reviews [3, 27-35] and extensive lists in standard medical textbooks (110 interacting drugs listed in the United States Pharmacopeia Dispensing Information [USP DI]) [36] attest to the widely held belief that drug interactions with warfarin are common and potentially harmful. However, reports on drug interactions are replete with small case series, single case reports, and extrapolation of in vitro or animal data. This evidence, when judged by the usual evaluative scales for therapy, is of lower quality [21, 37, 38]. Given that expensive clinical trial resources are unlikely to be used to address the fine points of therapy such as drug interactions, it is essential to evaluate the quality of these studies. Thus, we evaluated the quality of reports about drug and food interactions with warfarin. We prospectively applied explicit, reproducible criteria for determining the strength of inferred causation. We recognized that a yes, did cause or a no, did not cause conclusion might not be possible and used an estimate of the probability of causation (level of evidence) by adapting previously described principles of causality assessment [39-42]. Methods Relevant studies were identified by searching the MEDLINE and TOXLINE databases from 1966 to the end of October 1993 using the Medical Subject Headings warfarin, drug interactions, and English only. Articles were considered eligible for evaluation if they contained original data about drug and food interactions with warfarin in humans. Bibliographies were also checked for additional pertinent studies. Reports on drugs not available in the United States or Canada were excluded. Eligible studies were evaluated independently by two authors according to three main categories: participants, description of interaction, and level of evidence. Participants We separated reports into those describing 1) patients who received the interacting drug during usual warfarin therapy and 2) healthy volunteers or patients prospectively entered into an experiment while receiving warfarin. Description of Interaction We noted the drug or food affected by the interaction, the type of interaction (potentiation, inhibition, or no effect), and the proposed or documented mechanism of interaction. Level of Evidence (Assurance) Each article was evaluated, with yes or no responses given according to seven criteria previously approved by a panel of experts in the fields of thromboembolism, clinical pharmacology, and clinical epidemiology (Appendix 1). Appendix 1. Criteria for Establishing a Drug or Food Interaction with Warfarin Responses to four criteria (A to D) required additional guidelines in order to be specific to the evaluation of drug interactions with warfarin. To meet criterion A, for patient-based studies, before the potentially interactive substance was started, the warfarin dose and intensity of anticoagulation must have been stable. Further, the potentially interacting substance had to be used in usual doses for enough time to attain a substantial plasma level. For volunteer-based studies, participants had to have received warfarin alone and with the interacting drug for similar periods. For criterion B, in patient-based studies, the coagulation variable had to be outside the therapeutic range, whereas for volunteer studies, a change of at least 20% in coagulation variables was required. To satisfy criterion C, we required some indication that medical conditions, especially liver disease, as well as other drug therapy and diet (notably dietary vitamin K intake) were constant. However, for healthy volunteers we assumed that these confounders were absent, even if not explicitly stated. For criterion D, other objective evidence refers to changes in plasma levels of warfarin or of vitamin Kdependent factors II, VII, IX, or X. The level of evidence (assurance) that a drug or food interaction with warfarin could occur was then determined as outlined in Appendix 1 (Table 3). Level 1 evidence obtained from patient-based and volunteer-based reports was considered definitive evidence of an interaction. Inter-rater agreement was assessed using a weighted statistic [43]. Wherever reported, we considered data from individual participants rather than relying on summary results. If individual data were not available, the summary statistics were used. If several studies assessed the same drug, the interaction supported by the highest level of evidence was considered the final warfarin interaction. If a study in volunteers showed a different type of interaction than did a study with the same level of evidence in a patient-based report, the latter was considered to be the final warfarin interaction. Level 1 studies were further reviewed with regard to the severity of effect of the interaction. Two thresholds were established: The first was clinically evident hemorrhage or thrombosis and the second was a doubling or halving [for inhibition] of coagulation measurements. Articles were also appraised for any description of the mechanism of interaction. Pharmacokinetic data suggesting altered drug clearance or changes in pharmacodynamic data, such as clotting factor or vitamin K levels, were considered adequate evidence. Because warfarin is a racemic compound, we classified the interaction as 1) being stereoselective if a differential change was noted in enantiomer concentrations, 2) being nonstereoselective if both enantiomers changed substantially but by a similar proportion, or 3) affecting clearance by an unknown mechanism if only plasma warfarin levels without its enantiomers were measured. Results Of 793 citations retrieved, 120 contained original data on 186 interactions. The weighted statistic for the level of evidence evaluation was 0.67, representing substantial agreement [43]. Disagreement about the scores of 38 articles was resolved by repeat review and discussion until a consensus was reached. Forty-three of 86 different drug and food interactions appraised were judged highly probable (level 1 evidence): Sixteen had a potentiating effect, 10 had an inhibiting effect, and 17 had no effect. These drugs were classified into the following six categories: antibiotics, anti-inflammatory agents or analgesics, cardiac drugs, gastrointestinal drugs, drugs active on the central nervous system, and miscellaneous drugs or foods (Table 1). The reported interactions for another 18 drugs were judged probable (level 2 evidence): 14 potentiating, 1 inhibiting, and 3 with no effect. Of the remaining 25 interacting medications, 16 were considered possible (level 3 evidence) and 9 were doubtful (level 4 evidence). All reports described interactions leading to some effect on warfarin therapy (as opposed to altering the effect of the other drug). Our summary of results listing the type of interaction by level of evidence according to drug categories is presented in Appendix 2. Table 1. Level 1 Evidence of Drug and Food Interactions with Warfarin* Appendix 2. Drug and Food Interactions with Warfarin by Level of Supporting Evidence and Type of Interaction* Drugs Potentiating the Effect of Warfarin Many antibiotics are reported to potentiate the effect of warfarin. The evidence was considered highly probable (level 1) for cotrimoxazole, erythromycin, isoniazid, fluconazole, miconazole, and metronidazole [44-50] and was probable (level 2) for ciprofloxacin, itraconazole, and tetracycline [51-58]. The evidence was much weaker for six other antibiotics [59-63]. Several cardiac drugs had highly probable evidence [64-73] that they potentiated the effect of warfarin: These included amiodarone, clofibrate, propafenone, propranolol, and sulfinpyrazone. Sulfinpyrazone's effect was biphasic, which means that an initial potentiation of the warfarin anticoagulant effect was noted, followed by inhibition of the effect. Quinidine, simvastatin, and acetylsalicylic acid had probable evidence that they potentiated warfarin [74-76]. Possible and doubtful evidence were reported for five other drugs [77-81]. Among the anti-inflammatory or analgesic drugs, phenylbutazone, piroxicam, acetylsalicylic acid, acetaminophen, and dextropropoxyphene had highly probable or probable evidence [74, 82-89]. The other medications with highly probable or probable evidence were cimetidine, omeprazole, alcohol (only if concomitant liver disease was present), chloral hydrate, disulfiram, phenytoin (late effect of inhibition), tamoxifen, anabolic steroids, and influenza vaccines [74, 90-105]. Drugs Inhibiting the Effect of Warfarin Fewer drugs inhibited the effect of warfarin, but the proportion with level 1 evidence was higher. Highly probable evidence was reported for nafcillin, rifampin, griseofulvin, cholestyramine, barbiturates, carbamazepine, chlordiazepoxide, sucra