Abstract
Databases are used to screen for warfarin-medication interactions. We measured the reliability and validity of 5 databases. Databases were queried for 30 medications identified as having varying degrees of interactions with warfarin in a recent systematic literature review. Reliability was measured by the percentage agreement between the databases for each interaction and validity was measured by agreement between each database and the systematic literature review. All of the databases and the systematic review agreed on 14 medications (47%). There were 5 medications (17%) where all databases reported an interaction, but no interaction was noted by the systematic review. The databases did not agree on the remaining 11 medications (37%). Four of these interactions were of moderate or greater clinical significance. These commonly used databases frequently do not agree on the occurrence of warfarin-medications interactions, and some interactions identified by the best clinical evidence were not identified in several databases.
Background
Warfarin is a common cause of adverse drug events that contributes to approximately 43 000 emergency department visits per year. 1 When the effects are therapeutic (International Normalized Ratio (INR) 2.0-3.0), the risk of a major bleeding episode is approximately 4.8 of 100 patient years. This risk increases to 40.5 of 100 patient years if the INR is greater than 4.5. 2
Although overanticoagulation is the most common concern with warfarin, undertreatment is not without risk; patients with a mechanical heart valve and an INR less than 2.5 are at increased risk for peripheral and cerebral embolism. 3 As both overanticoagulation and underanticoagulation carry substantial risks, patients using warfarin must be careful to maintain their anticoagulation in the optimum range.
Medication interactions can alter the effects of warfarin and increase the risk of bleeding or thrombotic complications. Many patients taking warfarin are prescribed medications that may alter the efficacy of anticoagulation.4,5 These interactions may be identified by screening new prescriptions for interactions using a commercial system. 6 Several commercial databases are available to identify these interactions, but the reliability and validity of these databases has not been well described. The objective of this study was to determine the reliability of these databases by comparing commonly used commercial databases in the detection of warfarin-medication interactions and, further, determine the validity of the information provided by these databases by comparing the interactions identified by the databases to the interactions identified in a systematic literature review.
Methods
Design
This was an observational study of five commercial databases commonly used to identify warfarin-medication interactions in the clinical setting (Drugdex, Lexicomp, ePocrates, Medscape, PEPID). A single investigator conducted searches between November 2007 and February 2008 using the versions of the database that were current at that time.
We selected 30 medications from a recent systematic review of warfarin-medication interactions. 1 The review categorized the medications by the direction of their effect on the anticoagulant effects of warfarin (potentiation or inhibition), strength of evidence for the interaction (highly probably, probable, possible, and highly improbable), and the clinical significant of the interaction (major, moderate, minor, and nonclinical). Major potentiation was defined as an interaction that led to life-threatening bleeding defined as the requirement of at least 2 units of blood transfused over less than 7 days. Moderate potentiation was defined as an interaction that required an adjustment of the warfarin dose, increased the INR to greater than 5, or increased the INR by 1.5 or greater. Minor potentiation was defined as an interaction that increased the INR but did not require dosing adjustment and the INR did not increase more than 1.5 or exceed 5 at any point. Nonclinical interactions were defined as those that increased warfarin concentrations but did not affect INR.
The medications we selected were stratified by the type of interaction (1) erythromycin, cimetidine, propranolol, isoniazid, metronidazole, omeprazole (highly probable for potentiation), (2) levofloxacin, tetracycline, phenytoin, tramadol, fluvastatin (probable for potentiation), (3) amoxicillin, orilistat, lovastatin, chloramphenicol, ticlopidine (possible potentiation), (4) ranitidine, metoprolol, fluoxetine, naproxen, atenolol, clopidogrel (highly probable for no effect), (5) ibuprofen, ketoconazole, atorvastatin (probable no effect), and (6) carbamazepine, rifampin, trazodone, griseofulvin, nafcillin (highly probable inhibition).
For our analysis, medications classified as highly probable for no interaction or probable for no interaction were grouped as “no interaction,” while all other classifications were categorized as “interaction.” The literature review results were then compared to each database, and the interactions were compared among the databases. Reliability was measured by determining the percentage of interactions where all of the databases agreed on the presence or absence of an interaction. Validity was measured by determining the percentage of interactions where the databases correctly classified the interactions as present or absent compared to the systematic literature review. We also calculated the 95% confidence intervals for the percentage agreement among the databases and between the databases and the systematic review.
Results
All of the databases and the systematic review agreed on 14 (47%, 95% CI 28% to 66%) of the medications (atorvastatin, cimetidine, carbamazepine, erythromycin, fluvastatin, griseofulvin, lovastatin, levofloxacin, metronidazole, metoprolol, nafcillin, phenytoin, rifampin, and ticlopidine). There were 5 medications (17%, 6% to 35%) where an interaction was reported by all of the databases, but no interaction was noted by the review article (ibuprofen, ketoconazole, clopidogrel, fluoxetine, and naproxen).
The databases were not in agreement about the remaining 11 (37%, 95% CI 20% to 56%) warfarin-medication interactions. Three medications (omeprazole, propranolol, and isoniazid), classified as highly probable potentiation by the gold standard, were not identified by 1 database but were identified by the other 4. Two databases identified a warfarin-medication interaction with atenolol and 1 identified a warfarin-medication interaction with ranitidine (both identified by the review as not causing interactions). Four databases failed to identify a highly probable inhibition of warfarin by trazodone. The remaining medications had a warfarin-medication interaction reported by one or more databases and were rated by the review as having possible interactions (amoxicillin, chloramphenicol, and orilistat) or probable interactions (tetracycline and tramadol).
The interactions missed by 1 or more databases were of varying clinical significance. Although 4 (propranolol, omeprazole, tetracycline, and isoniazid) were rated as clinically insignificant, 1 was mild significance (amoxicillin), 3 were moderate significance (orilistat, chloramphenicol, and tramadol) and 1 was rated as a major interaction (trazodone).
Discussion
Our study found that the 5 databases inconsistently identified approximately one third of the evaluated warfarin-medication interactions. Although the clinical significance of these interactions varied, we believe that these results bring the reliability of the databases into question. It seems irrational that clinicians trying to identify interactions would reach different conclusions depending on the database they searched. One possible explanation would be that some databases elected not to list less significant interactions. However, in several cases, it appears one or more databases missed clinically significant interactions that could lead to overcoagulation or underanticoagulation.
The implications of the discrepancies between the databases and the systematic review are less clear. Our initial plan was that the systematic review would serve as a gold standard. We believe that this is a reasonable approach for medications where the review identified interactions that were rated as moderate or greater probability. The definition of moderate or severe used by Holbrook would include clinically relevant interactions. As several of the missed interactions were of moderate significance (one database missed a major interaction), we believe that the validity of at least some databases for identification is questionable.
The implications of interactions identified by the databases but not the systematic review are of less certain significance. The review used rigorous criteria including timing of the interaction, documentation of changes in INR, evaluation of potential confounders, rechallenge, dose-response, and other evidence to determine the probability of the interaction. It is likely that many case reports do not meet enough of these criteria to be rated as higher than possible. This definition could result in some medication interactions being underreported. Although most clinicians would consider an interaction probable if it was repeatedly identified in case reports or in a large case series, the review requires more rigorous reporting to rate an interaction as more than possible.
Three (ibuprofen, clopidrigel, and naproxen) of the five interactions reported in the databases but not in the systematic review are due to inhibition of platelet function. These interactions would not be reflected by changes in INR, so it is not surprising that these were not identified as significant interactions by the review. The first systematic studies identifying an interaction between warfarin and fluoxetine were published after the search dates for the systematic review, 7 so it is also not surprising that this interaction was not identified. The final interaction identified by the databases but not the systematic review was with ketoconazole. The basis for this interaction appears to be a single case report, 8 but the interaction was not confirmed in a small volunteer study. 9 Therefore, the interaction between ketoconazole and warfarin identified in the databases is of questionable validity.
Our results are limited by the use of a single systematic review to identify warfarin-medication interactions. As noted above, this review had very stringent criteria to report an interaction, and this may have led to underrecognition of some interactions. It is possible that an alternative gold standard may have identified these interactions and increased the measured agreement. One could also argue that Holbrook’s definition of mild and moderate interactions were simply a change in INR (without an identified increased rate of bleeding) and are of minimal clinical significance. However, we believe that these definitions are reasonable and that clinicians are interested in interactions that alter the INR to this degree. Another limitation is that we did not investigate all medications listed in the Holbrook article. It is possible that the interactions for the drugs we investigated do not reflect the overall performance of the databases.
There are several implications to misidentification of warfarin-medication interactions. If a significant interaction is missed, the patient will be at increased risk of bleeding or thrombosis. Even if there are no clinical complications, once the change in coagulation is identified, the patient will require warfarin dosing adjustment and increased monitoring. This will increase medical costs.
Although these databases identified the majority of interactions, several significant interactions were not detected. Patients on warfarin should be instructed that these databases may not identify all interactions that they must monitor themselves for evidence of changes in their coagulation status after starting a new medication. It is also reasonable to check several data sources for an interaction and increase the frequency of monitoring if there is any potential for an interaction.
In summary, we found discrepancies among 5 commonly used databases and found that several warfarin-medication interactions were not identified in drug interaction databases. Health care providers should be aware that the use of these databases does not guarantee that all interactions will be identified, and patients taking warfarin should be instructed to monitor themselves for any signs of bleeding whenever a new medication is started.
Footnotes
Dr. Heard is an employee of the Rocky Mountain Poison and Drug Center (RMPDC). RMPDC has consulting contracts with Thompson Publishing which produces Drugdex. Dr. Heard received only his salary for work performed on this project, and RMPDC received no funding from Thompson for this project.
The project described was supported by Award Number K08DA020573 from the National Institute On Drug Abuse. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute On Drug Abuse or the National Institutes of Health.