The timing of INR values required for inclusion into the study groups is shown. Not every INR in the preindex period was between 2.0 and 3.5. Only the first INR in the follow-up period was included for analysis of outcomes.
eTable. Potential Study Antibiotics
Clark NP, Delate T, Riggs CS, Witt DM, Hylek EM, Garcia DA, Ageno W, Dentali F, Crowther MA, for the Warfarin-Associated Research Projects and Other Endeavors Consortium. Warfarin Interactions With Antibiotics in the Ambulatory Care Setting. JAMA Intern Med. 2014;174(3):409-416. doi:10.1001/jamainternmed.2013.13957
The effect of antibiotic coadministration on the international normalized ratio (INR) in a relatively stable, real-world warfarin population has not been adequately described. Case reports and studies of healthy volunteers do not account for the potential contribution of acute illness to INR variability.
To compare the risk of excessive anticoagulation among patients with stable warfarin therapy purchasing an antibiotic (antibiotic group) with the risk in patients purchasing a warfarin refill (stable controls) and patients with upper respiratory tract infection but not receiving an antibiotic (sick controls).
Design, Setting, and Participants
A retrospective, longitudinal cohort study evaluated patients receiving warfarin between January 1, 2005, and March 31, 2011, at Kaiser Permanente Colorado, an integrated health care delivery system. Continuous data were expressed as mean (SD) or median (interquartile range). Multivariable logistic regression analysis was used to identify factors independently associated with a follow-up INR of 5.0 or more. A total of 5857 (48.8%), 5579 (46.5%), and 570 (4.7%) patients were included in the antibiotic, stable control, and sick control groups, respectively. Mean age was 68.3 years, and atrial fibrillation was the most common (44.4%) indication for anticoagulation.
Warfarin therapy with a medical visit for upper respiratory tract infection or coadministration of antibiotics.
Main Outcomes and Measures
Primary outcomes were the proportion of patients experiencing a follow-up INR of 5.0 or more and change between the last INR measured before the index date and the follow-up INR.
The proportion of patients experiencing an INR of 5.0 or more was 3.2%, 2.6%, and 1.2% for the antibiotic, sick, and stable groups, respectively (P < .001, antibiotic vs stable control group; P < .017, sick vs stable control group; P = .44, antibiotic vs sick control group). Cancer diagnosis, elevated baseline INR, and female sex predicted a follow-up INR of 5.0 or more. Among antibiotics, those interfering with warfarin metabolism posed the greatest risk for an INR of 5.0 or more.
Conclusions and Relevance
Acute upper respiratory tract infection increases the risk of excessive anticoagulation independent of antibiotic use. Antibiotics also increase the risk; however, most patients with previously stable warfarin therapy will not experience clinically relevant increases in INR following antibiotic exposure or acute upper respiratory tract infection.
Among the challenges in providing optimal anticoagulation with warfarin is an extensive list of drug-drug interactions that can result in fluctuating international normalized ratio (INR) values.1- 3 Variability in the INR and low time within the therapeutic range (ie, <65%) have been associated with increased risk of bleeding and thrombosis during warfarin therapy.4- 6 Warfarin interactions with antibiotics are particularly problematic because of their intermittent use and short duration of exposure, which creates the potential for more INR variability and the need for additional INR monitoring.7
Despite widespread recognition, surprisingly little is known regarding the scope or typical INR effect during antibiotic coadministration in a stable, real-world warfarin population. A systematic review1 concluded that further analyses of warfarin interactions with other drugs are urgently needed. Even authoritative consensus guidelines, such as the ninth edition of the antithrombotic guidelines from the American College of Chest Physicians,7 are unable to inform clinical practice because of insufficient documentation of the clinical consequences of warfarin interactions with antibiotics. As a result, there is a lack of agreement among clinicians and drug interaction references as to which warfarin interactions with antibiotics are most important, what monitoring should be undertaken, and how the interactions should be managed.8
A study9 using a Medicaid database identified patients receiving long-term warfarin therapy and concomitant fluoroquinolones, sulfonamides, and azole antifungals and compared them with a control population of patients receiving warfarin and antibiotics not thought to interact substantially with warfarin (amoxicillin and cephalexin). Although the study found an increased risk of hospitalization for gastrointestinal bleeding in the study group, increased bleeding overall was observed in both groups. These findings suggest that infection-related factors such as hepatic inflammation, fever, reduced vitamin K intake, or use of over-the-counter remedies— not just interactions with antibiotics—contributed to adverse anticoagulant therapy outcomes. Another retrospective study10 compared INR outcomes in patients receiving warfarin who were receiving concomitantly prescribed azithromycin, levofloxacin, or trimethoprim-sulfamethoxazole with outcomes in a control group in which terazosin was coprescribed. Increased INRs were observed across antibiotic groups, but the effect of concurrent illness on INR response could not be ruled out because the controls were not acutely ill.
To our knowledge, no study has attempted to quantify the effect of illness with or without warfarin-antibiotic interactions on INR response. The purpose of this study was to determine whether antibiotics coprescribed to patients receiving long-term warfarin therapy are more likely to result in clinically significant INR increases compared with patients receiving long-term warfarin therapy who are not ill (stable controls) and those who are acutely ill but do not receive concomitant antibiotics (sick controls).
This retrospective, longitudinal cohort analysis was conducted at Kaiser Permanente Colorado (KPCO), an integrated health care delivery system providing medical care for approximately 530 000 patients. Electronic medical, pharmacy, and laboratory records and the anticoagulation database (DAWN-AC; 4S Systems, Ltd) were used to identify patients, treatments, and outcomes for this study. All study activities were reviewed and approved by the KPCO institutional review board.
Comprehensive anticoagulation services at KPCO are provided by a centralized clinical pharmacy anticoagulation and anemia service (CPAAS). Details regarding operational aspects of this service have been described previously.11 Providers are alerted regarding drug interactions when ordering new medications within the electronic medical record. When antibiotics interacting with warfarin are deemed necessary, the standard operating procedure at CPAAS involves continuing warfarin without a preemptive dosage change and retesting the INR 3 to 7 days after antibiotic initiation.
Patients receiving long-term warfarin therapy between January 1, 2005, and March 31, 2011, were eligible for study inclusion. The study inclusion index dates for the antibiotic, sick control, and stable control groups were the dates of antibiotic purchase, prescription antitussive purchase or upper respiratory tract infection diagnosis, and warfarin refill purchase, respectively. All patients purchased warfarin within the 90 days before the respective index date. At least 2 INR measurements during the 120 days before the index date were required, with the 2 values most proximal to the index date being between 2.0 and 3.5 to limit the influence of INR variability during the observation period.
Antibiotic group patients had to have purchased a prescription for an oral antibiotic (Supplement [eTable] includes a list of potential study antibiotics) and had at least 1 INR measured 3 to 15 days after the purchase (defined as the follow-up INR). Sick control group patients had to have either purchased guaifenesin with codeine or have a coded provider visit for upper respiratory tract infection (codes available from the authors) and a follow-up INR measured within 15 days. Stable control group patients purchased a warfarin refill and had a follow-up INR measured within the next 3 to 30 days. Only the first INR recorded during the specified follow-up period for each group was included in the analysis of outcomes. All included patients were aged 18 years or older and had continuous KPCO health plan membership with pharmacy benefits for at least 6 months before the index date. Patients were monitored for 30 days or until health plan termination or death. Patients purchasing an additional antibiotic (antibiotic group) or any antibiotic (sick and stable control groups) within 30 days of the index date were excluded.
The inclusion criteria and timing of follow-up INR values for each of the 3 cohorts are shown in the Figure. The use of different follow-up INR time frames among groups was necessary to isolate the influence of antibiotics by allowing at least 3 days of coadministration before INR inclusion. Conversely, the follow-up INR for sick controls included the index date (ie, date of antitussive purchase or clinic visit) to allow for inclusion of INR values closest to when patients were evaluated and received treatment for an illness. The follow-up INR timing in the stable controls was one of convenience and extended to 30 days to increase the sample size.
The primary study outcomes were the proportion of patients experiencing a follow-up INR of 5.0 or more and the change between the last INR measured before the index date (defined as the preindex INR) and the follow-up INR. Secondary outcomes included the proportions of patients experiencing a follow-up INR of more than 3.5 (which would typically trigger a warfarin dosage change in patients receiving antibiotics), clinically relevant bleeding (defined as the occurrence of bleeding resulting in hospitalization or an emergency department visit regardless of severity), objectively confirmed thromboembolism, or death within 30 days of the index date. Predictors of a follow-up INR of 5.0 or more were identified, and the contribution of body temperature above 38°C and the mechanism of drug interaction on the risk of developing a follow-up INR of 5.0 or more were explored a priori.
Bleeding and thromboembolic complications were identified through queries of KPCO electronic claims using predefined International Classification of Diseases, Ninth Revision (ICD-9), codes (available on request) in the primary or secondary positions. Potential events were adjudicated by 2 reviewers (N.P.C., T.D., or C.S.R.) blinded to cohort designation. Disagreements between reviewers were resolved by a third reviewer (D.M.W.). Deaths were identified from KPCO electronic membership data. Information on baseline comorbidities (alcoholism, diabetes mellitus, hypertension, heart failure, prior venous thrombosis, prior arterial thrombosis, prior stroke, and active cancer [excluding squamous and basal cell carcinoma]) were identified from electronic data on KPCO clinic visits using predefined ICD-9 codes (available on request). Information on the primary indication for anticoagulation, target INR range, INR values, INR test dates, and date of warfarin therapy initiation were obtained from the anticoagulation database. Information on antibiotic, guaifenesin with codeine, and warfarin purchases were obtained from KPCO electronic pharmacy records using generic product identifier codes (available on request). Body temperatures recorded during medical office visits in the 7 days before and inclusive of the respective index dates were identified using data from the KPCO electronic medical records.
Age was calculated as of the respective index date. Antibiotics were categorized into 1 of 3 potential interaction mechanism types: disruption of vitamin K synthesis, inhibition of cytochrome P450 metabolism, or no interaction reported. Interaction mechanism was determined by cross-referencing 2 widely used drug interaction references.12,13 Body temperatures were categorized as 38°C or less and above 38°C . The preindex INR was categorized as 2.0 to 2.4, 2.5 to 2.9, and 3.0 to 3.5.
Continuous data were expressed as a mean (SD) or median with interquartile range and were compared using parametric or nonparametric analysis of variance as appropriate. Categorical data were expressed as percentages and compared using the χ2 test of association or Fisher exact test as appropriate. Post hoc analyses were performed when statistically significant differences were identified across the 3 groups. Multivariable logistic regression analysis was used to identify factors independently associated with a follow-up INR of 5.0 or more. All characteristics occurring in at least 1% of each group with P < .20 were included in the model. Data analyses were performed using SAS, version 9.1.3 (SAS Institute, Inc). The α levels were set at .05 and .017 for comparisons across 2 and 3 groups, respectively.
A total of 12 006 patients met the inclusion criteria. Of these, 5857 (48.8%), 5579 (46.5%), and 570 (4.7%) patients were included in the antibiotic, stable control, and sick control groups, respectively. Overall, the mean (SD) cohort age was 68.3 (14.4) years, and 49.9% of the participants were male (Table 1). Most patients had a target INR range of 2.0 to 3.0, and the median preindex INR was 2.5 (interquartile range, 2.2-2.9). The most common indication for anticoagulation was atrial fibrillation (44.4%), followed by venous thromboembolism (31.8%). Hypertension (47.2%) was the most common comorbidity, followed by histories of venous thromboembolism (22.8%) and heart failure (14.6%).
Compared with patients in the sick and stable control groups, patients in the antibiotic group were older and more likely to be receiving anticoagulants for atrial fibrillation (Table 1) (both P < .001). Prior venous thromboembolism was more common in the stable control patients (26.6%) than in the sick control and antibiotic patients (17.9% and 19.7%, respectively; P < .017 for all comparisons). No other baseline characteristic varied by more than 5.0% across the groups. The most frequently prescribed antibiotic was amoxicillin (8.6%) followed by ciprofloxacin (8.2%) and cephalexin (7.5%) (Table 2).
All patients had follow-up INRs measured within the time frame specified for inclusion into their group. The percentages of patients experiencing a follow-up INR of 5.0 or more were 3.2%, 2.6%, and 1.2% for the antibiotic, sick control, and stable control groups, respectively (Table 3). The risk of an INR of 5.0 or more was greater among the antibiotic and sick control groups compared with the stable control group (P < .017). There was a numerically greater risk of a follow-up INR of 5.0 or more in the antibiotic group compared with the sick control group, but the result was not significant (P = .44). Although a statistically significant increase in the mean INR change was observed for the antibiotic group compared with the stable control group, it was not clinically significant (ie, the mean increase was <0.1 INR units). The percentage of patients with a follow-up INR greater than 3.5 was higher in both the antibiotic (13.5%) and sick control (15.1%) groups compared with the stable control group (9.2%) (both P < .001).
Clinically relevant bleeding and thromboembolic outcomes were infrequent and similar across groups (Table 3) (all P > .05). Antibiotics interfering with warfarin metabolism were more likely to be associated with an INR of 5.0 or more (8.6%) than were those disrupting vitamin K synthesis (3.1%) or not known to interact with warfarin (2.1%) (P < .001) (Table 4).
There were 5031 patients with a body temperature recorded on the index date or within 7 days earlier, including 62.7% (n = 3670), 13.0% (n = 74), and 23.1% (n = 1287) of patients in the antibiotic, sick control, and stable control groups, respectively. Among patients with a recorded temperature, 7.9% (n = 293), 6.8% (n = 5), and 3.0% (n = 38) of those in the antibiotic, sick control, and stable control groups, respectively, had at least 1 temperature reading above 38°C (P < .001 across groups, with P < .001 between antibiotic and stable patients). Patients with a body temperature above 38°C were twice as likely to have a follow-up INR of 5.0 or more compared with those with a temperature of 38°C or less (6.6% vs 3.2%; P = .001).
Factors independently associated with a follow-up INR of 5.0 or more included patients receiving an antibiotic (odds ratio [OR], 2.46) and sick controls (OR, 2.12) compared with stable controls (Table 5). In addition, patients with a preindex INR of 2.5 to less than 3.0 (vs preindex INR 2.0 to <2.5; OR, 1.65), preindex INR 3.0 or more (vs preindex INR 2.0 to <2.5; OR, 2.30), females (vs males; OR, 1.46), and diagnosis of active cancer (vs no diagnosis of active cancer; OR, 2.20) were associated with an increased likelihood of a follow-up INR of 5.0 or more. Conversely, an indication for warfarin of venous thromboembolism (vs atrial fibrillation; OR, 0.69) was associated with a decreased likelihood of having a follow-up INR of 5.0 or more.
Limited methodologically rigorous research is available describing the nature and scope of the challenge presented when antibiotics are coprescribed with warfarin in clinical practice.7 The contribution of illness to INR variability has been suspected by clinicians, but no previous studies have quantified this effect.9 The present study addressed both of these gaps in the literature. We observed that patients receiving warfarin who were evaluated in a medical office for upper respiratory tract infection and/or purchased a prescription antitussive (sick controls) but did not purchase an antibiotic were more likely than stable control group patients to have a follow-up INR of 5.0 or more. Antibiotic group patients were also more likely to have a follow-up INR of 5.0 or more than were stable control patients, but the difference between the antibiotic and sick control groups was not statistically significant. We identified that the likelihood of an INR of 5.0 or more varies substantially depending on the antibiotic prescribed and the warfarin interaction mechanism. Antibiotics inhibiting warfarin metabolism were more frequently associated with follow-up INRs of 5.0 or more compared with those that disrupted vitamin K synthesis or had not previously been reported to interact with warfarin. However, most patients in either the antibiotic or sick control groups did not have follow-up INRs that would have necessitated a change in the warfarin dosage, and similar 30-day rates of thromboembolism, bleeding, and death were observed across all study groups. We conclude, therefore, that the absolute risk of harm associated with coprescribing antibiotic and warfarin therapy is low.
Potential explanations as to why an upper respiratory tract infection without antibiotic prescription might increase the risk of excessive anticoagulation include reduced oral intake and resultant decreased consumption of vitamin K–rich foods,14 the effect of acetaminophen-containing cough and cold remedies that can increase the INR,15,16 or increased clotting factor catabolism associated with fever.17,18 In our study, patients with a body temperature above 38°C were twice as likely to have a follow-up INR of 5.0 or more than were patients with temperatures below this threshold.
Previous studies19,20 evaluating preemptive warfarin dosage reduction to mitigate the risk of excessive anticoagulation among patients prescribed antibiotics have reported mixed results. Warfarin dosage reduction during trimethoprim-sulfamethoxazole coadministration resulted in fewer INRs higher than 4.0 without an increase in subtherapeutic anticoagulation compared with a control group not receiving a dosage reduction. Although encouraging, these findings were limited by a small sample size (N = 8).20 Among patients receiving doxycycline or levofloxacin, preemptive warfarin dosage reduction increased the risk of subtherapeutic anticoagulation.19,20 Transient, short-term INR elevation does not appear to present a substantial risk for major hemorrhage.21,22 In addition, our findings reveal that the mean INR change among patients with stable warfarin therapy who receive antibiotics is minimal, arguing against routine use of preemptive warfarin dosage reduction in this setting.
Our study was strengthened by inclusion of a large number of real-world patients, uniform practice for addressing antibiotic prescriptions during warfarin therapy, thorough patient follow-up, and blinded independent adjudication of outcome events by multiple reviewers. Carefully defined cohort designation and a run-in period of relatively stable warfarin therapy minimized background INR fluctuation that could have influenced the primary outcome. In some cases, out-of-range preindex INRs may have produced warfarin dosage changes increasing INR variability during follow-up. Reliance on ICD-9 codes for identification of clinically relevant bleeding and thromboembolic outcomes could have resulted in underestimation of these outcomes. In addition, although standard CPAAS procedure for managing antibiotic interactions involves continuing the warfarin dose unchanged with repeated INR monitoring in 3 to 7 days, adherence to this standard may not have been 100%. Deviations from this protocol that included a preemptive warfarin dose reduction could result in underestimation of the true interaction INR effect. However, if this scenario occurred, it was rare and unlikely to have significantly altered our observations. Several baseline characteristics were significantly different between groups. Although these were included in the multivariable logistic regression analysis for the primary outcome of an INR of 5.0 or more, demographic disparities could have influenced secondary outcomes. We were unable to identify infection type within the antibiotic group. Different infections or infection locations could have resulted in varying risks for excessive anticoagulation. Finally, although the risk of an INR of 5.0 or more was numerically higher only in the antibiotic group than in the sick control group (3.2% vs 2.6%; P = .44), the smaller sample size within the sick control group (n = 570) may have limited our ability to detect a significant difference.
Patients receiving warfarin who develop an acute upper respiratory tract infection are at increased risk of excessive anticoagulation with or without antibiotic exposure. The risk of excessive anticoagulation for individual antibiotics varied according to the interaction mechanism with the greatest risk presented by antibiotics interfering with warfarin metabolism. Characteristics associated with an increased risk of an INR of 5.0 or more include female sex, active cancer, and elevated baseline INR. Timely INR monitoring may be particularly important when one or more of these factors is present. Studies are needed to determine whether less-frequent INR monitoring can be safely used during coadministration of antibiotics less likely to interact with warfarin therapy.
Accepted for Publication: November 16, 2013.
Corresponding Author: Nathan P. Clark, PharmD, Kaiser Permanente Colorado, 16601 E Centretech Pkwy, Aurora, CO 80011 (firstname.lastname@example.org).
Published Online: January 20, 2014. doi:10.1001/jamainternmed.2013.13957.
Author Contributions: Drs Clark and Delate had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Clark, Delate, Riggs, Witt, Hylek, Ageno, Dentali, Crowther.
Acquisition of data: Clark, Delate, Riggs.
Analysis and interpretation of data: Delate, Witt, Hylek, Garcia, Ageno, Crowther.
Drafting of the manuscript: Clark, Delate, Riggs, Witt, Dentali, Crowther.
Critical revision of the manuscript for important intellectual content: Clark, Delate, Riggs, Witt, Hylek, Garcia, Ageno, Dentali.
Statistical analysis: Delate, Riggs, Witt, Hylek.
Administrative, technical, or material support: Clark, Garcia.
Study supervision: Clark, Ageno, Crowther.
Conflict of Interest Disclosures: Dr Crowther serves as an advisor to Alexion, Artisan Pharma, Bayer, Boehringer Ingelheim, CSL Behring, Leo Pharma, and Pfizer; has prepared educational materials for CSL Behring, Octapharm, and Pfizer; and has provided expert testimony for Bayer. Dr Crowther’s institution has received funding for research projects from Boehringer Ingelheim, Leo Pharma, Octapharm, Pfizer. Dr Hylek serves as an advisor to Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Daiichi Sankyo, Johnson and Johnson, and Pfizer and participates in clinical symposia sponsored by Bayer, Boehringer Ingelheim, and Bristol-Myers Squibb. Dr Garcia served as an advisor to Boehringer Ingelheim, Bristol-Meyers Squibb, CSL Behring, and Daiichi Sankyo. No other disclosures were reported.