Warfarin-associated bleeding events and concomitant use of potentially interacting medicines reported to the Norwegian spontaneous reporting system

Abstract

AIMS

To study warfarin associated bleeding events reported to the Norwegian spontaneous reporting system and evaluate the differences in assessment of potentially interacting medicines between reporters and evaluators.

METHODS

Data on bleeding events on warfarin were retrieved from the Norwegian spontaneous reporting system database. Key measurements were time to bleeding, use of concomitant medications and the evaluation done by reporters.

RESULTS

In 289 case reports a total of 1261 medicines (median 4.0 per patient, range 1–17) was used. The evaluators (authors of this article) identified 546 medicines including warfarin (median 2.0 per patient, range 1–7) that could possibly cause bleeding alone or in combination. Reporters assessed 349 medicines (median 1.0 per patient, range 1–4) as suspect. Evaluators identified 156 pharmacokinetic and 101 pharmacodynamic interactions, compared with 19 pharmacokinetic and 56 pharmacodynamic interactions reported as suspected by the reporters. Time to bleeding was stated in 224 reports. Among the early bleeding events, the reports on warfarin without interacting medicines showed the highest INR (international normalized ratio). Heparin was used in 17/21 reported bleeding events during the first week on warfarin. Among the late bleeding events, reports with pharmacokinetic interacting medicines had the highest INR.

CONCLUSIONS

Concomitant use of potentially interacting medicines was involved in the majority of the warfarin-associated bleeding events reported to the Norwegian spontaneous reporting system. Reporters assessed mostly warfarin as the only contributor to bleeding. In particular, pharmacokinetically interacting medicines were not suspected as contributing to bleeding.

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT

• Bleeding is the main adverse drug reaction of warfarin use.

• Several medicines, herbal products and dietary supplements may interact with warfarin.

• A high percentage of patients on warfarin receive potentially interacting substances.

WHAT THIS STUDY ADDS

• Potentially interacting medicines were used in more than 50% of the warfarin-associated bleeding events reported to the Norwegian spontaneous reporting system.

• Only a minority of the concomitant medicines were reported as suspected or interacting in the reporters' assessments.

• There was a higher degree of under-reporting of pharmacokinetically than of pharmacodynamically interacting medicines.

Introduction

Vitamin K antagonists, such as warfarin, are effective in thrombosis prevention, but are also frequently implicated in adverse drug reactions (ADRs), mainly bleeding events. In Norway, warfarin has, for several decades, been the medicine most often associated with ADRs with fatal outcome [1, 2]. Bleeding during treatment with vitamin K antagonists is also frequently reported in several other countries [3–7]. Previous studies have shown that risk factors for vitamin K antagonist-associated bleeding events are the initial phase of therapy, high international normalized ratio (INR), co-morbidity, interacting medicines and advanced age [8–12].

There is an increasing focus on bleeding complications associated with the combined use of various antithrombotic medicines. Observational studies indicate that antiplatelet and anticoagulant combinations, with concomitant use of non-steroidal anti-inflammatory drugs (NSAIDs), selective serotonin re-uptake inhibitors (SSRIs) or antibacterial drugs increase bleeding risk [11, 13–17]. Up to 80% of anticoagulant users are described as using potentially hazardous combinations of medicines [18, 19]. Studies of anticoagulant-related bleeding events from spontaneous reporting systems also show frequent use of interacting medicines, but the types of interacting medicines have not been described in detail [1, 3]. Further, growing literature describes the lack of awareness among health care providers and patients regarding drug–drug interactions. Several studies have found that the ability of prescribers to recognize potential drug–drug interactions is low [20–22]. To our knowledge, there are no published studies on using the spontaneous reporting system for the characterization of the awareness of physicians of warfarin–drug interactions.

We hypothesized that physicians reporting warfarin associated bleeding events to the Norwegian spontaneous reporting system mostly recognize a single medicine (i.e. warfarin) as the cause of bleeding. However, we suspect that many of the bleeding events actually are caused by a combination of medicines that increases the antithrombotic effect.

We also hypothesized that INR would be different between patients using pharmacokinetically interacting medicines, patients using pharmacodynamically interacting medicines and patients not using interacting medicines at the time of bleeding: Patients using warfarin in combination with pharmacokinetically interacting medicines were expected to have a higher INR at the time of bleeding compared with patients using warfarin alone or together with medicines causing a pharmacodynamic interaction. The aim of this study was therefore to analyze reports from the years 2003–05 focusing on INR and the use of potentially interacting medicines and to evaluate the differences in assessments of interacting medicines by reporters and evaluators.

Methods

Data collection

In Norway, it is compulsory for physicians and dentists to report all serious or unexpected ADRs via the regional pharmacovigilance centres (RELIS) to the national spontaneous reporting system. An adverse drug reaction form is completed by the reporter and sent to RELIS. The report must include at least one suspected medicine, any other suspected medicines and any concomitant medicines. There is a box for ticking if an interaction is suspected.

All relevant information from each report is stored in the database at the Norwegian Medicines Agency (NoMA). The reports contain information about the patient, medication use, suspected ADR, outcome, assessment done by reporter, causality assessment by RELIS, reply to reporter and administrative data. The ADRs are translated to Medical Dictionary for Regulatory Activities terms (MedDRA), which is a hierarchical system that categorizes medical terms into body-system organ classes [23]. Staff at RELIS perform a causality assessment based on the WHO criteria, but this assessment is only done for medicines that the reporters have assessed as suspect or interacting [24].

Data search

A data search identified all spontaneous reports on bleeding events received during a 3-year period from 1 January 2003 to 31 December 2005 and stored in NoMA. All reports where vitamin K antagonists were mentioned were further analyzed. Duplicates were excluded. From each report the age and sex of the patient were recorded, together with all the medications documented in the reports, the duration of use, INR values if stated, the indication for warfarin use and the outcome. The profession of the reporter was recorded and whether the reporter had listed the drug(s) as suspected of having caused the bleeding or interacted with warfarin, or as concomitant medication not related to the ADR. The localization of the bleeding was categorized as cerebral, gastrointestinal or originating from other localizations (e.g. haematoma, injection site bleeding, pleural haemorrhage, epistaxis and retinal haemorrhage). If more than one bleeding location was stated, only the most severe bleeding was included.

Medications

The Norwegian National Formulary (Legemiddelhandboken) was searched for medicines that may potentially interact and increase the effect of warfarin [25]. For medicines acting pharmacokinetically with warfarin the search results were similar to the corresponding interaction chapter in the British National Formulary (BNF) and Martindale [26, 27], except for antimicrobial agents, which are all defined as possibly interacting with warfarin in Legemiddelhandboken. For medicines acting pharmacodynamically with warfarin such as antiplatelet medicines (acetylsalicylic acid, NSAIDs and SSRIs), heparins and fibrinolytic agents, there were more inconsistencies between BNF, Martindale and Legemiddelhandboken. Legemiddelhandboken was used as a reference for interacting medicines since this information is readily available to the prescriber. In particular, medicines used for a short period, such as antibacterials or pain relieving medicines were judged important. Selective cyclo-oxygenase 2 inhibitors (COX-2 inhibitors) were included because they may increase the risk of gastrointestinal bleeding [28]. The medicines were classified according to the Anatomical Therapeutical Chemical (ATC) classification system [23]. Topical applications such as eye drops or steroid creams of these medicines were not considered. Bisphosphonates, with erosive effects on gastric and oesophageal mucosa, were not included.

In this analysis all the potentially interacting medicines were included if used at the time of bleeding, not excluding warfarin because of low INR or other medicines because of short duration of use. The medicines were categorized as having either mainly pharmacodynamic interactions or mainly pharmacokinetic interactions. The patients were divided into four groups as using warfarin 1) without interacting medicines, 2) with pharmacodynamically interacting medicines, 3) with pharmacokinetically interacting medicines or 4) with both pharmacokinetically and pharmacodynamically interacting medicines.

Since all but one patient used warfarin (one patient used dicoumarol), all the patients were called warfarin users. The dicoumarol case was also included in the statistics because dicoumarol has a similar interaction potential to warfarin. Since studies have shown an increased risk of bleeding within the first 90 days of warfarin among people 65 years and older [8], the reports were categorized into short term use of warfarin (<3 months before bleeding) and long term use (<3 months before bleeding).

Statistics

All statistical analyses were performed using the statistical program SPSS version 16.0 (SPSS Inc., Chicago, IL, USA). The Mann–Whitney test and t-test were used for testing continuous variables. The Pearson chi-square test was used for testing categorical variables. P values less than 0.05 were taken as statistically significant.

The Regional Ethics Committee and the local patient ombudsman approved the study.

Results

This study is an analysis of 289 case reports of warfarin-associated bleeds received by the Norwegian Pharmacovigilance System during the 3 year period 2003–05, constituting 47% of the 616 reports of bleeding events. The characteristics of the reports are presented in Table 1. Hospital doctors accounted for 271 reports (93.8%) and 282 (97.6%) of the ADRs were characterized as serious. The number of reports increased with age, closely reflecting the number of users per age group according to data from the Norwegian Prescription Database (NorPD) in 2004 (Figure 1). A higher proportion of the cerebral bleeding events (109 of 174 cases, 62.6%) than gastrointestinal bleeding events (24 of 69 cases, 34.8%) and other bleeding events (6 of 46 cases, 13.0%) was fatal. The difference was statistically significant (P < 0.01). Otherwise there were no differences between adverse events with and without fatal outcome. Information about time to event was available for 224 patients. The number of bleeding events during the first 3 months accounted for 37 (16.5%) reports, compared with 187 (83.5%) reports after more than 3 months of warfarin use. The majority of the early bleeding events occurred during the first week (21/37) and the majority of the late bleeding events occurred after more than 1 year of warfarin use (166/187).

Table 1.

Characteristics of reports

Variable	Number of cases (%) n= 289
Age (years) mean (range)*	75.9 (41–93)
Female	117 (40.5)
Fatal cases	139 (48.1)
Indication for warfarin use
Atrial fibrillation	161 (55.7)
DVT or PE	47 (16.3)
Chronic heart disease	37 (12.8)
Cerebral infarction, TIA	10 (3.5)
Other reasons	7 (2.4)
Not indicated	27 (9.3)
Medicines, mean (range)	4.4 (1–17)
INR, mean (range)†	4.4 (0.9–10.0)

^*Information available in 287 reports (99.3%).

^†Information available in 250 reports (86.5%). DVT, deep vein thrombosis; PE, pulmonary embolism; TIA, transitory ischaemic attack; other reasons, Factor V Leiden, atherosclerosis.

Figure 1.

Number of bleeding events reported in Norway during 2003–05 related to age and outcome. The inset shows the corresponding prescription of warfarin in Norway in 2004 related to age (data from the Norwegian Prescription Database). Non-fatal (); Fatal ()

The characteristics of medicine use

The 289 patients used a median of 4.0 medicines each (range 1–17 medicines) representing 210 different substances from the ATC-code system. Among the total number of 1261 medicines reported, use of herbal products or dietary supplements was stated in only two reports. In 226 (78.2%) reports more than one medicine was used.

Interacting medicines

The reports included a total of 1261 medicines. Of these, the evaluators (authors of this article) identified 546 medicines (median 2.0 medicines per patient, range 1–7) including warfarin that could possibly cause bleeding alone or in combination. In contrast, the reporters only suspected 349 medicines (median 1.0 medicines per patient, range 1–4) (P < 0.01). The difference between reporters and evaluators in assessment of the number of medicines suspected to contribute to bleeding is shown in Figure 2. Table 2 shows the identified medicines with a potential to interact with warfarin and increase the bleeding risk, and the corresponding assessment done by the reporters. NSAIDS/COX-2 inhibitors, heparins and antibacterials were the medicines most frequently identified as suspect or interacting by both evaluators and reporters, but only medicines with a possible pharmacodynamic interaction were identified by the reporters in more than 50% of the evaluator identified reports (fibrinolytics 4/4 reports (100%), heparins 19/25 (76.0%), acetylsalicylic acid 11/18 (61.1%), and NSAIDs/COX-2 inhibitors 21/36 (58.3%)). In 19 reports (6.6%) warfarin was not considered by the reporter as suspect or interacting. In those reports heparin (8 reports), NSAIDs/COX-2 inhibitor (7), fibrinolytics (3), antibacterials (2), temozolomide (1) and adalimumab (1) were suspected. Figure 3 shows the assessment done by the evaluators and reporters of medicines with potentially pharmacokinetic and pharmacodynamic interactions. The difference in assessment of interacting medicines between reporters and evaluators was statistically significant (P < 0.01).

Figure 2.

Number of medicines suspected to contribute to bleeding assessed by evaluator and by reporter. Reporter (); Evaluator ()

Table 2.

Medicines suspected to contribute to bleeding

Medicines (generic names as stated in the reports)	Number of medicines assessed by evaluators	Number of medicines assessed by reporters
Vitamin K antagonists	289	270
Pharmacodynamic interactions (according to Legemiddelhandboken)
NSAIDs/COX-2 inhibitors (diclofenac, aceclofenac, piroxicam, ibuprofen, naproxen, ketoprofen, celecoxib, rofecoxib, etoricoxib, nabumetone, valdecoxib),	36	21
Heparins (heparin, dalteparin, enoxaparin)	25	19
Acetylsalicylic acid*	18	11
Antidepressants; SSRIs (citalopram, escitalopram, paroxetine, sertraline)	18	1
Fibrinolytics (alteplase, reteplase)	4	4
Pharmacokinetic interactions (according to Legemiddelhandboken)
Antibacterials and antiprotozoals (amoxicillin, ampicillin, penicillin, cefalexin, ceftriaxone, cefuroxime, ciprofloxacin, clarithromycin, dicloxacillin, doxycycline, erythromycin, meropenem, metronidazole, pivmecillinam, trimethoprim + sulfamethoxazole, tobramycin)	42	13
Fluvastatin, simvastatin	22	0
Allopurinol	22	1
Paracetamol	21	1
Corticosteroids	16	1
Omeprazole	13	0
Amiodarone	4	1
Other pharmacokinetic interacting medicines (dextropropoxyphene, phenytoin, thyroid hormones, tramadol, venlafaxine)	16	2
Other medicines (not stated as interacting in Legemiddelhandboken) (alendronic acid, temozolomide, infliximab, adalimumab)	0	4
Sum	546	349

Legemiddelhandboken: The Norwegian National Formulary. NSAIDs, non-steroidal anti-inflammatory drugs; COX-2 inhibitors, cyclo-oxygenase 2 inhibitors; SSRIs, selective serotonin re-uptake inhibitors.

^*One person used dipyridamole in combination with acetylsalicylic acid.

Figure 3.

Number of medicines with pharmacokinetic and pharmacodynamic interactions assessed by evaluator and by reporter. Suspected by reporter (); Suspected by evaluator ()

Data from 198 reports (68.5%) were informative for both the time to bleeding and INR. There were differences between the short term and long term warfarin users with respect to interacting medicines affecting INR (Figure 4). Among the bleeding events during the first 3 months of warfarin treatment, the reports on warfarin alone showed the highest INR (median 6.6, range 1.6–8.1), which is significantly higher than the INR (median 2.2, range 0.9–5.5) among the bleeding events with suspected pharmacodynamic interactions (P= 0.017). There was also a trend towards a higher INR in the reports on warfarin alone compared with the reports with both pharmacodynamic and pharmacokinetic interactions (P= 0.098), and when the extreme value was excluded, this result was significant (P= 0.032). Of the reported bleeding events during the first week, 17/21 patients used potentially interacting medicines and all 17 used heparins. In addition, the following 15 medicines were used: acetylsalicylic acid (5 patients), fibrinolytics (3), SSRI (1), antibacterials (3), allopurinol (1) and paracetamol (2) (data not shown). Among the late bleeding events there was a diversity of medicines involved, and the reports with suspected pharmacokinetically interacting medicines had a significantly higher INR (median 4.7, range 2.1–8.0) at the time of bleeding compared with the reports with warfarin used alone (median INR 3.5, range 1.1–8.1) (P= 0.021). There was also a trend towards a higher INR in the reports on warfarin with both pharmacodynamically and pharmacokinetically interacting medicines compared with warfarin used alone (P= 0.097), but this group consisted of only 18 reports.

Figure 4.

Box and whisker plot shows the median, interquartile range and extreme value (*) of INR according to use of interacting medicines (pharmacodynamic interactions, pharmacokinetic interactions and both) compared with warfarin used alone for (A) short term (P= 0.017, NS and P= 0.098, respectively) and (B) long term warfarin users (NS, P= 0.021 and P= 0.097, respectively). NS, not significant

Discussion

This analysis of anticoagulant-related bleeding events from the Norwegian spontaneous reporting system describes the use of potentially pharmacodynamically and pharmacokinetically interacting medicines used concomitantly with warfarin prior to reported bleeding. The analysis shows that potentially interacting medicines were used in more than 50% of the cases. The most frequently used potentially interacting medicines were antibacterials, NSAIDs/COX-2 inhibitors and heparins. Among the reported ADRs there were more medicines with potentially pharmacokinetic interactions than medicines with pharmacodynamic interactions. This is in agreement with other studies on potential warfarin interactions where pharmacokinetic interactions have been reported to be more common than pharmacodynamic interactions [19, 29]. However, the type of most frequently prescribed concurrent medicines varies between studies and countries, probably because of different therapeutic traditions and selection of medicines [18, 19, 29].

The use of potentially interacting medicines among anticoagulant users varies substantially between studies (up to 80% is seen in prescription database studies and around 50% in clinical case control studies) [10, 18, 19, 29]. Different study designs and time aspects of the studies may explain some of the differences between the study results. In addition, different definitions of interacting medicines along with the ever-expanding number of medicines reported to interact with warfarin might explain much of the differences. In two studies from Scandinavian spontaneous reporting systems the use of potentially interacting medicines was seen in 25% and 17%, respectively, of anticoagulant-related bleeding events [1, 3]. However, the types and number of medicines with potential interaction varied between the two studies. The relatively low number of interacting medicines in studies from the spontaneous reporting system could also reflect that physicians report the suspected medicine only (warfarin), without mentioning the concomitant medicines used.

In our study, the reporters assessed most frequently warfarin as the only suspected medicine and an interacting medicine in less than 20% of the reports. This low reporting rate of interacting medicines is also seen in other studies of ADRs from spontaneous reporting systems [5, 6]. In this study, when a co-suspected or an interacting medicine was reported, 75% of the interactions were pharmacodynamic and 25% were pharmacokinetic. All the pharmacodynamic interactions, except from SSRIs, seemed to be well known among the reporters. In contrast, of the possible pharmacokinetic interactions, except from antimicrobials, co-suspected or interacting medicines were only reported sporadically. This could reflect a lack of physician awareness about pharmacokinetic interactions. Other possible explanations for these reporter-evaluator differences might be that the physicians report only what they regard as the most important or probable cause of the bleeding and that they often are short of time [30]. The large difference in assessments between reporter and evaluator indicates that there is a systematic underreporting of concomitant medicines in reports of warfarin ADRs, especially those interacting pharmacokinetically. This difference between evaluations done by reporter and evaluator is important to recognize if studies on data from the spontaneous reporting systems draw conclusions based on the reporters' evaluation only [31]. Pharmacokinetic interactions are not intuitive and each interaction has to be remembered or looked up unless automated drug alerts are available in the electronic prescription systems. Thus, such drug alerts in addition to the use of anticoagulation clinics, patient self-monitoring, and development of risk-assessment tools, should be considered for better recognition of drug interactions [22].

Several studies have shown that the initial phase of anticoagulant therapy conveys the highest risk of bleeding [1, 8, 12]. However, in our study the great majority of the patients had been treated with warfarin for more than 1 year at the onset of bleeding. The high number of bleeding events after long term use of warfarin was also seen in a recent Swedish study on fatal ADRs, which found that most bleeding events occurred after more than 1 year of warfarin use [3]. An explanation for this might be that the high risk at the start of therapy lasts only a few weeks, while a low monthly risk during several years accounts for most bleeding events in absolute numbers. Prospective studies that address this issue often miss these late bleeding events because the follow-up lasts for only a couple of months or up to 1 year. It is also possible that the initiation of treatment was done with more caution over the past years or that there are changed patterns of reporting.

In this study we hypothesized that patients using warfarin in combination with pharmacokinetically interacting medicines were expected to have a higher INR at the time of bleeding compared with patients using warfarin alone or together with medicines causing a pharmacodynamic interaction. Our results show that the hypothesis is valid for the late bleeding events. It might be that addition of a medicine with a potential pharmacokinetic interaction was not followed up with INR testing so that the increase in INR was not recognized before bleeding. In the late bleeding events nearly half of all reports had an INR above 4.0 suggesting difficulties in monitoring the treatment or simply poor follow-up. We do not know whether the high INR is a result of lower vitamin K dietary intake, reduced health condition or an interaction, but doctors and patients should be cautious and do more frequent INR testing when adding any new medication, including non-prescription medicines, to patients on warfarin.

A reason for pharmacokinetically interacting medicines not influencing INR in the early bleeding events might be that the majority are also using another anticoagulant with no effect on INR (i.e. heparins) or that the interactions are being adjusted for by frequent INR measurement. Of note is the high INR among patients using warfarin alone, suggesting this group might include patients with pharmacogenetic polymorphisms making them vulnerable to standard doses of warfarin.

Limitations of the data

The spontaneous reporting system for ADRs has several limitations such as under-reporting, even for events that are compulsory to report. It is a collection of voluntary reports rather than a systematic observation of a defined patient population. The reports submitted do not establish causality, only that the reporters suspect that a causal relationship might exist. Most of the reporters are hospital doctors. They have limited access to the record of medicines prescribed to the patients, and have to rely on information given by the patient or his/her relatives. Our data from the spontaneous reporting system are incomplete regarding type, dosing and time relations for all medicines reported, especially for concomitant medications and non-prescription medicines, and we do not know if the reported medicines have actually been taken.

Regarding herbal medicines or dietary supplements, this was stated in only two reports. This likely represents an under-reporting. Several studies have shown a high use of non-prescription medicines, dietary supplements and herbal products among elderly persons, especially among prescription medication users [32, 33]. It has also been shown that disclosure of such use to conventional medical professionals is low, usually about 30% [34–36]. This should raise concern, since there are an increasing number of reports describing clinically serious interactions between warfarin and herbal products/non-prescription medicines [37, 38]. Even though we have no data on the use of herbal products, dietary supplements or non-prescription medicines among warfarin users in Norway, we would expect that there are several users among the reported ADR cases, in particular among the late bleeding events. This could contribute to an underestimation of the number of medicines interacting with warfarin.

In conclusion, concomitant use of potentially interacting medicines may contribute to the majority of the reported warfarin bleeds. In our study, only a minority of the concomitant medicines are suspected in the reporters' assessments. In particular, pharmacokinetically interacting medicines are not reported. This difference between evaluations done by reporter and evaluator is important to recognize if studies on data from the spontaneous reporting systems draw conclusions based on the reporters' evaluation only.

Competing Interests

There are no competing interests to declare.