Introduction

Oral anticoagulation is a high-risk therapy: Incidence of bleeding complications has been reported as 10–17 per 100 patient-years (any severity) and 2–5 per 100 patient-years (major bleeding) [1]. In Germany, some 172.5 million defined daily doses (DDD) of phenprocoumon were prescribed in 2000 [2],corresponding to approximately 473,000 anticoagulated patients and some 10,000–80,000 patients with bleeding complications per year.

Phenprocoumon, the drug almost exclusively used for oral anticoagulation in Germany[2, 3], is a vitamin K antagonist similarly structured to warfarin (with a much longer half-life: 6.5 vs 1.5 days) [4], but considered to be less prone to drug interactions.[5] Despite the differences in molecular structure and pharmacokinetics, the metabolism of both warfarin and phenprocoumon depends (to some extent) on the CYP2C9 isoenzyme of the cytochrome P450.[6] The exact pathway of the phenprocoumon metabolism, however, remains to be clarified. Interindividual differences in drug response caused by variable metabolizing enzyme activity due to genetic variants of CYP2C9 have been described. Next to a wild-type allele CYP2C9*1, two variant alleles with decreased enzyme activity have been described: people with a *1/*2 polymorphism have a somewhat reduced enzyme activity, carriers of *1/*3 or *2/*2 genotypes have a significantly reduced metabolism, and *2/*3 and *3/*3 carriers are classified as slow metabolizers [7].

In 1999, Aithal et al. reported that selected anticoagulated patients presenting genetic variants of CYP2C9 often needed very low doses of warfarin, and that low-dose patients had more than a threefold increased risk of major bleeding complications (a direct association of bleeding risk and presence of a variant allele was, however, not demonstrated)[8]. Margaglione et al. [9] and Higashi et al. [10] have shown a two- to threefold increased bleeding risk in warfarin-anticoagulated patients with any variant CYP2C9 allele who were followed up in specialized clinics. In contrast to the UK, where patients are usually followed up in hospital-based anticoagulation clinics, in Germany most patients with oral anticoagulation are managed in general practice (an approach which has shown to be feasible in the UK also) [11, 12].

The aim of our study was to determine whether CYP2C9 polymorphisms are relevant for patients taking phenprocoumon, and to evaluate their feasibility and validity as risk markers in a (unselected) cross-sectional sample of anticoagulated general practice patients [9, 13, 14, 15, 16].

Methods

The study was approved by the Göttingen University ethics committee.

The prevalence of *3 polymorphisms in general populations is reported to be 7–10% [17, 18]. On this basis a sample size of 170–500 would be necessary to demonstrate a sixfold increased bleeding risk with a 5% significance level and a power of 80%.

All 25 teaching general practices in the city of Göttingen, as well as the university outpatient clinic for cardiovascular diseases, were invited to participate. Five of the twelve participating practices (response rate 48%) were run by one GP, 6 had two GPs, and 1 was a group practice with four partners. Respondents were asked to recruit all their patients taking phenprocoumon. Eligible patients were determined using the practice computers. After obtaining written informed consent, an extra 9 ml of venous blood (EDTA-filled vacutainer) was drawn during a routine INR control, sent to the local institute of human genetics, and frozen until genotyping of CYP2C9. Samples were identified by a patient code number to insure blind analysis. Additionally, patients and GPs (or practice staff) were asked to complete a short questionnaire on the indication, duration frequency of INR controls, past bleeding, complications (site and severity), concurrent chronic disorders and treatments, as well as a history of drug intolerance. Patients and GPs were also asked to provide information on dosage schedules and past INR values (copy of the patients' anticoagulation pass or relevant chart documentation). The questionnaires were returned by mail or collected by one of the researchers. Questionnaires and blood samples were matched by means of the patient code number.

Phenprocoumon dose requirements were determined as follows[12]: The six most recent INR values within the therapeutic range (target INR as recommended for each indication [19]) were determined in each patient's documentation.

Blood samples were analyzed using PCR-based procedures as described by Stubbins [20]. DNA fragments were examined for restriction fragment length polymorphisms (RFLPs); incomplete digestion was ruled out by application of controls in every batch and double-checking suspect samples. Genotyping was blind with regard to patient identity and questionnaire responses; results were matched with the questionnaires by means of the patient code number in the department of GP. Descriptive statistics, t-tests, odds ratios (OR), and 95% confidence intervals (CI) based on 2×2 contingency tables were calculated, relating occurrence of bleeding with presence of variant alleles, phenprocoumon requirement, target INR, and comedication. To assess the influence of variant alleles on bleeding compared with other possible factors, multivariate (logistic regression) models were determined using SAS Version 8.1 [21] Accuracy of genotyping as a diagnostic test was calculated according to Sackett. [22].

Results

From August 1999 to February 2000, 185 patients (137 general practice and 48 hospital outpatients) were included into the study. Fifty-eight general practice patients could not be included because they did not contact their practice within the research period, refused consent, or were not recruited by their GP for logistic reasons. Information on these patients could not be gathered due to the particularities of the German healthcare system (no gatekeeping, no practice lists). Five practices had included all their eligible patients, 5 practices between 50 and 100%, two practices had recruited less than half of their patients. A mean of 14 patients (range 2–31) were recruited per practice. Patient characteristics and indications for anticoagulation are summarized in Table 1.

Table 1. Patient characteristics (n=185)
Full size table

Phenprocoumon dosages could be retrieved for 155 patients (83.8%), documentation of INR and/or dosages was incomplete in the others. A vast majority of patients (88.1%) took other long-term drugs in addition to phenprocoumon (mean number of drugs 3.6, 95% CI 3.3–3.9). Comorbidity was documented in 159 patients (86%, average number of codiagnoses 2.6, 95% CI 2.3–2.9) 92.5% of whom were treated (the most frequent comorbidities and comedications are shown in Table 2).

Table 2. Comorbidity and comedication, frequent disorders and drugs (n=185, multiple answers possible)
Full size table

Genotypes

Genotyping of CYP2C9 alleles was performed for 179 (of the 185) patients (blood samples did not arrive at the institute of human genetics in six cases: four hospital outpatients and two general practice patients). Wild-type (*1/*1) alleles were most common (n=132, 73.7%); 26.3% had CYP2C9 polymorphisms: *1/*2 genotype in 32 patients (17.9%), *1/*3 genotype in 14 patients (7.8%), and *2/*3 genotype in 1 (0.6%) patient. Mean treatment duration did not differ significantly between wild-type or variant allele carriers, nor did target INR or age. Significantly less *2 carriers were on comedication compared with wild-type carriers (76 versus 92%, P=0.01), the difference between wild-type and *3 carriers was not significant (80% vs 92%). There was no significant difference in the mean phenprocoumon doses required by patients with wild-type, (15.29 mg/week, 95% CI 14.13–16.45), *1/*2 (16.02 mg/week, 95% CI 13.53–18.51), and *1/*3 (13.29 mg/week, 95% CI 10.45–16.13) genotype (no dosage information was available for the only *2/*3 patient); details on dose distributions are shown in Fig. 1.

Fig. 1.
figure 1

Weekly doses of phenprocoumon according to the CYP2C9 genotype. Means (cross), medians (horizontal bar), quartiles (box) and 10–90 percentiles (whiskers)

Full size image

However, patients who had bled required significantly lower weekly doses than patients without hemorrhagic complications: 12.58 mg (95% CI 10.92–14.23) vs 15.90 mg (95% CI 14.80–17.00). Comorbidity and comedication were not associated with increased bleeding risk, nor were individual codiagnoses or drug classes (details not shown).

Carriers of a CYP2C9*2 polymorphism did not have an increased bleeding risk (OR 0.35, CI 0.10–1.24). However, a threefold increased bleeding risk (OR 3.10, CI 1.02–9.40) was detected in the 8.4% of our patients carrying *3 polymorphisms (*1/*3 and *2/*3), resulting in a positive predictive value of 40%, i.e., four of ten patients with *3 polymorphisms will suffer bleeding complications; details are shown in Tables 3 and 4.

Table 3. *3 CYP2C9 polymorphism as a diagnostic indicator for bleeding
Full size table
Table 4. Test characteristics of*3 CYP2C9 polymorphism as a diagnostic indicator for bleeding
Full size table

Risk identification by CYP2C9*3 genotype persisted in a multivariate (logistic regression) analysis with sex, age (less than 60 years or 60 years and older), target INR, presence of drug intolerance, comedication, and presence of a *3 polymorphism as predictive variables model (OR 3.64, 95% CI 1.08–12.26). Reporting drug intolerance was also associated with an increased bleeding risk (OR 3.57, 95% CI 1.34–9.51), as well as age over 60 years (OR 3.37, CI 1.12–10.20). The other predicting variables had no significant influence.

Discussion

Identifying carriers of variant CYP2C9 alleles proved to be feasible in general practice, as genotyping could be performed on a blood sample taken during a routine INR control, and is relatively inexpensive. The prevalence of variant alleles was lower in our cross-sectional sample (27%) than in Aithal's control groups (48% in anticoagulated patients, 40% in community controls) [8], and frequencies reported in European volunteer populations (33–35%) [17, 18].

In a relatively small sample of unselected general practice patients, screening for variant CYP2C9 alleles could identify patients with a high bleeding risk: carriers of rare genotypes *1/*3 and *2/*3 proved to have a 3 times higher risk of bleeding complications than patients with wild-type or the more common *1/*2 variant alleles. This is in line with previous reports of CYP2C9*2 carriers having only a moderately reduced enzyme activity (which is in some contrast to in vitro studies [23]), while CYP2C9*3 carriers have significantly reduced function[7]. Phenprocoumon, like warfarin, is a substrate for CYP2C9 [6], and genetic variance in the metabolism appears to have significant influence on patients' bleeding risk. Consequently, a revision of the recent proposal to use phenprocoumon instead of warfarin in patients presenting variant CYP2C9 alleles should be considered [24]. The third vitamin K antagonist used for anticoagulation, acenocoumarol, does not seem to be a safe alternative either [25]. More studies may be necessary to determine and compare the metabolic pathways and pharmacogenetic implications for all three drugs.

Bleeding risk may be affected by several factors including vitamin K intake, which is difficult to measure reliably as questionnaires on dietary habits are of low validity [26]. Patient compliance as well as the quality of care (e.g., anticoagulation control, dose adjustment) are likely to influence the individual bleeding risk. However, assessing the quality of primary care and its influence on patient outcomes was not the focus of this study. To be useful in general practice patients, risk markers have to be prominent in patients with variable dietary habits, comorbidity, and compliance levels. Despite the sensitivity of "genetic testing" for bleeding being only 17%, our results indicate that, due to the high specificity (94%), CYP2C9 screening for *3 polymorphisms can identify 8.4% of patients—a subgroup with a particularly high risk of 40%.

However, as all anticoagulated patients have (or should have) regular INR controls, the additional value of genetic screening and the impact on the management of high-risk patients remains to be determined. Closer monitoring of drug interactions and INR values could be a consequence, with self-control of INR being a management option [27, 28]. To maximize the benefit, genotyping should probably be performed before the induction of anticoagulation. A slower, low-dose induction under close supervision of INR would then be chosen for patients at risk. Indications for anticoagulation or drugs interacting with anticoagulants (particularly those metabolized via CYP2C9) may be set more restrictively for patients with variant alleles, and target INR values may have to be adjusted according to individual risks for bleeding and thrombotic events. Risk adjustment for genotype could also be included in computer systems for dosage advice [29], though these devices are only occasionally used in Germany. Effectiveness of these measures with regard to bleeding and thromboembolic risk would have to be subject of further studies. Larger population-based trials are necessary to determine the number of bleeding episodes that could be prevented by genetic screening of patients requiring anticoagulation.

Some methodological problems may influence the validity of our study (some of them typical for studies in a general practice setting). For confidentiality and feasibility, patient recruitment was not centrally organized, but was done by individual GPs. Though GPs were instructed to include all their patients on phenprocoumon into our trial, the inclusion rate was variable and incomplete in many practices, as can be commonly observed in studies where patients are recruited by their GP, particularly in countries without a list system [30, 31]. For most patients, detailed information on the induction period of phenprocoumon could not be provided, nor could difficulties in dose adjustment be reliably recalled. In Germany, induction of anticoagulation often takes place when patients are still in hospital and is then not documented in the patient's pass or the GPs' charts.

As frequently seen in general practice-based surveys, documentation quality was variable between and within practices. However, recall or documentation bias is unlikely to affect the study outcome, as genotyping was performed blind with regard to patients' identity or questionnaire responses. As there was not contact between the researchers and patients, we had to rely on patients' answers on the questionnaires, which had been kept short and concise in order to ensure a high acceptance. For example, instead of asking patients how often and when they had experienced bleeding in several years (which seemed likely to elicit recall bias rather than reliable data), patients were asked whether there had been bleeding episodes at all. This approach did not allow for expression as bleeding episodes per patient year (and thus for controlling bleeding risk for duration of therapy). However, in our population, bleeding risk was of similar magnitude to that reported in other surveys [32],and there was no significant difference in duration of treatment between patients with or without bleeding complications.

In our cross-sectional study, patients having suffered adverse events of phenprocoumon may be underrepresented due to death or switch to other forms of treatment. If these (early) adverse events were related to rare CYP2C9 variants, it seems possible that the overall importance of the CYP2C9 genotype for phenprocoumon-induced bleeding was underestimated (and carriers of rare variant alleles underrepresented). The mean length of treatment being 5 years, patients in our study can be expected to be stabilized on phenprocoumon, but still the CYP2C9 genotype stands out as a risk factor for bleeding.

The high rate of comorbidity, typical for (mostly) elderly patients in general practice, explains the high comedication of anticoagulated patients [33, 34] and highlights the need for monitoring with regard to drug interactions and susceptibility to complications. Previous studies did not report details on comedication [35], they excluded patients "with drug interactions as apparent cause for low-dose requirements"[8] or adjusted for any [9] or CYP2C9-dependent [10] comedication. Judicious use of genetic testing may be justified in patients starting oral anticoagulation or presenting additional risk factors, including consumption of drugs that may increase hypocoagulation or have a CYP2C9-dependent metabolism. Intolerance of losartan as well as diclofenac (and other NSAIDs) reported by a few patients merits further studies, because metabolism of losartan and diclofenac depends on CYP2C9. [36, 37] We cannot exclude, however, that the association of bleeding risk and reported drug intolerance was confounded by patients reporting GPs' recommendations to avoid certain drugs (such as aspirin) as "drug intolerance."