Effects on bleeding complications of pharmacogenetic testing for initial dosing of vitamin K antagonists: a systematic review and meta-analysis


  • Manuscript handled by: P. de Moerloose
  • Final decision: F. R. Rosendaal, 18 June 2014



Although warfarin and other vitamin K antagonists (VKAs) are the most widely used oral anticoagulants for the prevention and treatment of thromboembolic events, a number of factors hamper their manageability, the most important being the inter-individual variability of the therapeutic dose requirement. Following the discovery of the influence of CYP2C9 and VKORC1 polymorphisms on VKA dose requirements, there has been interest in genotype-guided VKA dosing in order to reduce the risk of over-anticoagulation at the time of therapy initiation and hence the risk of bleeding, particularly prominent during the early days of treatment. To assess the impact on clinical outcomes of pharmacogenetic testing for initial VKA dosing, we have performed a systematic review and meta-analysis of the literature.


MEDLINE, EMBASE and Cochrane databases were searched up to March 2014. Only randomized controlled trials comparing genotype-guided vs. clinically-guided warfarin dosing were included.


Nine trials including 2812 patients met the inclusion criteria and were pooled for meta-analytical evaluation. Risk of bias, assessed according to the Cochrane methodology, showed a low risk for the majority of domains analyzed in the included trials. A statistically significant reduction in the risk ratio (RR) for developing major bleeding events was observed in the pharmacogenetic-guided group compared with the control group (RR = 0.47; 95% CI, 0.23–0.96; P = 0.040).


The results of this meta-analysis show that genotype-guided initial VKA dosing is able to reduce serious bleeding events by approximately 50% compared with clinically-guided dosing approaches.


Warfarin and other vitamin K antagonists (VKAs) are highly effective for the treatment and prophylaxis of a wide range of thrombotic cardiovascular, cerebrovascular and hematologic diseases [1, 2]. Although there is an increasing use of the latest generation of oral anticoagulants (e.g. dabigatran, rivaroxaban and apixaban) [3, 4], after more than six decades VKAs, and mainly warfarin, continue to be the mainstay of anticoagulation and the most prescribed oral anticoagulants worldwide [5]. However, VKAs narrow therapeutic window and wide inter-individual variability represent challenges for their clinical utilization [6, 7]. There has been considerable interest in the identification of genetic and non-genetic factors accounting for the wide variability in VKA dose [8-10]. Several studies have consistently documented the influence of pharmacogenetic testing in determining the VKA dose on the basis of polymorphisms in two genes, the vitamin K-epoxide reductase complex unit 1 (VKORC1, involved in the vitamin K cycle) and the cytochrome P450 2C9 enzyme (CYP2C9, involved in warfarin metabolism) [11-23]. Overall, these genotypes account for about 30‒40% of individual variability in VKA sensitivity [10]. On the basis of these observations and according to the indications emerging from systematic reviews and meta-analyses [24-28], several pharmacogenetic-guided dosing algorithms, incorporating clinical variables together with CYP2C9 and VKORC1 genotyping, have been developed to guide therapy with VKA [29-31], particularly during the initial period, which is known to be associated with a higher risk of bleeding [32]. However, almost all the published trials were focused primarily on a laboratory surrogate such as the percentage of time in the therapeutic International Normalized Ratio (INR) range (TTR), whereas the most important and relevant clinical outcomes, such as the incidence of hemorrhage and thrombosis, were only analyzed as secondary endpoints. Thus, in order to assess the clinical impact of genotype-based VKA dosing we have conducted a systematic review and meta-analysis of the published randomized controlled trials that have compared at the time of starting warfarin and other VKAs a dose-selection strategy based upon pharmacogenetic testing with one that did not. Our main goal was to assess whether or not the two approaches resulted in different rates of clinically relevant events such as bleeding, thrombosis and death.


A protocol was prospectively developed, detailing the specific objectives, criteria for study selection, risk of bias assessment, outcomes and statistical methods.


The goal of this systematic review and meta-analysis was to collect studies concerning the effectiveness of pharmacogenetic information for the prediction of optimal initial VKA dosing in patients at risk of thromboembolism. Warfarin was the most widely, but not the only, VKA used in these studies, which were all randomized controlled trials comparing a genotype-guided group with a control group assisted by standard procedures.

Search strategy

We tried to identify all published studies that evaluated the influence of CYP2C9 and VKORC1 genotypes on warfarin and other VKA dose requirements, using the MEDLINE (1980 to March week 1, 2014), EMBASE (1980 to March week 1, 2014) and Cochrane Central Register of Controlled Trials (CENTRAL) electronic databases. The search strategy was developed with no language restriction and used the keywords and subject headings presented in Table S1. Our search was supplemented by manually reviewing abstracts from the main meetings on thrombosis and hemostasis plus the reference lists of all retrieved articles, and manually searching recent issues of thrombosis and hemostasis journals plus recent reviews for additional published or unpublished trials on this topic.

Selection criteria

Study selection was performed independently by two reviewers (MF and MC), with disagreements resolved through discussion and on the basis of the opinion of a third reviewer (CB). Abstracts of all studies identified by the initial search were reviewed and irrelevant studies were excluded. We included all randomized controlled trials that supplied data on clinical outcomes in a pharmacogenetic dosing group, based upon the use of common genetic variants of CYP2C9 (CYP2C9*2 and CYP2C9*3 alleles) and/or VKORC1 (VKORC1 3673G→A), and in a group based upon a dosing algorithm that did not incorporate genetic testing. Eligible patients were adults commencing anticoagulation with VKA irrespective of the underlying condition (i.e. atrial fibrillation, deep vein thrombosis, pulmonary embolism, heart valve replacement and secondary prophylaxis of venous thromboembolism).

When multiple reports of the same study had been published, we decided to use the latest publication and to supplement it if necessary with data from earlier publications. There were no restrictions on inclusion on the basis of patient characteristics, publication type (journal article, abstract or conference proceedings) or publication language.

Data extraction

Two reviewers independently extracted the study (year of publication, design and study center) and patient characteristics (number of subjects studied, mean age, gender and race). The primary endpoints were the incidence of major bleeding, thrombosis and death during the initiation of warfarin and other VKA therapy. According to the International Society on Thrombosis and Haemostasis (ISTH) [33] a bleed was defined as major if it met at least one of the following criteria: clinically overt bleeding associated with a drop in hemoglobin of ≥ 2 g dL−1; clinically overt blood loss needing transfusion of ≥ 2 units of whole blood or erythrocytes; bleeding involving critical anatomical sites (intracranial, intraspinal, intramuscular with compartment syndrome, intraocular, retroperitoneal, pericardial and atraumatic intra-articular bleeding) and fatal bleeding.

Secondary endpoints included the time to reach a therapeutic INR (defined as the time to the first INR within target range providing that a subsequent INR ≥ 1 week later was also within target range), the percentage of TTR, the time to reach a stable warfarin or other VKA dose (defined as the INR within target range for a specified dose for a period of at least 3 weeks with < 10% change in dose), the percentage of time spent at supra-therapeutic INR (4.0 or higher), the percentage of time spent at sub-therapeutic INR (< 2.0) and the number of days spent in hospital. As the mean follow-up period varied widely between different studies, ranging from 22 days to 90 days, all the above-mentioned variables were referred to a 30-day follow-up period, when this information was available. Disagreement was resolved by consensus and on the basis of the opinion of a third reviewer. We did not attempt to contact the authors of the studies to obtain unpublished data.

Bias assessment

Two authors (MC and MF) independently assessed the studies fulfilling review inclusion criteria for methodological quality given in The Cochrane Handbook for Systematic Reviews of Intervention [34]. The risk of bias was assessed in individual studies across six domains: random sequence generation and allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessors (detection bias), incomplete outcome data (attrition bias) and selective outcome reporting (reporting bias). We categorized these judgments as ‘low risk’, ‘high risk’ or ‘unclear risk’ of bias. The blinding of outcome assessors was divided, according to the outcome and its measurement, into based on judgments and not based on judgments.

Statistical methods

A conventional meta-analytical pooling was performed. In the case of no heterogeneity (I² = 0), studies were pooled using a fixed-effect model. Where values of I² were greater than zero, a random effect analysis was undertaken. Statistical heterogeneity was evaluated with the Cochran's Q test and I2 (Higgins’ index of inconsistency). The following outcomes were evaluated: major bleeds, thromboembolism, deaths, patients with any INR ≥ 4 and TTR (%). Major bleeds, thromboembolism, deaths and patients with any INR ≥ 4 were counts of events or counts of patients’ variables, and the evaluated effect size was the risk ratio (RR). TTR was a continuous variable, and the chosen effect size was the weighted mean difference (WMD), which assures a sensible scale for an evaluation by the medical reader. The standardized mean difference (SMD) was also evaluated on this outcome. Moreover, the mean TTR was separately evaluated for the pharmacogenetic and control groups by one-arm meta-analytical procedure random effects.


Identification and characteristics of studies

Figure 1 shows the pattern of study choice from identification to final inclusion. A total of 1427 citations were identified in the literature search. Of these, nine met the inclusion criteria and were pooled for meta-analytical evaluation [11-23]. The characteristics of each study are shown in Table 1. Overall, 2812 patients were analyzed (1411 in the pharmacogenetic arm and 1401 in the control arm). There was no difference between pharmacogenetic and control groups regarding sex (percentage of men, 55.2% vs. 51.4%) and age (mean age, 64.8 years vs. 64.3 years). Regarding the intra-studies genotype frequency distribution, a substantial balance between the two different arms was observed (Table S2).

Table 1. Characteristics of the randomized controlled studies included in the analysis
Author, year [ref]Study groupNo. Follow-up* (days)Men (%)White race (%)Age*(years)Warfarin indication (%)Primary outcomes, n (%)Secondary outcomes
AFVTEJRValveOtherMajor bleedsTE eventsDeathsTTR* (%)Time to first therapeutic INR* (days)Time to stable warfarin dose* (days)Patients withHospitalization* (days)
                  INR < 2 (%)INR ≥ 4 (%)
  1. PG, pharmacogenetic; SD, standard dosing; AF, atrial fibrillation; VTE, venous thromboembolism; JR, joint replacement; TE, thromboembolic; TTR, time INR in therapeutic range; NR, not reported; *Mean or median; CYP2C9 tested; CYP2C9 and VKORC1 tested.

Hillman, 2005 [33]PG1828451007017331722112 (11)0041NR NR NR33NR
SD202844100684515152053 (15)2 (10)041NRNRNR30NR
Anderson, 2007 [34]PG1014650946313196503NRNRNR69NRNRNR30NR
Caraco, 2008 [35]PG952246100573763000000454.8 22 NR NRNR
SD9640421005931690001 (1)00247.540NRNRNR
Burmester, 2011 [36]PG115605710067433802103 (3)3 (3)2 (2)28NR 29NR383
SD115606110069493801704 (3.5)1 (1)3 (3)28NR31NR353
Borgman, 2012 [37]PG136054100593854000000634.7 NR NR40 NR
Jonas, 2013 [38]PG559054805942470471 (2)00452534NR458
SD5490506555265700174 (7)3 (6)2 (4)492835NR4913
Verhoef, 2013 [21]PG27390639768831700004 (1.5)05222NRNR31NR
SD2759056996883170001 (0.4)5 (2)04722NRNR33NR
Kimmel, 2013 [22]PG51428537359235600114 (1)5 (1)2 (<1)45NRNRNR20NR
SD501284973572160001110 (2)4 (1)1 (< 1)45NRNRNR21NR
Pirmohamed, 2013 [23]PG2279064986772280000006731 44 NR27NR
SD22890589966722800001 (0.5)0602959NR37NR
Figure 1.

Flow chart of study inclusions.

Summary graphs of methodological quality items are presented in Figures S1 and S2 for the nine randomized controlled trials included in the analysis. The generation of the randomization process and allocation concealment was generally adequately described. There were three studies defined as double blind, three as single blind, and three were unblinded. The remaining items were graded as free of bias in all but one trial that was at high risk of reporting bias.


The main results of the meta-analysis are in Table 2. Primary outcomes of interest were rates of major bleeding, thromboembolism and death. Major bleeding was the most important complication of VKA therapy in our analysis, accounting for approximately half (31/69) of all oral VKA-associated adverse outcomes. There was lack of statistical heterogeneity for all the outcomes analyzed (I2 = 0) and a fixed effect model was used. For the outcome ‘major bleeds’, a statistically significant difference was found compared with the control, with an RR = 0.47 (95% CI, 0.23–0.96; P = 0.040) favoring the pharmacogenetic-assisted arms of the studies (Fig. 2). Accordingly there was a mean 52.5% reduction of major bleeding episodes if genetic information was used.

Table 2. Outcomes under evaluation
 RR95% Confidence interval P I 2
Lower limitUpper limit
Major bleeds, no. of events0.470.230.960.0400.0%
TE, no. of events0.980.452.110.9630.0%
Deaths, no. of events0.710.192.600.6150.0%
Patients with INR ≥ 4, no.0.920.811.040.2210.0%
 WMD95% Confidence interval P I 2
Lower limitUpper limit
  1. RR, risk ratio; TE, thromboembolism; INR, International Normalized Ratio; WMD, weighted mean difference; I2, Higgins’ index of inconsistency, for heterogeneity.

Time in therapeutic range, %4.25−1.9510.450.18089.4%
Figure 2.

Meta-analysis of the risk ratio of major bleeding between pharmacogenetic dosing and the control group (Forest-plot).

Secondary outcomes of this analysis included TTR, time to first therapeutic INR, time to stable warfarin, number of patients with INR < 2 or ≥ 4 and number of days in hospital. However, only for the variables TTR and number of patients with INR ≥ 4 was it possible to perform a statistical analysis. The continuous variable TTR showed a modest increase in the pharmacogenetic-assisted patients, the mean TTR being 50.9% (95% CI, 39.6–62.1%) in the pharmacogenetic group and 46.8% in the control group (95% CI, 36.0–57.7), with high heterogeneity (I2 = 99% and 98%, respectively). Also the SMD method failed to produce a statistically significant result for TTR (SMD = 0.22; 95% CI, −0.06–0.52; P = 0.13).


Following the demonstration that CYP2C9 and VKORC1 polymorphisms are important determinants of dose requirements for warfarin and other VKAs, a number of investigators have attempted to incorporate various environmental and genetic factors into algorithms aimed at developing models to predict the initial VKA dosage. However, in spite of the approval by the US Food and Drug Administration (FDA) of a labelling change for warfarin mentioning the effects on dose requirements of genetic variations in CYP2C9 and VKORC1 enzymes [41], the clinical adoption of genotype-guided administration of VKA is still uncertain. Interest in this issue has been renewed by the recent publication of three large randomized controlled trials. Verhoef and colleagues [21] found that the percentages of TTR during the first 4 weeks after the initiation of treatment with acenocoumarol and pheprocoumon in genotype-guided and clinically-guided groups were 52.8% and 47.5% (P = 0.02), respectively. Similarly, in the 12 weeks after the initiation of warfarin anticoagulation, Pirmohamed and colleagues [23] demonstrated that a genotype-guided algorithm yielded better TTR results than those achieved with an algorithm based on clinical variables (67.4% vs. 60.3%; P < 0.001). In contrast, no statistically significant difference was found by Kimmel and colleagues [22], who compared the TTR of the two different algorithms at 4 weeks (45.2% in the genotype-guided group vs. 45.4% in the clinically-guided group).

All these trials, similarly to the great majority of previous studies, used as a primary endpoint a laboratory parameter (i.e. the TTR). Although this surrogate phenotype correlates with the incidence of bleeding during VKA therapy [42], the right question for the clinician is not how much time patients spend within therapeutic range during the initial phase of VKA treatment but whether or not the use of pharmacogenetic testing reduces the incidence of clinical complications as compared with clinically-guided dosing. This issue is of particularly great relevance when treatment is started, because the incidence of adverse events as a result of VKAs, particularly bleeding, is highest during the initial phase (usually the first month) of treatment [32].

With this background, we conducted a systematic review and meta-analysis of the existing literature. In order to improve the quality of the data collected (and of the results) our analysis was limited to randomized controlled trials. The most striking finding of this meta-analysis, including 2812 patients from nine randomized controlled trials, was that pharmacogenetic analysis halved the risk of major bleeding during the initial phase of anticoagulation with warfarin or another VKA. This finding appears relevant, particularly considering the high consistency (I2 = 0%) of collected data (Table 2). The degree of statistical significance for major bleeds was not strong (P = 0.040), perhaps depending on the modest number of events observed; however, the effect size was not trivial if examined under a correct clinical prospective (RR = 0.47; 95% CI, 0.23–0.96). At variance with our meta-analysis, a systematic review conducted by Kangelaris and colleagues [24] found no statistically significant difference in bleeding rates between patients who used or did not use genetic testing before starting VKA. Perhaps the low number of randomized controlled trials considered by Kangelaris (three studies with totalling 423 patients developing eight major bleeds) did not allow them to achieve sufficient statistical power to demonstrate such an association.

Another issue analyzed by this meta-analysis was the effectiveness of genotype-guided dosing of VKA in reducing the degree of variability of anticoagulation compared with clinically-guided dosing. Although not statistically significant, we observed a trend towards improvement of the TTR in the pharmacogenetic arm, which is likely to be the mechanism for the reduced bleeding rate observed in this group. Perhaps the lack of statistical significance for this secondary outcome may be due to the great variability among the time periods considered in the different studies pertaining to the initial phase of oral anticoagulation with VKA (ranging from 22 to 90 days).

With regards to bias, the use of the Cochrane methodology showed that the studies analyzed by us had some methodological limitations. The generation of the randomization process and allocation concealment were adequately described in the majority of trials, with no systematic differences between baseline characteristics of the groups in each study. There were three double-blind and three single-blind trials. However, the clinical outcomes of our analysis and their measurement are unlikely to be influenced by lack of blinding, even though the latter would make more accurate the evaluation of the degree of severity of adverse events.

Other potential limitations of our meta-analysis include the use of VKAs other than warfarin in one trial [21], and the enrolment of a significant number of black individuals in another study [22]. This latter factor could be of clinical relevance, especially considering that differences between African American and Caucasian patients were detected regarding the mean TTR in genotype-guided vs. clinically-guided groups [22]. However, an analysis stratified by ethnic groups was not possible because relevant information was lacking in almost all studies.

In conclusion, this meta-analysis suggests that a genotype-guided approach appears to lower the risk of severe bleeding in the initial period of VKA anticoagulation. Considering the limitations of this meta-analysis and of meta-analysis in general, prospective trials based upon clinical rather than on laboratory outcomes are needed to firmly establish the efficacy of a pharmacogenetic approach in guiding initial VKA dose requirements. Health technology assessments are also required before recommending this approach in clinical practice.


M. Franchini conception and design of the study, acquisition of data, analysis and interpretation of data, drafting of the manuscript and statistical analysis. C. Mengoli conception and design of the study, analysis and interpretation of data, critical revision of the manuscript, supervision and statistical analysis. M. Cruciani conception and design of the study, analysis and interpretation of data, critical revision of the manuscript and supervision. C. Bonfanti acquisition, analysis and interpretation of data and critical revision of the manuscript. P.M. Mannucci conception and design of the study, analysis and interpretation of data, critical revision of the manuscript and supervision. All the authors provided final approval of the article.

Disclosure of Conflict of Interests

The authors state that they have no conflict of interests.