Predicting warfarin maintenance dose in patients with venous thromboembolism based on the response to a standardized warfarin initiation nomogram

Transcription

1 Journal of Thrombosis and Haemostasis, 7: DOI: /j x ORIGINAL ARTICLE Predicting warfarin maintenance dose in patients with venous thromboembolism based on the response to a standardized warfarin initiation nomogram A. LAZO-LANGNER, K. MONKMAN and M. J. KOVACS Division of Hematology, Department of Medicine, University of Western Ontario, London, ON, Canada To cite this article: Lazo-Langner A, Monkman K, Kovacs MJ. Predicting warfarin maintenance dose in patients with venous thromboembolism based on the response to a standardized warfarin initiation nomogram. J Thromb Haemost 2009; 7: Summary. Background: Polymorphisms in the VKORC1 and CYP2C9 genes influence warfarin requirements. It has been suggested that dosing algorithms incorporating them might outperform usual care. Standardized warfarin initiation nomograms are safe and effective and patientsõ responses to them could be used to predict warfarin requirements without the need for genetic testing. Objectives: To develop a model to predict warfarin dose requirements based on the response to a standard nomogram without using genetic testing. Patients/methods: We included 363 outpatients with acute venous thromboembolism who were started on treatment using a standardized warfarin nomogram and achieved a stable maintenance warfarin dose defined as a dose prescribed twice consecutively after two consecutive INR measurements between 2.0 and 3.0. Linear regression was used to derive equations predicting the maintenance dose and models were validated using non-parametric bootstrapping and tested in an independent cohort. Results: Three models were constructed for patients completing the nomogram until day 3 (warfarin dose (mg week )=Exp [ (INR 3 ))0.008(Age)]; R 2 adj = 0.462), day 5 (warfarin dose (mg week ) = Exp[ (INR 3 ) )0.285(DINR 5)3 )]; R 2 adj = 0.603) and day 8 (warfarin dose (mg week ) = Exp[ (INR 8 ) (cumulated warfarin dose until nomogram day 7)]; R 2 adj = 0.643), where Exp is the exponential function; INR 3 and INR 8 are the INR on days 3 or 8 of the nomogram, and DINR 5)3 is the difference in the INR on days 5 and 3. All models were internally and externally validated and were accurate to within 25% of the actual dose in >60% of patients. Conclusion: Maintenance warfarin dose can be accurately Correspondence: Michael J. Kovacs, Hematology Division, London Health Sciences Centre, Victoria Hospital, 800 Commissioners Rd E, PO Box 5010, Rm A2-401, London, ON, N6A 5W9 Canada. Tel.: (ext ); fax: Michael.Kovacs@lhsc.on.ca Received 6 March 2009, accepted 5 May 2009 predicted using individual response to a standard warfarin initiation nomogram without the need for costly genetic testing. Keywords: CYP2C9, dose prediction, nomogram, venous thromboembolism, VKORC1, warfarin. Introduction Warfarin is the most commonly prescribed oral anticoagulant for primary and secondary prevention of venous thromboembolism (VTE) but its management is complicated by a narrow therapeutic window and high variability in dose requirements. These limitations make necessary the use of close monitoring through frequent determinations of the international normalized ratio (INR), which is both resource-intensive and inconvenient. Moreover, despite numerous attempts to standardize warfarin management, a high proportion of patients are still inadequately anticoagulated [1,2]. The mechanisms underlying the variability in dose requirements are incompletely understood but numerous studies have evaluated different potential predictors of warfarin requirements. To date, several clinical factors have been found to be associated with warfarin requirements, including age, sex, nutritional status, presence of comorbid conditions, and concomitant medications [3,4]. More recently, a growing body of evidence has suggested that single nucleotide polymorphisms (SNPs) occurring in the genes encoding the vitamin K epoxide reductase complex (VKORC1) and the cytochrome P450 2C9 isozyme (CYP2C9) [5 9] are major determinants of warfarin dose requirements because they influence the sensitivity to the drug and its metabolism. These findings have resulted in the development of a number of models to predict warfarin requirements, all of which incorporate both clinical and genetic factors, with the goal of safely hastening the achievement of a stable maintenance dose and ultimately minimizing the risk of excessive anticoagulation and bleeding [10 16]. Moreover, in 2007 the US Food and Drug Administration approved an update to the warfarin prescribing information suggesting the use of genetic-based warfarin dosing [17].

2 Warfarin dose prediction based on a standard nomogram 1277 Despite the initial enthusiasm, most published prediction models incorporating genetic factors have accounted for only about half of the dose variability. This has been recently replicated in a large study including over 1500 patients, which found that a prediction model incorporating genetic information explained between 53 and 59% of dose variation [18]. On the other hand, genetic testing is currently not (and likely will not be) widely available, is time consuming and expensive and, because the INR is in great part determined by the genetic makeup, the INR response to initial warfarin doses could be regarded as a surrogate marker for both genetic and environmental factors. Therefore a potential approach is to use the individual patientõs response to standardized warfarin initiation nomograms in order to predict maintenance warfarin requirements. Given that the occurrence of an acute VTE is a random event, the use of a nomogram provides a convenient and practical way to rapidly initiate warfarin therapy compared with a strategy incorporating genetic information that would likely result in unnecessary delays because of the time required to complete the genetic testing and would certainly result in an additional cost. We have previously published a randomized trial showing that the use of a 10-mg standardized warfarin initiation nomogram for the ambulatory treatment of patients with VTE results in the rapid achievement of a therapeutic INR without a high incidence of recurrent thromboembolism or bleeding events [19]. Furthermore, we have demonstrated that this nomogram is safe and effective in the routine management of ambulatory patients with newly diagnosed VTE [20]. The nomogram provides a fixed warfarin dose during the first 2 days and subsequent doses are adjusted according to the INR values on days 3, 5 and 8; however, there is no clear strategy to guide dosing from day 8 onwards. Therefore, we sought to develop a model for predicting the warfarin requirements in ambulatory patients being treated for VTE based solely on clinical factors and response to the nomogram, without the need for genetic testing. Methods Patients, data and study endpoints We performed a retrospective cohort study of consecutive ambulatory patients treated in the thrombosis clinic of a tertiary care hospital between January 2004 and April All patients were diagnosed with an objectively confirmed VTE according to previously published criteria [21]. Patients were included if they: (i) received initial standard treatment with LMWH, (ii) initiated warfarin using a previously published fixed-dose warfarin nomogram (Fig. 1) [19], (iii) followed the nomogram at least until day 3, (iv) achieved a stable warfarin dose defined as a dose that was prescribed twice consecutively after two consecutive INR measurements between 2.0 and 3.0 [3], (v) had a baseline INR of <1.4, (vi) had a platelet count > L, (vii) were at least 18 years old, (viii) had not received treatment with oral anticoagulants within the previous 2 weeks, and (ix) were followed for at least 90 days. According to the nomogram all patients receive 10 mg of warfarin on days 1 and 2 and an INR is then obtained on days 3, 5 and 8 to guide subsequent dosing. Patients not using the 10 mg nomogram or treated with LMWH monotherapy were excluded from the study. In our centre, the majority of patients with known active malignancy receive LMWH monotherapy for venous thromboembolism based on the results of the CLOT study [22] unless thrombosis occurred in the setting of a central venous catheter [23]. Variables of interest were defined aprioriand included age, weight (kg), gender, active cancer defined as a malignancy requiring treatment within 1 year prior to diagnosis or diagnosed up to 90 days after initiating anticoagulation, creatinine (mg dl ),INRonday3(INR 3 ), INR on day 5 (INR 5 ), INR on day 8 (INR 8 ), difference between INR on day 5 and day 3 (DINR 5)3 ), difference between INR on day 8 and day 5 (DINR 8)5 ), difference between INR on day 8 and day 3 (DINR 8)3 ), total warfarin dose received according to the nomogram from days 1 to 4 (Dose 1 4 ), and total warfarin dose received according to the nomogram from days 1 to 7 (Dose 1 7 ). The clinical factors were chosen because they are readily available and have been suggested to be predictors in other studies [11,24,25]. The primary study endpoint was the stable weekly maintenance dose of warfarin (in mg) as previously defined. Secondary endpoints included: (i) the accuracy of the model defined as the difference between the predicted and observed warfarin maintenance dose, expressed as a percentage of the latter, and (ii) the percentage of patients under-dosed or over-dosed >25% of the actual observed warfarin maintenance dose. Statistical analysis All data for continuous variables are presented as mean ± standard deviation (SD). Ninety-five per cent confidence intervals (CI) for proportions were estimated using the Wilson score method [26]. The observed weekly maintenance dose was compared between patients grouped by categorical variables using a StudentÕs t-test for independent samples with a two-sided P- value <0.05 considered as statistically significant. For continuous covariates simple linear regression analysis was conducted with the log-transformed observed weekly maintenance dose as the dependent variable. Residual analyses showed that, in order to maintain the assumption of homoscedasticity, the best fit models required using a logarithmic transformation of the weekly maintenance dose as well as the reciprocal of INR 3 and INR 8. All covariates achieving a significance <0.1 were included in the multiple regression models. Stepwise multiple linear regression was then conducted using values for the probability of F and to enter terms in or to remove terms from the equation, respectively. Three prediction models were constructed in sequence assuming patients completed the nomogram until days 3 (model 1), 5 (model 2) and 8 (model 3). The final models were chosen based on residual analysis, assessment of collinearity, adjusted R 2 values and R 2

3 1278 A. Lazo-Langner et al Day 3 INR < Days 3, 4 Dose (mg) 15, 15 10, 10 Day 5 INR < > 3.5 Days 5, 6, 7 Dose (mg) 15, 15, , 5, 7.5 0, 5, 5 0, 0, , 5 5, 5 < > , 7.5, 7.5 5, 5, 5 2.5, 2.5, 2.5 0, 2.5, , 2.5 0, 2.5 < > 3.5 5, 5, 5 2.5, 5, 2.5 0, 2.5, 0 0, 0, 2.5 > 3.0 0,0 - Day 1 = First day of warfarin - All patients receive 10 mg day 1 and day 2 - INR in morning, drug given early evening < > , 2.5, , 0, 2.5 0, 2.5, 0 0, 0, 2.5 Fig. 1. Warfarin initiation nomogram. On days 1 (first day of warfarin) and 2 all patients receive 10 mg of warfarin. From day 3 onwards, dose is adjusted according to the nomogram. (Reproduced with permission from the American College of Physicians) changes. Models resulting in an R 2 change 0.05 were discarded. Collinearity was evaluated using variance inflation factors, condition indices and variance proportions. The models were internally validated through non-parametric bootstrapping, a technique that allows obtaining estimates of the variability of the b coefficients for the parameters in the regression equations [27]. We used random sampling with replacement to obtain 1000 bootstrapped samples of 363, 305 and 260 observations for models 1, 2 and 3, respectively, and used them to obtain estimates of the standard error and 95% confidence intervals around the parametersõ coefficients. Models were further tested for external validity in an independent cohort of 66 VTE outpatients started on warfarin using the same nomogram. In this independent cohort parameter estimates, 95% confidence intervals and standard errors were also obtained using non-parametric bootstrapping. Finally, the accuracy of the model was assessed by calculating the variation between the predicted and the observed doses, expressed as a percentage of the latter. All analyses were performed using Microsoft Excel 2003 (Microsoft Corp., Redmond, WA, USA) and SPSS Statistics Release (SPSS, Chicago, IL, USA). Results Between January 2004 and April patients were treated for acute VTE using our 10 mg warfarin initiation nomogram and were followed for at least 90 days. Of the 414 patients, we included in the analysis those who achieved a stable warfarin dose and: completed the nomogram until at least day 3 (for derivation of model 1; n = 363 [87.7%; 95% CI ]), completed the nomogram at least until day 5 (for derivation of model 2; n = 305 [73.7%; 95% CI ]), and completed the nomogram until day 8 (for derivation of model 3; n =260 [62.3%; 95% CI ]. For the 363 patients the mean age was 57.5 ± 17.8 years, and the mean weight was 85.2 ± 20.8 kg. One hundred and ninety patients were female (52.3%; 95% CI, ), and 56 had active cancer (15.4%; 95% CI, ). The mean weekly maintenance warfarin dose was 36.6 ± 17.8 mg. The mean dose for males was 40.6 ± 18.0 mg whereas for females it was 33.0 ± 16.8 mg (P < 0.001). Patients with active cancer required a lower weekly maintenance dose than those without cancer (29.07 ± 14.1 vs ± 18.1 mg; P < 0.001). On simple linear regression all covariates except DINR 8)3 achieved a significance <0.1 (Table 1) and were therefore included in the multiple linear regression models. The most predictive covariates were: Dose 1 7 (R 2 adj = 0.517), INR 5 (R 2 adj = 0.413), INR 3 (R 2 adj = 0.395), and Dose 1 4 (R 2 adj = 0.391). Stepwise multiple regression resulted in the following final models: Dose ðmg week 1 Þ¼Exp ½2:737 þ 1:896ðINR 1 3 Þ 0:008(Age)Š ð1þ

4 Warfarin dose prediction based on a standard nomogram 1279 Table 1 Simple linear regression analysis. Adjusted coefficients of determination, slopes and significance for independent variables. The log-transformed weekly warfarin maintenance dose is the dependent variable Variable R 2 adj Slope (b) SE P Age (years) ) <0.001 Male gender <0.001 Weight (kg) <0.001 Active malignancy ) <0.001 Creatinine (mg dl ) ) INR <0.001 INR <0.001 INR DINR 5) ) <0.001 DINR 8) <0.001 DINR 8) Dose <0.001 Dose <0.001 R 2 adj, adjusted coefficient of determination; SE, standard error; INR x, reciprocal of the international normalized ratio on days 3, 5 or 8; DINR x)y, difference of the INR values on days shown; Dose x y, total warfarin dose (in mg) received by the patient according to the nomogram during the indicated days. Dose ðmg week 1 Þ¼Exp ½2:261 þ 2:412ðINR 1 3 Þ ð2þ 0:285ðDINR 5 3 ÞŠ Dose ðmg week 1 Þ¼Exp ½1:574 þ 1:788ðINR 1 8 Þ ð3þ þ 0:024ðDose 1 7 ÞŠ where Exp is the exponential function. The final models had adjusted coefficients of determination of 0.462, and for models 1, 2 and 3, respectively. No collinearity was found. Internal validity of the models was then tested on 1000 bootstrapped samples and the estimates thus obtained showed that the models worked properly (Table 2). The predicted weekly warfarin doses obtained using the three models were plotted vs. the observed therapeutic weekly warfarin doses, finding a good correlation (Fig. 2). Models were validated in an independent cohort of 66 patients obtaining values for the adjusted R 2 of 0.417, and for models 1, 2 and 3, respectively. In this validation cohort all the parameters in the regression models were significant in the bootstrap analysis. Finally, the evaluation of the modelsõ accuracy showed that the variation between the predicted and actual observed dose was within 25% of the latter in 59.8%, 69.2% and 72.7% of the patients for models 1, 2 and 3, respectively (Table 3). The three models resulted in a predicted dose that was over 25% below the actual dose in about 18% of the patients. Model one predicted doses that were above 25% of the actual dose in 21.8% of patients whereas models 2 and 3 resulted in <15% of patients being overdosed >25% (Table 3). Discussion Since the discovery that polymorphisms in the CYP2C9 [5,6] and VKORC1 [8,9] genes are significantly associated with warfarin dosing requirements, several maintenance dose prediction models that incorporate genetic factors have been published. These models have been derived from patient populations receiving warfarin for atrial fibrillation [16], and varying [10,13,15] or unspecified indications [11], and have used different warfarin initiation schemes. Almost all models have included age, CYP2C9 status, and some measure of body size (body surface area, height, or weight). VKORC1 has been included in all models published after its discovery in The first prediction equations including both genetic factors and the response to initial warfarin dosing were published in 2007 [12,14]. Millican et al. published a model to predict the maintenance dose requirements for orthopedic surgery patients receiving venous thromboembolism prophylaxis with warfarin. Initial dosing was based on clinical factors and either the CYP2C9 genotype or the VKORC1 genotype, and calculated Table 2 Parameter estimates and coefficients of determination for the final multiple linear regression models obtained through multiple linear regression and non-parametric bootstrapping Parameter Slope (b) 95% CI SE P Model 1 (R 2 adj = 0.462, SE 0.379; R 2 adj bootstrapped = 0.465) Intercept , <0.001 INR , <0.001 Age )0.008 )0.10, ) <0.001 Model 2 (R 2 adj = 0.603, SE 0.326; R 2 adj bootstrapped = 0.606) Intercept <0.001 INR , <0.001 DINR 5)3 )0.285 )0.344, ) <0.001 Model 3 (R 2 adj = 0.643, SE 0.306; R 2 adj bootstrapped = 0.646) Intercept , <0.001 INR , <0.001 Dose , <0.001 The 95% confidence intervals and standard errors were calculated from 1000 bootstrapped samples of 363, 305 and 260 patients for models 1, 2 and 3, respectively, obtained through random sampling with replacement. The dependent variable in the models is the log-transformed weekly maintenance dose. CI, confidence interval; SE, standard error; R 2 adj, adjusted coefficient of determination; INR x, reciprocal of the international normalized ratio on days 3 or 8; DINR 5)3, difference of the INR values on days 5 and 3; Dose 1 7, total warfarin dose (in mg) received by the patient according to the nomogram during the indicated days.

5 1280 A. Lazo-Langner et al Predicted weekly warfarin maintenance dose (mg) Model 1 Model 2 Model Observed weekly warfarin maintenance dose (mg) Fig. 2. Scatter plots of observed vs. predicted weekly warfarin maintenance dose. The plots show the correlation between the observed weekly warfarin dose and the weekly dose predicted by the three prediction models in patients started on warfarin therapy using a standardized 10 mg warfarin nomogram. according to a pharmacogenetic algorithm that was made available online later at [12]. They found that the INR value after three warfarin doses accounted for 34% of the variability in the maintenance warfarin dose. Their final model accounted for 79% of the variation in dose requirements, and included the following eight variables: INR after three warfarin doses, CYP2C9*3 and CYP2C9*2 genotypes, first and second warfarin doses, estimated blood loss, smoking status, and VKORC1 genotype. After this report, the same group published another study on patients with diverse indications for anticoagulation, including atrial fibrillation and VTE. In this study they found that the pharmacogenomic dosing algorithm explained 53 54% of the dose variability [28]. Recently, the same group found that early INR values were significantly associated with stable warfarin dose and further inclusion of genetic information on both VKORC1 and CYP2C9 genes did not result in a great improvement in dose prediction. Furthermore, after the first week of treatment genetic information did not improve the coefficient of determination of the prediction model [29]. However, in spite of the increasing interest in this issue, only two adequately designed randomized trials comparing the use of a genetic prediction model with standard warfarin dosing have been published so far. The first one used information on both VKORC1 and CYP2C9 and found over a 90-day period that although there was no significant difference in the percent of out-of-range INRs between the two arms, the use of the genotype-based prediction model resulted in fewer dosage changes and INR tests, although this model only explained 47% of dose variability [30]. The second study tested an algorithm including only genetic information on the CYP2C9 gene compared to a computer-generated warfarin initiation algorithm. This study showed that the genetic based algorithm resulted in faster achievement of a therapeutic INR and stable anticoagulation but it was not designed to predict maintenance doses [31]. Although recently overshadowed by studies of genetic predictors, the principle of using the response to initial warfarin doses to predict future requirements is not new. The use of the early INR values to predict future warfarin dose requirements was first proposed over 30 years ago [32], and numerous publications have since shown a relationship between the initial response to warfarin and the required maintenance dose [33,34]. One of the first ones was published in 1984 and consisted of a nomogram for the initiation of warfarin dosing in venous thromboembolism that used the British comparative ratio (a predecessor of the INR) after three doses of warfarin to predict the maintenance dose [24]. In this study of 50 patients, Table 3 Accuracy of the regression models Variation between predicted and observed warfarin maintenance dose Percentage of patients (95% CI) Percentage of patients (95% CI) Model Within 10% Within 15% Within 20% Within 25% Under-dosed >25% Over-dosed >25% Model (19.61, 28.33) (33.7, 43.67) (45.84, 56.07) (54.66, 64.70) (14.80, 22.77) (17.83, 26.29) Model (22.22, 32.13) (37.52, 48.56) (51.44, 62.48) (63.79, 74.10) (13.83, 22.38) (9.78, 17.37) Model (29.82, 41.37) (43.59, 55.65) (59.02, 70.54) (66.98, 77.75) (13.19, 22.37) (6.92, 14.25) The table shows the variation between the warfarin maintenance dose predicted by the models and the actual observed dose, expressed as a percentage of the latter. CI, confidence interval.

6 Warfarin dose prediction based on a standard nomogram 1281 Table 4 Hypothetical clinical examples illustrating the use of the three warfarin dose prediction models Model 1 55-year-old patient. INR obtained on day 3 of the nomogram is 1.4 Model 2 Model 3 Dose ðmg week 1 Þ ¼ Exp ½2:737 þð1:896 1:4 1 Þ ð0:008 55ÞŠ ¼ Exp ½2:737 þð1:896 0:714Þ ð0:008 55ÞŠ ¼ Exp ½2:737 þ 1:354 0:44Š ¼ Exp ½3:651Š ¼ 38:52 mg week 1 INR obtained on day 3 of the nomogram is 1.6. INR obtained on day 5 is 2.5 Dose ðmg week 1 Þ ¼ Exp ½2:261 þð2:412 1:6 1 Þ ð0:285 ð2:5 1:6ÞÞŠ ¼ Exp ½2:261 þð2:412 0:625Þ ð0:285 0:9ÞŠ ¼ Exp ½2:261 þ 1:507 0:257Š ¼ Exp ½3:768Š ¼ 33:52 mg week 1 INR obtained on day 8 of the nomogram is 2.9. Total dose received from day 1 to day 7 of the nomogram is 40 mg Dose ðmg week 1 Þ ¼ Exp ½1:574 þð1:788 2:9 1 Þþð0:024 40ÞŠ ¼ Exp ½1:574 þð1:788 0:345Þþð0:024 40ÞŠ ¼ Exp ½1:574 þ 0:617 þ 0:96Š ¼ Exp ½3:151Š ¼ 23:35 mg week 1 Exp, exponential function; INR x, reciprocal of the international normalized ratio on days 3 or 8; DINR 5)3, difference of the INR values on days 5 and 3; Dose 1 7, total warfarin dose (in mg) received by the patient according to the nomogram during the indicated days. the predicted maintenance dose correlated well with the observed maintenance dose and the predicted daily dose was within 1 mg of the actual dose for 46 patients (92%). More recently, Siguret et al. [35] published a model for warfarin initiation in elderly inpatients that predicted the maintenance dose based on the INR value after three warfarin doses, also showing a good correlation between the predicted and observed warfarin maintenance dose. A problem found in previous studies including clinical and genetic information either alone or in combination, is that in most of them warfarin was initiated using different schemes and because INR value is obviously dependent on warfarin dose, the predictive models using early INR values might not be applicable to all patients. Therefore, a particular strength of the present work is that we developed a set of prediction models in patients starting warfarin therapy using a standardized nomogram in which all patients receive 10 mg of warfarin on days 1 and 2 and an INR is obtained on days 3, 5 and 8 to guide subsequent dosing using a prespecified algorithm, an approach that has been proven to be useful and safe in clinical practise [20]. Using a standard dosing scheme has the advantage of providing a uniform way to ÔchallengeÕ the patientsõ genetic makeup, thus making the individual responses of patients a true reflection of their genetic status (i.e. a surrogate marker). This is not possible if patients get different initial warfarin doses. Furthermore, previous studies have included patients with different indications for anticoagulation in both in- and outpatient settings, which makes it difficult to decide which model to use, and additionally, many published models included covariates that were likely to be collinear, and although some studies did specifically report on evaluation of collinearity, most studies did not, thus raising questions regarding the validity of the regression models. Finally, we also considered that developing three models was necessary in order to make them useful in clinical practise because not all patients follow the nomogram to completion. Several findings arose from our study. First, although age has been repeatedly found to be associated with decreased warfarin requirements [3,36] we found that it was only significant in the first model and in the second and third models it became non-predictive, almost certainly as a consequence of incorporating more informative covariates such as subsequent INR values or cumulative warfarin dose. To the best of our knowledge this has not been found previously, most likely because previous studies have focused on single INR measurements. Second, interestingly, models 2 and 3 were similarly efficient (i.e. a high coefficient of determination with a small number of variables) compared with each other and with adjusted R 2 values similar to previously published geneticbased models. We believe that for model number 2, which incorporated the first INR measurement (i.e. INR 3 )andthe difference between the second and the first INR values (i.e. DINR 5)3 ), this might be due to the fact that the INR 3 value reflects the sensitivity to the warfarin and the DINR 5)3 is likely a reflection of its metabolism. It is known that early INR responses are mainly influenced by VKORC1 and to a lesser extent by CYP2C9 [7,37] and our findings support the notion of using clinical surrogates for genotypic data. Not surprisingly we found that the total dose of warfarin prescribed in the nomogram is the covariate with the highest predictive value. This covariate was incorporated in model number 3 and likely reflects the fact that by day 5 of the nomogram most patients would have reached a therapeutic INR and an approximate maintenance dose [19]. In third place we found that all models were accurate within 25% of the actual observed maintenance dose in at least 60% of the patients. More importantly, models 2 and 3 resulted in overdosing >25% in <15% of patients. Finally, by explaining between 46 and 64% of dose variability our models performed very similarly to models including genotypic data. This suggests that they can be used to guide subsequent dosing in VTE patients started on anticoagulants using our nomogram without the need for genetic testing. Table 4 shows three hypothetical examples illustrating the use of the models described in this work. The models that we propose have two main advantages: first, they use readily available information that does not require any

7 1282 A. Lazo-Langner et al genetic testing, with its associated costs, and second, they can be easily implemented using widely available commercial or public domain software. Details for implementing the models are available from the authors upon request. Cost issue is particularly appealing because a recent study suggested that, in patients with atrial fibrillation, at the present time a pharmacogenomic approach is unlikely to be cost-effective [38]. Limitations of our study include the fact that it has not yet been prospectively validated and that due to its retrospective nature genetic information was not available. In addition, our findings should not be extrapolated to other warfarin initiation nomograms. Furthermore, because we only included outpatients treated for VTE and our patients were relatively young (although 24% of the patients included were at least 75 years old), our findings might not be applicable to all patient populations. We believe that given that the models developed herein performed quite similarly to genetic-based models, this warrants a direct comparison with a genetic model. Despite the appeal of genetic testing, the utility of the patientõs response to initial warfarin dosing as a simple and inexpensive predictor of warfarin requirements cannot be neglected. Many clinicians believe that a genetic-based dosing scheme is unlikely to result in a practical advantage due to a number of reasons, including timely access to genotyping, which might result in unnecessary treatment delays and subsequent prolongation of the parenteral anticoagulation, increased costs, and inability of genetic-based models to account for environmental factors. Furthermore, dosing adjustments are still to be made based on INR values that could be regarded as a surrogate marker of genetic information. Finally, although models (including genetic) may predict maintenance dose, they do not account for long-term dose variability, which is still the main limitation of warfarin therapy. In conclusion, we have derived a group of equations based on readily available information that predicts maintenance warfarin dose in VTE outpatients initiating anticoagulation with a standardized nomogram. The performance of these models is comparable with that published for genetic-based ones and it could be used in conjunction with our 10 mg initiation nomogram to guide warfarin dosing in this group of patients. Further studies are needed to compare these models with genetic-based initiation schemes. Disclosure of Conflict of Interests The authors state that they have no conflict of interest. References 1AnsellJ,HirshJ,HylekE,JacobsonA,CrowtherM,PalaretiG. Pharmacology and management of the vitamin K antagonists: American College of Chest Physicians Evidence-Based Clinical Practice Guidelines (8th Edition). Chest 2008; 133: 160S 98S. 2 Wittkowsky AK, Devine EB. Frequency and causes of overanticoagulation and underanticoagulation in patients treated with warfarin. Pharmacotherapy 2004; 24: Garcia D, Regan S, Crowther M, Hughes RA, Hylek EM. Warfarin maintenance dosing patterns in clinical practice: implications for safer anticoagulation in the elderly population. Chest 2005; 127: Hylek EM, Heiman H, Skates SJ, Sheehan MA, Singer DE. Acetaminophen and other risk factors for excessive warfarin anticoagulation. JAMA 1998; 279: Aithal GP, Day CP, Kesteven PJ, Daly AK. Association of polymorphisms in the cytochrome P450 CYP2C9 with warfarin dose requirement and risk of bleeding complications. Lancet 1999; 353: Higashi MK, Veenstra DL, Kondo LM, Wittkowsky AK, Srinouanprachanh SL, Farin FM, Rettie AE. Association between CYP2C9 genetic variants and anticoagulation-related outcomes during warfarin therapy. JAMA 2002; 287: Schwarz UI, Ritchie MD, Bradford Y, Li C, Dudek SM, Frye- Anderson A, Kim RB, Roden DM, Stein CM. Genetic determinants of response to warfarin during initial anticoagulation. NEnglJMed 2008; 358: DÕAndrea G, DÕAmbrosio RL, Di Perna P, Chetta M, Santacroce R, Brancaccio V, Grandone E, Margaglione M. A polymorphism in the VKORC1 gene is associated with an interindividual variability in the dose-anticoagulant effect of warfarin.blood 2005; 105: Rieder MJ, Reiner AP, Gage BF, Nickerson DA, Eby CS, McLeod HL, Blough DK, Thummel KE, Veenstra DL, Rettie AE. Effect of VKORC1 haplotypes on transcriptional regulation and warfarin dose. N Engl J Med 2005; 352: Caldwell MD, Berg RL, Zhang KQ, Glurich I, Schmelzer JR, Yale SH, Vidaillet HJ, Burmester JK. Evaluation of genetic factors for warfarin dose prediction. Clin Med Res 2007; 5: GageBF,EbyC,MilliganPE,BanetGA,DuncanJR,McLeodHL. Use of pharmacogenetics and clinical factors to predict the maintenance dose of warfarin. Thromb Haemost 2004; 91: Lenzini PA, Grice GR, Milligan PE, Gatchel SK, Deych E, Eby CS, Burnett RS, Clohisy JC, Barrack RL, Gage BF. Optimal initial dose adjustment of warfarin in orthopedic patients. Ann Pharmacother 2007; 41: Miao L, Yang J, Huang C, Shen Z. Contribution of age, body weight, and CYP2C9 and VKORC1 genotype to the anticoagulant response to warfarin: proposal for a new dosing regimen in Chinese patients. Eur J Clin Pharmacol 2007; 63: Millican EA, Lenzini PA, Milligan PE, Grosso L, Eby C, Deych E, Grice G, Clohisy JC, Barrack RL, Burnett RS, Voora D, Gatchel S, Tiemeier A, Gage BF. Genetic-based dosing in orthopedic patients beginning warfarin therapy. Blood 2007; 110: Sconce EA, Khan TI, Wynne HA, Avery P, Monkhouse L, King BP, Wood P, Kesteven P, Daly AK, Kamali F. The impact of CYP2C9 and VKORC1 genetic polymorphism and patient characteristics upon warfarin dose requirements: proposal for a new dosing regimen. Blood 2005; 106: Zhu Y, Shennan M, Reynolds KK, Johnson NA, Herrnberger MR, Valdes R Jr, Linder MW. Estimation of warfarin maintenance dose based on VKORC1 (-1639 G>A) and CYP2C9 genotypes. Clin Chem 2007; 53: Warfarin (Coumadin Ò ) prescribing information. Food and Drug Administration, August 2007; s105lblv2.pdf. Accessed 25 February Wadelius M, Chen LY, Lindh JD, Eriksson N, Ghori MJR, BumpsteadS,HolmL,McGinnisR,RaneA,DeloukasP.Thelargest prospective warfarin-treated cohort supports genetic forecasting. Blood 2009; 113: Kovacs MJ, Rodger M, Anderson DR, Morrow B, Kells G, Kovacs J, Boyle E, Wells PS. Comparison of 10-mg and 5-mg warfarin initiation nomograms together with low-molecular-weight heparin for outpatient treatment of acute venous thromboembolism. A randomized, double-blind, controlled trial. Ann Intern Med 2003; 138:

8 Warfarin dose prediction based on a standard nomogram Monkman K, Lazo-Langner A, Kovacs MJ. A 10 mg warfarin initiation nomogram is safe and effective in outpatients starting oral anticoagulant therapy for venous thromboembolism. Thromb Res 2009; DOI: /j.thromres The Columbus Investigators. Low-molecular-weight heparin in the treatment of patients with venous thromboembolism. NEnglJMed 1997; 337: Lee AY, Levine MN, Baker RI, Bowden C, Kakkar AK, Prins M, Rickles FR, Julian JA, Haley S, Kovacs MJ, Gent M. Low-molecularweight heparin versus a coumarin for the prevention of recurrent venous thromboembolism in patients with cancer. NEnglJMed2003; 349: Kovacs MJ, Kahn SR, Rodger M, Anderson DR, Andreou R, Mangel JE, Morrow B, Clement AM, Wells PS. A pilot study of central venous catheter survival in cancer patients using low-molecular-weight heparin (dalteparin) and warfarin without catheter removal for the treatment of upper extremity deep vein thrombosis (The Catheter Study). J Thromb Haemost 2007; 5: Fennerty A, Dolben J, Thomas P, Backhouse G, Bentley DP, Campbell IA, Routledge PA. Flexible induction dose regimen for warfarin and prediction of maintenance dose. Br Med J (Clin Res Ed) 1984; 288: Shine D, Patel J, Kumar J, Malik A, Jaeger J, Maida M, Ord L, Burrows G. A randomized trial of initial warfarin dosing based on simple clinical criteria. Thromb Haemost 2003; 89: Newcombe RG. Two-sided confidence intervals for the single proportion: comparison of seven methods. Stat Med 1998; 17: Henderson AR. The bootstrap: a technique for data-driven statistics. Using computer-intensive analyses to explore experimental data. Clin Chim Acta 2005; 359: Gage BF, Eby C, Johnson JA, Deych E, Rieder MJ, Ridker PM, Milligan PE, Grice G, Lenzini P, Rettie AE, Aquilante CL, Grosso L, Marsh S, Langaee T, Farnett LE, Voora D, Veenstra DL, Glynn RJ, Barrett A, McLeod HL. Use of pharmacogenetic and clinical factors to predict the therapeutic dose of warfarin. Clin Pharmacol Ther 2008; 84: Li C, Schwarz UI, Ritchie MD, Roden DM, Stein CM, Kurnik D. Relative contribution of CYP2C9 and VKORC1 genotypes and early INR response to the prediction of warfarin sensitivity during initiation of therapy. Blood 2009; 113: Anderson JL, Horne BD, Stevens SM, Grove AS, Barton S, Nicholas ZP, Kahn SF, May HT, Samuelson KM, Muhlestein JB, Carlquist JF. Randomized trial of genotype-guided versus standard warfarin dosing in patients initiating oral anticoagulation. Circulation 2007; 116: Caraco Y, Blotnick S, Muszkat M. CYP2C9 genotype-guided warfarin prescribing enhances the efficacy and safety of anticoagulation: a prospective randomized controlled study. Clin Pharmacol Ther 2007; 83: Routledge PA, Davies DM, Bell SM, Cavanagh JS, Rawlins MD. Predicting patientsõ warfarin requirements. Lancet 1977; 2: Sawyer WT, Poe TE, Canaday BR, Weiner JS, Williams DM, Webb CE, Ellison MJ. Multicenter evaluation of six methods for predicting warfarin maintenance-dose requirements from initial response. Clin Pharm 1985; 4: Sharma NK, Routledge PA, Rawlins MD, Davies DM. Predicting the dose of warfarin for therapeutic anticoagulation. Thromb Haemost 1982; 47: Siguret V, Gouin I, Debray M, Perret-Guillaume C, Boddaert J, Mahe I, Donval V, Seux ML, Romain-Pilotaz M, Gisselbrecht M, Verny M, Pautas E. Initiation of warfarin therapy in elderly medical inpatients: a safe and accurate regimen. Am J Med 2005; 118: Gurwitz JH, Avorn J, Ross-Degnan D, Choodnovskiy I, Ansell J. Aging and the anticoagulant response to warfarin therapy. Ann Intern Med 1992; 116: Wilke RA, Berg RL, Vidaillet HJ, Caldwell MD, Burmester JK, Hillman MA. Impact of age, CYP2C9 genotype and concomitant medication on the rate of rise for prothrombin time during the first 30 days of warfarin therapy. Clin Med Res 2005; 3: Eckman MH, Rosand J, Greenberg SM, Gage BF. Cost-effectiveness of using pharmacogenetic information in warfarin dosing for patients with nonvalvular atrial fibrillation. Ann Intern Med 2009; 150: