To the editor:
Li et al reported recently in Blood that much of the information provided by CYP2C9 and VKORC1 genotypes during warfarin initiation therapy in outpatients is captured by early international normalized ratio (INR) responses.1 This confirms and extends previous observations in hospitalized, heavily medicated patients, that dose-adjusted INR (INR/dose) at day 4 is the most important predictor of warfarin dose at day 14.2 However, the predictive power of the best regression models in both studies, expressed by the correlation coefficient (r2) values of 0.401 and 0.51,2 between predicted and observed doses, is not superior to that of several pharmacogenetic dosing algorithms without an INR covariate, including the algorithm that we developed for Brazilian patients.3-10 We now report that adding INR/dose as a variable in multistep regression modeling of the stable warfarin dose in the Brazilian cohort leads to a novel algorithm with greater predictive power. Details of the original study design and linear multiple regression modeling of the stable warfarin dose have been published.3
For the present model development, we used the first available INR/dose measurement from 260 patients chosen randomly from the 390 patients in our cohort; the remaining 130 patients were used for model validation. The most informative regression model retained the same covariates previously identified as associated with stable warfarin weekly dose in this cohort (age, weight, treatment indication, comedication with amiodarone or simvastatin, VKORC1 and CYP2C9 genotypes) but included also an INR/dose term (Table 1). The r2 for the correlation between observed and model-predicted warfarin weekly dose in the development and the validation sets, was 0.60 and 0.59, respectively (Table 2), compared with 0.51 for our previous algorithm, which did not include an INR term.3 The mean absolute difference between model-predicted and observed doses, 6.5 and 6.2 mg/week in the development and validations sets, respectively (Table 2), did not differ from 6.9 mg/week for our previous algorithm.3
Variable . | Partial regression coefficient . | P . | Partial R2 statistic, % . |
---|---|---|---|
Age, y | −0.0054 | 1.10−1 | 0.3 |
INR/dose (first available) | −2.9909 | 3.10−8 | 4.9 |
Weight, kg | 0.0157 | 3.10−5 | 2.7 |
Therapeutic indication | 3.8 | ||
Heart valve prosthesis | 0.5726 | 1.10−6 | |
Thromboembolic disease | 0.4333 | 3.10−2 | |
Simvastatin | −0.4442 | 10−2 | 1.0 |
Amiodarone | −0.7748 | 10−9 | 5.9 |
CYP2C9 *2/*3/*5/*11 | 5.7 | ||
One variant allele | −0.5115 | 2.10−6 | |
Two variant alleles | −1.1043 | 10−5 | |
VKORC1 3673G>A | 20.1 | ||
3673GA | −0.8352 | 5.10−7 | |
3673AA | −1.6841 | <10−12 |
Variable . | Partial regression coefficient . | P . | Partial R2 statistic, % . |
---|---|---|---|
Age, y | −0.0054 | 1.10−1 | 0.3 |
INR/dose (first available) | −2.9909 | 3.10−8 | 4.9 |
Weight, kg | 0.0157 | 3.10−5 | 2.7 |
Therapeutic indication | 3.8 | ||
Heart valve prosthesis | 0.5726 | 1.10−6 | |
Thromboembolic disease | 0.4333 | 3.10−2 | |
Simvastatin | −0.4442 | 10−2 | 1.0 |
Amiodarone | −0.7748 | 10−9 | 5.9 |
CYP2C9 *2/*3/*5/*11 | 5.7 | ||
One variant allele | −0.5115 | 2.10−6 | |
Two variant alleles | −1.1043 | 10−5 | |
VKORC1 3673G>A | 20.1 | ||
3673GA | −0.8352 | 5.10−7 | |
3673AA | −1.6841 | <10−12 |
The partial R2 statistics measures the degree of association between 2 random variables, with the effect of a set of controlling random variables removed.
Sample set . | Correlation coefficient, r2* . | Mean absolute difference, mg/wk . |
---|---|---|
Development set (n = 260) | 0.60 | 6.5 |
Validation set (n = 130) | 0.59 | 6.2 |
Sample set . | Correlation coefficient, r2* . | Mean absolute difference, mg/wk . |
---|---|---|
Development set (n = 260) | 0.60 | 6.5 |
Validation set (n = 130) | 0.59 | 6.2 |
Based on the following regression equation: Square root of warfarin weekly dose (mg/week) = 5.5691 − 0.0054 × (age in years) −2.9909 × (INR/ dose, mg/week) + 0.0157 × (weight in kg) + 0.5726 × 1 (if patient has heart valve prosthesis) or 0 (if no heart valve prosthesis) + 0.4333 × 1 (if patient has thromboembolic disease) or 0 (if no thromboembolic disease) −0.4442 × 1 (if prescribed simvastatin) or 0 (if no prescribed simvastatin) − 0.7748 × 1 (if prescribed amiodarone) or 0 (if not prescribed amiodarone) − 0.5115 × 1 (if patient has one CYP2C9 variant allele) or 0 (if not) − 1.1043 × 1 (if patient has 2 CYP2C9 variant alleles) or 0 (if not) − 0.8352 × 1 (if VKORC1 3673GA genotype) or 0 (if not) − 1.6841 × 1 (if VKORC1 3673AA genotype) or 0 (if not).
Correlation coefficient (r2) between weekly warfarin dose predicted by the dosing algorithm (predicted dose) and the dose actually taken by the patient (observed dose).
VKORC1 genotype remained the most important predictor of warfarin weekly dose in the novel algorithm (Table 1). This contrasts with the predominant contribution of INR-associated terms, and the relatively small contribution of VKORC1 and CYP2C9 genotypes in Li et al1 and Michaud et al2 Differences in population cohorts, clinical settings (eg, expertise in INR-guided warfarin dose titration),1 assessed outcomes and time of INR/dose measurements might account for this discrepancy. A distinct feature of our algorithm is that the individual INR/dose term does not represent a fixed time point after starting warfarin therapy, but rather the first measurement taken after admission of the patients in the anticoagulant unit. This feature is potentially useful for patients under continuous warfarin treatment, who had not reached stable dosing despite repeated dose adjustments.
In summary, we confirmed that inclusion of an INR-related term increased (from 0.51 to 0.60) the predictive power of warfarin-dosing pharmacogenetic algorithms for Brazilian outpatients under chronic warfarin therapy. However, INR measurements did not entirely capture the information provided by CYP2C9 and, especially VKORC1 genotypes, the latter remaining the most informative predictor of stable warfarin dose requirements in our cohort.
Authorship
Acknowledgments: This work was supported in part by grants from Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq; Brasília, Brazil), Fundação de Amparo à Pesquisa do Estado do Rio de Janeiro (Faperj; Rio de Janeiro, Brazil), and Financiadora de Estudos e Projetos (Finep; Rio de Janeiro, Brazil).
Contribution: G.S.-K. designed the research, analyzed data, wrote the manuscript, and obtained funding; J.A.P. contributed to data collection and analysis, and edited the manuscript; E.A.-S. recruited and followed the patients, and contributed to data collection; and C.J.S. contributed to data analysis and edited the manuscript.
Conflict-of-interest disclosure: The authors declare no competing financial interests.
Correspondence: Guilherme Suarez-Kurtz, MD, PhD, Divisão de Farmacologia, Instituto Nacional de Câncer, Rua André Cavalcanti 37, Rio de Janeiro 21230-050, Brazil; e-mail: [email protected].
This feature is available to Subscribers Only
Sign In or Create an Account Close Modal