Genetic, clinical and behavioural determinants of vitamin K-antagonist dose--explored through multivariable modelling and visualization.

Vitamin K antagonists (VKA) are highly effective anticoagulants but their use is hampered by multiple interactions with food and medicine and a narrow therapeutic range. The large variation in dose requirements has led to the development of several dosing algorithms based on pharmacogenetic and clinical variables. In contrast, evidence about the influence of behavioural (i.e. diet and exercise) and socio-psychological factors is sparse. To investigate the impact of pharmacogenetic, clinical, behavioural and socio-psychological factors on maintenance dose of VKA. In a cross-sectional study, we interviewed 250 consecutive patients from an anticoagulant clinic and subsequently measured pharmacogenetic and anthropometric variables. Statistical analyses were carried out using linear regression and multivariable models with visualization features. In both types of analyses, the strongest determinants of VKA dose were polymorphisms in the VKORC1 and CYP2C9 genes and age. Half of the variation in VKA dose could be explained by a linear regression model including four variables, while a multivariable model with 20 pharmacogenetic and clinical variables explained 60%. A multivariable model including 94 predictor variables was not notably better regarding predictive performance, but visualization of this model offered information about the correlation structure between predictor variables. The strongest determinants of VKA dose are well-known pharmacogenetic variables and age. The variables describing health-related behaviour and socio-psychological factors are strongly inter-correlated and not useful in dosing algorithms.