Other models include details of initial INR measurements, loading doses or drugs known to alter the effect of warfarin. Models comprising only demographic, loading dose or co-medication details use information that is readily available to the clinician and a few recent studies have derived such algorithms. More recently, dosing algorithms have also included genetic factors, specifically variants in the VKORC1 and CYP2C9 genes which have been shown to be associated with dosing requirements. The benefit of including genetic information in dose prediction still remains unproven, although the science is conclusive that a patient’s genetics alter their warfarin dose requirements. With a view to assessing how well previously published models predicted MD in a dataset outside the derivation dataset, we tested their predictive ability in two independent patient cohorts. This also allowed the performance of the models to be compared against each other. Further, it allowed us to evaluate how suitable the method of linear regression, the most commonly utilized method for deriving warfarin maintenance dosing algorithms, may be for patients at the maintenance dosing phase of warfarin therapy. The model presented by Le Gal et al. includes only clinical covariates, including INR measurements on day 5 and day 8 and the total dose of warfarin taken during the first week. The second model, proposed by Solomon et al. includes information on total loading dose, INR at the end of the loading phase, age and the use of the co-medication amiodarone, a well-known inhibitor of warfarin RO5185426 cost metabolism. These two models did not include any information on genotypes and, as a consequence, may have an advantage in that they are based on data readily available to the clinician, so can be used without having to attain a patients genotype information. Four of the included models utilise genotypes for variants in CYP2C9 and VKORC1. The models proposed by Anderson et al. and Wadelius et al. assume that CYP2C9 alleles are non-proportional, thus including a separate covariate for each possible genotype, whereas the models proposed by Sconce and Zhu assume an additive effect of the variant allele. The models proposed by Anderson et al. and Wadelius et al. calculated a total weekly dose of warfarin; consequently clinicians would have to divide the recommended weekly dose into seven daily doses as they consider appropriate. Anderson et al.’s model also included demographic, genotype, and comedication covariates. The model was applied in the randomized control trial and information on the models R-squared in the derivation cohort was not supplied. The model from Wadelius et al., contained the largest number of covariates incorporating demographic and co-medication information alongside genotype covariates. The model proposed by Sconce et al. included less covariates than most of the other pharmacogenetic models, with demographic information only on age and height being included along with information on the genotypes. Similar in composition, but including weight instead of height.