The reason for the inclusion of two Torin 1 mTOR inhibitor similar dosing equations was to assess whether the model developed by Sconce et al. derived at the University of Newcastle, United Kingdom had an advantage over models derived in other countries in explaining variability in the Liverpool prospective validation cohort. The strength of these two models in particular was that they contain a small number of covariates yet explain a large amount of variability in their respective derivation datasets. Several warfarin maintenance dosing algorithms have been published, many including pharmacogenetic information, in the form of linear regression models. Despite the number of published dose prediction regression models, they have rarely been integrated into standard clinical practice. This in part, is due to the fact that most of these algorithms have not been externally validated in an independent dataset and if they have, then replication has been poor. In this study, we have compared six different MD prediction linear regression models using two independent cohort of patients to test their predictive ability outside their original derivation cohorts. This was done by re-fitting each model in turn to two independent cohorts; a prospective patient dataset recruited in Liverpool, United Kingdom and the control arm from the EU-PACT trial. Unsurprisingly, the performance of all six models was worse in the validation cohort as compared to the derivation cohort. The diminished performance could be explained by several factors. The two validation cohorts demographics shown in Table 2 reveal two points of note, firstly, the LP patients have a lower mean therapeutic dose than those in EU-PACT. This may be due to a difference in clinical practise between the two cohorts, potentially in the trade-off decision between maximising efficacy and minimsing adverse events. Secondly, the VKORC1 wild-type genotype is the most common in LP patients yet in EUP patients the homozygous dominant is more prevalent. Again, this difference could be due to different clinical practise provdided, particularly EUP patients without a VKORC1 variant allele reaching a stable maintenance dose with greater ease. Examining the difference between our validation cohorts and the derivation cohorts for the six dosing algorithms investigated in this manuscript, there were differences that could contribute to differences in algorithm perfromance in validation. There is a range in the size of derivation cohorts ranging from Zhu et al.’s derivation cohort of 56 to Wadelius et al.’s derivation cohort of 850. The expectation would be for algorithms derived utilising smaller derivation cohorts to be able to explain less variability in a larger or more diverse cohort. This was not particularly evident in this study as Zhu et al.’s algorithm performed more strongly than Wadelius et al.’s algorithm in the larger LP validation cohort. In the EUP cohort the two algorithm’s performances were relatively indistinguishable. Differences in derivation cohort covariates can also contribute to altered algorithm performance in validation cohorts.