We read with interest the paper by Liesenfeld et al.  on the population pharmacokinetic analysis of the oral thrombin inhibitor dabigatran etexilate from patients in the RE-LY trial  with non-valvular atrial fibrillation. The paper demonstrates how the method of population pharmacokinetic analysis and simulation can be applied to answer some common what-if questions raised in clinical practise, and we look forward to reading similar population analysis work from the other newer anticoagulant agents, when available.
In this paper however, there were a couple of points regarding the analysis conducted that intrigued us, which we feel require further clarification and commentary from the study authors.
Firstly, we note that weight was explored as a possible covariate for dabigatran volume of distribution, but was not explored as a possible covariate for dabigatran clearance. This surprises us; 99% of the metabolic processes within the body, including clearance, takes place within lean tissues , and although obese individuals have a lower lean body weight (LBW)/weight ratio overall , their total LBW is greater than that of normally weighted individuals. One would therefore predict a relationship between weight and dabigatran clearance, in line with current thinking [5,6]. Han et al.  have recently stipulated that (i) absolute clearance is greater in obese individuals, (ii) clearance increases non-linearly with weight, and (iii) clearance correlates linearly with mechanistically derived  lean body weight. Given the weight of patients in the RE-LY study ranged from 32.7 to 222.3 kg, it seems to us that it would have been important to explore the relationship between weight (or lean body weight) and dabigatran clearance, particularly as it is not uncommon nowadays to have patients who are at the extremes of weight, and the question of the appropriateness of the standard dose in these patients is always raised. We therefore feel that until this relationship is explored, it is too early for the authors to draw the conclusion that no dabigatran dose adjustment is needed in the very high- or low-body-weight patients.
Secondly, we note that the base model developed in this study was derived from a prior population analysis, which has yet to be published. This approach also surprised us, given the rich (27 706 dabigatran plasma concentrations) and real-world data the authors had at their disposal from the RE-LY study. In our opinion, developing a base model from their own data and then comparing this base model with those of others seems a more robust approach. We acknowledge that we have not seen the data and model developed by Dansirikul et al. , and it may well be that utilizing this base model for their study is indeed appropriate; however, developing a RE-LY base model first would eliminate the inheritance of any bias or limitations, which may have been present from the base model that was borrowed from, particularly if the study population is not well matched to those patients in the RE-LY study.
We note that the study authors of this other piece of work are the same as some of the authors of this study, and we would therefore urge them to explore the relationship between weight and dabigatran clearance, if they have not done so already in their other study.
In addition, closer examination of the goodness of fit plots (Figure 1 in the paper) demonstrates that there is clear bias in the final model proposed, with the plots clearly showing the model under-predicting dabigatran concentrations. A well-developed model would have the observed concentrations being mirrored on the predicted concentrations side of the line of unity; this is not the case here, which certainly raises questions about how good the model being proposed is. The aforementioned reasons might go some way in explaining why this is not the case in this study.
From a clinical perspective, it was interesting to observe the simulation work of this study, as it illustrates the utility of modeling and simulation and how you can explore important what-if questions raised from clinical practise. This utility does, however, become redundant if the model itself is not robust, which we feel is the case here. One useful point the modeling analysis does highlight, is the important relationship between creatinine clearance and dabigatran clearance. The study found an 11% increase in systemic dabigatran exposure when creatinine clearance dropped from 100 to 80 mL min−1, whereas a further 20 mL min−1 decrease from 50 to 30 mL min−1 increased dabigatran exposure by 50%. Accepting all the limitations of the model developed, this finding does re-enforce the message of the importance of renal function for dabigatran clearance and will clearly be an important consideration for dabigatran’s use for stroke prevention in atrial fibrillation, where we know renal function declines with advancing age . Indeed, in the UK, the manufacturers of dabigatran have recently issued a Dear healthcare professional letter, advising of the importance of renal function monitoring in patients being considered for or already on dabigatran therapy .
In our opinion, the final model proposed by the study authors is not satisfactory. Whilst we welcome the opportunity to read the population analysis of the RE-LY study, we feel the modeling analysis conducted in this study requires significant revision and refinement before the authors can draw the conclusions that they do.