A repeated time-to-event model is superior to a count regression analysis in linking FVIII activity to bleeding events in prophylactically treated patients with severe haemophilia A
Garrit Jentsch1, Alexander Solms2, Joachim Grevel1,3, Anita Shah4, Dirk Garmann5
1BAST GmbH Heidelberg Germany; 2BAYER AG Berlin Germany; 3 BAST Inc Ltd Loughborough UK; 4 BAYER AG Whippany NJ USA; 5BAYER AG Wuppertal Germany
Objectives: The objective of this analysis was to investigate whether a PK/PD relationship could be established between the activity-time profile of factor VIII (FVIII) and the occurrence of (repeated) bleeding events in patients prophylactically treated with BAY 94-9027. Two modelling approaches were compared for this task. In particular, it was evaluated whether a repeated time to event (RTTE) analysis and a standard regression analysis relating the average time without optimal FVIII protection (independent variable) to the annual number of bleeds (dependent variable), would both identify a significant exposure response (ER) relationship. Additionally, a covariate analysis was conducted to evaluate whether sources of inter-individual variability (IIV) in the bleeding pattern could be identified.
Methods: Data from two phase 2/3 clinical trials were analyzed, in which patients with severe hemophilia A were prophylactically treated with BAY 94-9027, a PEGylated recombinant FVIII product currently under clinical development. Individual FVIII activity-time profiles were predicted using the individual post-hoc estimates from a previously developed population PK model. For the standard ER analysis, these profiles were used to calculate the average weekly times that the patients spent below a certain FVIII threshold level, which varied between 1 and 20 IU/dL. These times without optimal protection were used as independent variable in a count regression analysis [1]. Moreover, the individual activity time profiles were employed in the second part of the analysis in which an RTTE model using non-linear mixed-effects modelling [2] was developed. The RTTE model was qualified using visual predictive checks (VPCs), which compared the observation-based Kaplan-Meier curves of the first six bleeding events to the respective model-predicted survival functions. Moreover, covariate effects were visualized by generating VPCs stratified according to the identified covariate relationships.
Results: In the standard ER approach, a statistically significant relationship (p<0.01) could not be identified between the time spent without optimal FVIII protection and the annual number of bleeds. In the second part of the analysis, an RTTE model with log-normally distributed bleeding intervals was selected as a base model. An Emax model was used to describe the effect of the FVIII activity on the hazard. Accounting for the individual FVIII activity-time profiles in the hazard function was highly significant (p<0.0001). Furthermore, the number of reported bleeds in the past year prior to study begin (NSB) was identified as a significant covariate on the parameter mu of the log-normal hazard function. A power model with an exponent of -0.0169 was found to describe this relationship such that an increase in NSB leads to a decrease in the length of the bleeding intervals.
Conclusions: A standard regression approach, which neglects both the times of the bleeding events and the dynamics of the FVIII kinetics, failed to establish a significant ER relationship, whereas an RTTE model successfully established a relationship between the individual FVIII activity-time profiles and the occurrence of bleeding events. The medical history of the patients was identified as a significant source of IIV. Moreover, the FVIII kinetics of BAY 94-9027 has a quantitatively similar effect on the occurrence of bleeding events as the non-PEGylated FVIII BAY 81-8973 [3]. The estimated EC50 values of the two compounds were almost identical (8.15 and 9.14 IU/dL for BAY 94-9027 and 81-8973 respectively), which indicated that the effectivity of the compounds did not change and that any differences between the two compounds are PK driven.
Moreover, it is tempting to speculate that the standard regression approach, which neglects both the exact bleeding times and the dynamics of the FVIII kinetics, failed to establish a significant ER relationship because no placebo group was included in the analysis and the studied doses were selected such that all patients benefited from treatment.
In the field of prophylactic treatment of hemophilia A patients with recombinant FVIII, this work demonstrates the superiority of an RTTE approach compared to a classical ER endpoint approach. In addition, the RTTE model can be simulated in order to evaluate clinical relevant scenarios such as the effects of different dosing schedules.
Conflict of Interest: ASo, ASh, and DG are employees of Bayer Pharma AG. GJ and JG were payed consultants for Bayer AG.
References:
[1] Collins PW, et al. J Thromb Haemost. 2009;7[3]:413-420
[2] Holford N. CPT: Pharmacometrics & Systems Pharmacology. 2013;2(5):e43-. doi:10.1038/psp.2013.18
[3] Garmann, D., et al. Blood. 2015; 126:1095