J. Nijhuis, M. Centanni
Department of Pharmacy, Uppsala University, Uppsala, Sweden.
Objectives: Governmental organisations are continuously making difficult decisions regarding regulatory approval, balancing drug efficacy against toxicity and costs. Pharmacoeconomic (PE) models are employed to inform the cost-effectiveness analysis (CEA) of a drug. Differences in such models can give rise to differences in outcome and therefore influence regulatory decision making [1]. An example of such discrepancy can be found for sunitinib in gastrointestinal stromal tumours (GIST), where CEA based on incremental cost-effectiveness ratios (ICER) was estimated to range from 130.248 to 204.033 euro/quality-adjusted life years (QALYs) gained. Besides discrepancies in national healthcare costs, such differences in outcomes may be due to model structure, ranging from time-to-event (TTE) models (exponential, Weibull [2,3]) to Markov models (discrete, continuous [4,5]), with simple proportional odds (PO) models for the toxicities. To demonstrate potential issues related to these structures we re-estimated the given sunitinib PE models, using simulated data from a previously developed pharmacometric framework [6,7] to evaluate differences in predicted toxicity, survival and ICER between the models.
Methods: The project was performed in four steps: (I) The existing pharmacometric framework [6] was used to simulate a virtual population (n=1000) with corresponding outcomes, including the toxicity, tumour progression and overall survival (OS), over a time span of 102 weeks. Patients were simulated twice: once with sunitinib (daily, 37.5 mg) and once without (placebo). (II) PE model parameters were estimated in NONMEM and the final models were used to simulate the occurrence of progression-free survival (PFS), OS and toxicities for each patient (n=1000) under both treatment and placebo. (III) From the results, the total costs and QALYs were calculated, and ICER was determined by comparing the treatment and placebo group as previously described [7]. (IV) The predicted outcomes of the PE models were compared to the pharmacometric framework.
Results: All PE models, including the toxicity PO models, were successfully estimated and corresponding simulations were performed. (1) Regarding the toxicity, the PO models predicted a stable incidence in occurrence and were not able to capture the time-varying aspects of adverse events, such as hand-foot syndrome (HFS), which in the pharmacometric framework simulations demonstrates a delay in appearance with subsequent decrease due to dose reductions (7.3% vs 0% – baseline, 7.5% vs 11.8% – 18 weeks, respectively). (2) The OS (treatment group, 102 weeks) simulated by the discrete Markov and exponential TTE model was found to be lower than the OS simulated by the Weibull TTE, the continuous Markov model and the pharmacometric framework. (3) The calculated ICER from each PE model ranged from 1.6 to 2.5 times the ICER simulated by the framework. Where the highest ICER was predicted by the exponential TTE, whereas the discrete Markov model was the closest to the ICER of the framework (130.248 euro per QALY gained). In all PE model simulations, the relatively largest fraction of the costs were attributable to drug costs.
Conclusions: (1) PO models assume a stable frequency over time and are therefore not able to represent the potential time-varying changes in toxicities. This could have consequences when frequent toxicities would normally lead to dose reductions, or when drug costs are extrapolated based on a single cycle. (2) OS was underestimated for the exponential and discrete Markov model, most likely due to rigidity in the model structure. (3) The final ICER predictions demonstrated a positive bias and large deviation from the framework value, indicating that the PE model structures have a great influence on sunitinib CEA. The variability in the outcomes highlights the need for PE models that take mechanistic considerations into account and to not simply assume one structural model to describe the data. This is particularly relevant for the field of oncology, where outcomes can differ due to the frequent occurrence of toxicities and the minimal gain in life-years. The variation in the results highlights the need for a mechanistic approach to perform a CEA of drugs to aid governmental bodies in their decision making.
Acknowledgement: This work was supported by the European University Consortium for Pharmaceutical Sciences.
References:
[1] Frederix et al. Pharmacoeconomics (2014);32(1):47-61.
[2] Chabot et al. Eur J Cancer (2008);44(7):972-7.
[3] Bond et al. Health Technol. Assess (2009);13 Suppl 2:69-74.
[4] Paz-Ares et al. J Clin Pharm Ther (2010);35(4):429-38.
[5] Contreras-Hernández et al. (2008);98(11):1762-8.
[6] Centanni et al. Clin Cancer Res (2020);26(17):4590-4598.
[7] Centanni et al. Front Pharmacol (2020);11:316.
Reference: PAGE 30 (2022) Abstr 10166 [www.page-meeting.org/?abstract=10166]
Poster: Drug/Disease Modelling - Oncology