Evaluating the impact of treatment discontinuation on the outcome of clinical trials for weight management: A simulation study
Fenja Klima1,2,3, Charlotte Kloft2,3, Anders Strathe1, Selma El Messaoudi1
1Novo Nordisk A/S, 2Department of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, 3Graduate Research Training Program PharMetrX
Introduction. Obesity presents a major challenge to global public health with an estimated global prevalence of nearly 20% by 2030 [1]. Several treatment options for weight loss, such as incretin analogues, are currently investigated in numerous clinical trials. Even though current protocols for phase 3 trials in weight loss management allow more flexibility in dose adaptions, significant proportions of treatment discontinuation are still observed. This may lead to misinterpretations when evaluating treatment effect, especially when off-treatment data is disregarded. For that reason, the treatment policy estimand [2] (intention-to-treat principle) is frequently used for primary analysis of study endpoints, such as weight loss. Currently, strategies for its robust prediction, i.e. integrating treatment discontinuation, are lacking. As a mean to improve the prediction of outcomes for future clinical trials, this simulation study aimed to evaluate the impact of treatment discontinuation on weight loss under different hypothetical treatment arms. Methods. First, 1000 patients were simulated using a multinormal distribution based on published data from a phase 3 study [3]. To simulate treatment discontinuation, a hypothetical time-to-event (TTE) model was assumed, using a Weibull distribution as hazard function over a treatment period of 68 weeks. To evaluate the impact of treatment discontinuation, several TTE model parameters were tested: typical parameters, i.e., chosen to match commonly observed proportions of treatment discontinuation [3,4], decrease of the scale parameter by 25% and 50% to account for increased numbers of events, and no discontinuation. Treatment discontinuation was assumed to be independent of weight loss. Next, weight loss was simulated based on a previously published weight loss model including both a slow and immediate effect [5] for three hypothetical treatment arms: once-weekly administration of placebo, hypothetical compound A, and hypothetical compound B, assuming a 25% increase in slow maximum effect compared to compound A. For each TTE model parameter, the scenarios were compared in terms of mean percentage weight loss from baseline. To account for variability, 500 replicates of the dataset were simulated, and 95% prediction intervals (PI) were provided. Results. The hypothetical TTE model assumed a scale parameter of 2.25×10e-4 1/d and shape parameter of 0.8 to describe the hazard function. The simulated proportions of treatment discontinuation differed between the different parameters. The typical treatment discontinuation [95% PI] at week 68 was simulated to be 15.2% [13.3-17.4], while the proportion increased to 18.7%, [16.7-21.1] and 25.0% [22.8-27.7] for a 25% and 50% change in the scale parameter, respectively. Moreover, the simulated mean weight loss at week 68 was impacted by treatment arms. Assuming no discontinuation, the simulated mean weight loss [95% PI] was 3.18% [2.79-3.68], 16.1% and 18.9% [18.2-19.3] with placebo, compound A and compound B, respectively. Assuming treatment discontinuation, the percentage weight loss with compound A would drop to 14.6% [14.1-15.2], 14.3% [13.7-14.3] and 13.7% [13.0-14.2] with the typical parameters, 25% and 50% change, respectively. With compound B, the weight loss was predicted to be 17.1% [16.4-17.7], 16.6% [16.0-17.2] and 15.9% [15.3-16.5] for the typical parameters, 25% and 50% change, respectively, suggesting that treatment policy estimands of compounds with higher efficacy were stronger affected by treatment discontinuations. Conclusion: A simulation framework for assessing the impact of treatment discontinuation on weight loss was developed, taking into account different hypothetical treatment arms. It showcased the importance of considering treatment discontinuations for accurate clinical trial simulations to improve the prediction of treatment policy estimand, and ultimately efficient clinical trial designs. Future TTE analysis with covariate inclusion could provide a strong framework for clinical trial simulations.
[1] World Obesity Federation. World Obesity Atlas 2022 [https://www.worldobesity.org/resources/resource-library/world-obesity-atlas-2022, accessed 05 March 2025] [2] Akacha et al. CPT Pharmacometrics Syst. Pharmacol. 2021 [3] Wilding et al. N. Engl. J. Med. 2021 [4] Davies et al. Lancet. 2021 [5] Strathe et al. Diabetes Obes. Metab. 2023