gms | German Medical Science

Introduction: Regulatory authorities have recognized the importance of Central Statistical Monitoring (CSM) in clinical trials as it enhances trial quality, efficiency and validity. Ongoing research has led to the development of several statistical methods to achieve a rigorous CSM approach. In a previous study, we implemented an approach based on the comparison of the Grand Mean (GM) to individual centers. This approach was validated on several statistical models covering continuous, binary, ordinal as well as count data. In the current research, we investigate whether this approach can be extended to cover time-to-event endpoints. Established methods for survival analysis, such as Weibull and Cox proportional hazards models, are investigated whether they control the type I error when performing multiple comparisons to the GM. We set the probability of false rejection of the null hypothesis, i.e., the familywise error rate (FWER), to 5%. Real-World-Data (RWD) from the German Multiple Sclerosis Registry (GMSR) was used to confirm the model’s pertinence in practice.

Methods: A plethora of statistical models dealing with time-to-event data exist. In this research, the Weibull model (WE) and the Cox proportional hazard model (PH) are examined under data generated by the Weibull distribution. In a Monte Carlo simulation study, we investigate whether both models can control the type I error under different parameter settings when performing comparisons to the GM. Parameter settings include various scenarios for the shape γ and scale λ parameters of the Weibull distribution. Simulations cover balanced and unbalanced designs to mimic scenarios found in clinical trials. The performance of both models was tested on 1000 simulation runs. Balanced and unbalanced scenarios include the same number of centers i=10. The number of subjects n_i per center ranges between 20, 50, 100, 150, 500 and 1000. For the unbalanced scenario, only a single center had half of n_i compared to the other nine centers. Simulations cover right censored scenarios in which the maximum exposure time for an event was 3 years for balanced and unbalanced designs.

Results: The control of FWER for both models is similar for balanced and unbalanced scenarios. Irrespective of γ and λ parameters, both models achieved a reasonable control of type I error as soon as n_i>50. Both models were more liberal for small n_i<50 specifically for higher values of γ and λ parameters. As soon as γ ≥ 1.5 with λ ≥ 1, higher violations in controlling FWER could be observed. However, for almost all scenarios the Weibull model seemed to be inferior to the Cox PH model. The GM multiple comparison was implemented on time to initiate an immunomodulatory multiple sclerosis treatment between centers within the GMSR. The comparison identified the centers deviating from the average time to initiate treatment and pinpointed whether these centers tend to initiate treatment early or whether they have a delay period.

Conclusion: Both models can be utilized for GM comparison. However, for datasets with small sample sizes, alternative methods need to be investigated. In addition, both models need to be studied further for other types of distributions to assess their robustness when assumptions are violated.

Competing interests: Firas Fneish had no personal financial interests to disclose other than being an employee of the German MS Registry. David Ellenberger had no personal financial interests to disclose other than being an employee of the German MS Registry. Niklas Frahm is an employee of the German MS Registry. Moreover, he received travel funds for research meetings from Novartis. Alexander Stahmann has no personal financial interests to disclose, other than being the leader of the German MS Registry, which receives (project) funding from a range of public and corporate sponsors, recently including G-BA, The German MS Trust, German MS Society, Biogen, Celgene (Bristol Myers Squibb), Merck, Novartis, Roche, Sanofi and Viatris. Gerhard Fortwengel has nothing to disclose. Frank Schaarschmidt has nothing to disclose.

The authors declare that an ethics committee vote is not required.