gms | German Medical Science

The choice of endpoints is important for clinical trials. Due to the development of more and more effective drugs patients live longer, which often results in longer follow-up times until the clinical endpoint can be measured. In order to shorten studies as well as reduce costs and sample sizes we often look out for surrogate endpoints because the results from the surrogate are often more readily available. Expensive measurements and invasive methods to derive the clinical endpoint are also a reason to search for surrogate endpoints. The validation of a surrogate endpoint is a very long and complex procedure involving several steps.

We will focus on the two statistical levels of evidence for surrogate endpoint validity. Individual-level surrogacy describes the association between the surrogate endpoint and the true endpoint for an individual patient. Trial-level surrogacy describes the association between the treatment effects on the surrogate endpoint and the true endpoint across several trials. The aim of my thesis is to better understand the model proposed by Buyse et al. [1] for continuous normally distributed endpoints and the measures for individual- and trial-level surrogacy. The two stage approach from Buyse et al. [1] was investigated in a simulation study using the R-package “Surrogate”. My focus was on analysing the association between the true correlation of the surrogate endpoint and the true endpoint on individual as well as on trial level with the respective measures. We varied the true correlations, the number of trials and the number of patients per trial resulting in several scenarios. For each scenario we estimated the individual- and trial-level surrogacy values and compared them to the respective true values. We also investigated the influence of the missing data patterns MCAR (missing completely at random), MAR (missing at random) and MNAR (missing not at random) on the estimated values for individual- and trial-level surrogacy.

The results of the simulation studies showed that the measure for individual-level surrogacy proposed by Buyse et al. [1] is an unbiased estimator for the correlation on individual level. On trial-level, however, our results indicated that the estimator highly depends the number of trials, the number of patients per trial as well as the correlation between the surrogate endpoint and the true endpoint on individual level. The dependency made it difficult to interpret the results. This especially occurred for small numbers of trials and small numbers of patients per trials. We also compared the complex calculation of the measurements for trial-level surrogacy proposed by Buyse et al. [1] with a simple estimation of the correlation of the mean values of the trials.

The investigation of the missing values showed different results depending on the missing pattern. As expected, MCAR values in a dataset did not have a big influence on the calculations for surrogacy. However, MAR values and MCAR values led to biased estimates. In addition, the results also depended on how the missing data was handled.

The authors declare that they have no competing interests.

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