gms | German Medical Science

Biomarker data plays an important role in the stratification of patients in precision medicine. Although there is a vast variety of candidates, finding suitable profiles is still challenging due to the lack of reproducibility and statistical power [1]. In addition, biomarker data is frequently left-censored, and it is yet unclear how best to handle data where a non-negligible proportion has values under a given detection limit [2]. In fact, many approaches have been suggested in the literature that differ with regard to assumptions and application settings. Also, they have been investigated in different settings as defined, for instance by sample sizes, percentage of non-detects, and the underlying distribution of the data, making the practical choice difficult. In this work we therefore target this issue by summarizing published theoretical and simulation studies in which methods for the analysis of left-censored data are suggested and compared with each other. We performed a systematic review following the PRISMA statement [3] to derive an overview of the existing methods and their applicability in different settings. The identified methods were then compared in an extensive simulation study which was based on left censored candidate biomarker data from the ALLIANCE cohort of the German Center for Lung Research [4]. The distributions of the datasets were all right skewed, and the proportion of censored values ranged from 0% to 90%. The results will help to guide researchers to select the most suitable method for their specific application.

The authors declare that they have no competing interests.

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