Abstract
Study Designs for the Evaluation of Biomarkers in Ovarian Cancer: A Systematic Review
Maria Olsen,1 Mona Ghannad,1,2 Patrick M. Bossuyt1
Objective
We documented the study design of recently reported evaluations of biomarkers in ovarian cancer to identify potential deficiencies.
Design
We performed a systematic search in PubMed (MEDLINE) for reports of studies published in 2015 evaluating biomarkers in ovarian cancer using a combination of “ovarian cancer,” “biomarker,” and clinical performance measures and outcomes (eg, “survival,” “prognosis,” “prediction,” and “AUC”). Eligible were reports of clinical studies that had included adult patients; evaluated samples or data from women diagnosed, screened, treated, or monitored for ovarian cancer; and reported a clinical performance measure with no restrictions to study type (eg, diagnostic, predictive, and prognostic). We used the 1998 National Institutes of Health definition of a biomarker. Screening of titles and abstracts (level 1) and full text (level 2) was done in duplicate by 2 independent reviewers. Disagreements were solved through discussion, and a third reviewer was consulted if consensus was not reached. Data extraction from included study reports was done by 1 researcher and validated by a second. Using a structured data extraction form, we identified study design features and classified designs such as the use of single or multiple study groups, single or multicenter study, sample size, and characteristics of sample acquisition.
Results
Our search resulted in 1026 studies: 516 (50%) and 345 (34%) were included from level 1 and 2, respectively, of which we evaluated the first 200 (starting January 1, 2015). We observed a wide diversity of study designs. In our preliminary results of 70 manuscripts, 36 (51%) were single group studies, whereas 27 (38%) included multiple groups (the remaining 8 [11%] being unclear); studies often used healthy controls (19 [27%]); and 8 (11%) used extreme phenotypes. Fifty-nine (84%) had used pre-existing samples, while only 7 (10%) relied on dedicated acquisition of specimens and data. The sample size was limited in many studies.
Conclusions
Our findings show a variability in study designs for the evaluation of biomarkers in ovarian cancer and confirm the presence of suboptimal elements in recent evaluations of the clinical performance of biomarkers, such as the inclusion of healthy control individuals, as well as limitations in the generalizability, with most studies being single-center. These limitations may contribute to failures in the translational phase of biomarker development, observed by other authors.
1Academic Medical Center Amsterdam, Dept. of Clinical Epidemiology, Biostastistics and Bioinformatics, Amsterdam Public Health Research Institute, Amsterdam, the Netherlands, m.olsen@amc.nl; 2Centre de Recherche Épidémiologie et Statistique Sorbonne Paris Cité, Université Paris Descartes, Centre d”épidémiologie Clinique, Hôpital Hôtel-Dieu, Paris, France
Conflict of Interest Disclosures:
None reported.
Funding/Support:
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 676207.
Role of Funder/Sponsor:
The funder played no role in this work.