Rapport   2015

Holden, Marit; Günther, Clara-Cecilie; Holden, Lars



Norsk Regnesentral



Antall sider: 36


FULLTEKST: http://publications.nr.no/1443529126/verificationOfTest-Holden.pdf

BLOBREC is a test for distinguishing breast‐cancer patients from population‐based controls described by
Dumeaux et al. We have performed a quality control of the methods and procedures used for developing this
test. Besides reproducing results obtained using exactly the same datasets as Dumeaux et al., we examined
how sensitive the test results are to the approach selected when preprocessing the data, and whether the test
results are influenced by drug use, smoking, or stress due to a potential cancer diagnosis. We also examined if
the test results for the breast‐cancer patients depend on whether the patient participated in the screening
program. A dataset intended for examining the effect of stress was used as a validation dataset for the test.
Our analyses confirm the results obtained by Dumeaux et al. We obtain comparable results when using
different approaches for preprocessing the data. Prediction performance for the datasets used when
developing the test is clearly better than for the validation dataset. Batch effects and other differences
between the datasets are the most likely explanations for this difference. However, the validation dataset
consists of many different subgroups of individuals with a limited number of individuals in each subgroup,
making the interpretation uncertain.  
We were not able to show that the test is influenced by stress, drug use or smoking, but again, the datasets
are too small to draw any firm conclusions.