Data for: Representing Oncology in Datasets: Standard or Custom Biomedical Terminology?

Published: 07-05-2019| Version 1 | DOI: 10.17632/xzbzz6czr7.1
Pablo López-García,
Philipp Daumke,
Stefan Schulz,
Martin Romacker


We collected 250 cancer-related records that had already been coded using a custom terminology at Roche Inc (ROCHE). The purpose was to annotate substances of pharmacological interest, taking both anatomy (Roche.Anatomie) and histology (Roche.Histology) into account. 150 records were given to either of two medical students (CoderA and CoderB). Coders were asked to identify representational units that expressed the same meaning in four target terminologies (SNOMED CT, NCIt, ICD-10 + ICD-O, and MedDRA) with as few codes as possible. Fifty of the cases were overlapping (double-coded), in order to enable the computation of inter-rater agreement. For evaluation, we used the following definitions: Hit=At least one code was provided for that terminology. Agreement=Both coders provided exactly the same codes. Therefore, an absence of codes by both counted as an agreement, and a same primary code but different secondary code counted as a disagreement.