Biobanks represent key resources for translational research and personalized medicine (Thasler et al., 2013). The Biobank under Administration of Human Tissue and Cell Research Foundation HTCR at University of Munich Medical Centre is routinely collecting remnant tissues and blood samples from treatments of patients at the Clinic for General, Visceral, Vascular and Transplantation Surgery as well as the Department for Thoracic Surgery.
However, to be a valuable resource for translational biomedical research these human biospecimens have to be annotated with clinical data, currently only available from free text records.
Access to these biospecimens and data, e.g. over web-based query tools, like the transinstitutional metabiobank CRIP (Schröder et al., 2011) or p-BioSPRE (Weiler et al., 2014), demands that information from various sources gets integrated into harmonized and structured data to enable stratified, parameterized queries (Ambert and Cohen, 2009).
So far, analysis of free text sources and structured data entry in the data protection compliant biobank information system (Müller and Thasler, 2014) is done manually, and regarding diagnostic classification of cases, a biobank specific classification was developed. This classification however was not matched to ICD codes so far. Based on automated free text analysis this coding can now be amended with international classifications, e.g. ICD-10 or ICD-O-3, to facilitate project queries.
The automated knowledge extraction software CRIP.CodEx was designed to identify and extract information in free text medical records and assign corresponding codes (see Figure 1). We used CRIP.CodEx slightly off its designedfor purpose, which is free text medical reports. Instead we used CRIP.CodEx to recodify the textual description of the specialized HTCR biobank classification with ICD (see Figure 2). The initially existing catalogue of diagnoses was based on the actual collection of cases in the biobank, reflecting indications and surgical treatments and therefore being primarily aligned along a list of organs affected by the surgical treatment. To build a catalogue of diagnoses suitable for biospecimen research, in a first step, this catalogue had to be reorganized towards distinct pathological findings across affected organs. The resulting categories therefore were composed of two separate elements, a diagnosis from pathological report and the affected organ. These categories, 65 in total, were then manually referred to the ICD-10 catalogue resulting in 82 codes.
After ensuring that the final categories were sufficiently selective, we analyzed them in a first run using CRIP.CodEx for ICD-O-3 as well as ICD-10 classifications. The result was checked manually and discussed with a Pathologist.
However, the discussion of the first run showed that the problems identified relied not only to CRIP.CodEx, but also to the diagnostic catalogue, mainly the categories' composition as separate elements of diagnosis and organs affected by the surgical treatment. Both sources could be tackledand therefore both, the software configuration and also the categories have been worked on to reach improved results in the second run.
The configuration of CRIP.CodEx was optimized including these aspects: adding variants of local wording/synonyms, e.g. 'Carzinom' instead of 'Karzinom'/'carcinoma', into the internal dictionary, increasing the number of allowed multiple matches when detecting combined and hyphenated words with synonyms, as well as restricting ICD-10 neoplasm code assignment depending on detected ICD-O-3 classification.
The categories of the biobank specific classification (see Table 1) have been amended mainly by integrating the pathological diagnosis and the primarily diseased organ in to one descriptive text. Further amendments include specifying the affected organs and information on primary and secondary tumors more consistently, as well as adding tumor types in some categories while shortening the description of others and dividing into subcategories, accompanied by also amending the ICD-O-3 code assignments with an expert from the regional tumor registry towards the WHO "Blue Book" (Bosman et al., 2010) as international standard.
After both, optimizing the configuration of CRIP.CodEx, as well as further amending of the categories, we performed a second run for analyzing the categories.
Diagnosis 3 Results
The reorganized catalogue of diagnoses contained 64 categories and has been automatedly matched to classifications by CRIP.CodEx. For each of the categories CRIP.CodEx assigned matching classifications in ICD-10 or ICD-O-3. All together there have been 237 correct code assignments (true positives), while we identified 6 wrong assignments (false positives) and 12 missing codes (false negatives).
Due to the traceability of CRIP.CodEx's each individual code assignment, it was possible to identify the causes for all false positives and negatives. All but two causes could be resolved by the outlined optimization in the CRIP.CodEx configuration and the read in coding guide respectively.
The configuration of CRIP.CodEx as well as the original reorganized biobank specific catalogue of diagnoses has been optimized and amended as outlined above. Then again the local catalogue has been automatedly matched to ICD classifications by CRIP.CodEx. The final catalogue of diagnoses contains 73 categories, for each of them the software assigned matching classifications in ICD-10 or ICD-O-3 in the final second run. All together there have been 442 correct code assignments (true positives) by CRIP.CodEx, while we identified zero wrong assignments (false positives) and 14 missing codes (false negatives).
Since diagnostic information contained in medical free text is extracted and codified by the automated CRIP.CodEx software, we also showed that the information contained in the nomenclature of the individual biobank specific catalogue of diagnoses is sufficient for a mapping towards ICD-10 and basically also ICD-O catalogues. As a remaining issue however, for categories such as e.g. "Carcinoma of the gallbladder", which summarize a wide range of different morphologies, ICD-O code extraction from nomenclature is too general, and amending the nomenclature with a listing of these morphologies is also dissatisfying.
For the implementation of the restructured and amended biobank-specific catalogue in the Biobanks Database or the HTCR Web Application however, the categories have been now structured as a table by key features that can be maintained:
"Primary tumor vs. secondary tumor vs.
no tumor" "affected Organ" "originating organ in case of secondary tumor" "included morphologies" So as a next step, by tapping complementary data sources, mainly extracting diagnostic information from pathology reports by CRIP.CodEx, in addition to coding of specialized local classification, we deliver an automated matching of two different classification systems.
Even if all automated classifications have to be checked thoroughly according to an individual request from a research project, before providing samples and data, this initial, very effective automated classification to great extent facilitates case related database research as well as data export for display of the collection in biobank registries and even metabiobanks. Thereby we have not only enhanced the biobank's availability for translational research but also proposed a general protocol for matching internal codes with international classifications and standards.
The authors specially thank Dr. Jens Neumann from Institute of Pathology, Ludwig-Maximilians-University Munich and Dr. Gabriele Schubert-Fritschle of Tumor Registry Munich for discussing the results of CodEx code assignments and the amended categories and manual code assignments.