Assessment and improvement of the drug data structuredness from electronic health records to enable secondary usage and ensure semantic interoperability (Preprint)

Abstract

Digitization offers a multitude of opportunities to gain insights for current diagnostics and therapy from retrospective data. In this context, real-world data and its accessibility are of increasing importance to support unbiased and reliable research on big data. However, routinely collected data are not readily usable for research due to the unstructured nature within healthcare systems and the lack of interoperability between these systems. This challenge is very evident in drug data.

In this research, we present an approach that identifies and increases the structuredness of drug data while ensuring standardization according to ATC.

Our approach is based on available drug prescriptions and a drug catalog and consists of four steps. First, we perform an initial analysis of the structuredness of local drug data to define a point of comparison for the effectiveness of the overall approach. Second, we apply three algorithms to unstructured data that translate text into ATC codes based on string comparisons in terms of ingredients and product names, and similarity comparisons based on Levenshtein distance. Third, we validated the results of the three algorithms with expert knowledge based on the 1000 most frequently used prescription texts. Fourth, we performed a final evaluation to determine the increased degree of structuredness.

Initially, 52.3% of the 1,768,153 drug prescriptions were classified as unstructured. With the application of the three algorithms, we were able to increase the degree of structuredness to 85.18% based on the 1000 most frequent medication prescriptions. In this regard, the combination of Algorithm 1, 2, and 3 resulted in a correctness level of 100% (with 57,264 ATC codes identified), Algorithm 1 and 3 of 99.6% (with 152,404 codes identified), and Algorithm 1 and 2 resulted in 95.9% (with 39,472 codes identified).

As shown in the first analysis steps of our approach, the availability of a product catalog to select from during the documentation process is not sufficient to generate over structured data. Our four-step approach reduces the problems and reliably increases the structuredness automatically. Similarity matching shows promising results, especially for entries with no connection to a product catalog. However, further enhancement of the correctness of such a similarity matching algorithm needs to be investigated in future work.

no trial registration The research was approved by the Ethics Committee at the Technical University of Dresden as a retrospective, observational, non-interventional, non-human subject study (SR-EK-521112021).