Harmonization of Computable Oncology Regimens and their Application to Australian Cancer Treatment Data

Background: Observational research into patterns of variation in chemotherapy delivery and their effects on outcomes is greatly facilitated by having a defined baseline treatment protocol prescription against which observed drug exposures can be compared. This requires accurate identification of the intended treatment protocol for each patient, as well as detailed abstraction of its constituent components (i.e. drugs, dose, frequency, duration). With this, it becomes possible to derive measures such as reduced dose intensity, early treatment termination or adjustments to the treatment schedule. Methods: In this work, we first derive a computable form of Australian chemotherapy regimens as published on the eviQ web site. This abstraction is then mapped to the existing US HemOnc ontology, before being applied to a real-world drug delivery dataset to demonstrate its utility for detecting variation in cancer care. Results: It was possible to resolve relationships between these vocabularies with high completeness and accuracy (93%). Further to this, real-world drug delivery data was able to be matched to eviQ protocols for up to 87% of delivered regimens using an episodic data model. Of the delivered regimens mapped to eviQ, 92% of a validation subset with known protocol prescriptions matched their nominated regimen. Conclusions: By applying relatively simple, rule-based algorithms to computable protocol abstractions, it is possible to achieve matching between reference regimens and as-delivered treatment with a high degree of accuracy and completeness. These techniques provide a foundation of baseline protocol definitions for future work to characterize patterns of variation between patients’ prescribed and delivered systemic anti-cancer treatments.