Development and Validation of a Computable Radiotherapy Phenotype.

Background: This study aims to develop a robust methodology using structured and semi-structured health data to identify patients who have undergone radiation therapy, thereby facilitating future research on treatment outcomes.

Methods: In this retrospective cohort study, we identified Veterans receiving radiation oncology care through documentation of referrals, encounters, and billing codes from 2014 through 2023. We classified administrative codes based on process of care and type of radiation received, and then analyzed utilization patterns. Unstructured data analysis was performed using keyword search of relevant clinical notes in the metadata fields and categorizing those notes by radiation oncology processes. To validate our algorithm, we compared the cohort we developed using existing data sources to a cohort that was chart reviewed.

Results: The final cohort included 589,318 Veterans with radiation oncology care. Among these, 355,276 had codes indicating radiotherapy delivery. The most common treatments were image guided (IGRT), three-dimensional conformal (3DCRT), intensity-modulated (IMRT), and stereotactic radiosurgery/stereotactic body (SRS/SBRT). Anatomy-specific billing codes were underutilized. Clinical note analysis identified 1,341 unique labels for radiation oncology content, with 947,928 notes found for 204,064 patients. Validation against chart-reviewed datasets showed strong concordance and confirmed the accuracy of our algorithm in identifying radiation oncology care.

Conclusion: Automated extraction of medical records can be used to identify cohorts of patients who have undergone radiotherapy. Employing this algorithm may facilitate more precise phenotyping of radiotherapy cases and thus significantly enhance our understanding of these cohorts.