3687 - Validation of Claims-Based Algorithms to Characterize Thoracic Radiation Therapy Treatment Courses: Are Claims Enough to Study Radiation Therapy?

Purpose/Objective(s): Routinely collected administrative data offer insights into healthcare utilization and outcomes but lack detailed clinical information—such as the specific site of radiation therapy (RT)—which limits the study of RT toxicity and efficacy in claims databases. This study aims to develop and validate claims-based algorithms to accurately identify thoracic RT in administrative databases.

Materials/Methods: Patients at our institution with an International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) lung cancer diagnosis and any RT Current Procedural Terminology (CPT) code from 10/2015-1/2024 were analyzed. RT claims for each treatment episodes were extracted. RT details were manually abstracted from the treatment planning system and electronic health record to classify episodes as thoracic or non-thoracic RT. A priori algorithms were defined as the presence of respiratory motion management (RMM) codes, more than 14 treatment codes (except for stereotactic body radiation therapy courses), and exclusively thoracic ICD-10-CM codes during the episode (i.e., no claims for bone metastases). Decision tree models using Classification and Regression Tree algorithms were developed on 70% of the cohort and validated on the remaining 30%. Performance metrics (sensitivity (Sn), specificity (Sp), positive predictive value (PPV), and negative predictive value (NPV)) were calculated, with an acceptable algorithm defined by a lower-bound Clopper-Pearson 95% confidence interval for PPV exceeding 70%.

Results: Of 3491 patients and 4715 RT episodes initially identified, 131 episodes (3%) and 738 (16%) episodes were excluded for missing claims data and absence of external beam RT claims, respectively, leaving 3846 episodes for analysis. Performance metrics for the tested algorithms are shown.

Conclusion: Clinically-informed and decision tree–based models can accurately identify thoracic RT in claims data, achieving high PPVs for IMRT and SBRT courses, while limiting false negatives. These algorithms can be applied in claims databases to assess RT toxicity and effectiveness.

Abstract 3687 - Table 1: Algorithm performance