Presentation (CTI)

OBJECTIVES: The development and validation of ICD-10 algorithms for cancer diagnosis, treatment, and procedures is necessary to support real-world evidence studies. However, disparities in the number of validated algorithms across different cancer types remain unknown. This study provides a descriptive analysis of the number of validated ICD-10 algorithms for cervical and uterine cancer.
METHODS: Two systematic reviews using similar search strings were conducted to collate algorithms for the two cancer types. The search was limited to January 2016- July 2024 to focus on ICD-10 codes. Articles likely to include ICD code sets or algorithms for patient identification were included. The total number of algorithms was compared between cancer types as well as the presence of validation statistics.
RESULTS: For cervical cancer, 339 articles were screened, 50 articles included, 30 reported algorithms for patient identification, and 21 articles reported ICD-10 code algorithms. Seven unique coding algorithms for patient identification were found and only 1 of the articles reported validation statistics. For uterine cancer, 347 articles were screened, 46 articles included, 19 reported algorithms for patient identification, and 12 articles reported ICD-10 code sets identifying 11 unique algorithms for patient identification with 3 of the articles reporting validation statistics. The coding sets vary in their level of detail, from only a single code to the more comprehensive algorithms using combinations of diagnosis, procedure and prescription codes.
CONCLUSIONS: ICD-10 algorithms were identified in 70% and 63% of reviewed articles reporting patient identification methods for cervical cancer and uterine cancer, respectively. Validation statistics were only provided in 5% and 25% of the papers providing ICD-10 algorithms for identifying patients with cervical and uterine cancer, respectively. This illustrates that many research papers do not report computable operational definition details necessary to reproduce their findings. Also, most reported algorithms don’t come with validation statistics.