Registry versus claims-based index dates for studies of cancer diagnosis in administrative data

Abstract

Purpose: Studies of healthcare encounters leading to cancer diagnosis have increased over recent years. While some studies examine healthcare utilization before the cancer registry date of diagnosis, relevant pre-diagnosis interactions are not always immediately prior to this date due to date abstraction guidelines. We evaluated agreement of a registry date with a claims-based index and examined Emergency Department (ED) involvement in cancer diagnosis as an example of possible pre-diagnostic healthcare misclassification that could arise from improper date choice. Methods: We implemented an algorithm to define a claims-based index as the date of the earliest International Classification of Diseases code for the cancer in Medicare and estimated agreement with the date of diagnosis from a North Carolina registry for patients diagnosed aged 66 or older with 16 cancer types from 2008 to 2017 (n = 92,056). We then classified whether each cancer was initially diagnosed through care originating in the ED using each date. Results: The index date was identical to the cancer registry date for 47% of patients and preceded the registry date for 28%, with extent of agreement varying by cancer- and patient-specific characteristics. Agreement in ED-involved diagnosis classification using each date varied by cancer site, with sensitivity of classifications using the registry date relative to the index having a minimum of 86% for prostate and kidney cancer. Conclusion: Studies assessing healthcare utilization proximal to cancer diagnosis should carefully consider the relevant assessment window and be aware that the use of cancer registry versus claims-based dates may impact variable classification.