Abstract
BACKGROUND: Routinely collected data are increasingly being used for cancer research and health service evaluation. For both purposes, accurately identifying metastatic disease at diagnosis is essential. We developed an approach to identify metastatic disease at time of primary diagnosis according to national hospital administrative data (HAD) in patients identified with colorectal cancer (CRC) in the English national cancer registry (CR). METHODS: A national cohort of CRC patients diagnosed between 2013 and 2018 in England identified in CR data were linked to HAD. Metastatic disease was assumed to be present at diagnosis according to HAD if at least one of a set of pre-specified diagnostic ICD-10 codes appeared in a record of a hospital admission between one month before and six months after CRC diagnosis date. RESULTS: Of 186,236 patients, 40,421 (21.7%) had metastatic cancer according to HAD, 42,843 (23.0%) according to CR data, 49,827 (26.8%) according to either data source, and 33,437 (18.0%) according to both. Metastatic information was missing in CR data in 14,065 patients and 1,930 of these (13.7%) had metastatic cancer according to HAD. 1-year mortality was 59.3% (95%-CI: 58.8 - 59.8%) in patients with metastatic disease and 7.4% (7.2 - 7.5%) in patients without if HAD and CR data agreed. Mortality fell between these results if HAD and CR data disagreed. High mortality was seen in patients with missing metastatic data in the CR: 74.4% (72.4 - 76.3%) in patients with metastatic disease and 45.2% (44.3-46.1%) in patients without metastatic disease according to HAD. CONCLUSIONS: HAD should be linked to CR data to provide more accurate information on metastatic CRC at diagnosis including sites of metastasis. Linkage to HAD increased the number of patients identified with metastatic CRC by 14%, compared to CR data alone. Patients with metastatic disease at diagnosis in either data source had mortality outcomes expected for patients with metastatic cancer. CRC patients with missing metastasis data in CR data are likely to have metastatic disease and linkage to HAD provides important prognostic information.