THE VALUE OF USING CANCER REGISTRY DATA SETS LINKED TO PRIMARY CARE

Problem

The linkage of patient records across different healthcare sectors has enhanced research, particularly in the area of cancer. Clinical Practice Research Datalink (CPRD) provides primary care data linked to several National Cancer Registration and Analysis Service (NCRAS)) data sets, enabling researchers to study the clinical management and treatment of cancer patients. However, it is currently unclear how these data sources complement each other. Understanding this is crucial to determine the appropriate combination of data sources when planning a study.

Approach

This was a cross-sectional study using linked UK data from CPRD [CPRD GOLD primary care, NCRAS Cancer Registration Data and the recently available NCRAS Systemic Anti-Cancer Therapy dataset (SACT)]. Cases included women with incident breast cancer registered between 2013-2015. Treatment was defined as a record of receiving any form of chemotherapy drug. Concordance of recording was calculated by estimating the proportion of patients with (i) the same breast cancer diagnosis code recorded in the cancer registry and SACT, (ii) a record for systemic treatment in the cancer registry and SACT, and (iii) Body mass index (BMI) recorded in CPRD GOLD and SACT.

Findings

3,579 female patients with incident breast cancer and a treatment event recorded in the cancer registry data, were included [mean age: 61 years (standard deviation: 13.6)]. 990 of these patients also had a primary diagnosis recorded in SACT, of which 43% were an exact match with the record in the cancer registry data. Of 3,579 patients, only 9% had a record for receiving a named systemic treatment in the cancer registry. Of the 3,579 patients in the cohort, 894 (25%) had a record of receiving a named systemic treatment in the SACT data. Only 17% of patients had a record in both data sources. There was lower concordance (6%) for records of a specific systemic treatment.Of the 894 patients with a complete record of treatment in SACT, 86% had the information required to calculate BMI in the associated record. This data was available for 96% of the same cohort in CPRD GOLD. Of the patients with a complete BMI record in SACT, 46% of patients had a BMI record in CPRD GOLD which was within 1 kg/m2 of their record in SACT. For the remaining 54%, most of the recorded BMIs were higher in primary care. Outcome of treatment was missing for 95% patients with a record in SACT. Adjusted analyses and further analysis regarding the feasibility of determining outcome of treatment in CPRD GOLD (when this information was not available in SACT) will be explored.

Consequences

These results demonstrate the limitations of using these data sources in isolation and how combined they can help bridge data gaps in the individual data source.