Comparison of Epidemiological Estimates for 19 Cancer Types Using Electronic Health Record Databases in England: An Analysis of CPRD Aurum and CPRD GOLD Databases with Linked Hospital Episode Statistics and Cancer Registry Data

Anna B Chaplin,1 Olia Archangelidi,1 Katrina Wilcox Hagberg,2 David Neasham,1 George Kafatos1 1Center for Observational Research, Amgen Limited, Uxbridge, UK; 2Boston Collaborative Drug Surveillance Program, Lexington, MA, USACorrespondence: Anna B Chaplin, Center for Observational Research, Amgen Ltd, Uxbridge Business Park, 4 Sanderson Road, Denham, Uxbridge, UB8 1DH, UK, Tel +44 0 7741 601 593, Email [email protected]: To compare epidemiological estimates of 19 different cancer types in Clinical Practice Research Datalink (CPRD) Aurum and CPRD General Practice Online Database (GOLD) databases against linked secondary data sources in England to understand best use of these data sources for research.Methods: The source population comprised patients in CPRD Aurum or GOLD (separately) who were eligible for linkage to Cancer Registry (CR), Hospital Episode Statistics (HES), and Office for National Statistics (ONS). We selected patients who had an incident cancer diagnosis recorded in ≥ 1 data sources (CPRD Aurum or GOLD, HES, CR) between January 1, 2011 and December 31, 2018. We estimated incidence rates (IR) and counts by cancer type and data source, and survival probability by cancer type among patients with an ONS death record recorded between January 1, 2011 and April 30, 2020.Results: The highest incident case capture resulted from CPRD Aurum or GOLD linked to HES and CR. In the fully linked CPRD Aurum-HES-CR and CPRD GOLD-HES-CR datasets, cancers typically diagnosed and managed in primary care (eg, breast, prostate, and lung) had the highest IRs and more complete case capture compared with other data sources, whereas HES and CR had higher IRs for cancers diagnosed in secondary care settings (eg, gastric, renal, and bladder). Cancers with broad definitions (eg, head and neck) had wider variations in IRs across data sources than cancers with narrower definitions. Survival estimates were generally higher for cancer-related deaths versus all-cause deaths.Conclusion: Findings highlight variation in cancer recording across different data sources. Researchers using CPRD data should assess the benefit of incorporating linked data on a study-by-study basis. For studies of breast, prostate, and lung cancers, CPRD Aurum or GOLD alone may be sufficient; however, linkage to HES and/or CR is recommended where a more complete case capture is required.Plain Language Summary: Purpose: To compare the number of patients with 19 different cancer types recorded in databases from general practitioner (GP) practices in England with other databases. The aim was to help researchers understand and interpret results from these databases.Methods: The GP practices databases, Clinical Practice Research Datalink (CPRD) Aurum and GOLD, were compared with the Cancer Registry (CR), Hospital Episode Statistics (HES), and Office for National Statistics (ONS) death records. Patients in CPRD Aurum or GOLD (separately) were included in the study if their medical record could be linked to CR, HES, and ONS. We selected patients who had their first ever (incident) cancer diagnosis recorded in ≥ 1 data sources between January 1, 2011 and December 31, 2018. We estimated the number and rate of new cancer diagnoses by cancer type and data source. We estimated the probability of survival by cancer type among patients with a death record between January 1, 2011 and April 30, 2020.Results: The number of new cancer cases varied depending on the type of cancer. Linking GP records to hospital and cancer registry data resulted in the highest count of new cancer cases. GP data alone captured more diagnoses for cancers typically diagnosed and managed in primary care, while CR and HES linked datasets had higher rates for cancers diagnosed in secondary care.Conclusion: The results show that cancer records differ between various data sources. Researchers using CPRD data should carefully consider whether to use linked data for each specific study.Keywords: CPRD Aurum, CPRD GOLD, UK cancer registries, Hospital Episode Statistics, cancer incidence