• Validity of chemotherapy information derived from routinely collected healthcare data: A national cohort study of colon cancer patients

      Boyle, J. M.; Kuryba, A.; Braun, Michael S; Aggarwal, A.; van der Meulen, J.; Cowling, T; E.Walker, K.; Department of Health Services Research and Policy, London School of Hygiene and Tropical Medicine, London, UK; Clinical Effectiveness Unit, Royal College of Surgeons of England, London, UK. (2021)
      Background: We used a structured approach to validate chemotherapy information derived from a national routinely collected chemotherapy dataset and from national administrative hospital data. Methods: 10,280 patients who had surgical resection with stage III colon cancer were included. First, we compared information derived from the national chemotherapy dataset (SACT) and from the administrative hospital dataset (HES) in the English NHS with respect to receipt of adjuvant chemotherapy (ACT). Second, we compared regimen and number of cycles in linked patient-level records. Third, we carried out a sensitivity analysis to establish to what extent the impact of ACT receipt differed according to data source. Results: 6,012 patients (58 %) received ACT according to either dataset. Of these patients, 3,460 (58 %) had ACT records in both datasets, 1,649 (27 %) in SACT alone, and 903 (15 %) in HES alone. Of the 3,460 patients with records in both datasets, 3,320 (96 %) had matching regimens. There was good agreement on cycle number with similar proportions of patients recorded with a single cycle (6 % in SACT vs. 7 % in HES) and slightly fewer patients recorded with more than 8 cycles in SACT (32 % in SACT vs. 35 % in HES). 3-year cancer-specific mortality was similar for patients receiving ACT, regardless of whether a patient received ACT according to SACT alone (16.6 %), according to HES alone (16.8 %), or according to either SACT or HES (17.1 %). Conclusion: Routinely collected national chemotherapy data and administrative hospital data are highly accurate in recording regimen and number of chemotherapy cycles. However, chemotherapy information should ideally be captured from both datasets to avoid under-capture, particularly of oral chemotherapy from administrative hospital data, and to minimise bias.