Impact of variation in cancer registration practice on observed international cancer survival differences between International Cancer Benchmarking Partnership (ICBP) jurisdictions

Background: International cancer survival comparisons use cancer registration data to report cancer survival, which informs the development of cancer policy and practice. Studies like the International Cancer Benchmarking Partnership (ICBP) have a duty to understand how registration differences impact on survival prior to drawing conclusions. Methods: Key informants reported differences in registration practice for capturing incidence date, death certificate case handling and registration of multiple primary tumours. Sensitivity analyses estimated their impact on one-year survival using baseline and supplementary cancer registration data from England and Sweden. Results: Variations in registration practice accounted for up to a 7.3 percentage point difference between unadjusted (estimates from previous ICBP survival data) and adjusted (estimates recalculated accounting for registration differences) one-year survival, depending on tumour site and jurisdiction. One-year survival estimates for four jurisdictions were affected by adjustment: New South Wales, Norway, Ontario, Sweden. Sweden and Ontario's survival reduced after adjustment, yet they remained the jurisdictions with the highest survival for breast and ovarian cancer respectively. Sweden had the highest unadjusted lung cancer survival of 43.6% which was adjusted to 39.0% leaving Victoria and Manitoba with the highest estimate at 42.7%. For colorectal cancer, Victoria's highest survival of 85.1% remained unchanged after adjustment. Conclusion: Population-based cancer survival comparisons can be subject to registration biases that may impact the reported 'survival gap' between populations. Efforts should be made to apply consistent registration practices internationally. In the meantime, survival comparison studies should provide acknowledgement of or adjustment for the registration biases that may affect their conclusions.