Validity of registry data: Agreement between cancer records in an end-stage kidney disease registry (voluntary reporting) and a cancer register (statutory reporting)
Article first published online: 19 FEB 2010
© 2010 The Authors. Journal compilation © 2010 Asian Pacific Society of Nephrology
Volume 15, Issue 4, pages 491–501, June 2010
How to Cite
WEBSTER, A. C., SUPRAMANIAM, R., O'CONNELL, D. L., CHAPMAN, J. R. and CRAIG, J. C. (2010), Validity of registry data: Agreement between cancer records in an end-stage kidney disease registry (voluntary reporting) and a cancer register (statutory reporting). Nephrology, 15: 491–501. doi: 10.1111/j.1440-1797.2010.01297.x
- Issue published online: 25 MAY 2010
- Article first published online: 19 FEB 2010
- Accepted for publication 13 November 2009.Accepted manuscript online 19 February 2010.
- epidemiology and outcomes;
Aims: End-stage kidney disease registries inform outcomes and policy. Data quality is crucial but difficult to measure objectively. We assessed agreement between incident cancer reported to the Australian and New Zealand Dialysis and Transplant Registry (ANZDATA) and to the Central Cancer Registry (CCR) in New South Wales.
Methods: ANZDATA records were linked to CCR using probabilistic matching. We calculated agreement between registries for patients with ≥1 cancers, all cancers and site-specific cancer using the kappa statistic (κ). We investigated cases where records disagreed and compared estimates of cancer risk based either on ANZDATA or on CCR using standardized incidence ratios (indirect standardization by age, sex and calendar year).
Results: From 1980 to 2001, 9453 residents had dialysis or transplantation. ANZDATA recorded 867 cancers in 779 (8.2%) registrants; CCR 867 cancers in 788 (8.3%). ANZDATA recorded 170 patients with cancer that CCR did not, CCR recorded 179 patients that ANZDATA did not (κ = 0.76). ANZDATA had sensitivity 77.3% (confidence interval (CI) 74.2–80.2), specificity 98.1% (CI 97.7–98.3) if CCR records were regarded as the reference standard. Agreement was similar for diagnoses while receiving dialysis (κ = 0.78) or after transplantation (κ = 0.79), but varied by cancer type. Agreement was poorest for melanoma (κ = 0.61) and myeloma (κ = 0.47) and highest for lymphoma (κ = 0.80), leukaemia (κ = 0.86) and breast cancer (κ = 0.85). Artefact accounted for 20.8% of the non-concordance but error and misclassification did occur in both registries. Estimates of cancer risk based on ANZDATA or CCR records did not differ in any important way.
Conclusion: Agreement of cancer records between both registries was high and differences largely explicable. It is likely that both ANZDATA and CCR have some inaccuracies, for reasons that are now more explicit, with themes similar to those likely to be experienced by other registries.