What lessons can be learned for cancer registration quality assurance from data users? Skin cancer as an example

In cancer registration, data cleaning (i.e. amendments made by data users to datasets released by registries) is potentially informative for quality assurance, but generally underreported.

To assess the scope for learning lessons about cancer registration quality assurance from a data user (using skin cancer as the example).

The main design features were: (i) A descriptive study identifying, qualitatively and quantitatively, the breadth, depth, and impact of quality assurance issues raised by a user cleaning Merseyside and Cheshire Cancer Registry skin cancer data. Errors were rectified and pitfalls for interpretation were identified. (ii) A nested validation of morphology and site coding on random samples of cutaneous malignant melanomas, basal cell carcinomas (BCC), and squamous cell carcinomas. The 33132-record dataset comprised: all registered skin lesions, except metastases; most recorded variables (about patient, lesion, treatment, outcome); for Merseyside and Cheshire residents diagnosed 1970-1991.

(i) Ineligible cases represented 0.3% (97/33132), and were detected best by morphology checks. Most quality assurance issues identified related to local custom and practice, staff training, and computerization, being particularly illustrated by problematic BCC registration practice (e.g. records written over unchallenged by range checks; and idiosyncratic use of variables). (ii) Post-cleaning, morphology coding errors were minimal in the random samples.

There is great scope for data users to contribute to cancer registration quality assurance. Ultimately, the study dataset appeared fit for epidemiological analysis and important quality assurance messages emerged. Shared explicit standard guidelines for data preparation and validation are needed by users, whose insights could and should be better recognized by cancer registries.