The development and validation of oral cancer staging using administrative health data

BACKGROUND: Oral cancer is a major global health problem. The complexity of histological prognosticators in oral cancer makes it difficult to compare the benefits of different treatment regimens. The Taiwanese National Health database provides an opportunity to assess correlations between outcome and treatment protocols and to compare the effects of different treatment regimens. However, the absence of indices of disease severity is a critical problem. The aim of this study was to ascertain how accurately we could assess the severity of oral cancer at the time of initial diagnosis on the basis of variables in a national database. METHODS: In the cancer registry database of a medical center in Taiwan, we identified 1067 histologically confirmed cases of oral cancer (ICD9 codes 140, 141 and 143–145) that had been first diagnosed and subjected to initial treatment in this hospital. The clinical staging status was considered as the gold standard and we used concordance (C)-statistics to assess the model’s predictive performance. We added the predictors of treatment modality, cancer subsite, and age group to our models. RESULTS: Our final overall model included treatment regimen, site, age, and two interaction terms; namely, interactions between treatment regimen and age and those between treatment regimen, site, and age. In this model, the C-statistics were 0.82–0.84 in male subjects and 0.96–0.99 in female subjects. Of the models stratified by age, the model that considered treatment regimen and site had the highest C-statistics for the interaction term, this value being greater than 0.80 in male subjects and 0.9 in female subjects. CONCLUSION: In this study, we found that adjusting for sex, age at first diagnosis, oral cancer subsite, and therapy regimen provided the best indicator of severity of oral cancer. Our findings provide a method for assessing cancer severity when information about staging is not available from a national health-related database.