Predicting risk of bladder cancer in the UK

Abstract

We were interested to read the paper by Shephard et al1 on the clinical features of bladder cancer in primary care, as we have previously published a paper2 in the BJGP where we reported on symptoms associated with renal tract cancer (of which around 80% were bladder cancers) in primary care. Both studies evaluated the risk or positive predictive value of symptoms for detecting cases of cancer, and both used UK primary care databases, but the study designs were different with Shepherd et al using a case-control approach while we used a cohort design. A key finding of both studies is that haematuria is a strong predictor of bladder/renal cancer, with a stronger association at younger ages. Indeed we reported that around three-quarters of renal cancer cases had reported haematuria before their diagnosis. Both studies also found at least a doubling of risk in people consulting with abdominal pain. A strength of our study design was that we were able to combine the risks associated with the different symptoms and clinical features with the established risks associated with smoking, into an algorithm that can predict absolute risk in individuals presenting with symptoms. We also accounted for the steep increases in risk with age and for different risks in men and women, since rates in men are at least three times higher than in women. Crucially we validated this algorithm in a separate cohort of patients, and found that it discriminated extremely well between cancer cases and the remainder of the cohort (ROC values of 0.95 in men and 0.91 in women). The 10% of patients with highest risks contained 87% of all renal cancers diagnosed within 2 years. It is unclear how many cancer cases would be detected using the approach of Shepherd et al, nor is there any validation on an external dataset which is an essential step needed to determine whether the results are valuable.3