Risk of cancer following presentation with new-onset atrial fibrillation using data from UK national databases

Aims Atrial fibrillation (AF) and cancer are both highly prevalent conditions and are known to be associated. Our aim was to identify predictors and develop models for all cancer types, in men and women, and for the four most common cancer types in the AF population using linked primary and secondary care data from the UK. Methods and results We included 163 549 patients diagnosed with AF between January 1998 and May 2016, and no previous history of cancer. Following TRIPOD methodology, we developed a ridge-penalized multivariable logistic regression model to predict 1-year cancer incidence after AF diagnosis, using 70% of the data for derivation and 30% for validation. Age was associated with an increased risk across all cancer types. Socioeconomic deprivation, smoking, excessive alcohol intake, family history of cancer, chronic kidney disease, anaemia, and several cancer-related symptoms and clinical signs (e.g. rectal bleeding, loss of appetite) were associated with an increased risk in one or more cancer types. The prediction models showed moderate-to-good discrimination in the validation set, with c-statistic of 0.69 (0.68–0.70) for all cancer in men, 0.63 (0.62–0.65) for all cancer in women, 0.70 (0.68–0.73) for lung cancer, 0.70 (0.66–0.73) for colorectal cancer, 0.59 (0.53–0.65) for breast cancer, and 0.78 (0.72–0.84) for prostate cancer. Conclusion Most of the identified potential risk factors for cancer in the AF population are also associated with cardiovascular disease. The 1-year cancer prediction models showed moderate to good predictive performance and may help improve the management of patients with AF.