Identification of recurrences in women diagnosed with early invasive breast cancer using routinely collected data in England.

Abstract

Background: Breast cancer is the commonest cancer in the UK, with around 55,000 women diagnosed annually. Information is routinely available on breast cancer mortality but not on recurrence.

Methods: We used a database compiled by the West Midlands Cancer Intelligence Unit during 1997-2011 to develop and train a deterministic algorithm to identify recurrences in routinely collected data (RCD) available within NHS England. We trained the algorithm further using 150 women with stage II-III breast cancer who were recruited into the AZURE trial during 2003-2006 and invited to approximately 24 clinic follow-up visits over ten years. We then evaluated its performance using data for the remaining 1930 women in England in the AZURE trial.

Results: The sensitivity of the RCD to detect distant recurrences recorded in the AZURE trial during the ten years following randomisation was 95.6% and its sensitivity to detect any recurrence was 96.6%. The corresponding specificities were 91.9% for distant recurrence and 77.7% for any recurrence.

Conclusions: These findings demonstrate the potential of routinely collected data to identify breast cancer recurrences in England. The algorithm may have a role in several settings and make long-term follow-up in randomised trials of breast cancer treatments more cost-effective.