C113
A comparison of three sources of information to guide the Bayesian design of a trial in the rare cancer Merkel-cell carcinoma
Mark Pritchard1, Lucinda Billingham1, Jeremy Marsden2, Neil Steven1
1Cancer Research UK Clinical Trials Unit, University of Birmingham, UK, 2University Hospitals Birmingham NHS Foundation Trust, Birmingham, UK
Background
Merkel-cell carcinoma (MCC) is a rare cancer of the skin. Its rarity poses practical problems for recruiting enough patients for a properly powered phase III trial, and therefore little evidence exists to guide adjuvant treatment. Bayesian design and analysis of trials has been suggested as a possible solution but has been criticized because inappropriate prior distributions can distort trials conclusions. We here describe three methods used to generate prior distributions for the effect of adjuvant chemotherapy treatment for MCC.
Method
(1) A systematic review of the literature on MCC was conducted using the methodology of Tan and colleagues (BMJ 2003;327:47) to incorporate all available evidence, weighted according to criteria of quality and relevance.
(2) Opinions of consultant oncologists and dermatologists with an interest in MCC were collected at a national meeting.
(3) West Midlands population-based cancer registration data were collected for patients diagnosed with MCC between 1980 and 2008.
Results
(1) Only one randomised study has previously investigated treating MCC with chemotherapy but over 800 papers report on series of patients or individual case reports with at least 300 containing relevant details of patient treatments and outcomes. A combined probability distribution represents the current evidence.
(2) Ten consultants provided opinions. Distributions for each consultant and a combined analysis represent current clinical opinion.
(3) Registration data included 189 patients diagnosed with MCC: eight were recorded as having adjuvant chemotherapy. This was insufficient for generating a probability distribution.
Conclusion
The discordance of consultants opinions suggests that the combined prior distribution is highly subjective and unlikely to be widely accepted. A documented, reproducible systematic search of the literature provides a prior distribution that may be of benefit in the design of trials for rare cancers such as MCC.
Acknowledgments
Cancer registration data were provided by the West Midlands Cancer Intelligence Unit, Birmingham.
