Bayesian predictions of final outcomes: regulatory approval of a spinal implant

Abstract

We describe a randomized controlled trial of an investigational spinal implant. The investigational device has an obvious benefit in comparison with control in that it precludes the need for harvesting bone graft and the pain and morbidity associated with it. Therefore, the principal comparison is one of noninferiority. The primary endpoint is overall success at two years. The “noninferiority margin” is 10%. Waiting for two years after the last patient's surgery may not be necessary depending on earlier measurements of success. We model the relationship between one-and two-year results. Our Bayesian analysis considers all available information, including some patients who have both one-and two-year results and some patients who have only one-year results. Our study provides an example in which Bayesian predictive modeling provided earlier information than otherwise and therefore it shortened the time line of the development of a therapeutic strategy.

Get full access to this article

View all access options for this article.

References

1. Berry DA . Statistics: A Bayesian perspective. Belmont, CA: Wadsworth Publishing Company, 1996.

2. Spiegelhalter DJ , Abrams KR , Myles JP . Bayesian approaches to clinical trials and health care evaluations (statistics in practice). New York: John Wiley & Sons, 2004.

3. Little RJA , Rubin DB . Statistical analysis with missing data. New York: John Wiley, 1987.