372 – Contributed Oral Poster Presentations: Biopharmaceutical Section
Statistical Assessment of Clinical Trials with Discordant Pairs of Observations
Dar Shong Hwang
B.R.S.I.
James Lee
Daiichi Sankyo Pharma Development
Chyi-Hung Hsu
Janssen R&D
In a clinical trial having patient's self assessment and investigator's assessment of the symptoms/signs of a disease, the result of a patient's assessment may not be totally in agreement with investigator's assessment. Intuitively the discordant pairs may not be reliable to judge whether the treatment is a success or a failure. Instead, the concordant pairs wherein both patient and investigator agree that the treatment is a success or a failure should be used. The situation is similar to "getting a second opinion" in medical practice. When 2nd opinion agrees with the 1st opinion, one would have more confidence on the recommendation. In this poster, we provide an example of analyzing a parallel group design with 2 treatment groups (Active vs. Placebo) wherein each patient having both patient's and investigator's assessments. Optimum statistical inference procedure turns out to be one that discards the discordant pairs of data, and only the concordant pairs are used. The result agrees with one's intuition. This is an interesting contrast to a matched pair design using Exact McNemar's test wherein only the discordant pairs are used in the comparison of the 2 treatments.