DESIGN OF EXPERIMENT / TEST

2 posts / 0 new
Last post
nedcampbell
nedcampbell's picture
DESIGN OF EXPERIMENT / TEST

I am wanting to validate that a natural language coding tool is at least 95% accurate compared to human analysis.  It processes radiology medical reports (40,000,000+) and specifically identifies if any of 7 different independent procedures were completed and documented by a radiologist in each report based on key word search and other complex analytics.  It is expected for these 7 procedures, that anywhere between 250,000 and 2,000,000 of these reports will have included these procedures and should be marked "positive" for this procedure.
What is an appropriate DOE / test to perform to confirm the tool is at least 95% accurate with a confidence level of 95%?
Thanks.
Ned

R Bishop
R Bishop's picture
topic moved to bioinformatics

topic moved to bioinformatics. sorry i dont have the answer but the folks in here might