Using Generalizability Theory for the Estimation of Reliability of a Patient Classification System