This work addresses some of the problems with development of diagnostic screening tests for apparently healthy patients. Maximization of screening cost-effectiveness translates naturally into optimization of test outcome decision algorithms with unequal error penalties. Research continued on the formulation of algorithms for training neural networks with unequal error weighting. A theoretical formulation was completed, and a list of candidate algorithms developed. Work has begun on the comparison of candidates. Advanced programming languages for symbolic mathematics open opportunities for more flexible data modeling programs, simultaneously available on many different computer architectures. The existing NIH program ALLFIT approximates data from families of experiments by mathematical formulas. Extension of ALLFIT using the Mathematica programming language is under development. by permitting the user to specify the model with Mathematica expressions, many different types of data analysis can be performed and compared without reprogramming.
