Recent research by Steinberg, Fine and Chappell presents methods for designing studies of diagnostic procedures when the goal is to prove, at a given level of confidence, that the method meets prescribed goals for Positive and/or Negative Predictive Value (PPV and NPV).
The paper shows how to derive optimal allocation of the study subjects between those who have the condition and healthy controls and also how to find the needed sample size.
The results are based on asymptotic formulas for making inference about PPV or NPV. The validity of these formulas for smaller samples can be checked by simulation.
This website provides a set of R functions that can be used for: checking the inference in small samples, calibrating confidence bound multipliers, computing optimal allocations and sample sizes, comparing inference for different possible allocations.
Click here to get the R functions.