Abstract
Purpose Contemporary approaches to clinical diagnosis have not adequately exploited state-of-the-art empirical techniques in deriving diagnostic criterion sets that are statistically optimal based on 1) relevant external indicators and 2) replicability across data sets. We provide a proof of concept that optimal criterion sets can be derived with respect to alcohol use disorder (AUD) diagnosis that are both more efficient and precise than current systems. Methods Using data from the National Epidemiologic Survey on Alcohol and Related Conditions we selected chronicity (i.e. persistence) of AUD diagnosis and comorbidity of AUD with other disorders as validation criteria on which to optimize the size of the AUD criterion set and the threshold for AUD diagnosis. We used cross-validation and consensus approaches for choosing a final solution. Results Cross-validation did not produce a solution that replicated across random subsamples or differed from conventional diagnosis. Alternatively, consensus produced a more global solution that was associated with greater validity than “conventional” diagnosis. Conclusion Such methods, if applied to extant diagnostic criteria and algorithms can generate simpler and more reliable rules and hold promise for greatly reducing misclassification of individuals in both research and applied clinical contexts.