Selection criteria for upper gastrointestinal examinations: attempts at improvement.

An attempt was made to improve upon selection criteria for the performance of upper gastrointestinal (UGI) series in three settings: a teaching hospital, a community hospital, and a health maintenance organization. Two statistical techniques, the polychotomous logistic model (to develop predictive algorithms for the identification of specific diseases) and the maximum attainable discrimination technique, were used to show the relationship between the percentage of patients with any disease detected and the percentage of UGI examinations performed. Results showed that neither technique improved significantly upon selection criteria for identifying patients with abnormal UGI series.