Accurate prediction of duodenal-ulcer healing rate by discriminant analysis

Abstract

Previous studies have shown that approximately 50% and 70% of duodenal ulcers heal after 2 and 4 wk, respectively, of cimetidine, and at least one-third heal after 4 wk of placebo. In order to identify these groups of ulcers before treatment, a two-phase study was performed, including an initial double-blind trial of cimetidine vs. placebo in 120 patients, and a subsequent open study with identical protocol of cimetidine vs. no cimetidine in another 60 patients. Forty clinical, personal, physiologic, and endoscopic characteristics were prospectively obtained in each patient, and were analyzed by stepwise discriminant analysis at the end of phase 1. This identified the discriminants against healing after 2 wk of cimetidine as late onset disease, body weight, and ulcer diameter; those after 4 wk of cimetidine as analgesic consumption, neurosis, low fasting serum gastrin, low pentagastrin D50 and ulcer diameter; and those after 4 wk of placebo as back pain, bleeding, and alcohol consumption. Based on the discriminant scores derived, the sensitivity, specificity, and efficiency of prediction for complete healing as determined endoscopically were 74.4%, 90%, and 82.3% for 2-wk cimetidine, 100%, 87.5%, and 97.5% for 4-wk cimetidine, and 85.7%, 83.3%, and 84.2% for 4-wk placebo treatment. In phase 2, correct predictions were made in 36 of 40, 39 of 40, and 17 of 20 patients treated, respectively, for 2 and 4 wk with cimetidine, and 4 wk without cimetidine. Accurate prediction of duodenal-ulcer healing rate with and without cimetidine is thus possible by discriminant analysis. As many medical and surgical modalities of treatment are now available, this approach should have the potential of selecting the appropriate form of treatment for a given patient.