OBJECTIVES: The objective of this study is to cross-validate a new questionnaire designed to predict the future risk of NSAID-induced gastrointestinal events against a standard questionnaire, the GI SCORE survey. METHODS: Four-hundred consenting, consecutive patients from a rheumatology clinic in the mid-western United States were administered both questionnaires. The questionnaires were not administered in any particular order. The new questionnaire contains questions that are similar to five of six questions found in GI SCORE, in addition to five questions not found in GI SCORE. The completed GI SCORE questionnaires were scored and the associated risk levels were determined. Feasible generalized least squares (FGLS) and multinomial logistic (MNL) regression were used to map the questions from the new questionnaire onto the scores and two different groups of risk levels, respectively, determined from GI SCORE. Based on the results of the FGLS analysis, a scoring scheme was created for the new questionnaire, allowing the prediction of risk levels similar to GI SCORE. Risk levels generated from both FGLS and MNL were compared to those predicted by GI SCORE. RESULTS: For FGLS, the new questionnaire predicted risk levels that matched those predicted by GI SCORE with 83% accuracy. When the original 4 risk levels predicted by GI SCORE were collapsed into 3 risk levels, with the 2 most severe risk levels becoming a single risk level, the predictions from the new questionnaire were 89% accurate. For MNL, the new questionnaire was 76% and 87% accurate for four and three risk levels, respectively. CONCLUSIONS: The new questionnaire appears to be reasonably accurate in predicting the same risk as those predicted by a standard survey. FGLS is slightly more accurate than MNL in predicting risk for both groups of risk considered.