Design of predictive models for positive outcomes of upper and lower gastrointestinal endoscopies in children and adolescents.

OBJECTIVES

To develop models to accurately determine the outcomes of diagnostic endoscopies performed in children and adolescents without known gastrointestinal disease.

MATERIALS AND

METHODS

Retrospective chart review of all endoscopies performed in children 2 to 18 years of age without known gastrointestinal disease from January 1 to December 31, 2000. The association between age, presenting symptoms, physical examination findings, laboratory investigations, and endoscopy outcomes was assessed. Predictive models for positive outcomes on endoscopy were estimated for upper and lower endoscopies separately by use of multiple logistic regression. Receiver operating curves were constructed to evaluate the performance of the models. A model with a sensitivity of 95% and specificity of 40% was considered clinically significant.

RESULTS

Positive findings on endoscopy were found in 191 (55%) of 346 and in 120 (59%) of 204 upper and lower endoscopies, respectively. Age above 13 years, vomiting, and hypoalbuminemia were significant predictors of positive upper endoscopies. Rectal bleeding, hypoalbuminemia, and elevated erythrocyte sedimentation rate were significant predictors of positive lower endoscopies. Extrapolating from the receiver operating curves, a sensitivity of 95% corresponded to a specificity of 10% for the upper endoscopy model and 30% for the lower endoscopy model.

CONCLUSIONS

In our population of children and adolescents, several clinical characteristics were predictive of positive upper and lower endoscopy outcomes. Predictive models composed of these clinical variables were statistically, but not clinically, significant. The inclusion of additional clinical characteristics that could be assessed in prospective studies will likely improve the clinical significance of endoscopy outcome prediction.