A Clinical Trial of TrenDx: An Automated Trend-Detection Program

We carried out a trial designed to assess the performance of a computer program called TrenDx at the task of diagnosing growth disorders in children, given a limited set of data. We compared the performance of TrenDx to that of physicians performing the same task. The task consisted of reviewing a growth chart of a patient and deciding whether the patient should be referred to a growth clinic for a possible growth problem, giving a preliminary diagnosis, and choosing the time which it would have been appropriate to refer the child. The test cases consisted of the height, weight, and bone-age data of 95 children that had been referred to the Boston Children’s Hospital. The patient cases were organized into packets of 10 and distributed to physicians. Twenty two (22) physicians participated. Two gold-standards were used, the medical record diagnosis and the opinion of a pediatric endocrinologist. A scoring algorithm was devised based on the responses of the pediatric endocrinologists. The performance of the experts compared to the medical record indicated that it was possible to accomplish the task that we had designed. When compared to the medical record diagnosis, TrenDx and the physicians performed similarly in terms of referring patients, but the physicians chose the correct diagnosis more often. Compared to the experts, the physicians performed better than TrenDx in terms of referral decision and score.