Background: Decision tree classification is a standard machine learning technique that has been used for a wide range of applications. Patients with inflammatory bowel disease (IBD) are at increased risk of developing low bone mineral density (BMD). This study aimed at developing a new approach to select truly affected IBD patients who are indicated for densitometry, hence, subjecting fewer patients for bone densitometry and reducing expenses.
Materials and methods: Simple decision trees have been developed by means of WEKA (Waikato Environment for Knowledge Analysis) package of machine learning algorithms to predict factors influencing the bone density among IBD patients. The BMD status was the outcome variable whereas age, sex, duration of disease, smoking status, corticosteroid use, oral contraceptive use, calcium or vitamin D supplementation, menstruation, milk abstinence, BMI, and levels of calcium, phosphorous, alkaline phosphatase, and 25-OH vitamin D were all attributes.
Results: Testing showed the decision trees to have sensitivities of 65.7-82.8%, specificities of 95.2-96.3%, accuracies of 86.2-89.8%, and Matthews correlation coefficients of 0.68-0.79. Smoking status was the most significant node (root) for ulcerative colitis and IBD-associated trees whereas calcium status was the root of Crohn's disease patients' decision tree.
Conclusion: BD specialists could use such decision trees to reduce substantially the number of patients referred for bone densitometry and potentially save resources.