Machine Learning Algorithms for Objective Remission and Clinical Outcomes with Thiopurines

• Published in: Journal of Crohn's and colitis 2017-07, Vol.11 (7), p.801-810
• Main Authors: Waljee, Akbar K, Sauder, Kay, Patel, Anand, Segar, Sandeep, Liu, Boang, Zhang, Yiwei, Zhu, Ji, Stidham, Ryan W, Balis, Ulysses, Higgins, Peter D R
• Format: Article
• Language: English
• Subjects: Adolescent; Adult; Algorithms; Area Under Curve; Azathioprine - metabolism; Azathioprine - therapeutic use; Drug Prescriptions; Female; Hospitalization; Humans; Immunosuppressive Agents - therapeutic use; Inflammatory Bowel Diseases - drug therapy; Inflammatory Bowel Diseases - surgery; Machine Learning; Male; Medication Adherence; Mercaptopurine - analogs & derivatives; Mercaptopurine - metabolism; Mercaptopurine - therapeutic use; Middle Aged; Original; Remission Induction; Retrospective Studies; ROC Curve; Treatment Outcome; Young Adult
• ISSN: 1873-9946; 1876-4479
• DOI: 10.1093/ecco-jcc/jjx014

Big data analytics leverage patterns in data to harvest valuable information, but are rarely implemented in clinical care. Optimising thiopurine therapy for inflammatory bowel disease [IBD] has proved difficult. Current methods using 6-thioguanine nucleotide [6-TGN] metabolites have failed in randomized controlled trials [RCTs], and have not been used to predict objective remission [OR]. Our aims were to: 1) develop machine learning algorithms [MLA] using laboratory values and age to identify patients in objective remission on thiopurines; and 2) determine whether achieving algorithm-predicted objective remission resulted in fewer clinical events per year. Objective remission was defined as the absence of objective evidence of intestinal inflammation. MLAs were developed to predict three outcomes: objective remission, non-adherence, and preferential shunting to 6-methylmercaptopurine [6-MMP]. The performance of the algorithms was evaluated using the area under the receiver operating characteristic curve [AuROC]. Clinical event rates of new steroid prescriptions, hospitalisations, and abdominal surgeries were measured. Retrospective review was performed on medical records of 1080 IBD patients on thiopurines. The AuROC for algorithm-predicted remission in the validation set was 0.79 vs 0.49 for 6-TGN. The mean number of clinical events per year in patients with sustained algorithm-predicted remission [APR] was 1.08 vs 3.95 in those that did not have sustained APR [p < 1 x 10-5]. Reductions in the individual endpoints of steroid prescriptions/year [-1.63, p < 1 x 10-5], hospitalisations/year [-1.05, p < 1 x 10-5], and surgeries/year [-0.19, p = 0.065] were seen with algorithm-predicted remission. A machine learning algorithm was able to identify IBD patients on thiopurines with algorithm-predicted objective remission, a state associated with significant clinical benefits, including decreased steroid prescriptions, hospitalisations, and surgeries.