Use of Big-Data Algorithms to Characterize Patients with T2D on Basal Insulin (BI) Who Add a Glucagon-Like Peptide-1 Receptor Agonist (GLP-1 RA) and Predict Their A1C Response

• Published in: Diabetes (New York, N.Y.) Vol. 67; no. Supplement_1
• Main Authors: ZIMMERMANN, ESTHER, LENART, ADAM, DA ROCHA FERNANDES, JOAO DIOGO, EGGERT, SARAH, RANTHE, MATTIS F.
• Format: Journal Article
• Language: English
• Published: 01.07.2018
• ISSN: 0012-1797; 1939-327X
• DOI: 10.2337/db18-100-LB

Machine learning allows extensive analysis of big complex data. This study had two aims: 1) characterize patients on BI who add a GLP-1RA and 2) identify predictors of ≥1% decline in A1C. Patients with T2D who were prescribed BI for ≥90 days but not GLP-1RA for 180 days beforehand (in the U.S. IBM Explorys database between 2010 and 2016) were included (N=80,019). For the A1C analysis, A1C readings ≤180 days before, and 180-360 days after initiating GLP-1RA were required (N=8731). Logistic regression with 23 pre-specified variables, and subsequent hypothesis-free machine learning models, with 155000 additional variables covering clinical, claims and billing data addressed both aims. GLP-1RA initiators were characterized by a BI duration of >180 days (vs. ≤180 days) estimated odds ratio (OR) 5.87 (95% CI: 5.49-6.27), receiving oral antidiabetic drugs(s) OR 1.70 (1.64-1.77) and co-medication(s) (both vs. none) OR 3.22 (2.96-3.50), a BMI >30 kg/m2 (vs. <30 kg/m2) OR 1.93 (1.84-2.03), age <75 years (vs. ≥75 years) OR 3.63 (3.37-3.92) and private insurance (vs. non-private) OR 2.2 (2.10-2.31). Variable selection via machine learning confirmed the importance of these variables. Baseline A1C was the only strong predictor of ≥1% decline in A1C, ORs (95% CI) compared with A1C <7% were 4.99 (3.29-7.57), 7.04 (4.77-10.39), 14.56 (9.98-21.24), 23.21 (15.92-33.85), 36.28 (25.05-52.54), 73.14 (50.32-106.32) for categories 7-<7.5, 7.5-<8, 8-<8.5, 8.5-<9, 9-<10, ≥10%, respectively. Machine learning, applying 155000 variables, confirmed the importance of baseline A1C. On average, patients who improved lowered A1C from 10.0% (interquartile range [IQR]: 8.6-11.0) to 7.7% (IQR 6.7-8.4). Patients with T2D on BI who added a GLP-1RA were likely to be <75 years old and had characteristics of progressed disease. Baseline A1C determined a ≥1% decline in A1C, suggesting patients on BI with high A1C would benefit from combination treatment with GLP-1RA. Disclosure E. Zimmermann: Employee; Self; Novo Nordisk A/S. Stock/Shareholder; Spouse/Partner; Novo Nordisk A/S. A. Lenart: Employee; Self; Novo Nordisk A/S. J. da Rocha Fernandes: Employee; Self; Novo Nordisk A/S, International Diabetes Federation. S. Eggert: Employee; Self; Novo Nordisk A/S. M.F. Ranthe: Employee; Self; Novo Nordisk A/S. Stock/Shareholder; Self; Novo Nordisk A/S.