Predicting Success with a First-Generation Hybrid Closed-Loop Artificial Pancreas System Among Children, Adolescents, and Young Adults with Type 1 Diabetes: A Model Development and Validation Study

Abstract

Background: Hybrid Closed-Loop (HCL) systems aid individuals with type 1 diabetes in improving glycemic control; however, sustained use over time has not been consistent for all users. This study developed and validated prognostic models for successful 12-month use of the first commercial HCL system based on baseline and 1- or 3-month data. Methods and Materials: Data from participants at the Barbara Davis Center (N = 85) who began use of the MiniMed 670G HCL were used to develop prognostic models using logistic regression and Lasso model selection. Candidate factors included sex, age, duration of diabetes, baseline hemoglobin A1c (HbA1c), race, ethnicity, insurance status, history of insulin pump and continuous glucose monitor use, 1- or 3-month Auto Mode use, boluses per day, and time in range (TIR; 70-180 mg/dL), and scores on behavioral questionnaires. Successful use of HCL was predefined as Auto Mode use ≥60%. The 3-month model was then externally validated against a sample from Stanford University (N = 55). Results: Factors in the final model included baseline HbA1c, sex, ethnicity, 1- or 3-month Auto Mode use, Boluses per Day, and TIR. The 1- and 3-month prognostic models had very good predictive ability with area under the curve values of 0.894 and 0.900, respectively. External validity was acceptable with an area under the curve of 0.717. Conclusions: Our prognostic models use clinically accessible baseline and early device-use factors to identify risk for failure to succeed with 670G HCL technology. These models may be useful to develop targeted interventions to promote success with new technologies.

Keywords: Artificial pancreas; Hybrid closed loop; Pediatric diabetes; Predictive modeling; Type 1 diabetes.

FIG. 1.

Predictive model with BDC data and prognostic parameters. (A) Model development with 1-month predictors. (B) Model development with 3-month predictors. BDC, Barbara Davis Center for Childhood Diabetes.

FIG. 2.

Validation of predictive model with Stanford data using 3-month predictors. (A) Validation model area under the curve. (B) Area under the curve comparison between the BDC development cohort and Stanford validation cohort.