Nefeli Dimitropoulou1, Cassy Dingena1, Lisa den Brok1, Michelle van Oeteren1, Stavroula Mougiakakou2, Bastiaan de Galan1, Giesje Nefs3
1CARIM Cardiovascular Research Institute Maastricht, Maastricht University, Maastricht, the Netherlands
2ARTORG Centre of Biomedical Engineering Research, University of Bern, Switzerland.
3 Radboud University Medical Center, Department of Medical Psychology, 6500 HB
nefeli.dimitrtopoulou@maastrichtuniversity.nl
Background: People with diabetes on intensive insulin treatment often struggle to achieve recommended glycemic targets. Artificial intelligence (AI) driven bolus advisors are a promising tool for assisting self-care. Our aim was to develop and validate a questionnaire measuring the (expected) benefits and burdens of AI-driven bolus advisors in people with type 1 or type 2 diabetes, using intensive insulin treatment.
Methods: In this mixed method study, literature review and qualitative interviews (n=16), which were thematically coded, were performed for item development. A quantitative online survey was conducted to determine the psychometric characteristics, resulting to ‘Benefits and buRdens of AI-driveN bolus advisors’ (BrAIn) questionnaire.
Results: Based on the qualitative part of the study, the initial questionnaire draft comprised 24 questions and one overall item. In total, 262 people with type 1 (n=141) or type 2 (n=121) diabetes completed the survey (67% men, median age 63.0 [IQR: 53.0-72.0] years, diabetes duration 20 [IQR: 10-30] years, 19% on an insulin pump, 96% using glucose sensor). An exploratory factor analysis based on the initial 24 items with forced one-factor solution resulted in 21 items with factor loading ≥0.4 (explained variance 31.9%; Cronbach’s α=0.90). However, an unforced exploratory factor analysis yielded a three-factor solution (42.8% explained variance). Removing all items not meeting criteria for simple structure left one main factor representing “Expectations” (9 items; Cronbach’s a=0.82). In linear regression analysis, this factor “Expectations” was positively associated with the overall item (β=0.087, p <0.001), even after adjustment for diabetes type and duration.
Conclusion: The BrAIn questionnaire comprising 9 items and one overall item shows acceptable psychometric characteristics. It can be applied for measuring the (expected) benefits and burdens of AI-driven bolus advisors in people with type 1 or type 2 diabetes in both research and clinical practice.