Neural network based clinical decision support system for the calculation the initial continuous subcutaneous insulin infusion settings

BACKGROUND: Despite existing recommendations for the initial calculation of insulin pump settings, the process is largely subjective and depends on the physician’s personal experience.

AIM: Development of a clinical decision support system (CDSS) that determines the initial settings of the insulin pump, which would have satisfactory agreement with the expert opinion of physicians.

MATERIALS AND METHODS: Neural network model developed using data (continuous subcutaneous insulin infusion (CSII) settings, age, weight, total daily dose, and HbA1c) from 2850 children with T1D who were switched to CSII and achieved optimal glycemic control according to glucose levels. CDSS utilizing the model implemented as a computer program in Python.

A prospective assessment of the agreement between the recommendations of the CDSS and the physician conducted on 35 data sets of children with T1D (median age 9.3 years [6.4, 11.5]), and 840 points for decisions were analyzed. 4 degrees of agreement were used: complete consistency, when the physicians agreed with the CDSS recommendations; partial consistency, when the physicians didn’t agree with the CDSS recommendations, but the difference was in the range of ±15%; complete inconsistency — the difference more than ±15%; acceptable consistency is the sum of full and partial consistency (± 15% error is clinically acceptable). The null hypothesis of the study was the absence of difference in consistency/inconsistency between physicians and the CDSS.

RESULTS: The frequency of full consistency between CDSS and physician recommendations for initiating insulin pump therapy is 29.8-43.8%, and inconsistency is 33.7-41.1%. Acceptable consistency is 58.9–66.3%. There were no significant differences in mean insulin pump parameters between CDSS and physicians.

CONCLUSION: The results obtained are consistent with previous studies. Proposed model demonstrates acceptable performance regarding initial CSII settings, without significant deviations between various parameters.

1. Walsh J, Roberts R, Bailey T. Guidelines for insulin dosing in continuous subcutaneous insulin infusion using new formulas from a retrospective study of individuals with optimal glucose levels. J Diabetes Sci Technol. 2010;4(5):1174-1181. https://doi.org/10.1177/193229681000400516

2. Laptev DN, Filippov YI, Emel’yanov AO, Kuraeva TL. Age-adjustment of insulin pump settings in children and adolescents with type 1 diabetes mellitus. Diabetes mellitus. 2013;16(3):109-115. (In Russ.) https://doi.org/10.14341/2072-0351-98

3. Bongiovanni M, Fresa R, Visalli N, et al. A Study of the Carbohydrate-to-Insulin Ratio in Pregnant Women with Type 1 Diabetes on Pump Treatment. Diabetes Technol Ther. 2016;18(6):360-365. https://doi.org/10.1089/dia.2015.0246

4. Streun GL, Elmiger MP, Dobay A, et al. A machine learning approach for handling big data produced by high resolution mass spectrometry after data independent acquisition of small molecules - Proof of concept study using an artificial neural network for sample classification. Drug Test Anal. 2020;12(6):836-845. https://doi.org/10.1002/dta.2775

5. Rong D, Xie L, Ying Y. Computer vision detection of foreign objects in walnuts using deep learning. Comput Electron Agric. 2019;162:1001–1010. https://dx.doi.org/10.1016/j.compag.2019.05.019

6. Holterhus PM, Odendahl R, Oesingmann S, et al. Classification of distinct baseline insulin infusion patterns in children and adolescents with type 1 diabetes on continuous subcutaneous insulin infusion therapy. Diabetes Care. 2007;30(3):568-573. https://doi.org/10.2337/dc06-2105

7. Encyclopedia of Information Science and Technology, Fourth Edition: / ed. Khosrow-Pour, D.B.A. M. IGI Global, 2018

8. Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: Machine learning in Python. J Mach Learn Res. 2011

9. Kingma DP, Ba J. Adam: A Method for Stochastic Optimization. 3rd Int Conf Learn Represent ICLR 2015 - Conf Track Proc. December 2014. http://arxiv.org/abs/1412.6980

10. Romanenkova EM, Eremina IA, Titovich EV, et al. C-peptide levels and the prevalence of islets autoantibodies in children with type 1 diabetes mellitus with different duration of the disease. Diabetes mellitus. 2022;25(2):155-165. (In Russ.) https://doi.org/10.14341/DM12843

11. Laptev DN, Bezlepkina OB, Demina ES, et al. Evaluation of FreeStyle Libre in pediatric t1dm: improved glycemic control, reduction in diabetic ketoacidosis and severe hypoglycemia. Problems of Endocrinology. 2022;68(3):86-92. (In Russ.)] https://doi.org/10.14341/probl12877

12. Laptev DN, Emelyanov AO, Andrianova EA, et al. The use of Flash glucose monitoring in children with type 1 diabetes mellitus in real clinical practice. Diabetes mellitus. 2021;24(6):504-510. (In Russ.) https://doi.org/10.14341/DM12817

13. Gajewska KA, Biesma R, Bennett K, Sreenan S. Barriers and facilitators to accessing insulin pump therapy by adults with type 1 diabetes mellitus: a qualitative study. Acta Diabetol. 2021;58(1):93-105. https://doi.org/10.1007/s00592-020-01595-5

14. Sherr JL, Schoelwer M, Dos Santos TJ, et al. ISPAD Clinical Practice Consensus Guidelines 2022: Diabetes technologies: Insulin delivery. Pediatr Diabetes. 2022;23(8):1406-1431. https://doi.org/10.1111/pedi.13421

15. Alemzadeh R, Hoffmann RG, Dasgupta M, Parton E. Development of optimal kids insulin dosing system formulas for young children with type 1 diabetes mellitus. Diabetes Technol Ther. 2012;14(5):418-422. https://doi.org/10.1089/dia.2011.0184

16. Bode BW, Sabbah HT, Gross TM, Fredrickson LP, Davidson PC. Diabetes management in the new millennium using insulin pump therapy. Diabetes Metab Res Rev. 2002;18 Suppl 1:S14-S20. https://doi.org/10.1002/dmrr.205

17. Nimri R, Dassau E, Segall T, et al. Adjusting insulin doses in patients with type 1 diabetes who use insulin pump and continuous glucose monitoring: Variations among countries and physicians. Diabetes Obes Metab. 2018;20(10):2458-2466. https://doi.org/10.1111/dom.13408

18. Rebrova OYu, Fedyaeva VK, Khachatryan GR. Adaptatsiya i validizatsiya voprosnika dlya otsenki riska sistematicheskikh oshibok v randomizirovannykh kontroliruemykh ispytaniyakh. Meditsinskie tekhnologii. Otsenka i vybor. 2015;19(1):9-17. (In Russ.)