Comparing artificial pancreas controlled by hybrid 'closed-loop' machine learning (ML) trained algorithm to multi-daily injection (MDI), insulin pump without CGM and 'sensor assisted' insulin pump therapies for Diabetes Type 1 (DT1) treatment

摘要

Aims/hypothesis Diabetes Type 1 (DT1) therapy by means of artificial pancreas consisting on insulin pump with CGM and hybrid "closed-loop" control algorithm trained with machine learning technology provides better glycemia control than multi-daily injection, insulin pump without CGM and "sensor assisted" insulin pump therapies. Methods Using Accu-Chek smart pix software to analyze the own data collected in-vivo by JC Peiro, author and DT1 patient, in the period August 2004 to August 2019, collecting glycemia control data using the different therapies. Accu-Chek smart pix has been used to collect 4.621 glycemia tests for a period of 1.241 days. The period measured with CGM contains data with +90% sensor coverage. Control graphics measure mean and median glycemia, standard deviation, time in range %, above range%, below range %, % hypoglycemia, high glucose blood index and low glucose blood index. Finally, analysis is validated in-silico using the UVA/Padova T1DMS simulator, with MATLAB and a population of 30 individuals under Multi-daily Injection and under "closed-loop" artificial pancreas therapies. Results The improvement of therapy with artificial pancreas with hybrid "closed-loop" control algorithm, when comparing to MDI reduces -70.7% the periods above range, reduces -67.2% periods below range, TIR% increases 75% reaching 84%, hypoglycemia's % reduced -91.2%, HGBI reduces -67%, LGBI reduces -73.8% reaching 2.3 and 1.6 respectively, reducing the mean glycemia in -16% reaching mean of 121mg/dL with SD 42mg/dL, reducing median glycemia in -20% reaching median of 115 mg/dL with SD 55 mg/dL and reducing the Glycated Hemoglobin (HbA1c) -20.1% reaching a 6.6% value. Conclusions/interpretation We can conclude that therapy for DT1 using "hybrid closed-loop" artificial pancreas controlled by ML trained algorithm provides statistically better glycemic control results. The improvement of the treatment is significantly better regarding all the analyzed therapies: MDI, insulin pump without CGM and "sensor assisted" insulin pump.