Assessment of the Safety and Efficacy of a Novel Machine-learning Algorithm in the Management of Glycaemic Control in Insulin-Dependent Diabetics: A Randomised Controlled Clinical Trial

23.1 Title of the research project

Assessment of the Safety and Efficacy of a Novel Machine-learning Algorithm in the Management of Glycaemic Control in Insulin-Dependent Diabetics: A Randomised Controlled Clinical Trial

23.2 Aim and objectives of the research project

AIM: To evaluate the safety and efficacy of a novel machine learning model for insulin dosage estimation in patients that use insulin as the main drug for diabetes management.

Objectives: 

• To assess the safety profile of the novel machine-learning algorithm
• To evaluate the glycaemic control achieved while using the algorithm
• To compare the reduction in HbA1c reduction attained 
• To measure patient satisfaction on using the application

23.3 Introduction and need for the research

The burden of diabetes, both globally, and particularly in India, is huge – is an understatement. Diabetes, once considered a disease of affluence, has now become a significant public health challenge in India. With its burgeoning population, diverse cultural practices, and rapid urbanization, India finds itself grappling with an alarming increase in diabetes cases. This silent epidemic is not only a threat to individual health but also poses substantial economic burdens and strains on the healthcare system.

This innovative invention aims to mitigate some of these challenges. Using machine learning models, we have developed a system whereby individuals can simply use an app-based interface which, when fed data related to their pre-prandial capillary blood glucose readings, can estimate the required insulin dosage to be administered by the individual, such that the dosage is:- 

• adequate to prevent blood glucose spikes post-meal, thereby preventing complications associated with hyperglycaemia
• safe to prevent any hypoglycaemic symptoms (provided the recommended calorie intake is ensured); 
• personalised to each individual by relying on their own glucose and insulin levels as feed-data for machine learning. This obsoletes the need to follow standard reference insulin charts, which is the current norm.

23.4 Methodology and research design

This will be a randomized investigator-blind controlled clinical trial. Patients meeting the inclusion criteria will be recruited and randomized into a trial group and a control group. The participants in the trial group will have their insulin dosages estimated by the machine learning algorithm, while the control group participants will have their insulin dosages predefined by the treating physicians, following contemporary clinical practices. Safety profile will be estimated on a real-time basis using a mobile application and efficacy will be evaluated based on Hba1c reduction in both groups.

23.5 Inclusion criteria

• Patients aged 18 years and above, but below 70 years of age
• Clinically diagnosed T2DM
• Patients already on insulin therapy as part of their DM therapy 

23.6 Exclusion criteria

• Known hypersensitivity to insulin
• Pregnant or lactating women
• Patients above 70 years of age, or below 18 years of age.
• Comorbidities – Chronic Kidney Disease
• T1DM patients on insulin therapy

23.7 Sample size, sampling technique and statistical analyses

Sample Size: 60, divided into two equal arms.

Sampling Technique: Non-probability convenience sampling

Statistical Analysis: Descriptive and Correlation Statistics

Software for Statistical Analysis: IBM SPSS

23.8 Potential risks and benefits

Potential Risks: Hypoglycaemia, Hyperglycaemia, Diabetic Ketoacidosis

Potential Benefits: Reduced hospital visits with better overall diabetic control within the comforts of one’s home.

23.9 Expected outcome 

The primary outcome will be the correlation of the Hba1c (glycosylated haemoglobin) levels of patients in the arm of the study that used the mobile application with the control group that does not use the app.

The secondary outcomes will be determine other demographic factors that correlate with the reduction in the Hba1c levels in the two arms of the study.

We expect that the mobile application will demonstrate superior performance and achieve a greater and stricter control of blood sugar in the population that uses it and will be evidenced by the greater reduction in the Hba1c level in that group. Thus, this study, if successful, would establish the safety, efficacy, and public utility of the mobile application for general use among the public diagnosed with insulin-dependent type-2 diabetes.

23.10 Limitations of the study

1. Small sample size may suffer from selection bias due to non-probability convenience sampling methodology used
2. Participant usage of the mobile application may not be regular resulting in sampling loss and thereby, affecting the overall performance of the app
3. Generalisability of the findings may be limited to healthy diabetics without added comorbidities such as nephropathy or neuropathy.

23.13 References

1. Inzucchi SE, Bergenstal RM, Buse JB. Management of hyperglycemia in type 2 diabetes, 2015: a patient-centered approach: update to a position statement of the American Diabetes Association and the …. Diabetes [Internet] 2015;Available from: https://diabetesjournals.org/care/article-abstract/38/1/140/37869
2. Owens DR, Monnier L, Barnett AH. Future challenges and therapeutic opportunities in type 2 diabetes: Changing the paradigm of current therapy. Diabetes Obes Metab 2017;19(10):1339–52.
3. Ritzel R, Roussel R, Giaccari A, Vora J, Brulle-Wohlhueter C, Yki-Järvinen H. Better glycaemic control and less hypoglycaemia with insulin glargine 300 U/mL vs glargine 100 U/mL: 1-year patient-level meta-analysis of the EDITION clinical studies in people with type 2 diabetes. Diabetes Obes Metab 2018;20(3):541–8.