DEVELOPMENT OF PREDICTIVE MODELS FOR DIABETICS IN ROUTINE LIFE AND EMERGENCY SITUATIONS

The global healthcare systems have been in a transformation phase, moving away from a reactive to proactive approach of medical practices which rely more on engineering and technology. This transformation has not only been brought about due to improvements in measurement technology but also because of systems engineering tools such as modeling, optimization, and control. Particularly, in the recent years, diabetes care has been investigated from a systems perspective in order to shift the conventional treatment strategies towards a proactive mode. Diabetes management includes safer maintenance of blood glucose (BG) levels on a daily basis and also management of various biochemical variables during emergency hospital admissions attributed to diabetic complications. In both these cases, development of reliable models to predict the required biochemical outcomes is crucial. These models should have the ability to mimic the physiological conditions, employ practically feasible input and output measurements, and comprise variables that are of clinicians¿ interest. This thesis features the development of predictive models for subjects with type 1 diabetes (T1D) on their routine lifestyle, and for diabetic patients admitted in hospital due to a short term complication, called, Diabetic Ketoacidosis (DKA). The first part of the thesis will focus on the two different classes of personalized models developed for prediction of BG dynamics in T1Ds for exercise, meal and insulin interventions. The second part deals with the identification of definitive and intermediate outcomes of DKA using the clinical data retrieved from SGH. Multivariate statistical modeling tools like partial least squares are used for this purpose.

Author dc:creator

• NAVIYN PRABHU BALAKRISHNAN

• Diabetes, Mathematical Models, Identifiability Analysis, Parameter Estimation, Diabetic KetoAcidosis, Hybrid Models