An Investigation On The Delay Of The Blood Glucose Level Forecast

Diabetes is a chronic disease affecting a large number of human population in developing and developed countries which involves elevated levels of blood glucose for prolonged periods.Patients with severe conditions require timely administration of insulin dose and cutoff to ensure blood glucose stay within safe levels.Consequently,accurate forecast of blood glucose data is a valuable tool to aid medical decision making in providing timely treatment.In the ideal scenario,an accurate forecast will be able to foretell hyperglycemic and hypoglycemic events and serve as an early warning system which enables swift execution of treatment to minimize threat and damage to the human body.To achieve this,a forecast of multiple time period is necessary.Blood glucose forecast methods has been reported by many different studies with primary focus on forecast accuracy.However,it is observed that in many reported results,there exist a delay when multiple period forecast is performed.Most studies did not focus on the forecast delay in the analysis and some studies used a subjective method to evaluate the forecast delay.It is apparent that there is lack of a comprehensive study which provides further understanding on forecast delay which has a direct impact on its practical application.Hence,in this study,an investigation is carried out to gain further insight in the delay of blood glucose level forecast.In order to achieve this,time series and machine learning method forecast methods were used,including a new method developed with the intention of reducing forecast delay.Forecasts were carried out using only blood glucose value of diabetes patients obtained from continuous glucose monitoring(CGM)systems.A standardize evaluation is carried out on actual diabetes patients' blood glucose data to rate the performance of different forecast methods in terms of accuracy and forecast delay.Furthermore,simulated data and processed data of actual patients were used to test the impact of different data characteristics on the performance of forecast delay.