Prediction of blood glucose level based on lipid profile and blood pressure using multiple linear regression model

• Main Author: Qurratu 'Aini Aishah, Ahmad Fazil
• Format: Thesis
• Language: English
• Published: 2021
• Subjects: TA Engineering (General). Civil engineering (General); TJ Mechanical engineering and machinery
• Online Access: http://umpir.ump.edu.my/id/eprint/34923/1/Prediction%20of%20blood%20glucose%20level%20based%20on%20lipid%20profile%20and%20blood%20pressure.ir.pdf; http://umpir.ump.edu.my/id/eprint/34923/

Type 2 diabetes mellitus (T2DM) refers to the inability to produce or respond to insulin, resulting in an elevated blood glucose level in the human body. Due to concerns over current diabetes screening and diagnostic procedures that require fasting, oral glucose consumption, and invasive nature (finger prick), the number of undiagnosed T2DM increases yearly. The increase is due to the hesitation of individuals to undergo screening tests as their routine check-up. As T2DM is closely related to blood glucose levels, a predictive model is developed to predict blood glucose levels, which can be used as an alternative for screening T2DM. Thus, the present study proposed a multiple linear regression equation for predicting the fasting blood glucose level based on independent parameters of lipid profile and blood pressure. It is widely known that high blood cholesterol and high blood pressure are the risk factors of T2DM. In this study, a set of 302 data was collected from UMP's retrospective data via the data directory of the University Health Centre from 2017 to 2018. The present study used 211 (70%) data to fit the predictive model, whereas another 91 (30%) of the data were used for selfvalidation of the model. Moreover, the overall model performance was observed by refitting the whole data set (n = 302, 100%) into the predictive model equation. The main outcome of the study showed that 46.8% (adjusted R2= 0.468, p-value < 0.05) of the fasting blood glucose level could be predicted using multiple linear regression based on high-density lipoprotein cholesterol, triglycerides, and systolic blood pressure levels without the standard fasting procedure. The prediction made by this model is acceptable with moderate accuracy (MAPE = 9.46%). This predictive model is easily adaptable to data changes (the difference of error metric values between the training data and testing data: MAE = 0.1836 mmol/L, RMSE = 0.1040 mmol/L, and MAPE = 3.93%). Thus, in order to increase the accuracy of the model, future research should consider a bigger and broader cohort from different comorbidities, which can be an alternative method in screening T2DM.