Applying Linear Regression Analysis Model to Predict HbA1C Values Using the Daily Estimated Average Glucose (eAG) Values from a 14-Month Period for a type 2 Diabetes Patient based on GH-Method: Math-Physical Medicine (No. 559)

Abstract

Since 1/1/2012, the author has been collecting various biomedical and lifestyle data related to his health conditions (~3 million data) which includes 4 categories of medical conditions, 4 chronic diseases consisting of obesity, diabetes (via finger-piercing glucose and HbA1C), hypertension, and hyperlipidemia (m1 through m4), along with 6 categories of lifestyle details, including exercise, water intake, sleep, stress, food, and daily life routines (m6 through m10). Starting on 1/1/2013, he accumulates 4 glucose data per day, one for fasting plasma glucose (FPG) and 3 for postprandial plasma glucose (PPG). In addition, beginning on 5/8/2018, his glucoses are automatically measured using a continuous glucose monitoring (CGM) sensor device to collect 96 glucose data per day. Over the past 12 years, he has tested for his HbA1C value each quarter since 2010. Through research work on type 2 diabetes (T2D) from 2015 to 2017, he developed several math-physical models to predict his HbA1C values prior to labtests for HbA1C at a clinic or hospital. He compares his predicted HbA1C values against the lab-tested HbA1C values and has achieved a 99% to 100% prediction accuracy. In this article, he investigates his predicted HbA1C values via the daily estimated average glucose (eAG) values using a linear regression model from the past 14-month period from 10/1/2020 to 11/24/2021. This report is part of a series in his research work based on statistical regression model for the month of November 2021. Several written regression articles were initiated from his body temperature (BT) which he started to measure on 10/1/2020. In order to compare some of his research results from this period, he selected the same 14-month window. In conclusion, the regression predicted HbA1C versus measured HbA1C, based on the daily glucose value, can be described by the following statistically derived equation: Regression predicted HbA1C (Y) = 0.0167*measured eAG (X) + 4.2327 This linear regression equation has the following conclusive key data: Correlation = 61% (not very high, but enough) Variance = 37% (sufficient) Significance F & p-value = 2% (<5%) Furthermore, when we put the 3 datasets (measure A1C, predicted A1C, and lab-tested A1C) together, we observe that all three average HbA1C values during this 14-month period are equal to 6.1, which achieved 100% prediction accuracy.