Estimativa da creatinina sérica em adultos hígidos e renais crônicos através de técnicas de aprendizado de máquina utilizando variáveis não invasivas

Abstract

Chronic kidney disease (CKD) is an important public health problem. It is estimated that currently 11.7 to 15.1% of the world population is affected by this condition. This disease has impacts on the quality of life as it is associated with kidney failure and other adverse outcomes. CKD is not an easy-to-detect disease, and the barriers to making the diagnosis go beyond late symptoms, such as the scarcity of nephrologists and difficult access to specific tests. Estimation of glomerular filtration rate (GFR) through the use of creatinine is the most used method in clinical practice, but it is obtained by biochemical analysis of a blood sample. In this scenario, in order to facilitate the diagnosis of this disease, Machine Learning techniques have been used, as their algorithms are capable of learning and analyzing patterns and using them to solve specific demands in the health field. Thus, the objective of this study was to develop a mathematical model of high reliability and easy handling, capable of estimating the serum creatinine to determine the GFR, only with non-invasive and low cost indicators. 116 healthy adults and chronic renal patients participated in this study. To test the normality of the data, the Kolmogorov-Smirnov test was used in the SPSS® software. To estimate creatinine, the multiple linear regression method was used with the input characteristics: gender, age, SBP and DBP, and Pearson's correlation to compare the estimated and actual values of creatinine and GFR, in the MATLAB® program. Descriptive statistics of the sample were presented as absolute values for gender and ethnicity, and as mean and standard deviation (SD) for: age, BMI, height, weight, waist circumference, SBP, DBP, serum creatinine and GFR. The mathematical model used showed a strong correlation for both creatinine: r=0.72 with SD=0.14; and GFR: r=0.87 with SD=13.2. The computational model implemented in this study was efficient in estimating the serum creatinine, showing a strong correlation between the estimated and the real value, the same occurred for the GFR, but with better performance. Therefore, the software developed in this study is able to estimate renal function and can be a great ally for preventive measures and early treatment of CKD in the low-income population.