Penalized Logistic Regression Model to Predict a Results of RT-PCR by Using Blood Laboratory Test

Abstract

Statistical modelling to determine the effect of several predictor variables on the binary response variable is known as multiple logistic regression model. The addition of a penalty function to the model is done to improve prediction accuracy. Penalized logistic regression shrinks the regression coefficient to zero. This penalized logistic regression model will be used to predict a result of RT-PCR by using the features of blood laboratory tests. This research uses LASSO and elastic net penalties function. This study aims to determine the prediction performance of the RT-PCR test using logistic regression with LASSO and elastic net penalties. The data from the RT-PCR test were used as the binary response variable. Patient age quantile and 27 features of laboratory blood test were used as predictor variables. The results of this research showed that prediction performance of a RT-PCR test using LASSO logistic regression was better than elastic net logistic regression. The LASSO logistic regression model had a good performance for predicting the RT-PCR test with 88% accuracy and 93% AUC. Based on the result of LASSO logistic regression model, the features of laboratory blood tests that affect a RT-PCR test were leukocytes, basophils, RDW and C-reactive protein.