Development of Logistic Regression–Based Equations to Predict Cancer Diagnosis, Stage, and Survival from Routine Clinical and Laboratory Data

Abstract

The prediction of cancer diagnosis, stage, and survival early in the course of disease using clinical and laboratory characteristics easily available at diagnosis can aid in decision-making in resource-limited settings. In this study, we presented several Logistic Regression Equations to estimate the Probability of Cancer Diagnosis, Survival, and Tumor Stage from Demographic, Hematological, Metabolic, and Tumor Marker variables. We looked at the blood samples of the cancer patients (N=1000) with the help of standard blood tests (CBC, blood glucose) and tumor markers (CA-125, PSA, CEA), along with the type of cancer, stage, treatment outcome, and survival status of the individual. The author fitted binary logistic regression for modelling cancer diagnosis and survival, and multinomial logistic regression for modelling tumour stage. The diagnostic model generated an explicit equation for the log-odds of cancer; the survival model generated an analogous equation for the probability of survival; and the multinomial model generated class-specific equations for tumor stage. Routine markers, including hemoglobin and CA-125, made valuable contributions to prediction and are in line with prior reports on blood-test–based cancer modeling. Logistic regression models based on standard clinical and laboratory variables can easily be expressed as simple mathematical equations. Such an equation can also be used to create transparent figures (ROC, calibration, and probability profiles). These tools can be used in risk stratification, especially where resources are low.