Comparison of the logistic regression and Cox regression model based on summer research with Memorial University
Abstract
In some epidemiological studies, the response measurement of interest is the time to occurrence of an event; however, the model of choice is the logistic regression model. In logistic regression model, the response variable is binary unlike survival regression models such as Cox model, where the response variable is the time-to-event.This talk aims to show that the logistic regression model and the Cox model may yield different results based on an application of a real data. A study conducted by Caplehorn and Bell in 1991 explored the use of methadone in a cohort of 238 heroin addicts, who were enrolled in either one of the two clinics (Clinic 1 or 2) offering methadone maintenance treatment. The two clinics differ in their live-in policies for patients and we would like to investigate which clinic is better at retaining individuals. The logistic regression model was used to assess the association of type of clinic with treatment cessation. Only the type of clinic and level of drug dose were obtained as associated with treatment cessation. The Cox regression model was used to study the effect of the type of clinic on the retention time of the clinic while adjusting for other significant factors, level of drug dose and prison record. Results from the Cox model with time-varying effects revealed no significant difference between the two clinics in retention times in the first 300 days after admission, but Clinic 2 tends to have longer retention times after 300 days.
