Comparison of Logistic Regression versus Propensity Score When the Number of Events Is Low and There Are Multiple Confounders

Abstract

The aim of this study was to use Monte Carlo simulations to compare logistic regression with propensity scores in terms of bias, precision, empirical coverage probability, empirical power, and robustness when the number of events is low relative to the number of confounders. The authors simulated a cohort study and performed 252,480 trials. In the logistic regression, the bias decreased as the number of events per confounder increased. In the propensity score, the bias decreased as the strength of the association of the exposure with the outcome increased. Propensity scores produced estimates that were less biased, more robust, and more precise than the logistic regression estimates when there were seven or fewer events per confounder. The logistic regression empirical coverage probability increased as the number of events per confounder increased. The propensity score empirical coverage probability decreased after eight or more events per confounder. Overall, the propensity score exhibited more empirical power than logistic regression. Propensity scores are a good alternative to control for imbalances when there are seven or fewer events per confounder; however, empirical power could range from 35% to 60%. Logistic regression is the technique of choice when there are at least eight events per confounder.

Effect of marital status on survival in glioblastoma multiforme by demographics, education, economic factors, and insurance status

“…In addition to multivariate Cox regression, PSM has been proposed as another method to reduce the impact of patient selection bias on observational data and mimic randomized controlled trials 23, 24. PSM has been widely used in several areas of medical research including studies to assess factors associated with cancer survival 25, 26, 27, 28, 29.…”

Food Insufficiency and Health Services Utilization in a National Sample of Homeless Adults

“…Candidate covariates were chosen based on prior hypotheses and literature review. In order to obtain stable estimates of effect and avoid model overfitting, 44,45 only covariates associated with the respective outcomes at a univariate significance level of p<0.20 were included in the final multivariable models.…”

Anti–tumor necrosis factor α therapy and the risk of serious bacterial infections in elderly patients with rheumatoid arthritis

“…We therefore developed propensity score models for more efficient estimation (27,28). Propensity scores were estimated using logistic regression models predicting the probability of using a specific DMARD compared with MTX as a function of all potential confounders listed above (C statistics, a marker for predicting treatment choice, varied between 0.65 and 0.75).…”