Propensity skóre v biomedicíně

Abstract

This thesis deals with the analysis of an observational study of patients who underwent colorectal surgery in the years 2001-2009 at the Hospital Ostrava. Different uses of propensity scores are analysed on these data. Propensity score is estimated in this thesis using logistic regression model and boosted model and then compared. The balance assessment of the dataset by the propensity score methods (matching, weighting) are assessed by standardized difference between groups, hypothesis testing and graphical visualization of the balance of each variable. By balancing the data using propensity scores, the dataset should match the dataset we would expect from a randomised trial study. In the original dataset, we evaluated the survival time between the groups of patients operated on by laparoscopic and open methods using the estimated survival function. Patient survival time is examined on the balanced dataset compared to the original data. Using a Cox model, the risk ratios of patients with respect to each variable are estimated. Risk ratios related to the choice of surgical technique are compared on the balanced datasets compared to the original data.