Applications of propensity score in observational studies in health sciences

The utilization of propensity score methods has seen a marked increase in observational studies in recent times. However, there is a lack of clear information about which method researchers should use in which data set. This thesis study aims to address this knowledge gap by examining the areas of use, advantages, and disadvantages of different propensity score methods. Additionally, it investigates the factors affecting method selection. To this end, a data set comprising 239 patients who underwent partial nephrectomy and obtained from the field of Urology, as well as the "Osteoarthritis" data set containing data from osteoporosis patients, which is openly available in the R program, were utilized for the study. To evaluate the effect of the treatment method on the outcome in the data, different scenarios were created. The impact of the binary or continuous nature of the outcome variable, the variation in sample size, and the alteration in response rate within the binary outcome variable on the efficacy of propensity score methods in estimating treatment effects was examined. The analyses performed revealed that, while the matching method is the most effective method in eliminating the imbalance caused by confounding variables, its success in estimating the treatment effect decreases in small sample sizes (high standard error and low classification rate). Conversely, the inverse probability weighting method demonstrated consistent and satisfactory performance across all scenarios. The stratification method merits particular attention for its capacity to utilize the entire population and attain a high area under the curve value. However, it is important to note that wide confidence intervals were identified as a notable drawback. Regression adjustment demonstrated notable efficacy, particularly in continuous outcome variables, and exhibited high classification success in low sample sizes. Consequently, as opposed to the most prominent and frequently favored the propensity matching method, it is advised to prefer alternative propensity score methods depending on the structural characteristics of the data and the objective of the study. Among these methods, if the outcome variable is binary, the inverse probability weighting or stratification methods should be employed. If the outcome variable is continuous, the regression adjustment or matching methods should be used to enhance the scientific value of the study by increasing the success of prediction.