Reassessing the Effectiveness of Right Heart Catherization (RHC) Procedure on the Short-Term Prognosis of Critically Ill Patients Using Targeted Learning: A Comparison of Targeting Maximum Likelihood Estimation with Propensity Score Matching Analysis
Abstract
Background: Target maximum likelihood estimation (TMLE) is an instrumental tool used in semiparametric or nonparametric models. One of the main advantages of TMLE is that it provides practitioners with targeted robust and very efficient unbiased estimators of parameter of interest for proposed scientific questions. To explore the usefulness of TMLE, we decided to assess a public health issue currently plaguing our community, which is the burden of heart disease. Heart disease includes multiple negative health outcomes which are related to atherosclerosis, which is a condition that develops when a substance called plaque builds up and narrows the walls of the arteries which can cause blood clot and hence can cause a heart attack or stroke. Right heart catherization (RHC) is a well-established procedure used in modern medicine as a diagnostic tool for patients with congenital and acquired right heart disease and to actively monitor patients in intensive care units with serious cardiovascular diseases. However, previous and current randomized controlled trials and other studies provide limited evidence that supports the clinical utilization of RHC in critically ill-patients’ management. This is a question that needs to be addressed because the advantages of RHC to the patients’ survival rates and better quality of life is one of the prime objectives of public health. Methods: To further investigate this method, we implemented TMLE method and Cox regression model using propensity score matching (PSM) analysis on a previously analyzed dataset for studies which investigated the effectiveness of treatment or diagnostic procedure on the outcome measures of participants using propensity score analysis. We used a prospective cohort SUPPORT study previously analyzed by Connors et al that assessed the effectiveness of right heart catherization in the initial care of critically ill patients from five US teaching hospitals between 1989 and 1994. Additionally, we used simulated dataset to assess the performance of the statistical methods. A range of covariates based on demographics, socioeconomic status, and clinical relevant variables were used to adjust for potential confounders. Results: Using the SUPPORT study, our findings showed that TMLE offers double robustness and offers protection against misspecification and mismodeling of acquiring survival estimates. The results show that there is a trend that suggests that participants without RHC had better survival outcomes than participants with RHC. The Cox model for propensity score matching also showed that the risk for mortality was greater for participants with RHC. Using the simulated dataset, the results from using TMLE method and PSM method in a Cox Regression Model, showed similar trends as seen in the SUPPORT dataset. Estimates from TMLE were like the true estimates from our known distribution of the simulated dataset, in comparison to the results produced from using PSM in a Cox Regression Model. Conclusion: TMLE can be used for investigating the relationship between outcomes and predictors with better estimates indicating precision and less bias due to non-assumptions of model selection. For future work, we plan to include treatment covariate interactions for investigating the relationship between survival outcomes and the set of predictors. TMLE should be implemented in studies and further research should investigate the benefits and restrictions this innovative method for the interpretation of epidemiological studies.
Subject Area
Biology|Public health
Recommended Citation
Akosile, Mary, "Reassessing the Effectiveness of Right Heart Catherization (RHC) Procedure on the Short-Term Prognosis of Critically Ill Patients Using Targeted Learning: A Comparison of Targeting Maximum Likelihood Estimation with Propensity Score Matching Analysis" (2017). Texas Medical Center Dissertations (via ProQuest). AAI10684927.
https://digitalcommons.library.tmc.edu/dissertations/AAI10684927