Please use this identifier to cite or link to this item:
http://hdl.handle.net/11375/20270
Title: | METHODOLOGICAL ISSUES IN PREDICTION MODELS AND DATA ANALYSES USING OBSERVATIONAL AND CLINICAL TRIAL DATA |
Authors: | LI, GUOWEI |
Advisor: | THABANE, LEHANA |
Department: | Clinical Epidemiology/Clinical Epidemiology & Biostatistics |
Keywords: | PREDICTION MODEL;OBSERVATIONAL DATA;CLINICAL TRIAL;DATA ANALYSIS |
Publication Date: | 2016 |
Abstract: | Background and objectives: Prediction models are useful tools in clinical practise by providing predictive estimates of outcome probabilities to aid in decision making. As biomedical research advances, concerns have been raised regarding combined effectiveness (benefit) and safety (harm) outcomes in a prediction model, while typically different prediction models only focus on predictions of separate outcomes. A second issue is that, evidence also reveals poor predictive accuracy in different populations and settings for some prediction models, requiring model calibration or redevelopment. A third issue in data analyses is whether the treatment effect estimates could be influenced by competing risk bias. If other events preclude the outcomes of interest, these events would compete with the outcomes and thus fundamentally change the probability of the outcomes of interest. Failure to recognize the existence of competing risk or to account for it may result in misleading conclusions in health research. Therefore in this thesis, we explored three methodological issues in prediction models and data analyses by: (1) developing and externally validating a prediction model for patients’ individual combined benefit and harm outcomes (stroke with no major bleeding, major bleeding with no stroke, neither event, or both stroke and major bleeding) with and without warfarin therapy for atrial fibrillation; (2) constructing a prediction model for hospital mortality in medical-surgical critically ill patients; and (3) performing a competing risk analysis to assess the efficacy of the low molecular weight heparin dalteparin versus unfractionated heparin in venous thromboembolism in medical-surgical critically ill patients. Methods: Project 1: Using the Kaiser Permanente Colorado (KPCO) anticoagulation management cohort in the Denver-Boulder metropolitan area of Colorado in the United States to include patients with AF who were and were not prescribed warfarin therapy, we used a new approach to build a prediction model of individual combined benefit and harm outcomes related to warfarin therapy (stroke with no major bleeding, major bleeding with no stroke, neither event, or both stroke and major bleeding) in patients with AF. We utilized a polytomous logistic regression (PLR) model to identify risk factors and then construct the new prediction model. Model performances and validation were evaluated systematically in the study. Project 2: We used data from a multicenter randomized controlled trial named Prophylaxis for Thromboembolism in Critical Care Trial (PROTECT) to develop a new prediction model for hospital mortality in critically ill medical-surgical patients receiving heparin thromboprophylaxis. We first identified risk factors independent of APACHE (Acute Physiology and Chronic Health Evaluation) II score for hospital mortality, and then combined the identified risk factors and APACHE II score to build the new prediction model. Model performances were compared between the new prediction model and the APACHE II score. Project 3: We re-analyzed the data from PROTECT to perform a sensitivity analysis based on a competing risk analysis to investigate the efficacy of dalteparin versus unfractionated heparin in preventing venous thromboembolism in medical-surgical critically ill patients, taking all-cause death as a competing risk for venous thromboembolism. Results from the competing risk analysis were compared with findings from the cause-specific analysis. Results and Conclusions: Project 1: The PLR model could simultaneously predict risk of individual combined benefit and harm outcomes in patients with and without warfarin therapy for AF. The prediction model was a good fit, had acceptable discrimination and calibration, and was internally and externally validated. Should this approach be validated in other patient populations, it has potential advantages over existing risk stratification approaches. Project 2: The new model combining other risk factors and APACHE II score was a good fit, well calibrated and internally validated. However, the discriminative ability of the prediction model was not satisfactory. Compared with the APACHE II score alone, the new prediction model increased data collection, was more complex but did not substantially improve discriminative ability. Project 3: The competing risk analysis yielded no significant effect of dalteparin compared with unfractionated heparin on proximal leg deep vein thromboses, but a lower risk of pulmonary embolism in critically ill medical-surgical patients. Findings from the competing risk analysis were similar to results from the cause-specific analysis. |
URI: | http://hdl.handle.net/11375/20270 |
Appears in Collections: | Open Access Dissertations and Theses |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
LI_GUOWEI_finalsubmission201608_PHD.pdf | 6.85 MB | Adobe PDF | View/Open |
Items in MacSphere are protected by copyright, with all rights reserved, unless otherwise indicated.