EMPIRICAL COMPARISON OF THE STATISTICAL METHODS OF ANALYZING INTERVENTION EFFECTS AND CORRELATION ANALYSIS BETWEEN CLINICAL OUTCOMES AND SURROGATE COMPOSITE SCORES IN RANDOMIZED CONTROLLED TRIALS USING COMPETE III TRIAL DATA
Date of Award
Master of Science (MSc)
Mathematics and Statistics
Roman Viveros-Aguilera, Gary Foster
Roman Viveros-Aguilera, Gary Foster
Background: A better application of evidence-based available therapies and optimal patient care are suggested to have a positive association with patient outcomes for cardiovascular disease (CVD) patients. Electronic integration of care tested in the Computerization of Medical Practices for the Enhancement of Therapeutic Effectiveness (COMPETE) Π study showed that a shared electronic decision-support system to support the primary care of diabetes improved the process of care and some clinical markers of the quality of diabetes care. On the basis of COMPETE Π trial, COMPETE Ш study showed that older adults at increased risk of cardiovascular events, if connected with their family physicians and other providers via an electronic network sharing an intensive, individualized cardiovascular tracking, advice and support program, enhanced their process of care – using a process composite score to lower their cardiovascular risk more than those in conventional care. However, results of the effect of intervention on composite process and clinical outcomes were not similar – there was no significant effect on clinical outcomes.
Objectives: Our objectives were to investigate the robustness of the results based the commonly used statistical models using COMPETE III dataset and explore the validity of the surrogate process composite score using a correlation analysis between the clinical outcomes and process composite score.
Methods: Generalized estimating equations (GEE) were used as a primary statistical model in this study. Three patient-level statistical methods (simple linear regression, fixed-effects regression, and mixed-effects regression) and two center-level statistical approaches (center-level fixed-effects model and center-level random-effects model) were compared to reference GEE model in terms of the robustness of the results – magnitude, direction and statistical significance of the estimated effects on the change of process composite score / on-target clinical composite score. GEE was also used to investigate thecorrelation between the clinical outcomes and surrogate process composite scores.
Results: All six statistical models used in this study produced robust estimates of intervention effect. No significant association between cardiovascular events and on-target clinical composite score and individual component of on-target clinical composite score were found between the intervention group and control group. However, blood pressure, LDL cholesterol, and psychosocial index are significant predictors of cardiovascular events. Process composite score can both predict the cardiovascular events and clinical improvement, but the results were not statistically significant- possibly due to the small number of events. However, the process composite score was significantly associated with the on-target clinical composite score.
Conclusions: We concluded that all five analytic models yielded similar robust estimation of intervention effect comparing to the reference GEE model. The relatively smaller estimate effects in the center-level fixed-effects model suggest that the within-center variation should be considered in the analysis of multicenter RCTs. Process composite score may serve as a good predictor for CVD outcomes.
Xu, Jian-Yi, "EMPIRICAL COMPARISON OF THE STATISTICAL METHODS OF ANALYZING INTERVENTION EFFECTS AND CORRELATION ANALYSIS BETWEEN CLINICAL OUTCOMES AND SURROGATE COMPOSITE SCORES IN RANDOMIZED CONTROLLED TRIALS USING COMPETE III TRIAL DATA" (2011). Open Access Dissertations and Theses. Paper 6431.
McMaster University Library