STATISTICAL AND METHODOLOGICAL ISSUES IN EVALUATION OF INTEGRATED CARE PROGRAMS

Abstract

<p><strong>Background </strong></p> <p>Integrated care programs are collaborations to improve health services delivery for patients with multiple conditions.</p> <p><strong>Objectives</strong></p> <p>This thesis investigated three issues in evaluation of integrated care programs: (1) quantifying integration for integrated care programs, (2) analyzing integrated care programs with substantial non-compliance, and (3) assessing bias when evaluating integrated care programs under different non-compliant scenarios.</p> <p><strong>Methods</strong></p> <p>Project 1: We developed a method to quantity integration through service providers’ perception and expectation. For each provider, four integration scores were calculated. The properties of the scores were assessed.</p> <p>Project 2: A randomized controlled trial (RCT) compared the Children’s Treatment Network (CTN) with usual care on managing the children with complex conditions. To handle non-compliance, we employed the intention-to-treat (ITT), as-treated (AT), per-protocol (PP), and instrumental variable (IV) analyses. We also investigated propensity score (PS) methods to control for potential confounding.</p> <p>Project 3: Based on the CTN study, we simulated trials of different non-compliant scenarios. We then compared the ITT, AT, PP, IV, and complier average casual effect methods in analyzing the data. The results were compared by the bias of the estimate, mean square error, and 95% coverage.</p> <p><strong>Results and conclusions</strong></p> <p>Project 1: We demonstrated the proposed method in measuring integration and some of its properties. By bootstrapping analyses, we showed that the global integration score was robust. Our method has extended existing measures of integration and possesses a good extent of validity.</p> <p>Project 2: The CTN intervention was not significantly different from usual care on improving patients’ outcomes. The study highlighted some methodological challenges in evaluating integrated care programs in a RCT setting.</p> <p>Project 3: When an intervention had a moderate or large effect, the ITT analysis was considerably biased under non-compliance and alternative analyses could provide unbiased results. To minimize the bias, we make some recommendations for the choice of analyses under different scenarios.</p>