Learn-As-you-GO (LAGO): An innovative adaptive design for multi-component intervention studies in cardiology and public health - The use of complex, multi-component interventions (CMCIs) is an increasingly important aspect of cardiovascular disease prevention, screening, and treatment. For example, this application’s co-Investigator Longenecker is the principal investigator of a CMCI trial nearing completion aimed at improving blood pressure control, EXTRA-CVD (XCVD), consisting of 4 components: 1. Nurse-led care coordination, 2. Nurse-managed medication protocols and adherence counseling 3. Home blood pressure (BP) monitoring, and 4. Electronic health record (EHR) support tools, to be compared with generic prevention education. Each component has a “dose”, e.g. number of days/week home BP should be recorded and reported, and number and duration of adherence counseling sessions. Implementation scientists, such as those conducting this trial, discuss the tension between fidelity to the original intervention protocol, and the need for tailoring, tweaking and adaptation, perhaps contextually driven, i.e. varying by facility size, composition of the provider workforce, or health status of the patient population served. Under our innovative Learn-As-you-GO (LAGO) design, as the trial evolves, the doses of these intervention components are adapted at pre-specified stages to maximize cost-effectiveness and reduce the ultimate risk of trial failure by achieving pre-specified statistical power, while at the same time preserving the nominal size of the hypothesis test for the overall intervention package effect and attaining a pre-planned study outcome goal, such as attainment of 80% of patients under blood pressure control. Because each intervention component is associated with different costs and effectiveness, it is difficult to specify the optimal intervention package, that is, the optimal intervention component “doses” along with the components themselves, before launching a trial, as standard methods require. LAGO designs for continuous outcomes, such as changes in blood pressure (mmHg) or cholesterol levels (mg/dL) as in XCVD; for repeated binary outcomes, such as per visit hypertension control; and which account for clustering of outcome rates within centers are not available. In this project, following on our 2021 Annals of Statistics publication establishing fundaments for logistic regression analysis of a single binary outcome in non-clustered data, we will derive the mathematical theory for LAGO designs for clustered binary and continuous repeated measures data, and compare the LAGO design to its standard non-adaptive factorial and two-armed alternatives where the intervention is fixed before the study commences. Once a LAGO study concludes, LAGO provides a means to design interventions for new centers, scaling up and out, subject to the same class of goals, perhaps contextually dependent. Methods will be applied to XCVD and PULESA-Uganda, another active CMCI trial led by co-Investigator Longenecker for which contact PI Spiegelman serves as study statistician, allowing for this trial to serve as a living laboratory for LAGO methods development. User-friendly, publicly available software will be produced and disseminated, to facilitate use of these designs by practitioners.
