“Learn-as-you-go” (LAGO) Designs

Together with Drs. Judith Lok, Daniel Nevo, and I, Daniel Nevo has been working on developing methodology for a novel study design: LAGO. LAGO is a multi-stage, multi-center design that simultaneously identifies the optimal intervention as well as estimates its impact. Studies with a LAGO design are carried out in stages, and after each stage, the results collected so far are analyzed, the intervention package is reassessed, and a revised version is put to use in the next stage. LAGO provides rigorous methods to inherently adapt to local, changing contexts, which is the usual setting in public health. LAGO aims to reduce chances of implementation failure for interventions whose efficacy has been proven in a smaller scale.

The statistical complication in analyzing LAGO design is that intervention in later stages depend on earlier stage outcomes. We have provided conditions under which a classical logistic regression analysis can be used. The main two conditions are that current stage data is independent of past data conditioned on the current intervention and that the recommended intervention converges to a limit as the per-stage data increase. We provide a 95% confidence set for the optimal intervention. We investigated the finite sample performance of our new methodology in simulations. We are finalizing a first manuscript describing the LAGO design, and have been exploring multiple collaborative opportunities to enhance the use of LAGO and to explore additional arising methodologic challenges. Drs Judith Lok, Daniel Nevo, and I all presented talks to disseminate our novel design.