The transfer of results from highly controlled CTs to a more generalizable community-based setting is called second stage translational research and is currently of great interest. By their nature RCTs are well monitored and highly non-representative of the way interventions are delivered in community-based settings. Because the randomization of participants to study arms is rarely feasible in community-based settings, non-randomized designs are often employed, with concomitant risk of imbalanced covariates creating bias that potentially undermines support for causality. Another related factor further complicates the rigorous evaluation of non- randomized interventions. Due to the logistics of dissemination, non-randomized designs often employ experimental units that are characterized by aggregate statistics, e.g. schools, rather than information from individual members, e.g. students. The combination of non-randomized designs and aggregate experimental units has seriously limited the evaluation of interventions administered in community-based settings. This has engendered uncertainty within the research community as to whether specific community-based interventions are truly effective. Statistical methods that rigorously evaluate the effects of interventions while controlling for bias are needed to address this pervasive concern.
Aim 1 will introduce the use of multivariate spatial methods to refine the evaluation of the Connecticut Collaboration for Fall Prevention (CCFP). CCFP was the first large- scale, longitudinal trial of a community-wide intervention designed to prevent injurious falls in older adults and demonstrated a rate of fall-related utilization from serious injury over a 5 year period that was 9.9% less in the intervention area relative to the usual care area.
In Aim 1 we simultaneously estimate the associations between the CCFP intervention and rates of head injury, hip fracture, and non-hip fracture.
In Aims 2 and 3 we analyze the Healthy Food Certification (HFC), a non-randomized community-based intervention that promoted adoption of voluntary state nutritional standards for all food sold in participating school districts in Connecticut. By exploiting the relative merits of propensity scores and spatial modeling, Aims 2 and 3 provide stronger evidence of causality for non-randomized interventions. The proposed innovation is calculation of propensity scores for experimental units of aggregate nature with subsequent application in matching and spatial regression adjustment. We will demonstrate the methods of Aims 2 and 3 by evaluating the HFC intervention in Connecticut's elementary, middle and high schools, and provide insight on childhood obesity.
The significance of this application is its provision of methodology that strengthens evidence of effectiveness and causality in the evaluation of second stage translational research and non- randomized studies. The proposed methods will allow a more rigorous evaluation of a wide range of non-randomized interventions ranging from clinical interventions for diabetes and heart disease to psychological interventions to reduce occurrence of domestic abuse. By pooling two reputable teams of researchers from Yale University and the University of Minnesota, the proposed statistical methods will be assessed by content experts whose experience, intuition and interpretation of model results will accurately gauge the utility of the proposed methods.
Showing the most recent 10 out of 14 publications
