WW14: MAKING MODELS BETTER

OBJECTIVES: In this workshop, we will (1) lay out the set of conditions for assessing the quality of models, (2) provide examples from the Duke Stroke Policy Model of how these conditions may be met, and (3) engage participants in a discussion of cases in other clinical areas. PARTICIPANTS WHO WOULD BENEFIT: Those who develop and use decision/cost-effectiveness models. In an ideal world all clinical decisions would be based on evidence from large comprehensive scientific studies. In the complex world of actual practice—the world of pharmacoeconomics—such evidence is often impractical or impossible to obtain. Models are used to help decision-makers make plausible and useful inferences in the absence of perfect information. Despite the broad application of models in this context, the process and conclusions of modeling exercises are frequently suspect. Concerns include: (1) failure to use the best available data; (2) inaccessibility of the model code; (3) lack of validation; and (4) difficulty interpreting the results of technical analyses. Despite recommendations from the US Preventive Services Task Force intended to address these concerns, models are still held in low regard. As part of the Stroke PORT project, a simulation model of stroke development and outcomes was developed. This model has since been expanded to accommodate a variety of potential interventions and data sources and is currently designated the Duke Stroke Policy Model (DSPM). One of the major objectives of the DSPM has been to address the above concerns. Our approach has been to (1) use the best data; (2) vigorously promote an open-code public-use application, available on the Internet; (3) perform rigorous validation; and (4) develop user-friendly applications based on the validated core code, tailored to different user groups. Researchers, teachers and public health professionals are encouraged to use the DSPM for academic and other non-commercial applications, and to engage with the developers to promote the continuing improvement of the model. In this workshop, we will (1) lay out the set of conditions for assessing the quality of models, (2) provide examples from the DSPM of how these conditions may be met, and (3) engage participants in a discussion of cases in other clinical areas.