Helping policymakers estimate the public health impact of precision medicine and targeted prevention interventions
The Prevention Impact and Efficiency (PIE) Model is a simple microsimulation modeling method and implementation, programmed in Stata, for analyzing the impact and efficiency of an intervention designed to prevent disease from occurring in a population. The microsimulation methodology allows easy estimation of population-level impact for complex, completely individualizable interventions and targeting of those interventions (i.e., Precision Medicine). Impact is expressed as the relative % reduction in disease events that occur, and Efficiency is expressed as the number needed to treat (NNT) with the intervention to prevent each disease event from occurring (lower NNT is more efficient). We provide downloadable template files for implementing this method and step-by-step instructions in our linked Github site. We hope the model will be used by guideline committees and policymakers looking to estimate the population-level impact and efficiency of their recommendations, and will archive submitted implementation files for published projects using the model on this website.
Public health impact can be achieved by 1) setting specific impact priorities, 2) planning and implementing policy designed to fulfill those priorities, and 3) measuring impact and repeating the process. The Healthy People program (e.g., Healthy People 2020) and others effectively support steps 1 and 3, and many diverse organizations attempt step 2, planning and implementing policies, but in our view there are not easy ways of estimating how any given policy might actually achieve specific public health impact goals. Public health initiatives typically specify evidence-based interventions that should move health indicators in the right direction, but by how much? Which interventions will help the most? What actual impact will they have on, on a population level, at reducing the incidence of a given disease outcome? And how efficient is each intervention strategy? If there are downsides to the intervention (e.g., cost, adverse effect risks), these usually scale with the number of individuals exposed to the intervention; so how many people need to be exposed to each intervention to prevent each event (i.e, what is the Number-Needed-to-Treat (NNT))? For example, if we want to reduce cardiovascular disease deaths, Healthy People 2020 Objective HDS-2, how would one compare the theoretical public health impact and efficiency of the following approaches?:
With the goal of developing a tool usable by guideline committees, we developed a simple model designed to produce estimates of theoretically-attainable public health impact and efficiency of targeted prevention strategies. The model requires only 1) an algorithm for estimating disease risk, 2) a decision rule for how to target a proposed intervention, 3) an algorithm for estimating how much risk is reduced for any given individual from the intervention, and 4) a dataset representative of the target population that includes any measurements required for steps 1-3 (NHANES often fulfills these requirements). Once you've programmed your algorithms, the model produces estimates of prevention impact (% reduction in events) and efficiency (number needed to treat to prevent each event) for any combination of intervention and targeting algorithm, and pre-packaged easily-configured code for tables and figures that illustrate the tradeoffs between impact and efficiency that are inherent in any targeted prevention intervention. The design work and coding that have gone into making these pre-packaged tables and figures is the value-add of the PIE Model that we hope will streamline analyses and produce estimates useful for policymakers and guideline committees.
You can download the PieModelTemplate_v1.0 (a set of Stata files that can be customized for your purposes) and detailed instructions on how to use the template from our linked Gitbut site. We encourage you to cite this website if you publish results that use these materials, and we also encourage you to send us a zipped file with all of your implementation files and your publication PDF and/or link to a Pubmed or other citation index, and we will archive your zipped file on this website. These archived files will be publicly available for download. We also encourage you to suggest and submit modified versions of our template files, and we will consider posting your modifications. Note: we are currently novice Github users, and are not currently utilizing all the version control and "forking" functionality on this site, but we are hoping to move in that direction.
Use these files at your own risk. If you find an error, please let us know!
Development of the initial model methods and template files was a collaborative effort that included Mark Pletcher, Michael Pignone, Tom Newman, Eric Vittinghoff, Andrew Moran, and Jamie Jarmul. Please direct questions about the model to Mark Pletcher at mpletcher@epi.ucsf.edu.