Content area
Abstract
Significant advances in medicine have been instrumental in improving population health in the US. This is evidenced by the decrease in the overall mortality rate and an increase in life expectancy observed from the 1940s to 2019. Though tremendous improvements have been made in the ability to diagnose, treat, and manage diseases, the burden associated with preventable diseases is still a major population and public health concern with huge associated economic costs. Many health interventions are designed and implemented to fight this disease burden at the clinical and public health level. Since those interventions can have different forms and can require a lot of resources to design, implement and maintain, it is important to be able to evaluate them to capture their benefits and effectiveness to help with future decision making.
Computer simulation modeling is a safe and inexpensive tool that can be used as a decision support tool for policy and medical decision makers in the context of intervention evaluation. Simulation modeling can leverage data from different sources (e.g., randomized control trials and observational studies) to model interventions and project the results over long term periods. As the goal of using simulation modeling for intervention evaluation is to inform decision making, model users need to keep in mind that like all models, simulation models inherently contain uncertainty. This uncertainty influences simulation model results and thus can influence recommendations. In the context of screening intervention evaluation using simulation modeling, patient behavior and patient characteristics are important elements to capture in the simulation models but modeling them is not a trivial task; lack of relevant evidence and data for modeling and/or the resulting increase in complexity makes it a challenging task. These data limitations and challenges can lead to the use of assumptions or crude approximations that result in different uncertainties in the simulation model and the simulated outcomes and can lead to erroneous recommendations.
This dissertation work was focused on investigating the impact of modeling decisions when simulating preventive health interventions; specifically, decisions pertaining to (1) the use of assumptions about patient screening behavior and (2) the inclusion of heterogeneity when evaluating screening interventions. This dissertation was divided into two parts to examine the effects of these assumptions in the context of screening intervention evaluation. The first part focused on considering different patient screening behavior assumptions in the context of colorectal cancer and the second part examined the impact of heterogeneity in disease modeling in the context of diabetic retinopathy. Colorectal cancer (CRC) and diabetic retinopathy (DR)simulation models were used along with different methodologies like Markov modeling and statistical methods to simulate long-term outcomes of related screening interventions with the different assumptions considered (i.e., patient screening behavior and heterogeneity assumptions)in order to capture their impact.
Findings show that assumptions about screening behavior and heterogeneity matter in simulation modeling and the choice of those assumptions can affect how effective a screening intervention seems to be. For a CRC screening colonoscopy intervention particularly, choices of assumptions about individuals screening behavior at the time of intervention can decrease the effectiveness of the intervention by as much as 8%. Investigating the impact of heterogeneity also showed that the results of the simulation with and without interventions can differ significantly when comparing outcomes for the whole cohort and subgroups across homogeneous versus heterogeneous assumptions made about the disease natural history.
Based on these findings, modelers should carefully consider their choice of assumptions regarding patient screening behavior and heterogeneity and reduce uncertainty about them when evaluating interventions. Policy recommendations taken from these contexts should be carefully examined in light of these aspects.
