Search

Opioid-related mortality increased by nearly 400% between 2000 and 2018. In response, federal, state, and local governments have enacted a heterogeneous collection of opioid-related policies in an effort to reverse the opioid crisis, producing a policy landscape that is both complex and dynamic. Correspondingly, there has been a rise in opioid-policy related evaluation studies, as policymakers and other stakeholders seek to understand which policies are most effective. In this paper, we provide an overview of methodological challenges facing opioid policy researchers when evaluating the effects of opioid policies using observational data, as well as some potential solutions to those challenges. In particular, we discuss the following key challenges: (1) Obtaining high-quality opioid policy data; (2) Appropriately operationalizing and specifying opioid policies; (3) Obtaining high-quality opioid outcome data; (4) Addressing confounding due to systematic differences between policy and non-policy states; (5) Identifying heterogeneous policy effects across states, population subgroups, and time; (6) Disentangling effects of concurrent policies; and (7) Overcoming limited statistical power to detect policy effects afforded by commonly-used methods. We discuss each of these challenges and propose some ways forward to address them. Increasing the methodological rigor of opioid evaluation studies is imperative to identifying and implementing opioid policies that are most effective at reducing opioid-related harms.

Patient-centered care involves patients in decision-making about their medical care. Effective shared decision-making requires relevant evidence on the expected health outcomes of treatment, assessment of patient preferences for possible health outcomes, and a method to synthesize this information. Few tools exist to help clinicians and patients synthesize clinical evidence and personal preferences. We develop a statistical framework that combines multiple data sources on expected treatment outcomes with individual preferences to produce a personalized preference-weighted outcome score (PWOS). The PWOS can be calculated for multiple treatments and compared to determine which will provide the best balance of risks and benefits given a patient’s preferences. We demonstrate our method by evaluating adjuvant chemotherapy regimens for colorectal cancer. We begin by identifying heterogeneity in patient preferences for health outcomes associated with colorectal cancer disease and treatment, finding classes of patients who value efficacy and side effects differently. Then we fit a hierarchical Bayesian model to randomized clinical trials and produce posterior distributions of expected outcomes under three chemotherapy regimens. We use the PWOS to combine distinct sets of patient preferences with these expected outcomes. Our method preserves estimation uncertainty and accounts for correlation among outcomes. In a simulation study of applying PWOS to make decisions at the population level, we find that our methods are most useful when there are large differences in individual preferences and small distinctions in treatment efficacy (i.e., preference-sensitive settings).

Propensity score methods (e.g., matching, weighting, subclassification) provide a statistical approach for balancing dissimilar exposure groups on baseline covariates. These methods were developed in the context of data with no hierarchical structure or clustering. Yet in many applications the data have a clustered structure that is of substantive importance, such as when individuals are nested within healthcare providers or within schools. Recent work has extended propensity score methods to a multilevel setting, primarily focusing on binary exposures. In this paper, we focus on propensity score weighting for a continuous, rather than binary, exposure in a multilevel setting. Using simulations, we compare several specifications of the propensity score: a random effects model, a fixed effects model, and a single-level model. Additionally, our simulations compare the performance of marginal versus cluster-mean stabilized propensity score weights. In our results, regression specifications that accounted for the multilevel structure reduced bias, particularly when cluster-level confounders were omitted. Furthermore, cluster mean weights outperformed marginal weights.