How we are studying it
We are establishing a timely ADF-opioid mortality surveillance system (AOMSS) based on multiple data sources: death certificate, prescription drug monitoring program, autopsy, Medicaid claims, toxicology report, and coroner investigation reports. In addition, we will create online implementation in REDCap, establish or maintain data sharing agreements, and update a literature review. On an ongoing basis, we will review of AOMSS data for patterns of abuse; pathological changes; and associated diagnoses and medical conditions, and use AOMSS data as a sentinel to detect adverse effects to develop new statistical approaches. These findings will be compiled in guides for replication in other states.
In addition to the development of an electronically accessible database described above, we will conduct a prospective cohort study using AOMSS data of longitudinal progression-to-death and the role of ADFs in overdose prevention. We will use a prevalent new-user cohort design. The base cohort will include all ADF patients (including those who directly initiated the ADF drug as well as those who switched from comparator drugs), as well as users of a comparator drug(s). The initial cohort enrollment will be determined from PMP records of patients who had > 1 prescription billed to Medicaid during each year of the study period. We will use both prescription-based and time-based exposure denominators, adjusting for opioid prescription history for both number of prescriptions and length of use. We expect to use time-conditional propensity scores to identify comparator drug users similar to patients who switched to the study drug. Conditional logistic regression can be performed to compute time-conditional propensity scores for all exposure sets. Medicaid-based time-varying covariates included in the conditional logistic regression will be identified based on a predefined conceptual framework (DAG). Using time-conditional propensity scores, we plan to match new ADF users with non-ADF users, and employ Cox proportional hazards regression to model hazards of several opioid abuse outcomes. For fatal overdose outcomes, we will consider extended Cox multi-state transition models accounting for competing risks or informative censoring (i.e., cancer deaths).