Multi Parameter Estimation of Prevalence modelling: methods to estimate the prevalence of opioid dependence in Scotland

Background: New prevalence estimates were needed to interpret what was driving increasing trends in drug related deaths in Scotland. Previous estimates relied on "capture-recapture" methods, but this approach became unviable due to some of the datasets are no longer available. Recognizing the methodological limitations of capture-recapture, a shift to a Multi Parameter Estimation of Prevalence (MPEP) modelling approach was implemented to estimate prevalence of opioid dependence. The MPEP approach, previously piloted in England and Australia (Jones et al. 2020, Downing et al. 2023), uses administrative record linkage between opioid agonist treatment (OAT) prescriptions and adverse events (e.g., mortality) data. A Bayesian statistical model, employing simultaneous regressions on aggregated data, is applied to estimate prevalence.
 
Methods: As with any method for indirect estimation of population size, the MPEP approach relies on a number of assumptions. Careful thought, and discussion with experts of the specific administrative data sets used and how events are coded locally, is needed. In this presentation we will discuss steps taken and decisions made in our application of the MPEP approach to estimate the number of people with opioid dependence in Scotland in 2014–2019. We estimated prevalence jointly from opioid-related mortality and hospitalisations data. Key stages included (1) derivation of a “baseline cohort” for modelling, including all individuals in receipt of OAT, (2) identification of subsets of opioid-related deaths and hospitalisations that are believed to be highly specific to the community of interest (i.e. to occur only among people with opioid dependence). We further discuss an extension of our previously proposed approach, to allow for all cause mortality in the model. We show how prevalence estimates based on modelling each data source (mortality and hospitalisations) separately can be calculated and compared to further interrogate model assumptions and explore robustness of conclusions. 
 
Results and Conclusions:
Our estimates are consistent with evidence on the number of DRD and non-fatal hospital admissions and changes in the risk of DRD experienced by people with opioid dependence. MPEP builds on linkage between administrative databases which are increasingly available across Europe. Prevalence estimates from this first application of MPEP in Scotland, for 2014/15 – 2019/20, have been submitted separately. The focus in this talk is on the modelling approach and how other countries can adopt and adapt our methods. Critically, MPEP is a flexible approach, adaptable to local settings and circumstances and builds in a test of consistency of information on size of known population and  drug related harm. We strongly encourage careful discussion with local experts, clear documentation of decisions and assumptions made, and use of sensitivity analyses to explore robustness of results.