2. Estimating the prevalence of opioid dependence in New South Wales from multiple data sources: case study application of a Bayesian modelling approach

Multiplier methods are often used to estimate prevalence, but are subject to a number of important limitations. We recently proposed a Bayesian modelling approach to estimate prevalence from data on a large cohort of individuals from the target population, linked to mortality records, addressing many of these limitations.

We (i) apply the model to a new case study, to estimate prevalence of opioid dependence in New South Wales (NSW), Australia; (ii) apply the approach to alternative types of adverse events and compare prevalence estimates; and (iii) extend the approach to jointly model multiple data sources. We modelled 2014-2016 data from the Opioid Agonist Treatment and Safety study, a rich set of routinely collected data on all people receiving opioid agonist treatment in NSW. We modelled linked data on opioid-related mortality, hospitalisations, and charges for possession or use of opioids.

Estimates of the overall prevalence of opioid dependence in 2016 based on the three data sources were similar: 9.3 (95% credible interval [CrI] 8.0-11.1), 10.2 (95% CrI 9.5-11.0) and 9.5 (95% CrI 9.0- 9.9) per 1000 people aged 15-64 years, based on mortality, hospitalisations and charges data respectively. However, we observed differences in gender-specific prevalence estimates, with estimates based on hospitalisations being markedly higher than others. When mortality and charges data were combined in a joint model, prevalence of opioid dependence in NSW in 2016 was estimated at 9.4 (95% CrI 9.0-9.8) per 1000 people, or 47,440 (95% CrI 45,460-49,610) people aged 15-64.

We demonstrate the applicability of this novel methodology, and how it can be extended to model multiple types of data together. In the case of NSW, further evidence is needed to establish whether some of the hospitalisations might have occurred in women who were not opioiddependent, which could explain the discrepancies.