North America is currently experiencing an opioid overdose crisis, with estimates of life expectancy dropping for two consecutive years owing to the large number of deaths due to opioid overdose. This has been particularly pronounced in British Columbia, where illicit-drug overdose deaths have increased by 750% in the last ten years, primarily being driven by the potent synthetic opioid fentanyl. The province has responded by declaring a public health emergency in April 2016, and by introducing or scaling up a number of interventions such as take-home naloxone kits, overdose prevention sites, and by increasing prescriptions of treatment for opioid use disorder. Current metrics for evaluation have focussed on numbers of intervention received or distributed. However, there is a need to understand how these interventions have had an impact on the number of overdose-related deaths.
In this talk, I will present the use of probabilistic graphical models (a Bayesian analysis method) in order to estimate the impact of intervention. Using the methodology, I combined together into one model disparate data routinely collected by the province including ambulance-attended overdoses, deaths due to overdose, and intervention data. I was also able to incorporate other factors which may exacerbate the epidemic, including drug supply contaminant information collected through urinalysis, and the impact of weather. I will provide a few examples of the approach before providing a detailed description of the current analysis for British Columbia and its implications.
As this new public health crisis emerges, it is imperative that we have tools and methods to understand the impact the current interventions have had, as well as planning for the future. Large-scale public health data are increasingly being collected and centralised. Use of these Bayesian analysis methods enable these data to be fully utilised to understand how policy and intervention impact population health.