Examining correlations between opioid dispensing and opioid-related hospitalizations in Canada, 2007-2016

Abstract

Background: High levels of opioid-related mortality, as well as morbidity, contribute to the excessive opioid-related disease burden in North America, induced by high availability of opioids. While correlations between opioid dispensing levels and mortality outcomes are well-established, fewer evidence exists on correlations with morbidity (e.g., hospitalizations). Methods: We examined possible overtime correlations between medical opioid dispensing and opioid-related hospitalizations in Canada, by province, 2007-2016. For dispensing, we examined annual volumes of medical opioid dispensing derived from a representative, stratified sample of retail pharmacies across Canada. Raw dispensing information for 'strong opioids' was converted into Defined Daily Doses per 1000 population per day (DDD/1000/day). Opioid-related hospitalization rates referred to opioid poisoning-related admissions by province, for fiscal years 2007-08 to 2016-17, drawn from the national Hospital Morbidity Database. We assessed possible correlations between opioid dispensing and hospitalizations by province using the Pearson product moment correlation; correlation values (r) and confidence intervals were reported. Results: Significant correlations for overtime correlations between population-levels of opioid dispensing and opioid-related hospitalizations were observed for three provinces: Quebec (r = 0.87, CI: 0.49-0.97; p = 0.002); New Brunswick (r = 0.85;CI: 0.43-0.97; p = 0.004) and Nova Scotia (r = 0.78; CI:0.25-0.95; p = 0.012), with an additional province, Saskatchewan, (r = 0.073; CI:-0.07-0.91;p = 0.073) featuring borderline significance. Conclusions: The correlations observed further add to evidence on opioid dispensing levels as a systemic driver of population-level harms. Notably, correlations were not identified principally in provinces with reported high contribution levels (> 50%) of illicit opioids to mortality, which are not captured by dispensing data and so may have distorted or concealed potential correlation effects due to contamination.