Table 1.

Regression models for the relation between the square rooted annual mean opioid consumption and the listing of opioids in national essential medicines lists (EMLs)

	Univariant (n=137)			Multivariable* (n=133)			Multivariable (n=117)
	Coefficient	95% CI	P value	Coefficient	95% CI	P value	Coefficient	95% CI	P value
Consumption versus number of opioids in EMLs	0.17	0.09 to 0.6	0	0.05	−0.02 to 0.11	0.19	0.01	−0.009 to 0.03	0.27
GDP/100 per capita				0.006	0.003 to 0.009	0	0.00005	−0.0012 to 0.0013	0.93
Healthcare expenditure per capita				0.0001	0.0007 to 0.002	0	0.0004	0.0002 to 0.0005	0
Population							−1.35e-10	−5.05e-10 to 2.35e-10	0.47
Life expectancy							−0.009	−0.03 to 0.01	0.38
Human development index							1.33	0.012 to 2.64	0.48
Corruption perception score							0.007	0.0007 to 0.01	0.03
Region (Africa)
America Asia Europe Oceania							−0.065 0.12 0.32 0.16	−0.3 to 0.17 −0.09 to 0.33 0.04 to 0.59 −0.3 to 0.6	0.59 0.27 0.03 0.48

The assumptions for untransformed linear regression were not met. Thus, we used a square root transformation of the dependent variable (ie, opioid consumption in mg/person), which improved the model.

*we conducted this multivariable analysis first as it had the least amount of missing data and the variables had the strongest predictors of opioid consumption.

GDP, gross domestic product.