Table 3.

Multiple logistic regression analysis: predictors of sources of pharmaceutical opioids (“active” role in pain pill acquisition versus “passive” role) for non-medical use in the past 6 months (n=390).

Variable	Odds ratio	95% CI	p value
Gender
Male (vs. “female” as a reference group)	0.88	0.57–1.37	0.573
Race/ethnicity
White (vs. “other” as a reference group)	1.76	1.08–2.84	0.022
Duration of illicit pharmaceutical opioid use (in years, continuous variable)	1.06	0.95–1.18	0.304
Frequency of non-medical pharmaceutical opioid use
1–2 days/week (vs. “less than 1 day/week” as a reference group)	2.87	1.67–4.92	<0.001
3 or more days/week (vs. “less than 1 day/week” as a reference group)	4.96	2.58–9.54	<0.001
Non-medical use of OxyContin	1.76	1.04–2.97	0.034
Pharmaceutical opioid administration
Snorting ( vs. “oral use” as a reference group)	1.53	0.76–3.09	0.232
Reasons of use
To get high and to self-medicate (vs. “self-medicate only” as a reference group)	2.35	1.22–4.52	0.010
To get high only (vs. “self-medicate only” as a reference group)	2.62	1.28–5.37	0.009

Hosmer-Lemeshow test: X² =2.033, df=8, p=0.980*

^*The Hosmer-Lemeshow goodness-of-fit-statistic was used to assess if the model’s estimates fit the data at an acceptable level. Non-significant outcome (Hosmer-Lemeshow goodness-of-fit test statistic is greater than .05) indicates that the model prediction does not significantly differ from the observed, and thus the model has a good fit.