Table 3.

GEE Negative Binomial Regression Analysis Estimating the Association Between Medication Possession Ratio (MPR) and Violent/Nonviolent Crime (n = 11462^a)

	Variable	Unadjusted Rate Ratio	Adjusted Rate Ratio
		(95% CI)b	(95% CI)c
Violent offenses	MPR
	≤0.19	1.54 (1.41, 1.68)d	1.39 (1.27, 1.52)
	0.20–0.39	2.68 (2.39, 3.01)	2.28 (2.03, 2.56)
	0.40–0.59	2.58 (2.29, 2.91)	2.21 (1.96, 2.50)
	0.60–0.79	1.95 (1.74, 2.19)	1.73 (1.55, 1.94)
	≥0.80	Reference	Reference
	MPR (<0.80)	1.78 (1.64, 1.93)	1.58 (1.46, 1.71)
Nonviolent offenses	MPR
	≤0.19	1.50 (1.39, 1.62)	1.31 (1.21, 1.41)
	0.20–0.39	2.01 (1.83, 2.21)	1.64 (1.50, 1.80)
	0.40–0.59	1.86 (1.70, 2.03)	1.64 (1.50, 1.80)
	0.60–0.79	1.65 (1.53, 1.78)	1.48 (1.37, 1.60)
	≥0.80	Reference	Reference
	MPR (<0.80)	1.62 (1.52, 1.73)	1.41 (1.33, 1.50)

Note: GEE: generalized estimating equation; MPR: medication possession ratio; CI: confidence interval.

^aDemographic information was unknown for 1280 participants (11%): age n = 3 (<1%); ethnicity n = 573 (5%); and education level n = 1193 (10%). Unknown age was replaced by mean age. Unknown ethnicity and education level were included as separate categories in the multivariable model.

^b95% CIs and both unadjusted and adjusted rate ratios were estimated using robust standard errors.

^cEach multivariable GEE model controlled for age at enrollment (centered, age 18), gender (men and women), ethnicity (White, Indigenous, other, and unknown), education level (<Gd. 10, Gd. 10/11, Gd. 12, Vocational/University, and unknown), use of substance disorder-related services (continuous variable), number of offenses in the previous year (continuous variable), number of 120-day intervals (continuous variable), and duration of follow-up in days (offset variable).

^dBold indicates significant at P value <.001.