Table 3.

Factors associated with self-reported antiretroviral interruptions in the past 3 months among HIV-infected homeless and marginally housed adults in San Francisco (N = 125)

Variable	Crude			Adjusteda
	OR	95% CI		AOR	95% CI
		LCL	UCL		LCL	UCL
Biologic Sex = Male	2.88	0.61	13.55
Income	1.00	1.00	1.00
Race (White as baseline)
African-American	0.16	0.03	0.78
Other	0.98	0.10	1.58
Age at baseline	0.98	0.91	1.05
Physical health score on SF-36	0.97	0.92	1.02
Hospitalized in the past year	0.68	0.14	3.23
ER visit in the past year	3.20	1.00	10.19
Primary care (2 or more visits in past year)	0.74	0.08	6.66
Beck depression inventory score (OR per scale point)	1.07	1.02	1.12	1.08**	1.03	1.15
Housing status (homeless vs. marginally housed)	4.04	1.07	15.27
Methadone treatment in past year	0.28	0.04	2.24
Medicare Part D-covered	5.66	1.66	19.33	7.50**	1.96	28.58
Crack cocaine use	1.13	0.33	3.89
Methamphetamine use	3.02	0.83	11.02
Any injection drug use in past year	0.76	0.25	2.35
Incarceration in past year	3.51	0.81	15.20

^aFinal multivariable logistic regression model obtained using stepwise regression with P < 0.2 to enter and P < 0.05 to retain. For the full model: −2 Log L = 55.752, df = 10 and for the reduced model (Medicare Part D-covered and BDI): −2 Log L = 70.574, df = 2. The overall Diff = 14.822, df = 8 which is <15.51 (cut-off for χ² with df = 8 is 15.51.) By −2 log likelihood test, final model is sufficient. The adjusted R² for the full and reduced models are 0.4470 and 0.2536, respectively

^**P < 0.01