. 2017 May;107(5):732–739. doi: 10.2105/AJPH.2017.303693

TABLE 3—

Associations Between Administratively Recorded Sexual Assault Victimization and Mental Health and Career Outcomes Among US Army Women in the Historical Administrative Data System: 2004–2009

Variables	Partially Adjusted Model^a	Fully Adjusted Model^b
Dichotomous outcomes, OR^c (95% CI)
Mental disorder treatment
Any treatment	2.5^* (2.4, 2.6)	2.5^* (2.4, 2.6)
Specialty treatment	3.0^* (2.9, 3.2)	3.1^* (2.9, 3.3)
Inpatient treatment	2.9^* (2.6, 3.2)	2.8^* (2.5, 3.1)
Any PTSD treatment	6.7^* (6.1, 7.4)	6.3^* (5.7, 6.9)
Specialty PTSD treatment	8.1^* (7.2, 9.1)	7.7^* (6.8, 8.6)
Inpatient PTSD treatment	7.5^* (5.9, 9.4)	6.8^* (5.4, 8.6)
Suicide attempt	3.1^* (2.6, 3.8)	3.0^* (2.5, 3.6)
Attrition	1.2^* (1.2, 1.3)	1.2^* (1.1, 1.2)
Demotion	2.2^* (2.0, 2.4)	2.1^* (1.9, 2.3)
Mental disorder treatment intensity (no. of treatment days) among patients,^d est^e (95% CI)
Any treatment	0.4^* (0.4, 0.4)	0.5^* (0.4, 0.5)
Specialty treatment	0.2^* (0.2, 0.2)	0.3^* (0.2, 0.3)
Inpatient treatment	0.2^* (0.1, 0.2)	0.2^* (0.1, 0.2)
Any PTSD treatment	0.2^* (0.2, 0.3)	0.3^* (0.3, 0.3)
Specialty PTSD treatment	0.2^* (0.1, 0.2)	0.3^* (0.3, 0.4)
Inpatient PTSD treatment	−0.1^* (−0.2, −0.1)	−0.1^* (−0.2, −0.1)

Note. CI = confidence interval; est = estimate; OR = odds ratio; PTSD = posttraumatic stress disorder. The sample size was n = 25 428.

^{^a}

Partially adjusted models controlled for the number of follow-up months between the month of the assault (survival models) and the month of starting treatment (generalized linear models) and the end of the observation period.

^{^b}

In addition to controlling for follow-up months, fully adjusted models also controlled for the variables selected by 10-fold cross-validated forward stepwise regression.

^{^c}

Coefficient estimates represent ORs. A dummy predictor variable was used to represent administratively recorded sexual assault victimization (victim = 1). Accordingly, the reference group for all ORs is women with no administrative record of sexual assault victimization.

^{^d}

For all 6 continuous outcomes, the generalized linear model that assumed a logarithmic link function with an error variance proportional to the mean was selected as the best overall model based on mean squared error and coefficient prediction strength (though results were similar with other link functions and error distributions; see Table D and Figures A–F, available as supplements to the online version of this article at http://www.ajph.org).

^{^e}

Given the use of a logarithmic link function, parameter estimates from the generalized linear models can be interpreted as multiplicative effects. For example, the fully adjusted coefficient of 0.5 for intensity of any mental health treatment indicates that victims of sexual assault had (on average) exp(0.5) = 1.65 times (65%) more treatment days than controls.

Significant at the .05 level, 2-sided test.