Log in / Register
Home arrow Health arrow Sleep in the military

Associations Between Sleep Symptoms and Sleep-Related Behaviors and Outcomes Among Previously Deployed Servicemembers

Given our study's focus on the degree to which sleep problems are independently associated with servicemember health and functioning in the post-deployment period, analyses for aim 2 only included servicemembers with at least one prior deployment (N = 1,596). Specifically, we used multivariate regression models to assess the association between sleep measures and post-deployment outcomes, including probable depression, probable PTSD, perceived unit readiness, and physical health.

Since sleep problems and disturbances may vary as a function of many other known factors that are themselves risk factors for poor mental or physical health or lowered readiness (see Chapter Two), our multivariate regression models adjusted for several key covariates, including sociodemographic and military characteristics, and probable TBI and depressive symptoms (described in detail above), which may account for observed associations between sleep and the outcomes of interest. The models for physical health and unit readiness included all the control covariates listed. For the models with probable PTSD or probable depression as outcomes, we used a reduced set of these controls, given the low rate of occurrence for each outcome in the sample (12.4 percent and 8.7 percent, respectively). Specifically, for these models, we included a subset of sociodemographic (age, gender, race) and military characteristics (branch, officer status, combat exposure, shift work, total number of deployments), as well as TBI and depressive symptoms (except where probable depression was the outcome), as these covariates have shown consistent associations with sleep, as well as with PTSD and depression, in prior literature. We added sleep measures to the multivariate models separately to assess whether there were associations between each of the sleep measures and the outcomes above and beyond the control covariates already in the model.

To be more conservative about the number of hypothesis tests conducted, we assessed statistical significance at the 0.01 level and used joint F-tests to test whether categorical predictors were significantly associated with the outcomes. We fit linear regression models to standardized versions of the self-reported physical health item, and we assessed unit readiness and goodness of fit for these models using QQ plots of the residuals. We standardized these outcomes and the single continuous sleep predictor, total PSQI, by dividing by their standard deviation to obtain regression coefficients that represent the standardized effects of each sleep measure on physical health and unit readiness. As cited in the literature, standardized effects of less than 0.20 are considered small, effects of 0.20-0.50 are considered moderate, and effects greater than 0.50 are considered large (Cohen, 1988). Logistic regression models were fit to probable PTSD and probable depression. To better describe the magnitude of the associations observed in the logistic models, we computed recycled mean predictions using these models to estimate the adjusted means for each binary outcome within each level of the particular sleep measure.

Found a mistake? Please highlight the word and press Shift + Enter  
< Prev   CONTENTS   Next >
Business & Finance
Computer Science
Language & Literature
Political science