Our sample was comprised of women taking part in a wider study known as the Center for Maternal and Infant Outcomes and Research in Translation (COMFORT).12 The COMFORT study enrolls pregnant women veterans identified from 15 Veterans Health Administration (VHA) sites across the USA and consists of two telephone surveys: the first during pregnancy and the second after delivery of the infant. These telephone surveys collected information on sociodemographics, military-related characteristics, and pregnancy- and health-related data and included the Edinburgh Postnatal Depression Scale (EPDS) to identify depression symptoms in participants.13
Using the prenatal and postnatal COMFORT surveys, we identified demographic, military, pregnancy, infant outcomes, and healthcare utilization variables. We combined categories of race into white vs. non-white, black vs. non-black, and others. Military sexual trauma (MST) was identified through a screener universally adopted by the VA.14 The presence of an MST experience was identified through an affirmative response to any of the following: received uninvited and unwanted sexual attention while in the military; force or the threat of force was used to have unwanted sexual contact while in the military; or ever received counseling or treatment for military sexual trauma from a VA or non-VA provider. Urban/rural geographic status was identified using participant zip codes and the FY15 VA Planning Systems Support Group (PSSG) file, which assigns geographic regions of urban, rural, and highly rural to zip codes.
Additionally, we obtained data on service-connected disability, Operation Enduring Freedom/Operation Iraqi Freedom (OEF/OIF) status, and healthcare utilization by matching participant social security numbers from our COMFORT sample to VHA administrative files from the VA Corporate Data Warehouse (CDW). The CDW is a repository of data updated nightly from the VHA electronic medical records system for operations and research use. Healthcare utilization included visits for primary care, post-traumatic stress disorder, psychiatric, military sexual trauma, and substance abuse visits, identified through VA Stop Codes. We included any visit that occurred within a pregnancy window of 280 days prior to delivery, as has been done in previous work.15
Patient Perceptions of Integrated Care Survey
In addition to the prenatal and postpartum COMFORT surveys described above, participants were mailed the Patient Perceptions of Integrated Care (PPIC) survey following participation in the postpartum telephone survey. The PPIC survey has been previously validated and psychometrically tested.16 Survey items were developed to assess care coordination within and across care settings and integration of patient and family capabilities, needs, and preferences with a patient’s care. Survey development has been previously described.10 We modified the original PPIC survey to include a total of 54 items used to measure perceptions of integrated care among our sample of postpartum veterans, asking specifically about experiences with VA primary care provider offices, care from other staff at the VA primary care provider’s office, and care from outside obstetrical providers. Women veterans who completed the PPIC survey between November 2016 and June 2018 were included in the present analysis.
We began by examining descriptive statistics for our sample. Additionally, we calculated descriptive statistics for PPIC responses, including means and standard deviations (SDs) of raw item scores, as well as the percent of responses in the “top box” (the percentage of responses falling into the most positive response category), as has been done in previous PPIC and survey analyses in order to check for ceiling effects.16,17
Building on previous PPIC analyses, we conducted a confirmatory factor analysis (CFA) to verify associations between our observed survey responses and the underlying latent constructs developed by previous research.16 Our survey items loaded onto the following six factors: VA Staff Knowledge about the Patient’s History, VA Provider Support for the Patient’s Self-directed Care, Test Result Communication, VA Provider Knowledge of the Patient, support for medication and home health management, and obstetrician’s knowledge about the patient’s medical history. We ran our CFA using PROC CALIS in SAS 9.2 (SAS Institute, Inc., Cary, NC, USA). To utilize all of the available data from our sample, we specified a full information maximum likelihood (FIML) method. Our CFA was evaluated for goodness of fit, with the root mean square error of approximation (RMSEA) of 0.0583, the non-normed fit index (NNFI) of 0.8826, and the comparative fit index (CFI) of 0.9014, indicating a reasonable to adequate fit of our model (Table 1).
After confirming a reasonable fit of our model, we calculated factor scores as the unweighted average of the numeric score items in each factor, where the average was calculated as the mean of non-missing responses within the domain. We also adjusted each survey item for individual respondent response tendency by fitting a linear regression of each item’s score (modeled continuously) as a function of demographic variables, including postpartum age, marital status, ethnicity, race, and an overall health rating collected as part of the PPIC survey (“In general, how would you rate your overall health?” on a 0–5 E/VG/G/F/P scale). We then predicted scores for each respondent using the regression models and calculated the residuals (i.e., the differences between a respondent’s observed and predicted scores on each item), which we used as adjusted survey responses.
We examined descriptive statistics for each of our factors as well as correlations within and between factors, using Cronbach’s alpha for internal consistency and Spearman’s correlation coefficients to compare between factors. We then calculated quartiles to assign each participant a quartile level for each of the six factors, following previous analyses utilizing PPIC survey results.16 Quartile scores were used to conduct ordered logistic regression models, with odds ratios interpreted as the average odds of a participant providing a response in a higher quartile of perceived integration relative to responses in lower quartiles. As our variables of interest included mental health characteristics, we utilized these variables in bivariate ordered logistic regression models (measured as Excellent/Very Good/Good/Fair/Poor on the PPIC survey). Additionally, we ran models adjusting for postpartum age, ethnicity (Hispanic vs. non-Hispanic), and an overall health. These variables were selected as they were moderately to strongly statistically significant (p < 0.10) in univariate models comparing demographics to each factor score. Race and marital status were not included as they showed no difference in any model by factor (all p > 0.10). Finally, to adjust our findings for multiple comparisons, we calculated the false discovery rate (the expected rate of type I error) separately for adjusted and unadjusted analyses, to no more than 5% of all statistically significant results. All analyses were conducted in SAS (version 9.2).