Effect of maternal depression and anxiety on use of health services for infants

Study population

A postpartum cross-sectional survey was conducted using a sample of women drawn from a larger longitudinal, population-based cohort study known as the Prenatal Health Project (PHP).9 The PHP recruited pregnant women from 7 ultrasound clinics in London, Ont, between 2002 and 2005. Eligibility criteria for the PHP cohort required that all study participants be at 10 to 22 weeks’ gestation, at least 16 years of age, English speaking, currently living in the London-Middlesex region, and having a singleton pregnancy. All women in the PHP had previously completed a prenatal telephone interview. No intervention was applied in this observational study. Women with high-risk pregnancies were excluded from the PHP. The sample for this postpartum survey included all mothers whose infants were between 2 and 12 months of age between November 2004 and July 2005. Postpartum data were collected through scripted telephone interviews. This study was approved by the Research Ethics Board at the University of Western Ontario.

Measures

The outcome was HSU for infants, measured during the postpartum period using an adapted version of the Health Services Utilization Questionnaire.10 Main outcomes were number of PCP visits per month and whether WIC visits or ED visits were made for infants. Primary care provider visits were defined as visit to the infants’ regular sources of care, whether family physicians or pediatricians. The number of primary care provider visits per month was calculated as the total number of PCP visits made for the infant since birth, divided by age in weeks, multiplied by 4.3. Visits to EDs and WICs since birth were dichotomized as yes or no responses.

Depression and anxiety were assessed during the postpartum telephone survey. Depressive symptoms were assessed using the Center for Epidemiologic Studies Depression Scale (CES-D), a 20-item scale designed for the general population. A CES-D score ≥ 16 is indicative of depressive symptoms.11 The reliability and validity of the CES-D have been well established; calculated Cronbach α coefficients range from 0.84 to 0.93.12–14 Specific to populations of mothers at 2, 3, 6, 12, and 18 months after giving birth, the Cronbach α levels are all greater than 0.80.15 Maternal anxiety was measured using the State-Trait Anxiety Inventory (STAI), a 12-item scale created by Spielberger et al.16 This inventory has been widely used and a variety of estimates of its reliability and validity have been reported.17 For the purpose of this study the results were dichotomized and a STAI score ≥ 90th percentile was used to classify elevated maternal anxiety.

Table 1 provides a full list of all covariates included in this study. We used the behavioural model of HSU as a conceptual framework for this study, and all variables are classified as predisposing characteristics, enabling resources, or need variables.18 These variables were identified a priori from the literature as potential confounders to be tested for inclusion in any of the multivariable regression models.

Statistical analysis

All statistical analyses were performed using SAS 9.1 for Windows. Infants were excluded from analysis if they did not currently have PCPs. In consideration of the fact that infants are expected to have a high number of PCP visits in the first 6 months of life, analyses regarding infant PCP visits were stratified into 2 groups: infants younger than 6 months of age and infants 6 months of age or older. In anticipation of high correlation between CES-D scores and STAI scores,19 separate models were constructed for depression and anxiety.

Modeling was based on multivariable regression. For the 2 PCP outcomes, linear regression was conducted. For WIC and ED use, unconditional logistic regression was conducted. The analysis was performed for both hypothesized risk factors (maternal depressive symptoms and maternal anxiety) and all 4 infant HSU outcomes (PCP visits for infants < 6 months of age, PCP visits for infants = 6 months of age, ED use, and WIC use), resulting in a total of 8 hypothesis-testing models. Any variable that resulted in a change > 10% when removed from the full model was defined as a confounder and included in the final multivariable model. This study did not test for any interactions because the literature did not provide evidence of any a priori, and upon initial analyses the main effect was found to be modest.