Maternal Health Needs and Pediatric Settings

advertisement
Running Head: PSYCHOSOCIAL AND BEHAVIORAL HEALTH
1
Supplemental Digital Content 1
Method
Design
Women participating in the original study were recruited from the postpartum unit of a
community hospital in the southwestern United States. The purpose of the original study was
to provide a longitudinal picture of postpartum weight and its psychosocial and behavioral
correlates covered by the key data points of the first 1-3 days after birth and subsequently 6
weeks, and 3, 6 and 12 months postpartum. Because the focus was on postpartum weight
patterns among women with normal births, women with preterm births or other perinatal
complications were excluded. In addition, because low-income women are more prone to
obesity and have less economic resources to devote to weight management (e.g., healthy food
purchases, membership at recreational sports facilities), the original study recruited women
whose prenatal care was covered by Medicaid, a state-federal partnership that includes
coverage for women of lower income during pregnancy.
In the original study, psychosocial and behavioral data were collected by questionnaire
in the hospital and at the follow-up points thereafter. Questionnaires were mailed to women in
advance of the data collections at 6 weeks, and 3, 6 and 12 months postpartum and were
collected when women came to a research center for weight measurements at these time
points. In addition, dietary data and other measures of lifestyle were collected by interview at
each time point starting at the 6th week postpartum time point.
Sample
We chose to use the low-income sample from the Austin New Mothers Study for this
secondary analysis because of the social disparities and inequities faced by this group of
Running Head: PSYCHOSOCIAL AND BEHAVIORAL HEALTH
2
women, who we believed might be experiencing extensive poor psychosocial and behavioral
health. With regard to sample size for the analysis of risk factors of cumulative and individual
domain poor health, the sample size-predictor ratio of 419 women per 9 predictor variables
exceeded the ratio of 10 subjects:1 predictor that is a general rule.
Validity of Measures
The sources we drew on in regard to the validity of the Center for Epidemiologic Studies
Depression Scale (CES-D; Myers, & Weissman, 1980; Radloff, 1977; Weissman, Sholomskas,
Pottenger, Prusoff, & Locke, 1977), the The Body Cathexis Scale (BCS; Robinson & Shaver,
1973; Walker, Freeland-Graves, Milani, George, et al., 2004), and the Self-Care Inventory (SCI;
Walker, Freeland-Graves, Milani, George, et al., 2004; Wiebe & McCallum, 1986) corroborate
the construct validity of these scales. In addition, several of these (e.g., CES-D) have been
studied in relation to their usage as screeners or predictors of external and more rigorously
regarded measures of a health phenomenon, such as depression. Consequently, in such studies
the type of validity would be considered criterion-related validity. However, validity is now
viewed as a unified psychometric property rather than one emphasizing types of validity.
Consequently, we have used the unqualified term “validity” for the various measures in this
secondary analysis to avoid thorny and sometimes subtle differences in types of validity of
measures.
Data Analysis
We selected tertiles for several reasons in this secondary analysis based on our
experience with the CES-D with low-income women. We knew from the original study that a
median split on the CES-D would have somewhat artificially divided a large number of
women who were closely clustered around the median. Because we wanted to capture those
Running Head: PSYCHOSOCIAL AND BEHAVIORAL HEALTH
3
with high situational stresses for this life change, we chose a more stringent measure of
tertiles. We also wanted to use a common metric across domains for forming the poor health
scores in the absence of established cutpoints for the other domain scales.
Relationships among the psychosocial and behavioral scales were analyzed using
Pearson and Spearman correlations. Because these were not substantially different, only the
Pearson correlations are presented in Table 2 in the manuscript.
We examined risk factors for both the cumulative poor health scores as well as
individual domains in order to determine how these might vary across these situations. For
example, a risk factor might be just a strong predictor of one individual domain but not of
other domains, and thus only relate weakly to the cumulative poor health score.
Missing responses to items on the five measures of psychosocial and behavioral health
were accommodated by use of value substitution of pro-rated scores if less that 15 percent of
items were blank; for more extensive missing data on the five measures, cases were deleted
from the dataset for this study. Thus, all cases used in this study had available data on the five
psychosocial and behavioral health measures. For risk factor variables, there were a few cases
with missing data (see Table 1). Cases with missing data on risk factors were excluded from
certain analyses requiring complete data.
In regard to the Poisson regression model used to test risk factors for cumulative poor
psychosocial and behavioral health, both Deviance and Pearson Goodness-of-Fit tests indicated
slight over-dispersion of the model (Value/df = 1.335 and 1.126, respectively). Because of this
slight over-dispersion, a secondary Negative Binomial regression model was run in an attempt to
find a better model than the Poisson model. However, a -2 Log Likelihood difference test
showed that the Negative Binomial model did not fit the data better than the Poisson model
Running Head: PSYCHOSOCIAL AND BEHAVIORAL HEALTH
4
(difference = 1.99, df = 1, p = 0.16) as tested with a likelihood ratio test (Hosmer and Lemeshow,
2000). Thus, the decision was made to retain the Poisson model. For a review of models
appropriate for count data, see Long, 1997.
References
Hosmer, D.W. & Lemeshow, S. (2000). Applied logistic regression (2nd ed.). New York:
John-Wiley and Sons.
Long, J. S. (1997). Regression models for categorical and limited dependent variables.
Thousand Oaks, CA: Sage Publications.
Download