Methods
Theoretic Framework
This analysis uses the Andersen's Behavioral Model of Utilization of Care to predict factors influencing women's access to ICC. This model is well documented and widely used to determine predictors of access to care. The framework posits that access to and utilization of care can be predicted by a predisposition of people to use services, factors that enable or impede this use, a person's perception of need for care, and systems factors. Some of the specific 'predisposing' factors include age, gender, education, and ethnicity. 'Enabling' factors include having insurance; 'Need' factors include perceptions about health; and 'structural/systems' factors include such things as transportation and childcare. Recent developments in the tool have taken into account additional factors affecting vulnerable populations, including immigration status, acculturation, neighborhood conditions, psychological resources, housing mobility, mental illness, competing needs and food sources.
Study Population
We conducted a secondary analysis of subjects participating in a randomized clinical/behavioral trial of an interconceptional preterm birth risk reduction intervention in Philadelphia. Resident women experiencing a preterm birth at < 34 weeks of gestation were enrolled in the parent study. Women in the intervention arm received a series of intensive interventions designed to reduce their risks related to inflammatory pathways leading to a subsequent preterm birth. Six specific risks were addressed because of their common contribution to the inflammatory pathways to premature birth. These include genito-urinary infection, weight control, depression, housing inadequacy (stressor), smoking cessation and periodontal disease. Interventions on risks contributing to an inflammatory pathway (smoking, depression, infectious disease burden and maternal stress, and achieving an appropriate BMI), were introduced to decrease systemic inflammation and risk of repeat PTB.
Parent Study Data Collection
At the hospital visit, all participants were interviewed after delivery and prior to discharge to elicit demographic and other information. Once the maternal interview was completed, participants were randomly assigned to either intervention or usual care. For women assigned to the intervention group, the first study visit was scheduled within four weeks of discharge from the hospital. The one-month post partum visit for intervention group women was conducted at Drexel University. For women with multiple risk factors, interventions were delivered in stages over the 2-year intervention period. All intervention services scheduled and received were carefully documented and entered into tracking software. In addition, the services were delivered in a method that removed as many barriers as possible given existing knowledge of known barriers to care. All services were free of charge; women were provided with transportation, childcare and social support as needed; and appointment reminders were provided. In some cases, financial incentives were provided.
All women enrolled in the intervention arm of the parent study were included in this analysis. (n = 442). This study was approved by the Drexel University and the UNC-CH IRB.
The goals of this analysis were to identify and validate specific factors that adversely impact on women's ability to participate in interconceptionally delivered preterm birth prevention interventions, once all known barriers to care are addressed.
We used the modified Andersen Behavioral model to determine key predictors of access to interconceptional care for women at high risk of PTB. Specifically, we wanted to assess (1) how components of the Andersen Behavioral Model differed in this inner city urban population by level of participation after the known barriers were removed, and (2) to what extent the components of the Andersen Behavioral Model predict utilization of interconceptional care interventions for women at high risk of a subsequent preterm birth.
Independent Variables
The independent predictors include Predisposing factors, Enabling factors, Need factors and Systems factors. These will herein be collectively referred to as PENS.
Predisposing variables are operationalized using age (continuous), marital status (married/not married), education (< HS, HS grad, > HS), family size (1 or 2 members vs 2 or more) and substance use (Y/N for alcohol or drugs).
Enabling Factors are operationalized using insurance status (Y/N), income (categorical), availability of social support (Y/N), neighborhood safety and quality (safe/not safe), and perception of competing needs (Y/N).
Need Factors are operationalized by including perceived health (good/poor), and reported diagnosis of major health problems (Y/N).
System factors are measured as self-report of prior experience with providers (Good/Poor). Some Systems factors defined by the Andersen model are not included in this analysis because they are addressed by the parent study intervention (e.g. transportation, childcare), or because we did not have the data to assess them (homelessness length, language barriers).
Dependent Variable
Utilization of interconceptional care is measured as the number of visits completed divided by the number of visits scheduled between date of enrollment into the parent study and December 30, 2007. Only visits that required the woman to travel to a clinical setting were counted in the denominator. All home visits and phone interventions were excluded, thus this measure does not represent the overall or intervention-specific participation rates of the parent clinical trial. The utilization patterns for in-clinic visits were aggregated for the following 6 parent study interventions: Weight control, infection (vaginal), periodontal disease, housing, smoking cessation and depression. The aggregate counts were divided by the total aggregate number of scheduled visits and the resultant participation rates were grouped into 4 categories for analysis: "None" (did not attend any of the scheduled visits), "Some" (attended 1% to < 50% visits); "Most" (attended 50%–99%); or "All" (attended all (100%) of scheduled visits).
Statistical Analysis
Chi-square analyses were conducted to assess bivariate associations between interconceptional care participation levels and the predictive factors of the Andersen Model. Proportional odds models (POM) were proposed to assess the Andersen Model's predictability, however the proportionality assumptions were not met. Therefore, participation was analyzed as a nominal variable, and a generalized logit model for nominal outcomes was fit.
We assessed (a) the collective predictive power of the PENS factors in the utilization of each specific interconceptional care intervention, (b) which construct (predisposing, enabling, need or system) has the strongest significant effect, and (c) which specific individual factors within the sets are most predictive of utilization of care. Backward elimination was used to determine significant predictors, with a selection to stay criteria of 0.20. "Attended all visits" was used as the reference group for all analyses. All analyses were conducted using SAS 9.2.