Discussion
Summary of Key Findings
The Andersen Behavioral Model has been used previously to identify barriers to care participation and contains a comprehensive set of factors known to influence utilization of care. In this study, we applied the model to a population that differed from previous studies in two fundamental ways: (1) the population of women was receiving preventive care in the interconceptional period, and (2) as part of the care protocol, all traditionally known barriers to care were addressed to facilitate participation in the clinical trial. We expected to be able to identify which factors above and beyond those already addressed as part of the care protocol exerted impact on women's participation in interconceptional care.
While no factors clearly distinguished the women who missed every one of their scheduled visits compared to those who attended all visits, smoking acted as a barrier, significantly increasing the likelihood those women would miss over half of scheduled visits (compared to ALL visits).
Strengths and Limitations
The disadvantage of self-reporting lies in the possibility of reporting bias. We expect some underreporting of substance use and possibly in self-report of quality of prior experiences with health care providers, but there is no reason to suspect that the underreporting varies by level of care participation.
All women were not assigned to the same constellation of interventions and some were scheduled for more invasive appointments than others. Burden of intervention may be a factor in women's ability to attend. That is, the more appointments they are scheduled for, or the more invasive the intervention, the more likely they may be to miss some appointments. We did not consider the type of intervention, only the aggregate participation, and looked only at the number of visits attended as a proportion of the number scheduled. We assessed the correlation between number of scheduled visits and participation rates (data not shown) and found that there is not a linear relationship between the two. For example, the mean number of scheduled visits among those with a 0 participation rate was 2.26, while the mean was 8.94 scheduled visits for those who attended "some" visits, 9.77 for those who attended "most" visits, and 2.58 for those who attended "all" visits. Also, the number of scheduled visits is not a direct marker for intervention intensity because a subject may have had a second visit scheduled because she missed the previous appointment. As such, a high number of scheduled visits could represent several re-scheduled appointments. Our categorization of the outcome variable may also be a limitation. These groupings were used to delineate participation patterns that have some practical conceptual meaning (none, some, most, all). Fewer categories may have yielded a more parsimonious model, but would not improve conceptual interpretation. For example, aggregating the "Some" and "Most" groups is not conceptually valid because we would be looking at a difference between those who attend 1% and those who attend 99% of visits. We tested an aggregation of the "Most" and "All" categories to assess for differences. In this case, the mean visits and standard errors differ significantly between these categories so information would likely be lost by this aggregation. We did not find any significant changes in the findings reported. Finally, an inherent problem with the Andersen Behavioral Model is that it only predicts utilization of organized health care, which assumes that the intervention effectiveness is only based on what happens in the clinical setting. For many of the indicated interventions, follow-up activities are required in the home and the community, thus utilization of organized care may not fully predict impacts of the care received. In these analyses, we are not looking specifically at outcomes of care, only at utilization of organized, clinic-based interconceptional care.
The Andersen Behavioral Model is a holistic theoretic model and incorporates multiple levels of contextual and behavioral factors to explain health care utilization. In addition, this study took advantage of the unique opportunities presented by the parent study which applied a comprehensive set of evidence-based risk reduction approaches during the interconceptional period. This is the first study to measure predictors of participation in a large-scale, organized, comprehensive interconceptional care intervention among a mostly African American, urban and vulnerable population. In addition, an evaluation of participation under the best-case scenario (that is--where all known barriers to participation have been addressed) is also undertaken.
Interpretation
The parent clinical trial recruitment demonstrated that addressing barriers to health service access might generate high participation of mostly African American women in interconceptional care. However, when it comes to the long-term participation required for most forms of preventive care, we find that other unknown factors continue to affect women's willingness and ability to participate consistently in these services.
Barriers to types of preventive care such as ICC may differ considerably from the barriers to prenatal care. Pregnancy provides a motivational force and immediacy of concern for the baby's health to spur participation. Additionally, social norms exist which create a negative view of women who do not participate in PNC. No equivalent motivators or deterrents exist for interconceptional care and we have no studies that have identified barriers specific to this type of women's preventive care. For this study, we tested a priori assumptions that the same barriers would apply to both PNC and ICC participation. Studies delineating the barriers to PNC have evolved over time. Models of PNC participation started out including a conglomeration of individual factors describing populations at risk (young, less educated, single, large family size). These models were later modified to include behavioral factors (substance use, stress, low social support). Later, ecologic models were used, including factors which exist outside of the personal domain as predictors of utilization. These include contextual factors that influence women's ability to get away from competing demands (job demands, childcare needs),to get to the health care site (transportation, income), and factors influencing the quality of treatment once in the health care site (provider availability, wait times, hours of operation, discrimination). The Andersen Behavioral models may be the first to incorporate these multiple levels including barriers that exist outside of individual control. Inclusion of systemic and contextual factors as barriers or facilitators of the receipt of care is significant because it can potentially change the locus of intervention from the individual to systems. While theoretically, the domains of the Andersen model capture an appropriate and broad conceptualization of factors associated with care utilization, it is how the components are operationalized that ultimately define how well it will predict preventive care utilization. The specific factors included in the model must reflect the unique contexts and experiences of the population under study. While we operationalized all suggested model variables, and other known barriers were addressed as part of the care protocol, none of the currently known factors accounted for the non-participation of some women receiving free care nor do they explain the full participation (100%) of other women. We conclude that although we had the data to operationalize almost all variables suggested as relevant to vulnerable populations (we lacked only information on language barriers and homelessness length), these models still do not capture all of the barriers and facilitators specific to utilization of interconceptional care utilization for this population of urban, mostly African American women. These factors can only be identified through a deep contextual look at the daily lives of the women.