Introduction

Improving the psychological, social, and physical well-being among mothers, infants, and children is an important public health goal for the United States. Poor maternal health is associated with complications in pregnancy and childbirth, which, in turn, may result in suboptimal developmental outcomes across a child’s life course (Centers for Disease Control and Prevention 2016). Postnatally, the well-being of mothers (in addition to fathers and other co-parents) is an important factor predicting future public health challenges for families and communities (Office of Disease Prevention and Health Promotion 2016). Recognizing the importance of maternal, infant, and early childhood health and safety, the Maternal, Infant, and Early Childhood Home Visiting (MIECHV) program was enacted in 2010 through an amendment to Title V of the Social Security Act. Home Visiting pairs at-risk families with trained home visitors who provide ongoing education, screening, social support, and referral to additional community services. Although home visiting has traditionally focused on mothers, in part as conduits to the larger family unit, MIECHV funding does not generally restrict participation to mothers as the only primary caregivers.

Home visiting programs vary significantly in their models of service delivery, program content, and staff training and experience. Thus, outcomes also vary. Further, studies, mostly observational, that focus on individual home visiting models or specific types of target families, are not easily generalizable (MacMillan et al. 2009). Some studies have raised questions about the effectiveness or magnitude of effectiveness of home visiting programs (Gomby 2005; Gomby et al. 1999; Sweet and Appelbaum 2004), while others have demonstrated positive outcomes for mothers including decreases in parental stress and risk of mental health problems (Ferguson and Vanderpool 2013), reductions in child abuse and neglect (Avellar and Supplee 2013; Bilukha et al. 2005), and increases in positive mother–child interactions (de la Rosa et al. 2005). For child participants, home visiting programs have been associated with reduced incidence of teacher-rated aggressive behavior problems (Lyons-Ruth and Melnick 2004), fewer internalizing and externalizing difficulties (Lowell et al. 2011), and higher intellectual functioning (e.g., teacher-rated academic engagement and classroom social skills; Olds et al. 2004; Roggman et al. 2009).

The benefits of home visiting, however, cannot accrue unless at-risk families enroll and remain engaged in services. Most previous studies evaluating home visiting programs assessed the effect of participation dichotomously, as participating or not participating, with little attention given to variations in levels of engagement (Ciliska et al. 2001; Olds et al. 2002; Sangalang et al. 2006). However, both participation intensity and dosage were found to be positively associated with beneficial birth outcomes (Goyal et al. 2013; Slaughter and Issel 2012) and with maternal responsiveness and warmth toward their children (de la Rosa et al. 2005; Guttentag et al. 2014). Children in higher intensity home visiting programs also demonstrated higher levels of social engagement and more advanced language skills (Guttentag et al. 2014).

In previous research, various participant demographic characteristics, such as mothers’ age, employment status, and marital status, were found to be associated with program engagement (Ammerman et al. 2006; Daro et al. 2003; Goyal et al. 2014). Conversely, few studies have investigated the role of contextual factors in engaging at-risk families. Even less is known about how both family- and community-level characteristics concurrently affect program retention, dosage, and intensity.

Based on ecological and family system frameworks, McCurdy and Daro (2001) proposed a conceptual model of individual and contextual factors that influence parental involvement in family support programs. Community factors, such as poverty, crime, and instability, were suggested as risk factors that may hinder parents’ willingness to engage in support programs. Indeed, more recent empirical studies demonstrated that mothers living in communities characterized by socioeconomic deprivation were less likely to enroll in home visiting services (Goyal et al. 2014). Living in a county in which community health is poor and community violence is high was also associated with a reduced likelihood of engagement in home visiting (McGuigan et al. 2003).

Informed by an ecological framework (Bronfenbrenner 1986; Leventhal and Brooks-Gunn 2003) and the study of McCurdy and Daro (2001), we evaluated both family- and community-level characteristics associated with retention, dosage, and intensity of engagement in MIECHV-funded home visiting services among a general population of at-risk mothers living in Georgia. Based on previous empirical findings, we hypothesized that demographic characteristics of both the mother and the child would be associated with one or more program engagement outcomes. In addition, we hypothesized that adverse community environments, defined as material disadvantages and high-risk conditions for children, would have unique influences on program engagement outcomes, independent of family characteristics.

Method

Participants

In the fall of 2010, a state-wide needs assessment was conducted, directed by epidemiologists within the Maternal and Child Health Program of the Georgia Department of Public Health, and coordinated with Georgia’s Title V Maternal and Child Health Services Block Grant, Title II of the Child Abuse Prevention and Treatment Act, Community-Based Child Abuse Prevention Program, and Head Start Collaboration Office, and other state-wide stakeholders. Georgia’s 159 counties were assessed using maternal and child health, poverty, crime, high school completion, domestic violence, substance abuse, and child abuse and neglect indicators (Georgia Governor’s Office for Children and Families 2011). The top 25 at-risk counties were then ranked a second time based on estimates of their likely success in scaling up implementation of home visiting services within a comprehensive, community-based early childhood system. For example, counties that had prior, demonstrated commitment to early childhood prevention efforts, as well as experience with high-fidelity implementation of an evidence-based home visiting model, were given higher scores. Consideration was given to each county’s geographical location, with the goals of assuring equity across the state and learning as much as possible about implementation of home visiting in urban, suburban, and rural settings. In the end, after concurrence of state and county stakeholders, seven counties were chosen to participate in Georgia’s Maternal, Infant, and Early Childhood Home Visiting (MIECHV) program (see Table 1).

Table 1 Selected characteristics and risk factors for counties chosen for Georgia’s maternal, infant and early childhood home visiting program
Full size table

Federal funding of state, territorial, and tribal entities’ MIECHV programs was authorized through an amendment to Title V of the Social Security Act. The initial $1.5 billion over five years starting in 2010 represented a significant increase in federal funding of home visiting from $13.9 million in 2009. Funding for MIECHV was re-authorized in 2015 through an amendment to Medicare’s Sustainable Growth Rate Bill (Kelly 2015).

During the study period from January 1, 2012 through September 30, 2015, Georgia’s MIECHV program provided funding to implement three evidence-based home visiting models: Healthy Families America (HFA: http://www.healthyfamiliesamerica.org); Nurse-Family Partnership (NFP: http://www.nursefamilypartnership.org); and Parents as Teachers (PAT: http://www.parentsasteachers.org). In one county, implementation of all three home visiting models received funding through Georgia’s MIECHV program (Muscogee County) and in another county, two home visiting models were funded (Whitfield County, HFA and PAT). The remaining five counties each implemented one of the home visiting models through funding from Georgia’s MIECHV program (see Table 1).

Families potentially eligible for home visiting were screened for risk factors and eligibility through Georgia’s First Steps program, a community-based service that connects families to community resources appropriate for expectant mothers and families with children birth to five years of age. Based on availability of home visiting within a mother’s community, mother’s fit with a specific home visiting model (when multiple models were available), and verbal consent, the mother’s contact information and screening data were given to the appropriate community-based organization for initial home visiting engagement activities. Mothers’ written consent for receipt of services, gathering of ongoing screening and home visiting service-related data, and use of data for performance monitoring were collected at the initial enrollment visit. Our study, after review and approval by the University of Georgia’s Institutional Review Board, made use of this routinely collected data.

During the study period, 1486 expectant women and mothers with children under the age of five years old, enrolled in home visiting through one of the seven MIECHV-funded community-based organizations. These women and mothers identified themselves as the primary caregiver, as per Georgia’s MIECHV-program protocols at the time of the study. Mothers who continued their participation in home visiting past the study end date of September 30, 2015 were excluded from the study sample (n = 462) because their inclusion could have skewed our results. In the end, 1024 mothers who participated in home visiting between January 1, 2012 and September 30, 2015 were included in the final sample. There were no other eligibility criteria for inclusion in our study.

Procedure

Data are routinely collected on families participating in MIECHV-funded home visiting, primarily for federal and state mandated performance reports. Home visitors collect these data during family sessions. Upon the home visitor’s return to the community-based organization, the home visitor or other data entry staff enter the data into the Georgia Home Visitation Information System (GEOHVIS). In addition to data entry in GEOHVIS, the NFP programs also enter data into a stand-alone data system for nationwide model-specific reporting. Data exports from this system are routinely provided to Georgia’s Data Evaluation Coordinator to be combined with the GEOHVIS dataset. For our study, secondary data were exported out of GEOHVIS and aggregated into a single file for analysis.

Measures

Family engagement—Home visiting program participation

Three family engagement outcomes were assessed: (1) retention, the duration of participation in months; (2) dosage, the number of visits completed; and (3) intensity, the number of visits completed divided by the duration of participation. Using the program participation start date (i.e., the date of program enrollment) and the date of the last home visit, we first calculated the number of program participation days. Then, the calculated number of days were divided by 30 to generate program participation duration in months, rounding off the values to the nearest integer. Retention, the interval between the program start date and the date of the last home visit, was coded as a continuous variable. Participation less than thirty days was coded as 0. Dosage was measured as a count variable, indicating the number of home visits a family received. An intensity measure was further calculated by dividing dosage by the duration of participation to yield the number of home visits per month. Intensity was coded as a continuous variable.

Family demographics

Primary caregivers, all of whom were biological mothers, reported family demographic characteristics at enrollment. Although many of the mothers had more than one child, during the study period the MIECHV program policy was to collect data on one target child, except in the case of multiple births when data were collected on all multiples (e.g., both twins). Demographic variables included maternal age, relationship status, employment status, educational attainment, race/ethnicity, and primary language; child age and gender; number in household; and annual household income. The relationship status variable was coded as 0 (not living with a partner) or 1 (living with a partner) using mothers’ self-report on the presence of live-in romantic partners, including cohabiting (i.e., non-marital) and married partners. Maternal employment status was coded as 0 (not employed) or 1 (employed), and educational attainment was coded as 0 (less than 12 years) or 1 (12 years or more). Mother’s race/ethnicity was coded by three categories (Caucasian, African American, and Other) and included as independent variables in the model, resulting in regression coefficients that can be interpreted with reference to Caucasians. Mother’s primary language was coded as 0 (English) or 1 (other languages, which included Amharic, Burmese, Farsi, French, Nepali, Somali, Spanish, and Sudanese). The poverty threshold was set at 100% of the poverty limit according to U.S. Health and Human Services guidelines (Office of the Assistant Secretary for Planning and Evaluation 2011) and was dichotomously coded as 0 (above poverty level) or 1 (below poverty level), taking into account annual household income and number in household. Child gender was coded as 0 (female) or 1 (male). Child age at program enrollment was calculated using the mother-reported date of birth and then coded as 0 (prenatal), 1 (neonatal, age  <  1 month), 2 (1–12 months), 3 (13–24 months), or 4 (over 24 months).

Community-level factors

Community-level factors were assessed at the county level using data from the American Community Survey (ACS; U.S. Census Bureau 2012) and the Kids Count Data Center (Annie E. Casey Foundation 2012). Participants’ residential addresses were geocoded and matched to county level Census data. Four indicators of community disadvantage were assessed using 2012 ACS data: percentage of single-parent households, percentage of households receiving Temporary Assistance for Needy Families (TANF), percentage of unemployed adults, and median population income during the past 12 months (reverse coded). Three additional indicators were assessed using information from the 2012 Kids Count Data Center. These included proportion of low birthweight babies, unduplicated count of children with a substantiated incident of abuse or neglect per 1000, and number of deaths of infants less than 1 year of age per 1000. Following the example of prior studies that supported the concept of cumulative community risk and its links to both physical morbidity and psychological dysfunction (Brody et al. 2013; Cho and Kogan 2016), a cumulative index of community disadvantage was developed based on these seven indicators. Each indicator was coded dichotomously by median split (0 = absent, 1 = present). Scores were summed to indicate the extent of community disadvantage and risk for children, yielding an index that ranged from 0 to 7.

Data Analyses

Descriptive statistics and correlations were calculated for all study variables. Specifically, correlations between family demographic factors and community-level factors with program engagement outcomes were investigated using Pearson’s bivariate analysis. Next, because the study participants were clustered within counties, hypotheses were tested using two-level models in hierarchical linear modeling (HLM) implemented in Mplus, Version 7 (Muthén and Muthén 1998–2015) to correct the standard error estimates. Missing data were managed with full information likelihood estimation (FIML). FIML does not impute missing values; rather, it estimates model parameters and standard errors from all available data, minimizing potential bias that might influence results (Enders 2001). To produce a common scale and to reduce the likelihood of multicollinearity, the scores for all study variables were transformed into z-scores before the HLM analyses were conducted.

In the first model, influences of family demographics on each family engagement outcome were tested. In the second model, the community risk index was included, after controlling for family-level factors, using the HLM analysis command to adjust the parameter standard errors for interdependence in the data. In addition, all models controlled for home visiting model. Possible confounding associated with mothers’ year of enrollment was investigated, but no significant effects were observed (data not shown).

Results

Mothers ranged from 12 to 49 years of age at enrollment (M = 22.89, SD = 5.47); 87.8% (n = 899) were 18 years of age or older (see Table 2). At enrollment, 56.4% (n = 461) of mothers were married or living with a main romantic partner, and 60.3% (n = 488) enrolled in a home visiting program at or after the birth of the target child. A total of 815 target children were included in the study, of whom 54.0% (n = 440) were male. Retention ranged from less than 1 month to 41 months (M = 8.34, SD = 9.47). Dosage ranged from 1 to 103 sessions (M = 17.47, SD = 18.48). Intensity of engagement was calculated to average 2.32 home visits per month (SD = 1.19) and was not found to be significantly associated with the number of home visits observed for the study sample (r = .002, p = .960). A majority of the mothers were English-speaking (88.5%, n = 906), African American (62.2%, n = 634), had more than 12 years of education (62.0%, n = 629), and were unemployed (72.7%, n = 699). Of the mothers, 76.5% (n = 543) reported annual household income below 100% of the poverty line.

Table 2 Descriptive statistics for study variables
Full size table

In bivariate analysis, retention and dosage were significantly associated with maternal relationship status, education level, race/ethnicity, and primary language; child age; and family poverty status (see Table 3). Intensity was significantly associated with maternal age and employment status and child age. In addition, negative community environments were generally associated with low levels of the engagement outcomes.

Table 3 Bivariate associations between family/community factors and program participation outcomes
Full size table

In our hierarchical linear regression models, we examined family-level influences on program participation (Models 1, 3, and 5 in Table 4) and then the influence of the community risk index, after controlling for family-level influences (Models 2, 4, and 6 in Table 4). In Model 1, we observed that mothers who lived with a partner ( β = .15, p < .01) or spoke a primary language other than English ( β = .22, p < .01) were statistically more likely to remain engaged in home visiting for a longer duration than mothers who were not living with a partner or who spoke English as their primary language. Conversely, living below 100% of the poverty line ( β = −.09, p < .05) and child’s age at enrollment ( β = −.15, p < .01) were significantly and negatively associated with program retention. In Model 2, we added the community risk index. Although the association between community risk and retention was not statistically significant, the duration of program participation was observed to trend negatively with increases in community risk ( β = −.06, p < .10).

Table 4 Hierarchical regression results for family- and community-level factors’ effects on participants’ engagement in home visiting programs
Full size table

Next, we tested the effects of family-level factors on dosage (Model 3). Maternal relationship status ( β = .13, p < .01), primary language ( β = .19, p < .01), poverty level ( β = −.15, p < .05), and child’s age at enrollment ( β = −.15, p < .01) were significantly associated with dosage. Model 4 demonstrates that, after controlling for family-level effects, mothers living in negative community environments were more likely to complete fewer home visits than were mothers living in more positive community environments ( β = −.11, p < .05).

Finally, in Model 5, intensity of participation was negatively associated with maternal unemployment ( β = −.08, p < .05) and child age at enrollment ( β = −.09, p < .05). When the community risk index was included as a predictor in Model 6, the significant associations of maternal unemployment and child age with intensity were observed to disappear, while increases in community risk were significantly associated with decreases in intensity ( β = −.09, p < .05).

Discussion

In the present study, we investigated the influence of family- and community-level social determinants of ongoing engagement in home visiting for mothers participating through Georgia’s MIECHV program. Family-level factors included both maternal and child characteristics. Community disadvantage was measured via a cumulative risk index that captured variation in material adversity and circumstances affecting child health and safety. In prior studies, persistent developmental and health outcomes of cumulative community disadvantage were considerably stronger than were the independent effects of individual dimensions of community adversity (Cho and Kogan 2016). Hierarchical linear modeling was used to account for the clustering of families within counties.

We found that maternal relationship status, the primary language spoken in the home, family poverty level, and child age at enrollment in home visiting were significantly associated with program retention and dosage. Specifically, mothers who lived with romantic partners, who spoke a primary language other than English, whose family income was above the poverty level, and whose children were relatively younger at enrollment were more likely to engage in home visiting for a longer duration and complete a greater number of home visits. These results suggest that mothers who have relatively more relational and material resources provided through a stable partnership and income above the poverty level may be more likely to engage actively in a home visiting program. Mothers who do not speak fluent English and whose children are younger may have greater needs for family support programs, leading them, also, to participate more actively than other groups.

Our findings are consistent with previous studies that observed women with stable romantic relationships, sufficient social and economic resources, or heightened need for social support due to the age of their child (e.g., a new mother) were more likely to remain engaged in home visiting programs (Brookes et al. 2006; Daro et al. 2003; Dumka et al. 1997; Goyal et al. 2014; Navaie-Waliser et al. 2000). Although this association has not been universally supported (e.g., Ammerman et al. 2006), our findings suggest that participation is generally diminished among women at greater socioeconomic risk, specifically those who may need family support services the most. Interestingly, intensity of program engagement was associated only with maternal employment status. Employed mothers participated less frequently than did mothers who were not employed. This suggests that time constraints likely constitute a barrier to active engagement in home visiting. Other family-related socioeconomic factors did not predict program intensity.

In addition, we observed that cumulative community disadvantage was negatively associated with all aspects of program engagement. Mothers living in disadvantaged communities in which health and safety risks for children were relatively greater were more likely to be retained for shorter durations (trend), completed fewer home visits, and participated with less intensity than did mothers living in more positive contexts. These results demonstrate that community-level factors are associated independently with home visiting engagement beyond effects attributable to family-level factors. Our finding is consistent with the few previous studies that showed community violence and economic deprivation to be associated with program retention (Goyal et al. 2014; McGuigan et al. 2003). Stressful experiences in the community, such as inadequate job opportunities, residential instability, and concentrated poverty, may lead mothers to feel trapped, hopeless, and frustrated (McGuigan et al. 2003). This undermining of the family system’s mental and physical health resources creates barriers to program participation.

Limitations

Several potential limitations to the scope and generalizability of our results should be mentioned. First, we did not consider potential maternal psychological and behavioral characteristics that may be associated with both family- and community-level characteristics and may significantly influence program engagement outcomes. For instance, maternal depressive symptoms and health risk behaviors such as substance use have been found to be associated with home visiting participation (McCurdy et al. 2006; Raikes et al. 2006). Second, it is possible that mothers’ engagement is influenced by changes in family demographic characteristics over time (e.g., birth of a new child, change in employment status, etc.). Based on established service delivery protocols during the time of our study, demographic characteristics were recorded at enrollment and were not reassessed at a later timepoint. Consequently, we could not consider the potential influence of changes in demographic factors. Third, our focus on family and community-level factors overlooks potential impacts on family engagement that arise from the community-based organization and home visitors providing services. For example, home visitors’ education levels, training histories, responsiveness to families, and fidelity in providing home visiting model content were not included in our analyses. Previous studies have assessed the impact of home visitor characteristics using self-report measures of family satisfaction; fidelity to the home visiting model; and home visitor demographics, education, prior experience, and work style (Grange et al. 2011; McGuigan et al. 2003). Due to data restrictions, however, it was outside the scope of our current research to investigate the complexity that arises from main effects, interactions, and feedback between family, home visiting program, and community characteristics. Fourth, we did not consider the completed percentage of expected visits when assessing dosage. Although recommended visits are specified in each models’ curriculum, the strength of the recommendation varies by program. Further, there are factors that may influence the definition of recommended visits, for example: timing of enrollment (e.g., in relation to holiday periods), short-term events within a family’s life, transitions in home visitors, and home visitor discretion based on families’ needs or stated priorities. As a result, retrospective assessment of the completed percentage of expected number of visits, for a dataset of the size used in the study, would introduce significant complexity and an array of assumptions. We believe controlling for home visiting program captures much of the variation in model-specified recommended visits. Lastly, our study did not include fathers and others as primary caregivers, but our results did highlight the positive impact of a live-in partner on family engagement. Additional work is needed to specifically consider the potential factors that facilitate or hinder fathers’ participation in home visiting programs, recognizing their important role in parenting.

Despite these limitations, the present study makes a valuable contribution to existing research by elucidating the ways in which both family-level and community-level characteristics independently contribute to program engagement among at-risk families. More importantly, our investigation expanded upon previous research by considering public health aspects of community risk, specifically community-level threats to child health and safety, including the proportion of low birthweight babies and rates of infant mortality and child abuse or neglect. The findings of our study underscore that community disadvantage should be considered from multiple perspectives when investigating potential effects on outcomes of home visiting programs.

In an equitable society, families at highest risk should have the greatest access and receive the most potential benefit from family support programs, like home visiting. In 2013, the national preterm birth rate in the US was 11.4%, with associated medical costs and lost economic productivity estimated to be $4.33 billion (Trasande et al. 2016). In one study of home visiting, the rate of preterm birth for mothers enrolled in home visiting prenatally was 5.1%, as compared to 9.8% for mothers in the control group (Lee et al. 2009). In a second study of children 25 to 50 months of age, the risk of accidents and injuries was reduced by 40%, visits to the emergency department was reduced by 35%, and physicians recorded 45% fewer behavioral and parental coping problems, when measured against a comparison group (Olds et al. 1994). Home visiting programs’ potential benefits, however, are reduced when caregivers do not remain engaged.

At the program level, interventions to improve family engagement should be tested, explicitly considering the various states of community disadvantage in which families live. For instance, as part of the Home Visiting Cooperative Improvement and Innovation Network (HV-CoIIN), family engagement was initially designated as a compulsory innovation topic for all participating home visiting programs (Maternal and Child Health Bureau, Division of Home Visiting and Early Childhood Systems and Education Development Center, Inc. 2017). The large body of results obtained from the participating programs’ multiple Plan-Do-Study-Act cycles could be evaluated in greater depth, e.g., through realist synthesis, to move beyond “what works?” to “what works, for which families, under what circumstances?” (Rycroft-Malone et al. 2012).

At the policy level, it is crucial to ensure the resources needed to reach families at greatest risk are available. It is plausible that families with low socioeconomic status are more likely to live in disadvantaged communities, which may concurrently place a dual-burden and discouragement to be engaged in and helped by federal, state, and local family support policies and programs. Models of community disadvantage could be explicitly considered in future home visiting policies and funding decisions. Similar to Disproportionate Share Hospital adjustments made to payments by Medicaid and Medicare to hospitals that serve areas with larger proportions of economically disadvantaged patients (Centers for Medicare and Medicaid Services 2017), home visiting programs that provide services in areas of greater community risk could be funded at a higher level per participant. This could be accomplished by risk-adjusting home visitor caseload expectations in contracted performance deliverables. Increased time and resources would, in return, facilitate multifaceted approaches and continued innovation to engage our most at-risk families.