HSP intervention

HSP is significant in that it is one of the longest-standing programmes in Arizona and employs CHWs in 14 distinct Arizona counties to engage at-risk, low-income mothers in order to improve birth outcomes (figure 1). CHWs serve as the primary interventionist for the programme. In 2016, HSP CHWs provided services to 2534 unduplicated clients, conducted 16 698 home visits and facilitated 461 classes.28 Women are eligible to enrol in HSP if they (1) live in the targeted service area, (2) are pregnant or post partum with a child under the age of 2 years and (3) have one or more social or medical risk factors. Social risks can include but are not limited to: single-parent status, underserved racial or ethnic group, education equal to or less than high school level, annual income less than $40 000, and Medicaid or no insurance. Medical risks are broad and can include previous preterm birth, low birth weight, chronic disease, high maternal body mass index (BMI) and substance use. Women can be of any age and there are no income requirements to participate.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Arizona Health Start Program service area map, 2018. Map demonstrates the Arizona Health Start Program service areas within 14 counties across the state. Community health workers conduct regular home visits to under-represented pregnant women and their families in rural and urban communities. Map courtesy of and permission by Arizona Health Start Program, Arizona Department of Health Services. This map is not under copyright.

CHWs connect clients to prenatal care and increase client’s continuity of care during and after pregnancy. CHWs identify, screen and enrol eligible women; provide prenatal and postpartum education; provide referral and advocacy services; and emphasise timely immunisations and developmental assessments for children. Although not an exhaustive list, table 1 outlines the primary intervention activities conducted by the CHW. HSP CHW home visits are guided by an asset-based approach and two primary theories of behaviour change, the Trans Theoretical Model and the Social Cognitive Theory. Identifying assets acknowledges and supports the existing strengths and capabilities of individuals and resources to promote community-driven development and positive change.34 The Trans Theoretical Model assumes that behaviour modification in individuals is a multistage process in which people move through stages of readiness for change,35 and Social Cognitive Theory states that stages occur in the context of reciprocal relationships between the person’s environment, their behaviour and their cognition.36 CHWs are a community asset and well positioned to support HSP clients; they share both lived experiences and cultural knowledge of the community they serve. The home visiting sessions promote behaviour change through assessment, goal planning, referral, advocacy and follow-up activities, coupled with education through meaningful adult learning models. These interactions are designed to encourage personal agency of adult learners to integrate new knowledge and create a cognitive structure that makes sense of their own surroundings and situations.37 Through behaviour change theories and adult learning models, the HSP CHWs privilege the co-construction of knowledge among all participants, assume all are co-learners, and encourage critical thinking about self-sufficiency, empowerment, and personal agency related to the five HSP goals (table 1).

HSP CHW core competencies, roles and training

According to the HSP policy and procedure manual, CHWs must (1) live and work in the service area, (2) reflect the ethnic, cultural and socioeconomic characteristics of the communities they serve, (3) be able to read and write in English, (4) have a high school diploma or General Educational Development and (5) pass a criminal history background check within the Department of Public Safety records to be eligible to work for the state-funded programme. It is highly recommended (though not required) that CHWs have posthigh school training and education in MCH, early childhood development education, family studies, social work, nursing or a closely related field.28 Before a CHW can initiate unsupervised outreach or home visits, they must complete 40 hours of training in both the 10 CHW Core Competencies set forth by the CHW Core Consensus Project,38 which are recognised by the Arizona state legislature HB 2324 Voluntary CHW Certification,39 and the HSP Core Training.28 An additional 8 hours of home visit shadowing with a senior CHW are required.

Nationally recognised, the 10 CHW Core Competencies include: (1) Cultural and Systems Mediation; (2) Culturally Appropriate Health Education; (3) Care Coordination and Case Management; (4) Coaching and Social Support; (5) Advocacy; (6) Capacity Building; (7) Direct Service; (8) Individual and Community Assessments; (9) Outreach and (10) Research and Evaluation.38 HSP Core Training covers: (1) Essential Health Start Information (HSP basics, visits and community outreach); (2) Communication and Emotional Support; (3) Nutrition and Physical Activity (family nutrition and physical activity, infant nutrition and physical activity); (4) Health Education (healthy pregnancy, prenatal care, discomforts during pregnancy, labour and delivery, postpartum care and family planning, early childhood development and parenting skills, infant health and child health) and (5) Safety (home safety for infants and children, child abuse and domestic violence).28 CHWs are required to complete 12 hours of continuing education per year.

Intervention cohort

HSP administrative data from 2006 to 2015 is the primary source for identifying the retrospective intervention group. All Health Start clients enrolled during the 10-year observation period will be included in this study if their records are identified and linked from the HSP database to the Vital Records Birth Database (VRBD). Records will be linked based on the mother’s date of birth and first name. In order to be a candidate for the HSP study cohort, the mother’s date of birth must be an exact match while her first name must be at least 95% similar, using Jaro-Winkler similarity.40 Mother’s last name may change due to marriage; therefore, this criterion is not required to identify the intervention cohort. We will obtain the following information for each HSP study cohort mother: a unique identifier (ID), first name similarity percentage, last name similarity percentage, HSP enrollment date, programme closure information (ie, programme completion, reason for closure) and the child's birthdate. Using the process described above, 15 576 HSP records were linked to the VRBD.

Intervention cohort sample size

Our evaluation intervention group will include all HSP participants enrolled within 24 months of the date of birth of the child during 2006–2015. Of the initial 15 576 records identified through the HSP-to-VRBD data link, 5911 fall outside of the 24-month (either before or after) HSP enrollment window and will be excluded from all subsequent analyses. The resulting 9665 HSP-associated births constitute the basis of this study (figure 2). Because HSP participants can enrol before or after birth, we will limit the analysis for Aims 1 and 2 (newborn and maternal health outcomes) to those births for which the mother was enrolled during pregnancy. This final criterion results in 6493 HSP-attributed births for the evaluation of Aims 1 and 2. Aim 3 (child health outcomes) will be evaluated using the larger set of 9665 HSP-associated births.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Flow chart of intervention participant inclusion and exclusion criteria. Nine thousand six hundred and sixty-five Health Start Program births constitute the basis of this study. Fifteen thousand five hundred and seventy-six records were initially identified as Health Start Program matches; however, 5911 records were excluded because the child’s birth fell outside of the 24-month (either before or after) enrollment window. We evaluate Aims 1 and 2 with a subgroup: records for mothers enrolled in HSP prior to the child’s birth (6493 births). We evaluate Aim 3 using the larger set of 9665 HSP-associated births.

Synthetic comparison group

A comparison group of women not exposed to the HSP (non-HSP) will be created from all births occurring in Arizona during 2006–2015 (derived from VRBD). After identifying our study population we will use propensity score matching (PSM) to generate a statistically-similar synthetic control group that has, on average, the same observable preprogramme characteristics as the HSP mothers.41 The pool of potential comparators will come from all Arizona births that occurred over the study period (2006–2015). This process will be guided by HomVEE standards requiring that the covariates used to balance the treatment and control groups be associated with both treatment status and the outcomes of interest.42 Because the HSP eligibility criteria focus on social and medical risks, we will prioritise these types of measures in the PSM model, in addition to characteristics that have been shown to have strong associations with our outcomes of interest in previous empirical and theoretical work.

We will employ radius matching to identify comparison group mothers across the common support region.43 We will use the following measures in the PSM model: mother’s birth year, mother’s age at birth, county of residence. Additional indicator variables include: child’s birth order, maternal educational attainment, health insurance payer (Medicaid being a proxy for low-income status), race, ethnicity, availability of information for the father on the birth certificate, maternal country of birth, previous history of hypertension and median household income by zip code of residence. In addition to these demographic and socioeconomic characteristics, we will restrict potential comparators to mothers within the same fiscal years in order to account for economic conditions and any potentially shifting programme parameters. Imposing within-year matches will allow us to analyse the programme’s efficacy over time by cohort.

Comparison mothers may match to more than one HSP mother, based on the propensity score. Preliminary efforts to identify matches resulted in a potential synthetic comparison group of nearly 23 000 non-HSP mothers. Due to the respective sizes of the intervention and comparison populations, lack of statistical power is not a significant issue for this project.

Patients and the public were not involved in the design or planning of the study.

Outcomes

Data management

The four datasets that will be accessed for this study will be securely stored and protected through an honest broker. We designated the Center for Biomedical Informatics and Biostatistics’ Biomedical Informatics Services at the University of Arizona as the honest broker to facilitate the de-identification, transfer, and management of data, as well as maintain protected health information anonymisation and Health Insurance Portability and Accountability Act (HIPAA) compliance. In this role, the honest broker can identify individuals overlapping between relevant databases, and assign de-identified study codes that would enable cross-linking individuals between the systems.

Data monitoring

The honest brokers will link the HSP database to the VRBD to generate a comparison group. They will match both the HSP and non-HSP groups to HDD and ASIIS databases using personally identifiable information (eg, name, date of birth, social security number). The honest brokers will create a separate de-identified ‘limited data set’ for our analyses to compare the mean outcomes of HSP mothers to the comparison group mothers.

Statistical analysis

The motivation for using PSM to create a synthetic comparison group is to be able to observe the ‘counterfactual’ to HSP participation, that is, what would have happened in the absence of the programme. We will explore this by comparing outcomes between HSP mothers and those matched to them by the propensity score. More specifically, the average treatment effect generated by PSM will estimate the impact of the programme on the population of both HSP mothers and those who ‘look like’ HSP mothers by taking the difference in outcomes between HSP mothers and their matches, and vice-versa.

Our analytic population is of sufficient size to detect meaningful programme effects from low-frequency events, including preterm births, low and very low birth weights, maternal morbidity, and differences in immunisation and hospitalisation rates over a relatively long period. This is also true for specific subgroups served by HSP (eg, Hispanics, American Indian, economically disadvantaged).

Once we establish proper covariate balance between the intervention and matched-control groups, point estimates of the treatment effects will be estimated by comparing outcomes using Stata V.14 software and specifically the teffects command.44 Following Abadie and Imbens,45 46 this command considers the fact that propensity scores (ie, the parameter that determines the comparison population) are estimated when calculating the standard errors, and thus generates confidence intervals. The propensity scores will not be used as a covariate in traditional regression analysis because it is less effective in forcing baseline equivalence and assumes the relationship between the score and the outcome is linear.41

Both the HSP enrollment information and VRBD are administrative data sources, established and maintained for public health monitoring purposes. As such, we do not anticipate missing data to be a significant issue. We assume that such instances (as we find them) are very likely to be the result of human error and not any systematic issues with the data collection and/or reporting processes. Where missing-ness does occur in the variables that make up the propensity score model, we will control for these using dummy variables in place of the missing observations. In the case of missing outcome variables, we will restrict the analytic sample to the non-missing observations, and inspect control variables to verify that there are no systematic differences.