Prenatal and Infancy Nurse Home Visiting Effects on Mothers: 18-Year Follow-up of a Randomized Trial

We conducted a randomized clinical trial of NFP in a public system of obstetric and pediatric care in Memphis, Tennessee, with registration of the original sample completed between June 1990 and August 1991. We enrolled women <29 weeks’ gestation with no previous live births and at least 2 sociodemographic risk characteristics (unmarried, <12 years of education, unemployed). Ninety-two percent were African American, and at enrollment, 98% were unmarried, 64% were <18 years of age, and 85% were from households with incomes below the federal poverty guidelines.⁷  Participants completed informed consent procedures approved by the University of Rochester Institutional Review Board.

In Supplemental Table 3, we summarize the Consolidated Standards of Reporting Trials information. Eighty-eight percent (1138) of 1289 eligible pregnant women offered participation completed informed consent and were randomly assigned to 1 of 4 treatment conditions following a procedure that concealed assignment from individuals involved in gathering participant data.⁷  A total of 742 participants were assigned to 2 treatment conditions created to estimate program effects on postnatal outcomes: 514 were assigned to treatment 2 (control), and 228 were assigned to treatment 4 (NV), both described below. Sample size and assignment ratios were derived from statistical power calculations in the original phase of the trial.⁷  In Supplemental Table 3, we show those lost to follow-up because of miscarriage or child death before age 2 and maternal refusal to participate at earlier phases and show the number evaluated with maternal assessments at youth age 18. Some mothers, because of disabling conditions or refusal, did not complete all assessments at youth age 18. Numbers of completed assessments for each outcome are given in Tables 1 and 2. Repeated measures over time increased numbers for some outcomes and are noted by “All” under the “Assessment Timeframe” column.

The study design contrasted women assigned to 2 treatment conditions established to estimate the effects of NFP (described below) after delivery (Supplemental Table 3). Women in the T2 control group (n = 514) were provided free transportation for scheduled prenatal care plus developmental screening and referral for children at 6, 12, and 24 months of age. Women in the T4 NV group (n = 228) were provided the same services as the control group, plus NFP home-visiting through age 2.

NFP was designed to (1) improve pregnancy outcomes by promoting women’s prenatal health, (2) improve children’s health and development by promoting parents’ care of their children, and (3) enhance parents’ health and life courses by guiding women to reduce closely spaced subsequent pregnancies, complete their educations, and find work. Nurses focused on reducing closely spaced subsequent pregnancies to protect the health of the mother^27,28  and offspring^29–33  and to help women gain traction in education and employment.^4,34,35  Nurses linked families with needed services and, when possible, involved other family members (especially children’s fathers and grandmothers) in the visits.⁴  Program protocols were grounded in developmental epidemiology and theories of human attachment, human ecology, and self-efficacy.⁴ 

The program was implemented by the Memphis and Shelby County Health Department during a nursing shortage, leading to nurse turnover for 37% of the families.⁷  Nurses carried a maximum caseload of 25 families each, relied on detailed visit-by-visit guidelines structured around 62 visits, and used their clinical judgment to adjust visit dosage and content, including some telephone visits, to address individual family needs.^4,36,37  Visits were structured to coincide with particular maternal and child health issues likely to emerge at particular phases of gestation and the first 2 years of life that addressed program goals. Overall, nurses completed a mean of 7 home visits during pregnancy and 26 visits during the first 2 years postpartum.^7,37 

Descriptions of baseline and intervening data are provided in earlier publications.^7–12  All data for the current follow-up were gathered by staff masked to treatment assignment. Most maternal interviews were conducted after their first-born child’s 18th birthday (mean age: 18.67 years; SD = 0.95; range: 17.5–23.9). Interviews were completed between October 2008 and September 2014, with 618 of the 658 mothers whose children had not died before age 2, who had not died themselves, or had not refused participation before this assessment period. State administrative records for SNAP, AFDC and TANF, and Medicaid were reviewed by December 2015 for the 618 mothers interviewed at child age 18.

Primary, secondary, and exploratory (possible mediator) outcomes; the specific variables measured; and the bases for study hypotheses are shown in Supplemental Table 4. The primary outcome was mother’s total public benefit costs in 2009 dollars for SNAP, AFDC and TANF, and Medicaid from the first child’s birth through age 18. Secondary outcomes consisted of maternal substance abuse²³  and symptoms of depression that crossed borderline or clinical thresholds.²⁴ 

We assessed a set of exploratory outcomes primarily to examine their role in mediating the effect of the program on primary and secondary outcomes. They included a variable that characterized the cumulative number of years mothers reared subsequent children after birth of the first child (labeled subsequent-child years); rates of partnered relationships, including cohabitation, and marriage from child age 2 through 18; number of months marital partners were employed (at the18-year interview); cumulative number of months mothers reported working; maternal income derived from Social Security Administration (SSA) records (available at the time of review through child age 16); current use of illicit and illegal substances²⁵ ; borderline or clinical anxiety²⁶ ; and sense of mastery.¹⁷ 

Measures for the current report were derived primarily from intake⁷  and 18-year assessments, although intervening data^7–11  were integrated with 18-year measures to create longitudinal data sets.

Analyses are reported on all cases randomized insofar as outcome data were available (see Supplemental Table 3). The core statistical model consisted of a 2-level treatment factor (NV versus control), a 2-level maternal psychological resources factor (above versus below the sample median), their interaction, and 2 covariates: household poverty and maternal child-rearing attitudes associated with child abuse (CAA)³⁸  measured at intake. Both covariates were predictors of maternal life course outcomes. Examination of the maternal mastery outcome included intake mastery as an additional covariate. For quantitative outcomes on which we had multiple assessments (eg, public benefit costs), we analyzed data in mixed models that included, in addition to core model terms, women as levels of a random factor, a fixed repeated measures classification factor for time (first-born child age) of assessment, and all interactions of time with the other fixed classification factors. (Note that examination of SSA income [only available through 16 years after birth of the first child] included 3 additional covariates, a time-varying indicator of whether the mother was 19 or older for each year after the first child’s first birthday, smoking status, and years of education at intake, and that SSA income analyses were conducted by SSA staff, producing summary tables for this report.) For dichotomous outcomes, given small cell sizes, we examined outcomes in the 2-level treatment model, dropping the treatment × maternal psychological resources factor interaction, treating it as a simple classification factor; in all instances, this interaction term was not significant (P value >.20).

Key tests were focused on the treatment effect averaged over all other fixed classification variables, including those within subjects and the same treatment effect restricted to HPR mothers. The tables show the least squares (adjusted) means over time. For repeated outcomes, we report results averaged or summed over the entire period as well as the interval between child ages 12 and 18. To illustrate program effects on public benefit costs and subsequent-child years over time, we plot point estimates with SEs for the NV and control groups for every year after birth of the first child. As a means of comparing treatment differences on quantitative outcomes, we also show effects in SD units (mean difference divided by pooled SD), sometimes referred to as effect sizes (ESs).

Preliminary mediation analyses were conducted by using PROC CAUSALMED in SAS v9.4 (SAS Institute, Inc, Cary, NC).³⁹  Variables included treatment condition, the 2 covariates (CAA and household poverty), exploratory mediator variables reported below for which there were NV-control differences (examined one at a time), and the outcome variable.

The NV and control groups were similar on background characteristics for participants on whom 18-year follow-up assessments were conducted (Supplemental Table 5), with these exceptions: at intake, NV women, compared with control group, lived in households with less discretionary income, higher person-per-room density, higher scores on a household poverty index, and higher CAA.

As shown in Table 1, there were no NV-control differences in maternal reports of substance abuse or depression. There were, however, significant program effects on public benefit costs over the 18-year period after the birth of the women’s first children. NV women, compared with women in the control group, incurred $17 310 less in public benefit costs for SNAP, AFDC and TANF, and Medicaid (ES = −0.13; 95% confidence interval [CI]: −0.24 to −0.01; P = .03). This overall difference was driven by the intervention effect through child age 12, with no significant NV-control difference between ages 12 and 18, although differences in that age range contributed to the overall estimated effect through age 18. Although the overall difference was present for the sample as a whole, as shown in Fig 1, it was more pronounced ($28 847) for women with HPRs measured at registration (ES = −0.21, 95% CI: −0.38 to −0.04; P = .01). There were no overall NV-control differences in public benefit costs for women with LPR, but NV mothers with LPR received fewer public benefits during the 3-year period after the end of the program, that is, from child age 3 to 5 (Fig 2), which contributed to the overall estimate of intervention effect. In Supplemental Table 6, we show estimates of each component of the total benefit variable.

As shown in Table 2, there were no intervention-control differences in the number of months mothers worked since birth of their first child, substance use, or anxiety. As shown in Fig 3, however, NV women with HPR, compared with control group counterparts, had 4.64 fewer cumulative years rearing subsequent children after birth of the first child (ES = −0.22, 95% CI: −0.42 to −0.02; P = .03). There were no intervention-control differences for mothers with LPR (data not shown).

Although there were no intervention-control differences in whether women were in partnered relationships from child age 2 to 18, NV women, as a trend, were more likely to cohabit (36.8% vs 32.6%; adjusted odds ratio [aOR]: 1.20; 95% CI: 0.97 to 1.49; P = .09) and, as shown in Fig 4, were significantly more likely to be married over this period (19.2% vs 14.8%; aOR: 1.37; 95% CI: 1.01 to 1.85; P = .04). The treatment difference in marriage led to NV women living, at the age 18 follow-up, with spouses who were employed for 13.96 more months than control group women (ES = 0.22; 95% CI: 0.05 to 0.39; P = .012).

There were no intervention-control differences in mothers’ SSA-derived incomes from the first child’s birth through age 16, but NV women earned more than women in the control group in years 4 and 5 after birth of the first child (Fig 5). NV women also reported greater sense of mastery through age 18 (ES = 0.13; 95% CI: 0.03 to 0.23; P = .009).

We examined each of the significantly different outcomes (or trends) in Table 2 as possible mediators of program effects on public benefit costs. The only significant mediator was subsequent-child years, although all of the hypothesized mediators for which there were NV-control differences predicted public benefit costs as expected (data not shown).

The program had no effect on reports of maternal substance abuse, depression, substance use, months worked since birth of their first child, or anxiety but produced long-term effects on public benefit costs for SNAP, AFDC and TANF, and Medicaid. Program effects on public benefit costs were most pronounced for mothers with HPR, an effect explained by the reduction in subsequent-child years.

NV women, compared with women in the control group, had no increase in partnered relationships but had increased cohabitation (as a trend), marriage, and sense of mastery. Moreover, although there were no program effects on income, NV women earned more than women in the control group during years 4 and 5 after the first child’s birth. We have yet to fully understand whether improved sense of mastery, increases in early earnings, cohabitation, marriage, and marital-partner earnings, in aggregate, contribute to the program effect on public benefit costs.

Given that the program cost ∼$12 578 in 2009 dollars, the $17 316 discounted savings reduced public expenditures by $4738 per family. Other savings to government and society, such as reductions in first-born disability⁴⁰  and rates of low birth weight in second births,⁴¹  are not yet monetized.

The results found at this phase of the trial must be interpreted in light of their limitations. The first is that maternal substance use and abuse were evaluated with self-report, and extensive evidence indicates that surveys substantially underestimate substance use.^42–44  Note that NV mothers in the Elmira trial by the end of pregnancy became more accurate reporters of cigarette smoking than control group women, when evaluated with serum cotinine.⁴⁵ 

Second, the rate of depression reported in this study may underestimate its prevalence, given low rates of major depressive disorder reported by African American survey respondents compared with non-Hispanic white respondents⁴⁶ ; this may be traced to their reluctance to disclose vulnerabilities, given their historic identity in being strong in coping with discrimination.⁴⁷ 

Third, the Medicaid cost calculation combined administrative data indicating whether a mother received benefits for each month after the first child’s birth with maternal reports of subsequent children’s birthdates. This assumes that all children lived with mothers for all months after their births, which likely is imprecise.

Fourth, although marriage and cohabitation are associated with better well-being,^20,21  the findings on marriage and cohabitation presented in this article do not address relationship quality and stability, which are important predictors of partners’ physical and mental health.⁴⁸ 

Fifth, the return on investment needs to be interpreted in light of the extreme poverty, concentrated social disadvantage, and high rates of adolescent pregnancy found in this sample as well as the pre–welfare reform policy context in place when the trial began,⁴⁹  limiting generalizability of these findings.

In general, we found enduring program effects on public benefit costs through child age 18 but no program effects on maternal substance abuse and depression. These findings, along with an accompanying article on child outcomes,⁴⁰  support the potential of this program to promote child development and reduce public benefit costs when focused on very-low-income families living in impoverished communities.

We thank Evelyn Collins for tracing and engaging study participants and for managing the Memphis study office since 1991, Benjamin Jutson (University of Colorado Anschutz Medical Campus) for help in preparing this article, and Wendy Gehring (University of Colorado Anschutz Medical Campus) for help with data management. Evelyn Collins, Benjamin Jutson, and Wendy Gehring received compensation from research grants for their work.

AFDC

Aid to Families with Dependent Children

aOR

adjusted odds ratio

CAA

child-rearing attitudes associated with child abuse

CI

confidence interval

ES

effect size

HPR

high(er) psychological resource

LPR

limited psychological resource

NFP

Nurse-Family Partnership

NV

nurse-visited

SNAP

Supplemental Nutrition Assistance Program

SSA

Social Security Administration

TANF

Temporary Assistance for Needy Families