Long-term Effects of Prenatal and Infancy Nurse Home Visitation on the Life Course of Youths:  19-Year Follow-up of a Randomized Trial FREE

Preventive interventions for parents of young children that improve parental competencies and the early learning environment of the child hold considerable promise as a means of reducing health and developmental problems across the life course and associated costs to government and society.^1- 3 Randomized controlled trials of interventions that involve enriched child care and preschool education have shown consistent short-term effects on cognitive and academic performance,² although programs with fewer features of quality (eg, higher teacher to child ratios and less teacher preparation) have produced smaller effects.⁴ Follow-up studies^5- 7 of some programs into young adulthood have found important benefits, including increased rates of high school graduation, reductions in teen pregnancy, and fewer arrests, although some of these effects have been limited to subgroups such as females, low-birth-weight babies in higher weight ranges, and those at greater social disadvantage.

There has been much recent interest in home visitation services that focus on promoting maternal health and caregiving competencies during pregnancy and infancy.^8- 10 The promise of these programs has been only partially fulfilled, however, given mixed results from randomized controlled trials on maternal and child outcomes.¹¹ In addition, there have been no long-term studies of children in home visitation programs when they enter late adolescence and young adulthood.

A program of prenatal and infancy home visitation by nurses, known today as the Nurse-Family Partnership (NFP), in separate randomized controlled trials conducted in Elmira, New York; Memphis, Tennessee; and Denver, Colorado, has been found to improve aspects of maternal and child health through child age 15 years, including prenatal health, rates and timing of subsequent pregnancies, maternal economic self-sufficiency, maternal and child involvement with the criminal justice system, child abuse and neglect, and injuries.^12- 15

A follow-up of the Elmira participants at child age 15 years indicated that nurse-visited youths had significantly fewer arrests, convictions, and adjudications as persons in need of supervision for incorrigible behavior.¹³ The program did not affect less serious forms of delinquency or the 15-year-olds' sexual behavior.

The present article reports the results of a 19-year follow-up of children born to mothers in the Elmira trial and focuses on their graduation from high school, employment, sexual behavior, childbearing, substance use, and criminal behavior. Given the earlier impact of the program on maternal use of welfare, child maltreatment, childhood injuries, and criminal behavior on the part of mothers and children through child age 15 years, there was reason to expect enduring effects on youth functioning, especially their criminal behavior. The features of the child's behavior and family environment altered through age 15 years are consistent with the prevention of early-onset problem behaviors,¹⁶ which are significant contributors to adult criminal behavior.¹⁷ We expected fewer program effects on less serious problem behaviors.¹³

We also hypothesized that the program would affect other aspects of the youths' life trajectories, reflected in more high school graduations, lower rates of teen child birth, and less use of public assistance through age 19 years. Given the findings from analyses at earlier ages, we expected that these effects would be stronger for youths born to unmarried and low-income mothers.¹³ Such effects would be consistent with improvements in nurse-visited mothers' economic circumstances and the prevention of child abuse and neglect observed for these higher-risk mothers in the 15-year follow-up study.¹²

The present study is a 19-year follow-up of youths whose mothers enrolled in the Elmira trial of the NFP, a program of prenatal and infancy home visitation by nurses. Descriptions of the program and previous assessments can be found in previous articles^{12- 13,18- 21}; a summary is presented herein.

Mothers of the 19-year-old youths were recruited during their pregnancies from a free antepartum clinic sponsored by the county health department and the offices of private obstetricians. We actively recruited nulliparous women who were less than 25 weeks pregnant and who were young (<19 years old), unmarried, or of low socioeconomic status (SES). Nulliparous women without these risk characteristics were permitted to enroll. Between April 1, 1978, and September 30, 1980, 500 women were invited to participate, and 400 enrolled. Eighty-five percent of the enrolled participants were young, unmarried, or from low-SES households (A. Hollingshead, PhD, unpublished data, 1976). After completing informed consent forms and baseline interviews, women were stratified by sociodemographic characteristics and were randomized to 1 of 4 treatment conditions (see the following subsection). Approvals have been granted by institutional review boards at Cornell University, Ithaca, NY; the University of Rochester, Rochester, NY; and the University of Colorado, Denver.

Families in treatment group 1 (n = 94) were provided sensory and developmental screening for the child at 12 and 24 months of age. Based on the results of these screenings, the children were referred for clinical evaluation and treatment when needed. Families in treatment group 2 (n = 90) were provided these same screening services plus free transportation for prenatal and well-child care through the child's second birthday. There were no differences between treatment groups 1 and 2 in their use of prenatal and well-child care, and these groups were combined to form a single comparison group. Families in treatment group 3 (n = 100) were provided the same screening and transportation services offered in treatment group 2 but were also provided a nurse who visited them at home during pregnancy. Families in treatment group 4 (n = 116) were provided the same services as treatment group 3 except that the nurse continued visits through the child's second birthday.

At the home visits, the nurses had 3 goals: (1) to improve pregnancy outcomes by helping women improve their health-related behaviors, (2) to improve children's health and development by helping parents provide more competent care, and (3) to improve families' economic self-sufficiency by helping parents make appropriate choices regarding family planning, finishing their education, and finding work. In the service of these goals, nurses linked families with needed services and attempted to involve fathers, family members, and friends in the pregnancy, birth, and early care of the child. Nurses completed a mean of 9 visits (range, 0-16) during pregnancy and 23 (range, 0-59) from birth to the child's second birthday.

The present phase is a longitudinal follow-up of the children from the 400 randomized families when the children were 19 years of age. The 19-year assessments were conducted on 310 youths: 78% of the sample originally randomized and 88% of cases where the child was still alive, not adopted, and not mentally disabled (Table 1). There were no differences in treatment grouping; the mother's race, marital status, age, educational level, or SES at intake; or the child's sex for those assessed at the 19-year follow-up compared with those for whom data were missing (data not shown). There was no indication of attrition bias.

Youths completed a telephone interview that assessed their history of arrests and convictions, delinquent and criminal behavior, use of substances, educational achievement, pregnancies, births, and use of welfare. Those who reported being arrested were asked to provide dates for up to 6 arrests and convictions. We combined data from the present phase with those from the 15-year interview to calculate the number of arrests at particular ages. For the 15-year assessment, we recorded the number of arrests and the dates of their first and most recent arrests. All data were gathered by individuals masked to treatment.

We created variables to summarize whether the youths had been arrested or convicted in their lifetime and in the 12 months preceding the interview and whether they had used illegal drugs or engaged in binge drinking (>4 drinks in a row) during the 6 months preceding the interview. Count variables were created to assess the number of arrests and convictions. Following the National Youth Survey,²² variables were constructed to indicate whether the youth had engaged in the following criminal behaviors during the year preceding the 19-year interview: felony assault (aggravated assault, sexual assault, or gang fights), fraud, minor assault (hit parents, teachers, or students), minor theft (stole something worth <$50 or joyriding), illegal services (prostitution or sold marijuana or other drugs), vandalism or damaged property, and public disorder (panhandling or been loud, rowdy, or drunk in public). The incidences of robbery and felony theft were too low to serve as valid outcomes.

Variables also were created to summarize whether the youth had graduated from high school, was currently economically productive (ie, in school, the military, or job training or working full time at the time of the interview), had become or made a girl pregnant or become a parent, and used welfare, food stamps, or Medicaid. We also included a count of the youth's number of sexual partners in the past year and continuous measures of the frequency of birth control and condom use (on scales ranging from 1[never] to 7[always]).

We identified primary outcomes based on findings from earlier phases of this trial and from the literature on youth development and antisocial behavior. We included as primary outcomes variables that reflect arrests and convictions, self-reported violent behavior (eg, felony assault), high school graduation, teen childbearing, use of welfare, and current involvement in school, work, the military, or job training. We present secondary outcomes, such as self-reported minor delinquency, to provide the reader with a complete picture of program impact.

Analyses were conducted on all cases where outcome data were available, irrespective of the home-visited families' degree of program participation. A core statistical model was derived for the analysis similar to that used in earlier phases of this research. It consisted of a 3 × 2 × 2 factorial structure plus 2 covariates. The classification factors were treatment group (1 and 2 vs 3 vs 4), risk (unmarried and from low-SES families at registration vs married or higher SES), and the youth's sex. All interactions among the factors were included in the model. The youth's race/ethnicity (white vs nonwhite) and the mother's educational level at intake were included in the model without interactions.

Quantitative variables were examined in the general linear model. For dichotomous outcomes, we used the generalized linear model with a log link and binomial error distribution; for count data, we assumed a log link and negative binomial distribution. For dichotomous and count outcomes where fewer than 70 values were nonzero, we used simpler models due to concerns about the stability of treatment effect estimates.²³ Models for these outcomes included family risk, the mother's educational level at intake, and the youth's sex and race as classification factors without interactions.

The Cox proportional hazards method for survival analysis was used to estimate the hazard of first arrest for those in the comparison and nurse-visited groups.²⁴ These models included family risk, the mother's educational level at intake, and the youth's sex and race as factors, again without interactions. Growth curves for arrests across time were estimated for treatment groups by the youth's sex in a generalized mixed model with cubic age regressions, with log link and negative binomial error.

All of the treatment contrasts focused on the comparison of combined treatment groups 1 and 2 (the comparison group) with treatment group 4 (the pregnancy and infancy nurse-visited group) because we hypothesized that the greatest treatment effect would be exerted by the combination of prenatal and postnatal home visits.^18- 19 The results for the group that received prenatal home visitation only (treatment group 3) are included to examine whether that group, as predicted, had intermediate levels of functioning. Differential treatment effects based on the youth's sex and family risk are presented in the text where these factors interacted with treatment condition. Findings are reported as adjusted percentages or means controlling for the effects of the covariates. Risk ratios (RRs) are provided for dichotomous outcomes, incidence RRs (IRRs) for count outcomes, and mean differences for continuous outcomes; 95% confidence intervals (CIs) are also shown.

The treatment groups were essentially equivalent on background characteristics for the entire sample registered in the trial^{12- 13,18- 21} and for the sample assessed at 19-year follow-up (Table 2). Table 3 shows that among the 19-year-old youths, those visited by nurses during pregnancy and infancy were less likely to have ever been arrested or convicted than were those in the comparison group. Youths in the nurse-visited group also had fewer lifetime arrests and convictions than did those in the comparison group. Examination of treatment differences separately for boys and girls suggests that these effects were limited to girls. Specifically, nurse-visited girls (n = 44) were less likely to be arrested (10% vs 30%; RR, 0.33; 95% CI, 0.13-0.82) and convicted (4% vs 20%; 0.20; 0.05-0.85) than were girls in the comparison group (n = 73). There were no significant differences between nurse-visited boys (n = 47) and comparison boys (n = 67) in the lifetime likelihood of being arrested (46% vs 47%) or convicted (35% vs 39%). The intervention effect on number of arrests and convictions also varied by sex. Analysis of simple effects indicated that nurse-visited girls had fewer mean lifetime arrests (0.10 vs 0.54; IRR, 0.18; 95% CI, 0.06-0.54) and convictions (0.04 vs 0.37; 0.11; 0.02-0.51) than did girls in the comparison group; nurse-visited and comparison boys had similar rates of mean lifetime arrests (1.39 vs 1.37) and convictions (0.97 vs 0.91).

Examination of age at first arrest in the proportional hazards model indicated that nurse-visited youths had a smaller risk of first arrest than did youths in the comparison group (hazard ratio [HR], 0.63; 95% CI, 0.40-0.99). Consistent with the binomial arrest results, this effect varied by sex. Nurse-visited girls had a smaller risk of first arrest than did comparison girls (HR, 0.27; 95% CI, 0.10-0.72) (Figure 1). Nurse-visited and comparison boys were similar in their risk of first arrest (HR, 0.85; 95% CI, 0.50-1.46) (Figure 2).

It is instructive to examine the types of arrests being reported for girls in the present study. In the comparison group, 23 females were arrested by age 19 years. Of those, almost half (n = 11) involved violence or the threat of violence (eg, assault, harassment, domestic violence, and armed robbery). Of nurse-visited girls, 5 were arrested by age 19 years, and of the 3 for which we have data on reason for the arrest, none involved violence. Although not definitive, this pattern is consistent with the idea that the program is preventing more serious and more violent forms of antisocial behavior in the females in the study.

Figure 3 shows the fitted curves of the counts of arrests by child age, sex, and treatment condition and emphasizes that program impact on the number of arrests was limited to girls and that the treatment-control difference in girls was concentrated in mid-adolescence. The rate of arrests in control group girls declined during the latter half of adolescence to a rate comparable with that of their nurse-visited counterparts. The rate of arrests was higher in boys than in girls and remained high throughout all of adolescence once it reached its peak in mid-adolescence. There were no other program effects on the youths' self-reported criminal behavior or their use of alcohol or illegal drugs. There were no overall treatment effects on high school graduation, economic productivity, number of sexual partners, use of birth control, teen pregnancy or childbearing, and use of welfare, food stamps, or Medicaid (Table 4).

There was some indication that the program affected the fertility behaviors of youths born to high-risk mothers. Specifically, youths born to high-risk mothers in the intervention group reported higher rates of condom use (n = 38; mean, 4.13) than did similar youths in the comparison group (n = 58; mean, 3.11; mean difference, 1.01; 95% CI, 0.07-1.96). Boys in the intervention group born to high-risk mothers (n = 18), however, reported a greater number of sexual partners (mean, 3.07) than did similar boys in the comparison group (n = 23; mean, 1.53; IRR, 2.00; 95% CI, 1.16-3.45). Finally, girls in the intervention group born to high-risk mothers (n = 20) were less likely to use Medicaid than were their counterparts in the comparison group (n = 35; 18% vs 45%; RR, 0.40; 95% CI, 0.18-0.87), a difference largely attributable to high-risk intervention girls' reduced childbearing rate compared with that of high-risk girls in the comparison group (11% vs 30%; RR, 0.35; 95% CI, 0.12-1.02).

The program produced enduring effects on girls' involvement with the criminal justice system but, except for youths born to high-risk women, with no other effects on their life course. Girls in the nurse-visited group had almost no increase in the likelihood of a first arrest across time, whereas the likelihood of a first arrest increased steadily for girls in the comparison group after age 14 years. For boys, the likelihood of an arrest increased substantially for the intervention and control groups after age 12 years, and there were no significant treatment-control differences in arrests for boys through age 19 years. Although the pattern of results shown in Figure 3 is consistent with the intervention effects on arrests and convictions found at age 15 years,¹⁹ we had not previously found significant sex differences in these outcomes.

Girls in the nurse-visited group born to high-risk (unmarried and low-income) mothers had fewer children and were less likely to have received Medicaid than were high-risk girls in the comparison group. This pattern of effects for crime, childbirth, and use of Medicaid is consistent with earlier findings from this trial for the youths' mothers: high-risk nurse-visited mothers had fewer arrests, had fewer and more widely spaced subsequent children, and used less public assistance during the 15 years after the birth of the first child than did their comparison group counterparts.¹²

We intend to examine in future studies possible explanations for the coherence of intervention effects between mothers and daughters but not sons. We are particularly interested in the extent to which the program's effect on earlier risks (eg, child maltreatment, maternal arrests, and maternal use of welfare) predicts adolescent criminality equally for boys and girls. Given that this same program has been tested in other trials, we will have an opportunity to examine the extent to which the moderation of treatment effects on antisocial behavior by the child's sex occurs in other populations and contexts.^14- 15

We did not find that the intervention affected rates of high school graduation by age 19 years. Nineteen-year-old youths may simply be too young to observe patterns of educational achievement that include college attendance, or the program may not have affected educational outcomes for this population of youths in this semirural setting. Other early intervention programs serving different populations have not consistently shown program effects on these outcomes. For example, at age 23 years, there were no program effects of the Perry Preschool Program on high school graduation rates.⁵ Although the 21-year follow-up of the Abecedarian Program reported positive program effects on years of school completed,⁶ the Infant Health and Development Program, at age 18 years, found no program effects on high school dropout rates.⁷ We are conducting an adolescent follow-up of primarily black youths who participated in the Memphis trial of the NFP, which will give us additional insight into the long-term effects of the program on such educational outcomes and on criminal behavior, antisocial behavior, and economic productivity for youths whose families have lived in substantially different contexts than those in the Elmira NFP trial.

These findings must be tempered with recognition of their limitations. The most obvious is that the data were based on youth reports, which may be subject to treatment-related reporting bias. Although we do not have administrative data with which we can compare youth self-reports of arrests and convictions because youth arrest records are sealed, program effects on maternal arrests through child age 15 years were stronger when based on New York State criminal justice records than when they were based on self-report.¹² Mothers would be more prone to treatment-related reporting bias than would their children, however, given that the program ended when the children were 2 years old. Another limitation is the lack of racial diversity in the sample, which did not permit the examination of racial or ethnic differences in treatment effects. This was one of the principle rationales for the more recent replications of the Elmira trial in Memphis and Denver.

Overall, these findings suggest that the NFP program has the potential to produce lasting changes in criminal offending trajectories, early childbearing, and economic outcomes for girls born to low-resource mothers. Given that other long-term follow-up studies of model early-childhood interventions for infants and preschoolers have reported continued treatment effects with older adolescents and young adults, the impact of well-designed and implemented early interventions on crime reduction is promising.^25- 27 These findings also emphasize the need to direct more scientific attention to girls in observational and interventional research on criminal behavior and delinquency.²⁸

Age 19 years is too young to observe intervention effects on more established adult patterns of criminal behavior or to document accurately educational attainment, economic productivity, or longer-term patterns of childbearing. The follow-up study of the Elmira sample at age 27 years will give us a much clearer picture of the effect of the program on these and other indicators of young adult functioning.

Correspondence: John Eckenrode, PhD, Family Life Development Center, Cornell University, Beebe Hall, Ithaca, NY 14853 (jje1@cornell.edu).

Accepted for Publication: June 15, 2009.

Author Contributions: Drs Eckenrode, Campa, Luckey, and Olds and Mr Henderson had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Eckenrode, Henderson, Cole, Kitzman, Powers, and Olds. Acquisition of data: Eckenrode, Kitzman, Anson, Sidora-Arcoleo, and Powers. Analysis and interpretation of data: Eckenrode, Campa, Luckey, Henderson, Cole, Kitzman, Anson, Sidora-Arcoleo, Powers, and Olds. Drafting of the manuscript: Eckenrode, Campa, Henderson, Powers, and Olds. Critical revision of the manuscript for important intellectual content: Eckenrode, Campa, Luckey, Henderson, Cole, Kitzman, Anson, Sidora-Arcoleo, and Olds. Statistical analysis: Eckenrode, Campa, Luckey, Henderson, and Sidora-Arcoleo. Obtained funding: Eckenrode, Henderson, Cole, and Olds. Administrative, technical, and material support: Cole, Kitzman, Anson, and Powers. Study supervision: Eckenrode.

Financial Disclosure: None reported.

Funding/Support: This research was supported by grant 801-099 from the Smith Richardson Foundation. Support for earlier phases of this trial was provided by a Senior Research Scientist Award (Dr Olds) and by grants from the Prevention Research and Behavioral Medicine Branch of the National Institute of Mental Health, the Assistant Secretary for Planning and Evaluation, Health and Human Services, the Bureau of Maternal and Child Health (Department of Health and Human Services), the Robert Wood Johnson Foundation, the W. T. Grant Foundation, the Ford Foundation, and the Commonwealth Fund. Dr Olds' Research Center at the University of Colorado has a contract with the NFP National Service Office to conduct research on improving the NFP program model.

Disclaimer: Dr Olds has no personal financial interest in the NFP program.

Additional Contributions: John Shannon, PhD, supported the program and gathered data through Comprehensive Interdisciplinary Developmental Services, Elmira; Alise Mahr, BS, and Darlene Batroney, BA, traced and interviewed the families; and Terri Moffitt, PhD, provided valuable comments on an earlier draft of the manuscript.

Online-Only Material: This article is featured in the Archives Journal Club. Go here to download teaching PowerPoint slides.