Prenatal and Infancy Home Visiting in Germany: 7-Year Outcomes of a Randomized Trial

ProKind was implemented from 2006 to 2012 in 3 German federal states (Bremen, Lower Saxony, and Saxony) in 13 municipalities. In addition to (1) women with a first-time pregnancy in the 12th to 28th week of pregnancy, the inclusion criteria were (2) women with an economic risk (eg, receiving welfare benefits or being in debt), (3) women with at least 1 further social or personal risk (eg, being underage, lacking a school leaving certificate, or having experienced abuse or neglect), and (4) woman with at least a basic understanding of the German language. Overall, 755 mothers were randomly selected for the IG or control group (CG) condition (IG: 394, CG: 391). The participants were recruited through various disseminators (eg, gynecologists, youth welfare offices, or employment agencies) and completed informed consent procedures. The baseline randomization was stratified by municipality, maternal age (<18 vs >18 years old), and maternal nationality (German versus non-German). There were no statistically significant differences between groups in the frequency of social or personal risk factors at the baseline assessment (time 0).^12,13 

The ProKind program implemented the NFP core components (ie, target group criteria, the specified average number of visits and the average duration of visits, NFP guiding material translated into German language, a protocol of each home visit, and support for home visitors through supervision). In accordance with the NFP, visits began between the 12th and 28th weeks of pregnancy and continued until the child’s second birthday. Across this time frame, the frequency of home visits varied among weekly, biweekly, and monthly according to the NFP model for an overall maximum of 52 home visits with an average duration of 90 minutes each. Teaching materials and visit-by-visit guidelines, adapted from the NFP, structured the goals and contents of each home visit. Diverging from the original program, the German adaptation involves home visits conducted by social workers and state-licensed family midwives (midwives in Germany hold a 3-year apprenticeship degree consisting of 1600 hours of theory and 3000 hours of practical training) either alone (mainly family midwives) or in tandem (family midwife and social worker).

Families in the IG received an average of 32.7 home visits (SD 18.6). The average length of the home visits was 82 minutes. A total of 166 (42.2%) mothers in the experimental group dropped out of the intervention prematurely. The reasons for termination can be found in Fig 1. Previous analyses based on data collection at the follow-up points at 6 (time 2), 12 (time 3), and 24 months (time 4) after childbirth revealed positive effects on mothers’ feelings of parental self-efficacy, social support, and mothers’ knowledge about child-rearing.¹³  Furthermore, the intervention resulted in an improvement in cognitive development for girls, whereas no improvement could be found for boys.¹⁴  However, effects regarding parenting behavior (eg, breastfeeding) and children’s socioemotional development were absent at time 4.^13,15 

A face-to-face computer-administered personal interview, including a school performance test, and 2 computer-administered telephone interviews were scheduled. All interviews and tests were conducted by trained interviewers who were blinded with respect to participants’ intervention status. Data were collected between April 2015 and December 2017.

We present the results for the 8 preregistered primary hypotheses, which were evaluated from the perspective of the mother (9 outcome variables) and/or the perspective of the reference child (4 outcome variables).¹²  Table 1 gives an overview of the different domains, their operationalization, the relevant informant, and data source.

In total, 533 (response rate of 70.6%) of the original 755 randomly assigned families at baseline could be contacted again. There was no statistically significant difference in loss to follow-up between the CG (29.1%) and IG (30.6%) (P = .67). In Fig 1, the responses are summarized in a flowchart. Notably, only a small proportion of respondents refused to participate in the follow-up (3.4%). The loss to follow-up was mainly due to a lack of valid contact information and unsuccessful attempts to obtain this information (24.5%). At the point of the present follow-up (time 5), mothers were on average 29.6 years old (SD 4.36; range 22.1–48.1) and children were on average 7.55 years old (SD 0.75; range 5.44–10.19). Table 2 provides an overview of the descriptive statistics of the demographic characteristics and social, personal, and economic risk factors of the sample. Participants in the follow-up assessment from the IG and CG were comparable, except there was a higher prevalence of aggression and mental problems in the CG.

We standardized and recoded all continuous outcomes such that positive values corresponded to beneficial effects; the reported effect sizes (ESs) can be interpreted as group differences in the SDs of the CG. The confidence intervals (CIs) of the ESs were obtained by bootstrap methods with 5000 replications. To handle missing observations from families who did not participate in the follow-up and to take into account the imbalance in a few baseline risk factors (ie, maternal aggression and mental health problems) in the IG and CG at time 5, ordinary least squares regression was conducted by using inverse probability weighting (IPW).¹⁵  The IPW technique weights observed cases on the basis of the inverse of their probability of loss to follow-up, thereby decreasing the potential selection bias resulting from loss to follow-up. For this purpose, a logit model was fit to estimate the probability of participants completing the time 5 assessment, conditional on the baseline characteristics shown in Table 2 as well as the participant’s municipality at baseline (Supplemental Table 5 reveals predictors of attrition in the IG and CG). Afterward, the inverse of the predicted probabilities from the logit models were used as weights in the regressions, with the outcome variables as the dependent variables and the treatment indicator as the independent variable. Following the study protocol,¹²  we conducted 1-sided tests (P < .05) on the basis of the assumption that the intervention was not harmful, and we presented preregistered subgroup analyses of child sex. All analyses were performed with Stata version 15 (Stata Corp, College Station, TX).

The means and SDs of all primary outcome measures at time 5 may be found in Supplemental Table 6. Table 3 reveals ESs for the full sample at time 5 and separately by child sex. In Table 4, the results of the ordinary least squares regression are summarized by using IPW (for the full sample and separately by child sex). Supplemental Table 7 reveals results of IPW estimation based on stratification characteristics only.

Mothers from the IG reported fewer child total problem behaviors (ES = 0.21 [95% CI: 0.03 to 0.38]; β = .181, P = .02), internalizing behaviors (ES = 0.26 [95% CI: 0.08 to 0.44]; β = .200, P = .02), and externalizing behaviors (ES = 0.17 [95% CI: −0.01 to 0.34]; β = .148, P = .05) than those in the CG, and the difference in the latter was only statistically significant at an α level of 90%. In the preregistered subgroup analyses (child’s sex), mothers of boys in the IG reported fewer internalizing behaviors (ES = 0.37 [95% CI: 0.11 to 0.63]; β = .291, P = .01) and externalizing behaviors (ES = 0.28 [95% CI: 0.03 to 0.54]; β = .218, P < .05), as well as child total problem behaviors (ES = 0.24 [95% CI: −0.01 to 0.50]; β = .176, P = .08), than those in the CG, and the difference in the latter, again, was associated with a higher uncertainty. Regarding female child sex, no statistically significant intervention effect could be detected.

No statistically significant intervention effects were observed for children’s school performance (total sample: ES = −0.12 [95% CI: −0.32 to 0.08], β = −.81, P = .77; boys: ES = −0.16 [95% CI: −0.46 to 0.14], β = −.096, P = .74; girls: ES = −0.07 [95% CI: −0.36 to 0.21], β = −.070, P = .68) or child life satisfaction, neither from the perspective of mothers (total sample: ES = −0.04 [95% CI: −0.23 to 0.15], β = −.023, P = .59; boys: ES = −0.03 [95% CI: −0.28 to 0.23], β = −.020, P = .55; girls: ES = −0.03 [95% CI: −0.29 to 0.24], β = −.012, P = .53) nor from the perspective of children (total sample: ES = −0.02 [95% CI: −0.22 to 0.17], β = −.048, P = .67; boys: ES = −0.02 [95% CI: −0.31 to 0.26], β = −.080, P = .71; girls: ES = −0.03 [95% CI: −0.30 to 0.24], β = −.024, P = .56).

Mothers from the IG reported improved life satisfaction (ES = 0.25 [95% CI: 0.7 to 0.43]; β = .24, P = .004) and beneficial results regarding maternal mental health (ES = 0.25 [95% CI: 0.7 to 0.43]; β = .193, P = .02). Although no statistically significant group differences for mothers of girls could be found, there were significant intervention effects for mothers of boys in terms of mothers’ life satisfaction (ES = 0.31 [95% CI: 0.06 to 0.58]; β = .317, P = .006) and maternal mental health (ES = 0.38 [95% CI: 0.13 to 0.63]; β = .310, P = .008).

Although the mothers of the IG reported reduced rates of abusive parenting (ES = 0.19 [95% CI: 0.01 to 0.37]; β = .18, P = .03), this finding could not be confirmed from the perspective of the children. No statistically significant treatment effect was found for neglectful parenting from the perspective of mothers or children. Regarding adverse parenting, group differences were only statistically significant at an α level of 90% (ES = 0.13 [95% CI: −0.05 to 0.31]; β = .142, P = .09). In the preregistered subgroup analysis, mothers of boys in the IG reported less abusive parenting (ES = 0.36 [95% CI: 0.09 to 0.61]; β = .314, P = .005), neglectful parenting (ES = 0.47 [95% CI: 0.20 to 0.74]; β = .387, P = .002), and adverse parenting (ES = 0.35 [95% CI: 0.09 to 0.62]; β = .341, P = .008) than those in the CG.

In this follow-up of an RCT, we found that early home visits for disadvantaged families in Germany had lasting effects on mothers and children until primary school age. In particular, we found positive effects on the maternal ratings of child behavioral problems, maternal mental health and life satisfaction, and abusive, adverse, and neglectful parenting (for mothers of boys). However, we did not find any intervention effects on children’s life satisfaction or on children’s school performance.

The lack of an intervention effect on children’s school performance is in contrast to the findings from the evaluation of ProKind at time 4 (child age 24 months), in which Sandner and Jungmann¹⁴  reported a positive effect on the cognitive development of girls. In addition, significant effects on cognitive development, intellectual functioning, and school performance were also found in the NFP Memphis trial at child ages 6 and 9, with partly stronger effects for boys.^17–19  However, no effects on intellectual functioning or academic achievement were found in the Denver trial at child ages 6 and 9.²⁰  The overall pattern of the longitudinal ProKind effect development is in line with the findings from other early childhood programs with home visiting elements, for example, the Perry Preschool Project.²¹  In these studies, effects on cognitive development disappeared after school entry, whereas effects on child behavioral problems emerged at school entry.

Regarding potential sex differences, for most outcome domains, ProKind benefitted boys and mothers of boys significantly more than girls and mothers of girls. These results are partly inconsistent with the results of the US NFP studies. For example, in the Denver trial, effects on trajectories of externalizing behaviors were found among girls, but not boys, at ages 2, 4, 6, and 9 years.²²  Furthermore, findings from the Memphis trials reveled that home visits had a stronger impact on girls’ aggression at age 2 years.²³  Additionally, Heckman et al¹⁹  (Memphis trial) showed intervention effects only for girls, who had fewer conduct and attention problems at age 6.

How can this sex differential effect be explained? First, it should be noted that the majority of mothers in our study did not have stable partnerships over the course of the study.^13,14  Considering that single mothers seem to offer their sons less emotional support and are more likely to use violence against them rather than against their daughters,²⁴  it can be assumed that the intervention may be protecting boys from these negative effects. Accordingly, we found positive intervention effects for adverse, abusive, and neglectful parenting for mothers of boys, but not for mothers of girls, in our study. A reduction in these dysfunctional parenting practices can also explain the intervention effects on child behavior problems because harsh and authoritarian parenting, as well as inconsistent and permissive parenting behaviors, are linked to children’s behavioral problems.²⁵ 

Furthermore, sex differences in intervention outcomes must be considered in view of the health and welfare systems. Health care in Germany is generally free and includes prenatal and postnatal check-ups for pregnant women and new mothers. Because of the welfare system, mothers are not obliged to reenter the workforce until their children turn 3. Even after the child’s third birthday, unemployed mothers receive financed housing and welfare payments. In this context, Sandner²⁶  reported that IG mothers returned to work less quickly and received more welfare after child ages 2 and 3 years than CG mothers. These effects suggest that ProKind mothers spent more time raising their children than CG mothers. In this context, young boys are more vulnerable to stress than girls in prenatal, perinatal, and postnatal critical periods and, in particular, are more vulnerable to early maternal separation.²⁷  In summary, our results underline the importance of cross-cultural replication studies before transferring programs from the US context to Europe.

On the other hand, in line with our hypothesis, maternal life satisfaction improved. This effect is in contrast to the finding from the NFP that maternal life satisfaction did not improve in any trial. Additionally, the explanation for this finding may be found in the differences in the health and welfare systems between the United States and Germany. Although material deprivation may be lower and health care provision may be better in Germany than in the United States, disadvantaged mothers in a European welfare state are often isolated and suffer stress in the unknown situation of their first birth. In a European welfare state, home visitors may focus more on outcomes, such as maternal mental health, parenting, and child behavior than in the United States, where the avoidance of material deprivation and health care provision require more attention. The close supportive relation between the mother and home visitor may therefore be particularly beneficial for maternal life satisfaction, parenting, and child behavior in Europe.

There are many strengths associated with this follow-up study. It does, however, suffer from several limitations. One of these limitations is that most findings are based on self-report assessment. Although the questions stemmed from widely used questionnaires, it is still possible that the results could have a monomethodologic bias. However, there is some evidence that women who participate in the NFP become more accurate reporters of socially undesirable behavior.²⁸  This reporting may indicate that the effects of the self-rating inventories present lower bounds. Second, it should be noted that complete data regarding all times of measurement are not available for all cases. This means that some distorting effects cannot be absolutely excluded. And yet it should be noted that only a small proportion of respondents actually refused to participate in a further survey (∼6%), whereas the vast majority of the persons who were not reached because of missing addresses could not be determined. This mechanism speaks in favor of a missing-at-random pattern, which is generally less problematic than a missing-not-at-random pattern.²⁹  Overall, the response rate of ∼70% is comparable to that of other NFP trials at child age 6 years^17,22  and lies above the rate from another home visiting trial dealing with high-risk families in Ireland (European Union) at age 7 years.^30,31  In addition, we use modern procedures for handling missing data. Third, the number of analyzed outcomes raises challenges with multiple testing. In accordance with US NFP trials, we did not adjust for multiple comparisons to ensure that the results were comparable between US and European studies. Lastly, the number of home visits in the time 5 sample was slightly higher (35.8 vs 32.7) than in the baseline sample, which could contribute to positive confounding effects.

That the effects on mothers’ child abusive and neglectful parenting, as well as children’s behavioral problems, can be observed 5 years after the intervention is promising. The relevance of these findings are clear because long-term consequences of child maltreatment present an important risk factor for children’s health later in life, including highly costly clinical syndromes, high-risk behaviors, and chronic diseases.³²  Additionally, because behavioral traits are more relevant for later education decisions, employment, and wages than cognitive skills, the effects on child behavioral problems may also generate strong long-term fiscal impacts.³³ 

In summary, the results suggest that in a western European welfare state, a home visiting program targeting disadvantaged mothers has meaningful, lasting effects at primary school age in important outcome domains (ie, child behavior problems; adverse, abusive, and neglectful parenting; and maternal life satisfaction and mental health).

We thank all members of the staff of the ProKind follow-up. In addition, we thank all the participating families, whose cooperation enabled this research. We thank Anna Lohmann for helping prepare this article and Judith Hild for outstanding organizational support. This work is in memory of Peter F. Lutz, who supported the ProKind project through many years.

CG

control group

CI

confidence interval

ES

effect size

IG

intervention group

IPW

inverse probability weighting

NFP

Nurse-Family Partnership program

RCT

randomized controlled trial