The origins of adult mental and physical health problems can be traced to risk factors before birth and during early childhood (Barker, 1990; Goodman, Joyce and Smith, 2011; O’Donnell and Meaney, 2016). One of the most consistently identified risk factors for poor developmental outcomes is low family socio-economic status (SES) (Cobb-Clark, Salamanca and Zhu, 2016; Cunha, 2015). Early intervention programmes supporting low-SES families are recognised as a viable strategy for preventing long-term social and health problems (Heckman and Mosso, 2014). Typically, the mean impact of the intervention on all children is tested, thus ignoring one important source of variation: population heterogeneity (Nagin, 2005). By studying the differential impact of the intervention for children with different developmental profiles (i.e. population heterogeneity over time), it is possible to identify who benefits most from early intervention. This study, published in full elsewhere (Côté, Orri, Tremblay and Doyle, 2018), addressed this issue by testing the differential impact of a multi-component early intervention programme in Ireland for subgroups of children defined by latent trajectories of skills.

Two research questions were tested:

1. Did the intervention impact the probability of following a better developmental trajectory? Specifically, did children who received the treatment exhibit a lower probability of following a poor trajectory (or a higher probability of following a good trajectory), regardless of their initial levels of behavioural problems, cognitive skills, or propensity for health service use?

2. Within each trajectory, did treated children perform better than controls? Specifically, did children on poor developmental trajectories benefit the most from the intervention or, conversely, did children on good trajectories benefit the most?

Method

We used data from the Preparing for Life (PFL) randomised trial (Northside Partnership, Doyle and UCD Geary Institute PFL Evaluation Team, 2018; see also Doyle et al., 2015, 2017c), a programme aimed at promoting children’s development by supporting parents living in a disadvantaged community in Dublin from pregnancy until age four or five using three intensive parenting supports – a home visiting programme, a baby massage course, and the Triple P Positive Parenting Programme. The study included 233 families who were recruited during pregnancy and randomly assigned to a treatment group receiving the parenting supports or a control group receiving a reduced form of intervention.

The study outcomes were: internalised and externalising behaviours, general cognitive skills, vocabulary, and health service use, and these were repeatedly measured from six months to four years of age. First, the developmental trajectories for these outcomes were estimated using Latent Class Growth Analysis (LCGA), which allows for the identification of unobserved but distinct groups following similar developmentaltrajectories. Second, logistic regressions were used to test whether the treatment group compared to the control group were more likely to follow the high rather than the low trajectory. Third, Wald tests were used to compare treatment and control groups symptoms scores within each trajectory (high and low).

Results

Question 1. Did the treatment impact children’s probability of following a better trajectory?

Behavioural outcomes (Table 2A) between 2 and 4 years: For externalising behaviours, children followed either a low (n=104, 60 per cent) or a high trajectory (n=68, 39 per cent). The proportion of children in the treatment and control groups was similar in both trajectories (OR=1.10, CI=0.60- 2.04, NNT=40). For internalising behaviours, children followed either a high and slightly increasing (n=33, 19 per cent) or a low stable (n=139, 81 per cent) trajectory. The proportion of children in the treatment and control groups was similar in the two trajectories (OR=1.25, CI=0.59- 2.71, Number Needed to Treat [NNT]=28).

Cognitive outcomes (Table 2B) between 1 and 4 years: Children followed either a high (n=109, 62 per cent) or low (n=68, 31 per cent) cognitive skills trajectory. Treatment children were more likely to follow the high cognitive skills trajectory (OR=4.50, CI=2.22-9.65, NNT=4). Children followed either a high stable (n=62, 36 per cent) or low declining (n=111, 64 per cent) vocabulary trajectory. Treatment children were more likely to follow the high vocabulary trajectory (OR=2.02, CI=1.08-3.82, NNT=6).

Health service use (Table 2C) between 6 months and 4 years: For the number of health clinic visits, children followed either a high and increasing (n=33, 18 per cent) or low (n=148, 82 per cent) trajectory. Treatment and control children did not differ in their probability of following either trajectory (OR=1.68, CI=0.78-3.75, NNT=13).

Question 2. Within each trajectory, did treated children perform better than controls?

Behavioural outcomes (Table 3A): Within the high externalising behaviour trajectory, treated children had lower scores than controls at each time point, with medium to large effect sizes (Hedges’g: 0.45-to-0.58, p<0.05). Within the low externalising behaviour trajectory, treated and control children had similar externalising scores at all time points (Hedges’g: 0.03-to-0.07, p>0.05). No differences were found in the overall rate of change (slope) between the treatment and control groups in either the high (Hedges’g: 0.23, p<0.344) or low (Hedges’g: 0.03, p<0.911) trajectories. Similarly, no significant differences between the groups were found for the mean or the slope of the internalising trajectories for both the high (intercepts, Hedges’g: 0.46-to-0.56, p<0.05; slope, Hedges’g: 0.23, p=0.548) and low (intercepts, Hedges’g: 0.09-to-0.12, p<0.05; slope, Hedges’g: 0.23, p=0.548) trajectory groups.

Cognitive outcomes (Table 3B): For the cognitive skill trajectories, no differences were found between treatment and control children (in both the high and low trajectories) for the mean score at each time point (high trajectory, Hedges’g: 0.01-to-0.13, p<0.05; low trajectory, Hedges’g: 0.03-to-0.17, p<0.05), or the change over time (high trajectory, Hedges’g quadratic slope: 0.09, p<0.807; low trajectory, Hedges’g quadratic slope: 0.07, p<0.510). The same pattern was found for the vocabulary trajectories.

Health service use (Table 3C): For the health clinic visit trajectories, there were no differences in the mean number of visits for the treatment group compared to the control group across all time points for either trajectory (high trajectory, Hedges’g: 0.05-to-0.28, p<0.05; low trajectory, Hedges’g: 0.04-to-0.22, p<0.05). In addition, no differences were found in the rate of change over time for children in the low trajectory (Hedges’g linear and quadratic slopes: 0.20, p=0262, and 0.00, p<0.300). However, for the high trajectory, the control group increased their number of health visit significantly faster than the treatment group (Hedges’g linear and quadratic slopes: 0.37, p<0.041, and 0.26, p<0.075).

Discussion

We found that a five-year, multi-component early intervention programme providing education and social support to families living in economic deprivation had moderate-to-large positive effects on children’s behavioural and cognitive development and health services use between six months and four years of age. These resultwere in line with previous findings of the PFL trial (Doyle et al., 2015, 2017a, 2017b) and extended this body of work by testing the impact on children’s developmental patterns (i.e. population heterogeneity).

We found impacts on all three types of outcomes, which is in line with the focus of the PFL curriculum. First, the positive effect for externalising problems was limited to children in the high externalising trajectory: among children with the most problems, treated children had lower levels of externalising behaviour than controls, but no such difference was found among children with low levels of externalising problems. This result is in line with the compensatory hypothesis, suggesting that children with the highest levels of problems should gain the most from early intervention. Externalising behaviours (for example, physical aggression) are normative during early childhood, but children exhibiting the highest levels of such behaviours are at risk of continuing on a chronic trajectory (Côté et al., 2006; Tremblay, Vitaro and Côte, in press). Since children with high externalising behaviours are typically the most resistant to treatment, this study suggests that an intensive intervention from pregnancy to age four can prevent chronic externalising problems.

Second, concerning cognitive skills, our findings are in line with the average treatment effect hypothesis, whereby all children benefited from the intervention regardless of their initial levels of skills. The treatment increased children’s probability of following the high-level trajectory for both general cognitive skills and vocabulary, which suggests that early investment via PFL was an effective strategy for skill improvement. As demonstrated by the low NNT, the treatment effect was clinically relevant: four (cognitive skills) and six (vocabulary) children needed to receive the PFL treatment to see a clinical improvement in one child.

Third, we demonstrated that, within the trajectory of high health clinic visits, control children had faster rates of change, experiencing a substantial increase in the frequency of visits between six months and three years, and then a declining frequency. Thus, PFL treated families relied less on health services. Comparing our results to the literature is inherently limited due to PFL’s longer than average duration, multi-treatment nature, and the use of trajectory analysis. However, results are in line with those reported by the Healthy Families America (HFA) and Parents as Teachers (PAT), both five-year home visiting programmes, showing favourable effects on cognitive development after twenty-four months, but not before, while having little if no impact on child health at any time point (Anisfeld, Sandy and Guterman, 2004; Caldera et al., 2007; Drazen and Haust, 1993; Duggan et al., 2007; Kirkland and Mitchell-Herzfeld, 2012; Landsverk et al., 2002; Wagner, Cameto and Gerlach-Downie, 1996). Previous studies testing the impact of home visiting programmes on verbal skills have been inconclusive, with several reports of negative effects (Guttentag et al., 2014; Peacock, Konrad, Watson, Nickel, and Muhajarine, 2013). The HFA programme also reported positive effects on children’s internalising and externalising behaviour at twenty-four months (Caldera et al., 2007; Landsverk et al., 2002). The magnitude of the effects observed in PFL was overall larger than the magnitude of the effects reported in those programmes.

Strengths of the study include a rigorous trial design and frequently measured outcomes over four years allowing us to account for patterns of stability/change over time and address the heterogeneity in children’s developmental trajectories. Main limitations of the study are the small number of children in one of the two trajectories for some of the outcomes, which may have limited our power to detect significant effects (for example, internalising problems), and the reliance on parent-reported measures of children’s development, which may be subject to differential misreporting.

In conclusion, we found that PFL, a five-year prenatally commencing programme, was effective in improving children’s behavioural, cognitive, and health service use trajectories over the first four years of life. The programme was most effective for children with the most severe behavioural problems but was not effective in preventing problem behaviours from emerging. This is in line with the compensatory hypothesis and suggests that preventive interventions for behavioural problems may be most effective by targeting children with the most severe problems. Conversely, for cognitive skills, the intervention consistently placed all children on a better trajectory yet had no impact on the development within those trajectories. Thus, children benefited regardless of their initial levels of skills, which provides support for the average treatment effects hypothesis, suggesting that less-targeted interventions may be effective for cognitive development.

Acknowledgements

We would like to thank the Northside Partnership who provided funding for the Preparing for Life evaluation through The Atlantic Philanthropies and the Department of Children and Youth Affairs, and the Children’s Research Network for providing funding through the Prevention and Early Intervention Research Initiative Grant Scheme (CRN-PEI-2017). We are also grateful to all those who participated in and supported this research, especially the participating families and community organisations, the PFL intervention staff, and the Expert Advisory Committee. Thanks also to Prof. James Heckman, Prof. Colm Harmon, Prof. Cecily Kelleher, Prof. Sharon Ramey, and Prof. Craig Ramey for their guidance and advice throughout the project, and the Early Childhood Research Team at UCD Geary Institute for Public Policy for their contributions to the work.