Discussion
Our systematic review identified seven RCTs of parenting interventions that conducted a follow-up evaluation of the original trial cohort. Follow-ups were mostly short-term, within 1–3 years after programme completion; only two trials had long-term follow-ups (10+ years) that tracked cohorts from early childhood into adolescence or young adulthood.
Although there were consistent intervention benefits on multiple ECD and parent-level outcomes immediately after programme completion, follow-up results revealed a general fading of effects over time across all trials. The sustainability of intervention effects over time appeared to be associated with the magnitude of immediate postintervention effects on ECD outcomes. For example, with cognitive development, immediate impacts ranged from small effect sizes in four of the studies (SMD=0.2–0.3) to medium-to-large effect sizes for the remaining three studies (SMD=0.5–0.9). The three trials with larger immediate postintervention impacts showed significant sustained benefits in the short-term,19 23 24 whereas the other trials with small postintervention impacts did not show sustained benefits at any subsequent follow-up evaluation for any outcomes. Our results suggest there may be a threshold of immediate gains required—perhaps to the magnitude of at least moderate-sized postintervention effects (SMD>0.5)—in order to activate the potential for longer-term sustained benefits on ECD. Additional follow-up studies with larger samples are needed to confirm these trends, especially considering the wide CIs associated with most estimates.
The two interventions that achieved medium-to-large immediate gains in caregiving and parent-level outcomes were those that similarly had larger postintervention effects on ECD and subsequently sustained short-term benefits on ECD. More specifically, Yousafzai et al35 found large initial effects on maternal knowledge of ECD, stimulation and mother–child interactions, and sustained benefits on ECD and parent outcomes in the short-term. The trial by Muhoozi et al found medium-sized initial reductions in maternal depressive symptoms and sustained reductions in depression and improvements in ECD outcomes in the short-term.29 On the other hand, Attanasio et al,22 Cooper et al,28 Chang et al25 and Walker et al36 found small, if any, postintervention effects on maternal outcomes and no follow-up effects on any maternal or ECD outcomes over time. Given that improvements in parenting are generally the primary pathway through which these interventions improve child outcomes,4 if parenting behaviours are not meaningfully improved postintervention, then fadeout effects on ECD outcomes are even more likely. Our results highlight the importance of targeting and sufficiently improving parental behaviours and well-being in order to sustain longer-term programme impacts on ECD outcomes beyond the completion of parenting interventions.
While we identified a potential trend between initial impacts on ECD and parenting outcomes and sustainability of intervention effects of time, there are also a number of other factors that may explain the heterogeneity in follow-up results. First, intervention theories of change and target populations varied across trials. For example, half of the programmes enrolled birth cohorts and included components to enhance maternal sensitivity and responsiveness beginning during the postnatal period,24 28 36 compared with other interventions that focused primarily on increasing cognitive stimulation, distributed play materials to the households every week as part of the programme, and more directly engaged a broader and older age range of children between 9 and 24 months at enrolment.19 22 Variations in programme components and theories of change may reasonably explain why certain interventions did not improve particular ECD outcomes (eg, no impact of postnatal maternal sensitivity intervention on later child cognitive development outcomes37) and the null overall effects observed for behavioural development, which may require alternative interventions that have a stronger focus on social learning theory.38 The majority of interventions concluded prior to child age 2 years, with the exception of Attanasio et al22 that supported children up until age 3 years, and Grantham-McGregor et al19 that also engaged some children older than 3 years of age, depending on their initial age at enrolment. The transition to preschool is a critical developmental period, during which continued support for parents and children may confer additional advantages that may produce sustained effects on later outcomes.39
Second, intervention implementation characteristics also varied substantially in terms of dosage, duration, delivery agents and scale. For example, the original Jamaica Home Visiting programme was the most intensive and involved weekly 1-hour home visits for 24 months, delivered by community health aides, among a small sample in a relatively contained geographical area in the capital city,19 compared with a programme in Colombia that was much larger, integrated at scale into the existing conditional cash transfer programme and delivered by volunteer mothers through weekly home visits for 18 months.22 It has been suggested that more frequent and longer programme durations are associated with greater immediate postintervention effects of early childhood interventions.40 It is likely that sufficient programme exposure, as well as quality implementation, is even more crucial in order to produce longer-term enduring effects. In spite of these trends, it is worth noting that the intervention in Uganda, which had the shortest duration of 6 months, found sustained improvements in ECD outcomes and reductions in maternal depressive symptoms after a 2-year follow-up.26 29 These unique findings may be explained by the fact that this was primarily a research study (ie, outside of existing community service delivery platform) and used bachelor-level session facilitators that were likely substantially better trained and more skilled than lay community members used in other trials.
Third, characteristics of the study population and context varied widely. For example, the trials in Jamaica targeted stunted and low-birthweight children, and the trials in Colombia and Uganda targeted poor households. Prior studies have suggested that disadvantaged children may be more likely to benefit from early interventions.41 Others have suggested that interventions for disadvantaged children may increase likelihood of observing programme effects considering their additional vulnerabilities and already likely delayed developmental trajectories in the absence of any early intervention.42 At the same time, broader population-level socioeconomic deprivations can also undermine the sustainability of programme gains. For example, weak community health services, food insecurity or the lack of access to preprimary school education in low-income contexts can compromise the environments needed to subsequently sustain gains in children’s developmental skills.43
Taken together, our results highlight several gaps and considerations for future research. First, the majority of trials were relatively small efficacy studies, greatly limiting the ability to detect smaller effects in longer-term follow-ups. Moreover, many outcomes assessed in the follow-up rounds were not theoretically justified, and few parent-level outcomes were measured in the follow-up studies. Yet, behavioural changes in caregiving knowledge, skills and practices with their young child are a key theoretical pathway of parenting interventions.27 Our results emphasise the need for developing and applying theories of change to investigations of follow-up effects, which can inform decisions about which outcomes to assess and ensure hypothesised mechanisms are adequately captured.
Few trials have conducted post hoc analyses of potential mediators underlying intervention follow-up effects. Of notable exception, the trial in Pakistan found that sustained improvements in maternal scaffolding skills explained benefits of the intervention on children’s intelligence and executive functioning,44 and sustained improvements in maternal and paternal stimulations explained sustained intervention benefits to children’s cognitive and socioemotional development outcomes.45 Improved measurement of parenting outcomes across follow-ups and longitudinal mediation analyses are needed to understand common mechanisms that drive sustained treatment gains and identify processes that can be harnessed in future parenting interventions to increase the potential for longer-term impacts.
Although the current evidence for intervention effects on child or parent outcomes is limited in the short-term and even moreso inconclusive in the longer term, it is worth mentioning two additionally plausible interpretations of the present findings. Prior studies have suggested ‘sleeper’ effects with regard to potential long-term effects of parenting interventions.18 46 Sleeper effects refer to a phenomenon whereby an intervention produces no immediate postintervention effect (or a small effect) that is latent in the short-term, requires time to fully materialise and then gradually appears at a later follow-up.18 47 48 In addition, there may be potential effects that are not being captured using the current measures or for outcomes that were not assessed. Both of these possibilities support continued rounds of follow-up studies in order to explore whether sleeper or unmeasured effects might be a possible explanation for mixed short-term and seemingly null medium-term impacts. The trajectory of follow-up results from the Grantham-McGregor et al study indicated a large immediate postintervention impact, a null medium-term impact, but then a rebounding and sustained positive long-term effect. Based on these results, it appears possible that treatment impacts may fluctuate in the short-term to medium-term. Therefore, multiple waves of follows-ups are needed in order to determine longer-term patterns and potential trajectories of treatment effects.
There are several limitations of our review that are worth highlighting. First, longitudinal trials are often subject to loss to follow-up. The prevalence of loss to follow-up among the sample revisited by design ranged from 2% to 34%. Although some studies stated no observed differences between those who were reassessed and those who were lost to follow-up, others did not specify and therefore results may be subject to bias. Second, as already mentioned, most included trials were relatively small efficacy studies that did not present power calculations to determine whether the sample size was sufficient to detect follow-up treatment effects, which complicates interpretation of null results. Third, many studies did not report quantitative values for each stated outcome or provide details regarding measurement adaptation, reliability and validity. Fourth, quantitative data synthesis for effects over time on cognitive and behavioural development were exploratory in nature. Given the few trials represented and the heterogeneity in interventions, outcome measures and timing, pooled estimates should be interpreted with caution. Finally, our study only included published articles, which introduces the potential for possibly overestimating long-term effects, considering how initial null or weak findings are less likely to conduct follow-up evaluation and be disseminated by authors (ie, publication bias).