Age targeting of interventions is of enduring scientific interest – and is topical due to the recent policy and neuroscientific thrust towards early intervention. This policy thrust appears to stem from several sources, including neuroscientific evidence that development is more malleable during sensitive periods, and from economic data. Nobel prizewinning economist James Heckman aggregated data across different ages of children and youth, and multiple interventions, claiming greater effectiveness and cost-effectiveness of early compared to later interventions. However, his analyses are limited by collapsing data across interventions with very diverse goals and outcomes, and by confounding age with intervention type (e.g. early parenting vs. teen bootcamps), allowing no comparison of similar interventions at different ages. In the field of parenting interventions for conduct problems, investigators have tested age effects across similar interventions, using conventional meta-analysis across multiple randomised trials. However, these studies have yielded mixed findings, and are limited by this method only allowing for coding of moderating factors at trial aggregate-level. As a result, all information is lost about within-trial variability in age. Furthermore, no trials have included economic data on this question, by examining differential cost-effectiveness by age.
To overcome these limitations, we conducted individual participant data (IPD) meta-analysis of a near-complete set of European randomised trials of Incredible Years parenting intervention (IY;1799 children age 2-12; 14 trials, 6 countries). We used random effects modelling to separate individual from trial-level variation. Where outcome instruments differed, we harmonized data across trials using norm deviation scores for the primary outcome, parent-reported disruptive child behavior. Economic data were harmonised across a subset of 5 trials from UK and Ireland. Our complementary second systematic review used conventional aggregate-level (between trial) meta-analysis. This provides much greater generalisability, but at the cost of losing data on individual-level (within trial) variability. We combined trial-level data from 150 trials of parenting programs, analysed using robust variance estimation (396 effect sizes; N=16,345, mean trial age 2-10) to test whether, at trial-level, intervention effects on disruptive behavior are moderated by child age; and if developmentally-targeted programs are more effective.
We found that contrary to expectations from the literature, across both methods, there were no effects of age on intervention outcome; pooled economic analyses – albeit more modest in scope and age range – suggested that cost-effectiveness was greater at older ages. Thus there is potential for return on investment to be higher at older, not younger ages. For examining differential effects, our twin methods are ideal in combining strengths of generalizability across a wider range of interventions and settings in conventional meta-analysis, with the greater power, precision, and, especially, utilization of individual-level age information afforded by IPD. Examining differential cost-effectiveness is novel, and reaches very different conclusions from Heckman, when comparing effects of like-for-like interventions at different ages, in the field of parenting interventions.