From big data to causal insight: rethinking neurodevelopmental risk across the lifespan

Attention-deficit/hyperactivity disorder (ADHD) is increasingly treated with long-term pharmacotherapy, yet the randomised trials underpinning licensing seldom extend beyond months. The ADDUCE study, a five-year, multinational prospective cohort funded by the European Commission, generated uniquely detailed two-year safety data on methylphenidate in children and adolescents [1]. Even so, an investment exceeding five million euros could not deliver the statistical power needed for rare or delayed outcomes. This Campfire session asks what such bespoke cohorts can and cannot tell us, and how real-world data (RWD) from routinely collected electronic health records can complement them without inheriting their limitations or introducing new ones. The conversation then moves upstream to aetiology. Population-based studies of prenatal exposure to antidepressants [2] and antipsychotics [3] show how sibling comparisons, negative control exposures and propensity score methods can separate a causal signal from confounding by maternal illness. A recent multinational cohort of 3.6 million mother-child pairs on maternal diabetes [4] shows how far the infrastructure has developed: harmonised analyses across seven jurisdictions now provide the statistical power, replication and design flexibility that a single bespoke cohort, however carefully built, cannot achieve alone. Themes for discussion include: when a bespoke cohort is worth the cost and when RWD is the better instrument; whether self-controlled, family-based and target trial emulation designs deliver causal rather than merely associational answers; how findings should be weighed when trials, cohorts and RWD disagree; and what a lifespan view of neurodevelopmental risk demands of study design and clinical communication.