Familial confounding and causal inference in child and adolescent neurodevelopment and mental health.

Abstract

Recent announcements by the US Government linking paracetamol use during pregnancy to autism in offspring highlight the risks of misinterpreting observational research to inform policy; this is a clear example of the principle that association does not equal causation. Unmeasured familial confounding is a common bias in epidemiological studies, whereby shared genetic or environmental factors within families produce spurious associations between risk factors and outcomes. In this Viewpoint, aimed at clinicians from a range of disciplines working with children and young people with neurodevelopmental and mental health conditions, we discuss the concept of familial confounding and why it matters for causal inference. We illustrate the concept through several examples and outline study designs that can be used to minimise bias due to familial confounding, along with their strengths and limitations. We also highlight the potential consequences of ignoring familial confounding, both for causal inference and for policy making. Finally, we emphasise triangulation across complementary study designs as a key strategy for strengthening causal inference and informing policy decisions based on observational evidence.