How do we establish a biological marker for a behaviorally defined disorder? Autism as a test case.

Lancioni et al. (2011) wrote a roadmap. They asked, 'If someone invents a blood test or brain scan for autism, how would we know it works?'

The team listed three hard checks any biological screen must pass. No data were collected; this is a think-piece for researchers and funders.

The paper gives a three-step validity ladder. A marker must first separate autism from other groups, next predict real-life traits, and finally prove useful in everyday clinics.

Until a tool clears all three rungs, the authors say, don’t use it to diagnose or start services.

Bone et al. (2015) shows why the ladder matters. They tried to repeat flashy machine-learning autism classifiers and the models fell apart on new data—exactly the failure E et al. warned about when step-one checks are skipped.

Kremkow et al. (2022) looks forward, summarizing toddler apps and eye-tracking games that still sit at proof-of-concept stage. Those digital tools sit inside E’s framework; most have not yet climbed past rung one or two.

Philippe (2022) pushes the debate further, arguing that even gold-standard interviews miss rare genetic subtypes. This successor idea updates E et al.; validity may need a fourth rung—checking if a marker fits a biologically defined subgroup.

You will hear about ‘autism blood tests’ or AI apps that claim 90% accuracy. Before you refer a family or change your intake process, ask for evidence on the three-run ladder: separation from other diagnoses, link to daily skills, and clinic benefit. If the answer is ‘we’re still working on it,’ treat the tool as experimental and keep your current assessments in place.