\(H_0^I\): does the genotype-to-class mapping hold across diagnostic timing?

Definition

Null: the genotype-to-class association is invariant across strata of diagnostic era and age at diagnosis; the class-conditional polygenic scores (autism, ADHD, major depression, schizophrenia, educational attainment, IQ) and the rare-variant burden (de novo loss-of-function in constrained genes, rare inherited burden) show no class-by-stratum interaction. Alternative: the genotype-to-class mapping drifts, showing a class-by-stratum interaction on one or more genetic contrasts. Estimand: the class contrast on each genetic measure as a function of the axis.

Status

Waiting on data access. This test is gated on SFARI genotype access, not on any earlier rejection, and no genetic data is held yet. Nothing is run. Until the data is in hand, the phenotypic half of the headline question is answered by \(H_0^A\), \(H_0^B\), and \(H_0^C\), and this article is a placeholder for the genetic half.

Method

The planned test is a class-by-stratum interaction (a Wald or likelihood-ratio test) in a generalised linear model over each genetic measure, with cognitive impairment as a covariate. The autism class labels stay fixed at the reference, so no mixture is re-estimated: only the association between genotype and the fixed classes is tested for a change along the axis. Effect sizes are read against a resampling baseline. The read is deliberately powered as a test of a change in the mapping rather than of its baseline strength, because the genotype-to-class alignment is expected to be weak and diffuse.

This article will be completed once genotype access is granted and the stage is implemented and run.

See also

• Are the class profiles invariant, and is any drift small? (\(H_0^A\)), the phenotypic drift this genetic test is set beside.

• Does genetic drift track phenotypic drift? (\(H_0^J\)), the model comparison that reads this result against the phenotype.