organise_phenotypes - outlier detection and removal

[explodecomputer]

No description provided.

[explodecomputer]

• Use phen definitions to remove outliers due to measurement error
• Use windsorising to reduce influencers - GoDMC implements quartile +/- 3xIQR to detect outliers and windsorises them to re-value them at 5/95% value.

Update

This won't work with a cohort with multiple ages

1. Remove outliers based on definition
2. y ~ age + age^2 - get residuals
3. Windsorise residuals
4. Reconstruct phenotype as y_new = age * b1 + age^2 * b2 + new_residuals + intercept

Assumes constant variance with age - does using age quadratic hinder?

Good to try simulating