Skip to content

Model Fitting

The fitACE package handles statistical model fitting on simulated data. This page provides conceptual context; see the fitACE README for usage instructions.

Phenotype models

Continuous liabilities are mapped to age-of-onset phenotypes via time-to-event models:

Model Description
Frailty Proportional hazards with choice of baseline hazard (Weibull, Gompertz, lognormal, etc.). Liability scales hazard via \(z = \exp(\beta L)\). Given sufficient time, every individual eventually becomes affected.
Cure-Frailty Mixture model separating who gets the disease (susceptible vs. non-susceptible) from when (age-of-onset among susceptibles). Supports sex-specific prevalence.
ADuLT LTM Deterministic liability threshold model with logistic cumulative incidence proportion (Pedersen et al., 2023).
ADuLT Cox Proportional hazards with Weibull noise and rank-based CIP-to-age mapping (Pedersen et al., 2023).
Simple LTM Binary affected status from liability threshold, without age-of-onset or censoring.

Censoring

Two censoring layers mimic real-world data limitations:

  1. Age-window censoring -- per-generation observation intervals [left, right]. Events before left are left-truncated; events after right are right-censored.
  2. Competing-risk mortality -- death age drawn from a Weibull distribution, independent of disease liability. Individuals who die before onset are death-censored.

The combined effect: only a fraction of true cases are observed as affected.

Subsampling and ascertainment

The pipeline can restrict observed data, allowing the impact on downstream estimates to be studied:

  • Subsampling (N_sample) -- random subset of phenotyped individuals
  • Case ascertainment (case_ascertainment_ratio) -- cases sampled at higher rate
  • Pedigree dropout (pedigree_dropout_rate) -- random removal of individuals, breaking pedigree links and downgrading relationship types

Heritability estimation

The simulation validates several heritability estimation approaches:

  • Falconer's \(h^2\) -- from tetrachoric correlations between MZ and DZ twins
  • Tetrachoric correlations -- by relationship type and generation
  • Weibull frailty MLE -- pairwise survival-time correlation via Gauss-Hermite quadrature
  • EPIMIGHT -- external R package for heritability from family data
  • PA-FGRS -- polygenic risk scores from family history