A covariate-adjusted reference interval regresses the analyte on a covariate and takes percentile bounds from the residual SD, so the limits move with age, sex or size.
Many lab analytes have normal ranges that vary with age, sex, or another covariate (haemoglobin by sex, eGFR by age, hormone levels by menstrual cycle). A single reference interval misclassifies a substantial fraction of healthy subjects at the extremes of the covariate distribution.
Royston-Wright fits a regression model y = f(x) + ε, where f is typically a polynomial (or fractional polynomial). The reference bounds at any x are mean(x) ± z·σ_residual, where σ_residual is the within-x standard deviation.
For positively-skewed analytes, log-transform y first; the bounds are back-transformed for reporting. The implementation flags via Shapiro-Wilk on the residuals when normality is suspect.
Numeric measurement + covariate + polynomial degree (1–3) + reference level + log-transform toggle.
80-point smoothed mean curve with dashed upper / lower bounds + 8-point lookup table + Shapiro-Wilk on residuals.
Look up the bounds at the patient's specific covariate value to classify — not the marginal (covariate-ignored) bounds. The whole point is that ranges vary by covariate.