We then created each predictor vector by determining the total number of unique attributes for each gene, followed by post-processing of each field, including z-score normalization (see online Supplementary Material).
Model 1 simply included Z-score at each age.
We moreover included Z-score transformations of the NA and SI scales and tested for interaction effects.
To use both biological traits and economic costs as constraints in such prioritization analysis, we performed the same calculation of z-scores described above, including z-scores for land acquisition costs.
Models for the overall sample included z-scores calculated according to the means and SD for the overall sample; models for Eastern Europe and Western Europe included z-scores calculated with the mean and SD of each region, respectively.
More complete information, including Z-scores, lists of target genes for each regulator, associated mechanistic networks, and other upstream regulators are shown in Additional file 1: Table S3.
Each model generated was checked for various parameters which include Z score, GMQE (global quality estimation score) and QMEAN (model quality estimation) score to assess the accuracy of the model.
Age, disease duration, gender, smoking, baseline disease-modifying anti-rheumatic drug (DMARD) and glucocorticoid treatment as well as baseline DXR-BMD (including Z scores) were not univariately associated with radiographic progression at any time point.
DXA measures included Z-scores for total body BMD (TBMD, not excluding head), posterior anterior lumbar spine at L2-L4 (LBMD), femoral neck (FNBMD; only for subjects ≥20 years at study visit), and body composition (percent fat mass, total fat mass, LBM).
Explanatory variables included z-scores of 'proportion of optimal birth weight' (POBW), 'proportion of optimal birth length' (POBL) and 'proportion of optimal weight for length' (POWFL) (Blair et al, 2005); three measures of the appropriateness of intra-uterine growth (IUG) described previously (Milne et al, 2007).
