Bayesian graphical models in the analysis of HIV drug resistance data from the UK HIV Drug Resistance Database
Fraser Iain Lewis, Esther Fearnhill, Ronan J. Murray, D Pillay, David Dunn, Andrew Brown · UCL Discovery (University College London) · 2006
Deriving models from clinical data requires • relevant outcome measure • large enough curated dataset • appropriate models Data Required: pVL – Genotype – Therapy Change – pVL From UK HIV Drug Resistance Database select cases with 1. genotype ≤ 3 months before or ≤ 1 month after treatment change & 2. ≥ 1 viral load after treatment change & 3. ≥ 1 IAS mutation present = 3,202 records Time to breakthrough ≥ 1 IAS mutation Breakthroughs comprise a skewed distribution with long right tail Bayesian Mixture Model was fitted to time to failure. 2 separate Gaussian distributions were specified & MCMC estimates of optimal means and variances obtained.