Table 1.

Main criteria for data censoring

Objective	Exclusion definition
Remove non-adherent patients	All TCEs with a measure of adherence associated with the TCE or the patient indicating non-adherence (based on site-specific cut-offs).
	TCEs with baseline genotypes available that were predicted by the current RDI models and the Stanford HIVdb genotype interpretation system to respond to their new regimen but failed in the clinic.
Removal of TCEs involving single drugs	Any TCEs with a single drug at baseline or as the new regimen.
Removal of TCEs with drugs for which we had inadequate data	This varies from model to model according to the data available, but any TCEs including drugs that appear in fewer than 250 cases in a training set.
Removal or conversion of indeterminate or unrealistic viral load values	Viral load values of 0 or 50 are assumed to be <50 copies/mL but the precise value is unknown and they are coded as 50 copies/mL.
	Viral load data of the form ‘<X’ where X is >50 or 1.7 log10 copies/mL or ‘>X’ because the true value is unknown.
	Any TCEs with a viral load that is exactly 2.60 log10 or 400 copies/mL on the assumption that the values were from an assay with this as the lower limit of detection and that the true value is unknown.
Removal of unrealistic viral load values or CD4 counts	Plasma follow-up viral load values >9 log10 copies HIV RNA/mL. CD4 counts >2000 copies/mm3.