Oral and poster presentation at:
International Workshop on HIV & Hepatitis Drug Resistance and Curative Strategies
4th June 2013 - 8th June 2013
Toronto, Canada
Larder BA1, Revell AD1, Wang D1, Hamers R2, Tempelman H3, Barth R4, Wensing AMJ4, Morrow C5, Wood R5, van Sighem A6, Reiss P6, Nelson M7, Emery S8, Montaner J9, Lane HC10, on behalf of the RDI study group
1 HIV Resistance Response Database Initiative (RDI), London, UK
2 PharmAccess Foundation, Academic Medical Centre, Amsterdam, The Netherlands
3 Ndlovu Care Group, Elandsdoorn, South Africa
4 University Medical Centre, Utrecht, The Netherlands
5 Desmond Tutu HIV Centre, Cape Town, South Africa
6 Stichting HIV Monitoring, Amsterdam, The Netherlands
7 Chelsea and Westminster Hospital, London, UK
8 The Kirby Institute, University of New South Wales, Sydney, Australia
9 BC Centre for Excellence in HIV/AIDS, Vancouver, Canada
10 National Institute of Allergy and Infectious Diseases, Bethesda, MD, USA.
Background
In wellaEUR"resourced settings, genotypic resistance testing (GRT) plays a major role in the selection and sequencing of combination antiretroviral therapy (ART) for the long-term suppression of HIV. GRT is unavailable in most resource-limited settings (RLS) so optimal drug selection is a challenge. Here we report the development and evaluation of the latest computer models to predict response to ART without GRT.
Methods
Random forest (RF) models were trained to predict the probability of virological response (viral load <50 copies/ml) using baseline viral load, CD4 count, treatment history, new drug regimen and time to follow-up from 22,567 treatment change episodes (TCEs) following failure. The models were assessed during cross-validation, with an independent test set (n=1,000) and 100 TCEs from sub-Saharan Africa (100 SSA). The area under the ROC curve (AUC) was the main outcome. The predictive accuracy of the models was compared to GRT using genotypic sensitivity scores (GSS) generated by three rules-based interpretation systems. They were also used to identify potentially effective drug regimens for cases of virological failure following a treatment change in RLS.
Results
The models achieved a mean AUC of 0.82 during cross validation, 0.80 with the 1,000 test set and 0.78 for the 100 SSA set. Of the 1,000 test TCEs, 346 had genotypes available and the AUC for the GSS ranged between 0.56 and 0.57, which was significantly inferior to the accuracy of the models (p<0.0001). The models identified alternative 3-drug regimens, comprising locally available drugs, with higher probabilities of response for 96% of cases in RLS where the new treatment had failed.
Conclusions
These models, trained with the largest dataset so far, were able to predict response to HIV therapy significantly more accurately than genotypes with the most commonly used interpretation systems. Despite being trained with data mostly from well-resourced settings they were accurate for cases from RLS. The models have the potential to predict and avoid treatment failure by identifying effective, alternative, practical regimens and have potential clinical utility for RLS.
View Slides