The HIV Treatment Response Prediction System - using the experience of treating tens of thousands of patients to guide optimal drug selection

Oral and poster presentation at:

HIV Drug Therapy in the Americas - HIVAmericas

8th May 2014 - 10th May 2014

Rio de Janeiro, Brazil

Revell AD¹, Wang D¹, Reiss P, ² van Sighem A ², Hamers R ³, Morrow C ⁴, Gazzard B ⁵, Montaner JS ⁶, Lane HC ⁷, Larder BA ¹ on behalf of the global RDI study group

¹ HIV Resistance Response Database Initiative (RDI), London, UK;
² Netherlands HIV Monitoring Foundation, Amsterdam, The Netherlands;
³ PharmAccess Foundation, Academic Medical Centre of the University of Amsterdam, Amsterdam, The Netherlands;
⁴ Desmond Tutu HIV Centre, Cape Town, South Africa;
⁵ Chelsea and Westminster Hospital, London, UK;
⁶ BC Centre for Excellence in HIV/AIDS, Vancouver, Canada;
⁷ National Institute of Allergy and Infectious Diseases, Bethesda, MD, USA

Background
Optimising the selection of antiretroviral drugs for patients experiencing virological failure remains a challenge, particularly in resource-limited settings where the latest drugs may not be accessible and genotyping may not be affordable to guide drug selection. The RDI was set up in 2002 to develop computational models that predict virological response to different combinations of drugs, and to make those models freely available to assist optimal drug selection.

Methods
In order to collect sufficient data to develop such models, HIV cohorts and clinics around the world were approached for a range of longitudinal clinical, virological and treatment data for their patients. These data have been used to train a range of neural network, support vector machine, linear regression and random forest models to predict virological response to combination therapy. The models re tested with independent validation sets and their accuracy compared with that of genotyping with rules-based interpretation.

Results
Data from approximately 110,000 patients have been collected. The best performing models (random forests) routinely predict virological response/failure to a change of therapy with an accuracy of 80% or more, even without the use of a genotype. Moreover, they are significantly more accurate than genotyping itself. They are able to identify simple, less costly, alternative drug combinations that are predicted to be effective for the majority of cases that failed following a treatment change in the clinic. The models are used to power the free online HIV Treatment Response Prediction System HIV-TRePS.

Conclusions
HIV-TRePS, in essence, makes the distilled experience of hundreds of physicians treating tens of thousands of patients available to all. Use of the system to help guide antiretroviral treatment selection following virological failure has significant potential to improve patient outcomes and save costs following a switch to antiretroviral drug therapy, particularly in resource-limited settings.

View Slides   View Video