Poster presentation at:
XIX International AIDS Conference
22nd July 2012 - 27th July 2012
Washington DC, USA
Larder BA1, Revell AD1, Wang D1, Hamers R2, Tempelman H3, Barth R4, Wensing AMJ4, Morrow C5, Wood R5, DeWolf F6, Gazzard B7, Lane HC8, Montaner JM9 on behalf of the global 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 Netherlands HIV Monitoring Foundation, Amsterdam, The Netherlands
7 Chelsea and Westminster Hospital, London, UK
8 National Institute of Allergy and Infectious Diseases, Bethesda, MD, USA
9 BC Centre for Excellence in HIV/AIDS, Vancouver, Canada
Background In resource?limited settings (RLS), optimising therapy following treatment failure is particularly challenging due to limited treatment options and the general unavailability of resistance testing. In resource?rich settings (RRS), computational models can predict virological response accurately from a range of laboratory data including a genotype, and are useful web?based clinical tools. Here we develop models that do not require a genotype and assess their potential clinical utility using data from RLS.
Methods Random forest (RF) models were trained to predict the probability of virological response (viral load <400 copies/ml) following a treatment change, from baseline viral load, CD4 count, treatment history, new drug regimen and time to follow?up. Approximately 16,000 treatment change episodes (TCEs) meeting the entry criteria were extracted from our database of 85,000 patients (largely from RRS) and randomised into a training set of 14,891 and a test set of 800 TCEs. The models were assessed during cross?validation, with the 800 test set and with 231 cases from sub?Saharan Africa. The area under the ROC curve (AUC) was the main outcome.
Results The models achieved a mean AUC of 0.77 (95%CI 0.76,0.78) during cross validation and 0.77 (95%CI 0.73,0.80) with the 800 test set. For the RLS test sets, AUC ranged from 0.58?0.65. For South African TCEs, AUC= 0.62 (95%CI 0.53,0.71). The models correctly predicted approx. 60% of the RLS treatment failures and identified alternative 3?drug regimens comprising locally available drugs with higher probabilities of response for all clinical failures.
Conclusions Models trained with large datasets from resource?rich countries can predict virological response without a genotype with an accuracy that is comparable to that of genotyping with rules?based interpretation. The models have the potential to predict and avoid treatment failure by identifying effective, alternative, practical regimens. This approach has potential clinical utility for managing cases of treatment failure in RLS.