Modelling Treatment Response Could Reduce Virological Failure in Different Patient Populations
Andrew D Revell1*, Dechao Wang1, Gabriella d’Ettorre2, Frank DE Wolf3, Brian Gazzard4, Giancarlo Ceccarelli5, Jose Gatell6, María Jésus Pérez-elías7, Vincenzo Vullo8, Julio S Montaner9, H Clifford Lane10 and Brendan A Larder1
- *Corresponding Author:
- Andrew D. Revell, Ph.D
The HIV Resistance Response Database Initiative (RDI)
E-Mail: [email protected]
Received Date: May 22, 2012; Accepted Date: July 05, 2012; Published Date: July 07, 2012
Citation: Revell AD, Wang D, d’Ettorre GD, Wolf FDE, Gazzard B, et al (2012) Modelling Treatment Response Could Reduce Virological Failure in Different Patient Populations. J AIDS Clinic Res S5:008. doi:10.4172/2155-6113.S5-008
Copyright: © 2012 Revell AD, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Background: HIV drug resistance can cause viral re-bound in patients on combination antiretroviral therapy, requiring a change in therapy to re-establish virological control. The RDI has developed computational models that predict response to combination therapy based on the viral genotype, viral load, CD4 count and treatment history. Here we compare two sets of models developed with different levels of treatment history information and test their generalisability to new patient populations.
Methods: Two sets of five random forest models were trained to predict the probability of virological response (follow-up viral load <50 copies/ml viral RNA) following a change in antiretroviral therapy using the baseline viral load, CD4 count, genotype and treatment history from 7,263 treatment change episodes. One set used six treatment history variables and the other 18 - one for each drug. The accuracy of the models was assessed in terms of the area under the receiver-operator characteristic curve (AUC) during cross validation and with 375 TCEs from clinics that had not contributed data to the training set.
Results: The mean AUC achieved by the two sets of models during cross validation was 0·815 and 0.820. Mean overall accuracy was 75% and 76%, sensitivity 64% and 62% and specificity 81% and 84%. The AUC for each committee tested with the independent test set was 0.87 and 0.855. Mean overall accuracy was 89% and 87%, sensitivity 67% and 61% and specificity 90% and 87%. There were no significant differences between the two sets. The models correctly predicted 330 (92%) of the 357 treatment failures observed in practice and were able to identify alternative regimens that were predicted to be effective for up to 267 (75%) of the failures and regimens with a higher probability of response for all cases.
Conclusions: Computational models can predict accurately the virological response to antiretroviral therapy from a range of variables including genotype and treatment history even for patients from unfamiliar settings. This approach has potential utility as a useful aid to treatment decision-making and may reduce treatment failure