New global computer models published this week predict how patients whose HIV therapy is failing will respond to any new combination of drugs, with the highest accuracy to date.
The models, developed with data from tens of thousands of patients around the world, were significantly better at identifying potentially effective drug combinations than ‘local’ models trained with data from South Africa when tested with HIV patients from that country.
The publication describes three new sets of models: Global models trained with around 30,000 cases from around the world that predict response to HIV treatment without the need for a genotype (a test that reads the genetic code of the virus) and similar ‘local’ South African models, trained with around 3,000 cases from South Africa. These two sets of ‘no-genotype’ models were developed specifically for low-income settings where genotyping is unaffordable. Finally a set of global models that use a genotype in their predictions were developed.
The global and local ‘no-genotype’ models predicted responses to HIV treatment for South African patients with a similar level of accuracy (around 80%) but the global models were significantly better at identifying alternative combinations of drugs that were predicted to work for South African patients who failed the new treatment given in the clinic.
The global genotype models were marginally but not statistically significantly more accurate than the no-genotype models and all three sets of models were substantially more accurate predictors of treatment response than the genotype test itself (55-58% accuracy).
“The performance of the global models that do not require a genotype was very encouraging and provides further evidence that this could be a very helpful approach for selecting effective therapy in resource-constrained settings, such as South Africa,” commented Andrew Revell, executive director of the Response Database Initiative (RDI) and lead author on the paper.
The paper concludes “the global models have the potential to reduce virological failure and improve patient outcomes in all parts of the world, with particular utility in resource-limited settings. The models can provide clinicians with a practical tool to support optimised treatment decision-making in the absence of resistance tests and where expertise may be lacking in the context of a public health approach to antiretroviral roll-out and management.”
The new models are now available to be used by healthcare professionals as part of the RDI’s HIV Treatment Response Prediction System (HIV-TRePS), which is freely available online at https://www.hivrdi.org/treps/
Optimizing antiretroviral drug combination on an individual basis in resource-limited settings is challenging because of the limited availability of drugs and genotypic resistance testing. Here, we describe our latest computational models to predict treatment responses, with or without a genotype, and compare the potential utility of global and local models as a treatment tool for South Africa.
Global random forest models were trained to predict the probability of virological response to therapy following virological failure using 29 574 treatment change episodes (TCEs) without a genotype, 3179 of which were from South Africa and were used to develop local models. In addition, 15 130 TCEs including genotypes were used to develop another set of models. The ‘no-genotype’ models were tested with an independent global test set (n = 1700) plus a subset from South Africa (n = 222). The genotype models were tested with 750 independent cases.
The global no-genotype models achieved area under the receiver-operating characteristic curve (AUC) values of 0.82 and 0.79 with the global and South African tests sets, respectively, and the South African models achieved AUCs of 0.70 and 0.79. The genotype models achieved an AUC of 0.84. The global no-genotype models identified more alternative, locally available regimens that were predicted to be effective for cases that failed their new regimen in the South African clinics than the local models. Both sets of models were significantly more accurate predictors of outcomes than genotyping with rules-based interpretation.
These latest global models predict treatment responses accurately even without a genotype, out-performed the local South African models and have the potential to help optimize therapy, particularly in resource-limited settings.
Andrew D Revell, Dechao Wang, Robin Wood, Carl Morrow, Hugo Tempelman, Raph L Hamers, Peter Reiss, Ard I van Sighem, Mark Nelson, Julio SG Montaner, H Clifford Lane, Brendan A Larder