Dear HIV-TRePS user / RDI subscriber
On behalf of everyone at the RDI I hope you and your families are well and have not been too badly affected by Covid-19. We know from personal communications that many of you have been deeply involved in the global response and we would like to take this opportunity to thank you for all your hard work and dedication in fighting this pandemic.
We are writing to you today to inform you of the release of two new sets of models for predicting response to HIV therapy, without the need for a baseline CD4 count or viral genotype. A set of ‘Classifier’ models have been developed to predict the probability of the viral load becoming undetectable (<50 copies HIV RNA/mL) following the introduction of a new regimen. A second set of models has been developed to predict the absolute viral load following the introduction of a new regimen, for use in settings that employ a higher definition of virological response than 50 copies HIV RNA /mL.
Both sets of models performed well, in cross validation and independent testing. The classifier models achieved an area under the ROC curve in independent testing of 0.82 and overall accuracy of 75%. This represents just a 2% loss of accuracy compared with models that include the baseline CD4 count in their predictions. The models that predicted the absolute follow-up viral loads correlated highly significantly with the observed viral loads (coefficient of 0.69, p<0.0001) and a mean absolute error of 0.75. This compares well with the results of models developed in 2019 that use a baseline CD4 count, which achieved a correlation coefficient of 0.68 and a mean absolute error of 0.73.
The addition of these models to HIV-TRePS increases the flexibility of the system to make predictions for cases with different baseline data available and in settings that use different definitions of virological response and failure.
Finally, please let us know if you can introduce us to other groups that might have data to share with us to expand the size and heterogeneity of the database and improve the accuracy and generalisability of the models.
Date published: 18th June 2020