Unknown adherence issues and the possibility that hidden drug-resistant minority species impaired response to treatment are among the most likely, although not verified, reasons for prediction errors. The inclusion

of some currently obsolete therapies (e.g. use of nelfinavir or stavudine in five cases) and the lack of novel antiretroviral drug classes in the test data set may have been a limitation of the study. However, most of the therapies were not outdated and in addition are clearly relevant for most of the low- to middle-income areas where antiretroviral coverage has recently expanded. The free web service provided by the EuResist network may this website be particularly effective in these settings. Several high-genetic-barrier drugs such as darunavir, tipranavir and etravirine could not be considered for training the EuResist engine because of a shortage of data and thus could not be included in the study data set. The updated version of the EuResist

engine recently made available online (version 2.0) Stem Cell Compound Library can now also compute the response to these three drugs. It remains to be established how the expert system would perform with respect to human experts for these high-genetic-barrier drugs. This is clearly relevant because predicting the activity of such drugs is crucial in the current antiretroviral therapy situation, at least in Western countries. Also, drugs belonging to novel classes such as integrase inhibitors and coreceptor antagonists cannot be included in the computations because of the scarcity of available treatment cases and/or a lack of virus genotype information. The TCE definition itself had its own limitations. First, a short follow-up time was employed because EuResist was trained to predict response at 8 weeks. Short-term response is directly related to antiviral activity on the majority virus population and is usually less complicated by confounding

factors, such as adherence or toxicity, than long-term response. However, with the availability of novel well-tolerated long-lasting therapies, the goal L-gulonolactone oxidase shifts to prediction of longer-term response. While the aim of the study was to predict the 8-week response because the EuResist engine had been trained on that follow-up time, post hoc intention-to-treat analysis at 24 weeks (not shown) confirmed an accuracy of 0.78 for EuResist compared with an average accuracy of 0.71 for the human experts. The next update of the EuResist engine is also planned to focus on the 24-week response. Secondly, the definition of virological success was based on a single follow-up viral load measurement. In some cases, treatment success was reached at a later time-point under the same therapy (data not shown), making definition of the case as a failure questionable [15].

No related posts.