Slides presentation at:
Whistler HIV Update 2003
29th March 2003 - 31st March 2003
Brendan A. Larder PhD. Chair of the RDI Scientific Core Group, Cambridge, UK
HIV drug resistance testing has become a widespread part of HIV clinical management but significant shortcomings in interpretation currently limit its clinical utility.
Phenotyping is a time-consuming, expensive procedure, in which the susceptibility of a recombinant version of a virus is tested against individual drugs. The results are difficult to interpret: how much of change in susceptibility is clinically significant? Currently, various cut-off values are generally used, which do not necessarily equate to clinical response. Attempts have been made to define clinical cut-offs from clinical response data, but any categorical cut-off applied to a continuous biological process is somewhat arbitrary. In addition the number of patients involved in developing these estimates has been relatively small.
Genotyping is quicker and less expensive but, with over 200 different mutations affecting drug susceptibility, interpretation of complex mutational patterns remains a challenge. Many different sets of rules or algorithms have been developed, but these vary considerably and generally do not address the complex relationship between genotype and response. The hybrid virtual phenotype test, which averages the phenotypic results from genetically matched samples in a database, overcame the shortcomings of algorithms but not the inherent problems of phenotyping.
The new approach being pioneered by the RDI involves relating HIV resistance data directly to the virological response of patients to different combinations of antiretroviral drugs in clinical practice, using a substantial database. A range of sophisticated statistical analytical methods, including recursive partitioning and artificial neural networks (ANNs), is being used to enable the accurate prediction of viral load response to therapy from baseline genotype. The initial aim is to collect data from approximately 10,000 patients. ANNs have already been used with an initial dataset of a few hundred patients and produced statistically reliable predictions of virological response.
The RDI is a not-for profit initiative and the development of the database is an international collaborative effort involving a variety of private and public research groups. It is intended to make the database widely accessible to scientists and clinicians via the Internet. It is anticipated that this new approach will significantly improve resistance interpretation and treatment decision-making.