We aim to develop statistical methods that will help in understanding the relationship between predictor variables such as genetic sequence and treatment history, and outcome variables such as HIV drug susceptibility phenotype, immunological response, virological response and clinical response. Because antiretroviral drug resistance can frequently arise, thru both the direct effect of treatment, and the transmission of resistant virus, it is vital that routine patient management assess both the sensitivity of a patient to their current regimen, as well as the potential benefit of other regimens. The potential benefit of a regimen can be measured in terms of the development of future resistance, future immune response, duration of virological control and delay in onset of clinical outcome. One approach to assessing the degree of drug susceptibility makes use of HIV genotype sequences obtained from plasma; the cost of such assays has dropped to a level that permits their use in routine patient management. Investigating the relationship between genotype and HIV drug susceptibility phenotype, virological response and other outcome variables mentioned above require the development of new statistical methods for several reasons. One reason is because the genotype sequence is high dimensional in nature. Another is because some response variables may be censored time to event data. A third is because it is necessary to consider more than one response variables simultaneously. It is important to consider multiple outcome variables simultaneously in order to preserve the overall false-positive rate and thereby result in valid inference.
Our specific aims are: (1) To develop statistically rigorous and sound nonparametric methods to test the association between a high-dimensional predictor variable (genotype, treatment history) and a single response variable; (2) To extend the methods in S.A. 1. to accommodate incomplete response variables (censored time-to-event data); (3) To extend the methods in S.A. 1 to allow simultaneous inference with respect to two or more response variables. (4) To provide to the public, in a user-friendly form, software that executes this methodology.
| DiRienzo, A Gregory (2009) Flexible regression model selection for survival probabilities: with application to AIDS. Biometrics 65:1194-202 |
