Clinicians and researchers need to account in their longitudinal analyses for relationships in repeated measurements over time. Standard repeated measures models can be used to address relationships based on pre-specified study hypotheses. However, those analyses often do not fully capitalize on the extensive amount of information usually available in the complete data, including the various alternative predictors, possibly measured over multiple time points. Secondary analyses can also be conducted to fully utilize existing clinical and research data and so obtain deeper understanding of relationships within such data. The number of alternative models, though, will usually be quite extensive, making a thorough analysis challenging. Special methods are needed to adaptively search through alternative statistical models in order to identify an effective choice to apply to a given data set. We have developed a variety of such adaptive statistical modeling (ASM) methods for this purpose along with software to support their computation. These methods use k-fold likelihood cross-validation (LCV) for model evaluation together with rule-based heuristic search through parametric models of arbitrary dimension to generate a nonparametric fit to the data. Current LCV-based ASM methods support adaptive forms of classification, cluster analysis, factor analysis, and regression involving univariate or repeated outcome measurements with normal, logistic, or Poisson distributions. We have evaluated and tested these ASM methods using data from an antiretroviral adherence study of HIV positive participants, including analyses of electronically monitored adherence, self-reported adherence, viral load, and CD4 cell count, as well as with non-HIV-related data. We now propose to use ASM methods to conduct secondary analyses of longitudinal health outcomes for participants of the Multicenter AIDS Cohort Study (MACS) in order to assess the impact of drug use types, separately as well as in combination with other drug use types and with other available predictors, on how those outcomes change over time using both adaptive regression and classification methods. ASM methods currently provide only partial support for the proposed analyses, and so we further propose to extend ASM methods to provide full support for those analyses. These research goals will be achieved by addressing the following specific aims. 1. Assess the impact on longitudinal health outcomes for MACS participants of drug use types, separately as well as in combination with other drug use types and with other predictors related to AIDS and AIDS-related diagnoses, antiretroviral adherence, health care utilization/costs, HIV sero-status, medications, participant characteristics, quality of life, and sexual activity. 2. Extend ASM methods to provide full support for the analyses of Aim 1. The proposed research will enhance the understanding of the impact of drug use along with AIDS and AIDS- related diagnoses, antiretroviral adherence, health care utilization/costs, HIV sero-status, medications, participant characteristics, quality of life, and sexual activity on longitudinal health outcomes for men who have sex with men (MSM) through a comprehensive analysis of MACS data over time, thereby expanding the research base for the development of new approaches for HIV and drug use prevention interventions and for improved treatment and medical management of AIDS and AIDS-related conditions among drug-using MSM. Moreover, the methods to be used in this research apply more generally to other health science applications, and software to support the use of such methods will be made available for general use. ? ? ?
The proposed research will enhance the understanding of the impact of drug use along with AIDS and AIDS- related diagnoses, antiretroviral adherence, health care utilization/costs, HIV sero-status, medications, participant characteristics, quality of life, and sexual activity on longitudinal health outcomes for men who have sex with men (MSM) through a comprehensive analysis of MACS data over time, thereby expanding the research base for the development of new approaches for HIV and drug use prevention interventions and for improved treatment and medical management of AIDS and AIDS-related conditions among drug-using MSM. Moreover, the methods to be used in this research apply more generally to other health science applications, and software to support the use of such methods will be made available for general use.
