       Document 2320
 DOCN  M94A2320
 TI    Segmented Cox models can distinguish short and long term AIDS
       progression markers.
 DT    9412
 AU    Fore AJ; Kramer A; Grund B; Hannan P; University of Minnesota.
 SO    Int Conf AIDS. 1994 Aug 7-12;10(1):330 (abstract no. PC0256). Unique
       Identifier : AIDSLINE ICA10/94370255
 AB    OBJECTIVE: We studied the time-varying effects of CD4+ T-lymphocytes
       (CD4), immunoglobin A (IgA), and phytohaemagglutinin (PHA) on the
       development of AIDS. We used segmented Cox models to estimate the short
       and long term effects of the baseline covariates. METHODS: Using
       illustrative data from a longitudinal study, we examined 155
       HIV-seropositive homosexual or bisexual men with complete baseline
       information on the covariates under study. During the observation
       period, 50 men (32%) developed AIDS. Predictors of AIDS were first
       evaluated using traditional Cox model methods. We then estimated the
       contribution of the covariates to the hazard at each time point with
       Aalen's nonparametric linear regression model. An 'Aalen plot' was
       constructed by plotting the cumulative hazard function against time.
       From inspection of Aalen plots and the distribution of events, a break
       point of 1100 days was chosen for the segmented Cox models. Parameters
       in each segment (before and after 1100 days) were estimated. RESULTS: In
       traditional Cox analysis for the entire observation period, all
       covariates were significant predictors of progression. Using the
       segmented model for the first 1100 days, significant predictors were CD4
       (p = .048) and PHA (p = .007); IgA was marginally significant (p =
       .079). In the second segment, only CD4 was significant (p = .017).
       DISCUSSION AND CONCLUSIONS: In contrast to the traditional Cox model,
       the segmented Cox model enables us to investigate the possibly changing
       importance of baseline covariates over time. The Aalen plot provides a
       visual indication of time dependency and can suggest a suitable break
       point for segmented Cox models. Using these methods, short and long term
       predictors of AIDS progression can be identified. CD4, IgA, and PHA are
       all good short term predictors of AIDS. The predictive effect of
       baseline IgA and PHA appears to decay after approximately three years.
       Only CD4, a well established prognostic factor, is an important long
       term predictor.
 DE    Acquired Immunodeficiency Syndrome/DIAGNOSIS/EPIDEMIOLOGY/  *IMMUNOLOGY
       Adult  Bisexuality/STATISTICS & NUMER DATA  Homosexuality/STATISTICS &
       NUMER DATA  Human  HIV Infections/DIAGNOSIS/EPIDEMIOLOGY/*IMMUNOLOGY
       IgA/*BLOOD  Longitudinal Studies  Lymphocyte Transformation/*IMMUNOLOGY
       Male  Models, Statistical  Prognosis  Proportional Hazards Models  T4
       Lymphocytes/*IMMUNOLOGY  MEETING ABSTRACT

       SOURCE: National Library of Medicine.  NOTICE: This material may be
       protected by Copyright Law (Title 17, U.S.Code).

