THE USE OF AUXILIARY EVENTS TO IMPROVE THE ANALYSIS OF SURVIVAL FOR HIV-INFECTED PATIENTS - APPLICATION TO THE FRENCH PROSPECTIVE MULTICENTER COHORT (SEROCO)

A multicenter prospective cohort study, including 512 patients for who m date of HIV infection was known, showed that the use of an appropria te auxiliary event can improve the analysis of survival data and lead to an earlier detection of risk factors for HIV patients. Age at seroc onversion and primary symptomatic infection were used as risk factors. Two age groups were defined as age at seroconversion >30 years (n = 2 03) and less than or equal to 30 years (n = 309). Patients with primar y symptomatic infection PSI (n = 215) were compared with patients with out any clinical manifestation during primary infection (n = 297). Dea th was considered as the endpoint of primary interest and occurred in 76 patients in the study. Classical non-parametric methods (Kaplan-Mei er estimate and long-rank test) and parametric regression model (Weibu ll model) were used for a standard analysis of survival data. A parame tric approach using auxiliary information was used to estimate the sur vival function and to test the effect of age at seroconversion and PSI . We also applied a recently proposed distribution-free method to prod uce a non-parametric estimate of the survival function and to test age at seroconversion and PSI with respect to survival estimates. Both me thods are compared for two distinct auxiliary events (Karnofsky score below 75 and a first drop of CD4 lymphocyte counts below 200 cells/mm( 3)). The use of CD4 lymphocyte counts below 200 cells/mm(3) as an auxi liary event improved the analysis of survival data available in Decemb er 1994. For both methods incorporating CD4 counts below 200 cells/mm( 3) in addition to survival data, the effect of age at seroconversion o n survival was significant in April 1992 whereas it was not significan t with standard methods. For PSI exposure group, results shown in this work do not indicate any improvement in using auxiliary information. Conditions for using an appropriate auxiliary event as well as advanta ges and shortcomings of both methods are discussed. Methods used in th is work, with appropriate auxiliary information, are promising either through a reduction in the time to follow-up to detect risk factors fo r cohort studies or the time needed for drug development in clinical t rials.