Effect of ignoring the time of HIV seroconversion in estimating changes in survival over calendar time in observational studies: results from CASCADE.

OBJECTIVE: To compare estimates of changes in HIV survival over time derived from seroconverter and prevalent cohorts. DESIGN AND METHODS: Using pooled data from 19 seroconverter cohorts (CASCADE), the relative risk of death from HIV seroconversion by calendar time at risk from 1 January 1991 was examined. The analyses were repeated, ignoring knowledge of the time of seroconversion, but adjusting for the CD4 cell count at the time the participant came under observation, thus mimicking a prevalent cohort. Estimates from the 'prevalent' cohort approach were compared with those obtained from the seroconverter cohort. RESULTS: Of 5428 subjects at risk on 1 January 1991 or later, 1312 (24.2%) had died. In the analysis based on time from seroconversion, estimates of the effect of calendar year showed marked reductions in mortality in 1997-1999 only, with no evidence of a linear trend over the period 1991-1996 (P-trend = 0.85). Using the prevalent cohort approach a decrease in the relative risk of death was observed from 1991 to 1998-1999, with a statistically significant trend of a decrease in risk from 1991 to 1996 (P-trend = 0.002). Similar findings were observed when the analyses was repeated taking the start date of the cohort as 1 January 1988. CONCLUSION: Lack of knowledge of HIV infection duration may lead to biased and exaggerated estimates of survival improvements over time. The adjustment for duration of infection in prevalent HIV cohorts through laboratory markers may compensate inadequately for this.