When changes in the survival of HIV-1-infected persons are monitored over time, the inclusion of long-term survivors may lead to overestimation of survival in later calendar periods. This overestimation occurs in most seroprevalent cohorts of clinic attenders because laboratory markers may be inadequate to adjust fully for duration of infection. To adjust sufficiently for this effect, the time of seroconversion must be known.

Before the introduction of highly-active antiretroviral therapy (HAART) in 1996, there was little evidence that antiretroviral treatment would improve survival, and any apparent temporal changes were either not significant or could be explained by biases in study design and analyses.' Recent data from seroprevalent clinic populations have shown that the introduction of HAART as the standard of care has had a large impact on disease progression and survival.2,3 However, few studies of individuals whose time of HIV-1 seroconversion is known or could be estimated (seroconverters) have reported on the impact of HAART on survival,1 and none have provided current survival expectations. This is not surprising because seroconverter cohorts typically have a small sample size and many are underpowered to reliably detect changes in the most recent time periods.

We compared survival from seroconversion among persons at risk during 1986-96 (pre-HAART period) with those at risk in 1997-98 (HAART period) by use of pooled data on 5646 seroconverters from 17 cohorts (see The Lancet's website for details: www.thelancet.com) in 10 European countries, which included individuals exposed to HIVA through sex between men, injecting drug use, sex between men and women, and haemophilia. Using KaplanMeier methods and Cox proportional-hazards models and allowing for late entry, we estimated the distribution of time from seroconversion to death from all causes. For 5380 individuals at risk during 1986-96 (median date of infection, May, 1989), we censored follow-up at Dec 31, 1996. For 3757 individuals at risk during the second period, 1997-98 (median date of infection, October, 1990), we allowed persons to enter the risk set on date of seroconversion, date of entry into the original cohort, or Jan 1, 1997, whichever was latest. In this manner, a person was considered at risk in 1997-98 only from the time since seroconversion at Jan 1, 1997. For example, a person who seroconverted on Jan 1, 1987, will enter the risk set in the HAART period at 10 years, and a person who seroconverted on Jan 1, 1996, will enter that period at I year from seroconversion.

Of 5646 individuals included in the analyses, 1520 (26-9%) had died. The risk of death was reduced by 64% (95% CI 56-71) in 1997-98, compared with 1986-96 (figure). Estimated survival 10 years after seroconversion for the HAART period (1997-98) was substantially higher than estimates for the pre-HAART period (1986-96) (table).

Our findings show a large improvement in survival expectations in all age groups for those at risk during the period in which HAART became available. The continued long-term monitoring of seroconverters is crucial to assess whether the benefits from HAART are sustained because to date the follow-up period after its introduction is very short. Furthermore, the use of antiretroviral therapy closer to seroconversion, particularly for those diagnosed during acute infection, has unknown long-term implications.

Estimated proportions of individuals surviving from HIVA seroconversion In 1986-96 (pre-HAART period) and 1997-98 (HAART period)
Plots are adjusted to the median age at seroconversion (27 years).