Objectives: To show how a spurious association between the calendar year of seroconversion and HIV progression arises as a result of censoring the follow-up of individuals at their last visit, when the individuals' visits are intermittent.
Design: A notional cohort of 1140 seroconverters and a cohort study of 1270 HIV-infected individuals seroconverted between 1985 and 1994, and followed up to December 1995 (the Italian Seroconversion Study cohort).
Methods: Failure times and rate of the patients attending the clinic over the study period were simulated for the notional cohort. Three separate scenarios with different probabilities of making a visit were considered. Standard survival analysis techniques were used to assess the effect of the year of seroconversion on HIV progression. The progression to a CD4 cell count of 200 x 10(6)/l according to the calendar year of seroconversion in the Italian Seroconversion Study was assessed using different censoring strategies.
Results: A spurious effect of the year of seroconversion consistently appeared in 100 repeated simulations. When ignoring the visits occurring after the first year of follow-up in the Italian Seroconversion Study cohort, results supported the hypothesis of no effect of the year from seroconversion.
Conclusions: The choice of the censoring strategy is crucial when assessing the effect of year of seroconversion using survival analysis in cohort studies with intermittent visit structure. Different censoring strategies should be considered before firmly concluding that more virulent strains or the use of treatment are modifying the natural history of HIV disease from cohort studies of this nature.