Loss to follow-up effect in investigations of HIV-1 incidence

Fowke and colleagues (Nov 16, p 1347)1Fowke KR Nagelkerke NJD Kimani J et al.Resistance to HIV-1 infection among persistently seronegative prostitutes in Nairobi, Kenya.Lancet. 1996; 348: 1347-1351Summary Full Text Full Text PDF PubMed Scopus (389) Google Scholar ascribe an apparent decline in risk of HIV-1 seroconversion in a cohort of female sex workers in Kenya to “natural protective immunity to HIV-1”. They suggest that since the prevalence of HIV-1 among male clients increased over the study period, and risks for HIV-1 among sex workers did not decline, their risks for acquiring HIV-1 infection should have remained constant or increased. Instead, sex workers retained in follow-up seemed at declining risk for HIV-1 infection. They did not attempt to investigate natural immunity, but modelled the risk of HIV-1 with increasing cumulative exposure. Although natural immunity to HIV-1 may exist, another explanation was not sufficiently addressed: the effect of loss to followup on HIV-1 incidence in longitudinal studies. Among populations with high HIV-1 incidence, loss to follow-up has profound effects on observed incidence.2Heyward W Osmanov S Saba J et al.Preparation for Phase III HIV vaccine efficacy trials: methods for the determination of HIV incidence.AIDS. 1994; 8: 1285-1291Crossref PubMed Scopus (49) Google Scholar Bias can occur if individuals lost to follow-up have differing risks from those retained. In a cohort study of female sex workers in Thailand, we also noted striking declines in HIV-1 incidence rates over time.3Beyrer C, Khamboonruang C, Natpratan C, Celentano DD, Nelson KE. Incident HIV and STDs in direct and indirect commercial sex workers in Thailand. (Oral Presentation 038C, at the International AIDS Conference, Aug 12–17, 1994, Yokohama, Japan.)Google Scholar However, risk-factor analyses showed that women later lost to follow-up reported significantly higher HIV-1 risk behaviours than did women retained.4Beyrer C Brookmeyer R Natpratan C Ruckpaophunt S Nelson KE Khamboonruang C Measuring HIV-1 incidence in northern Thailand; prospective cohort results and estimates based on early diagnostic tests.J AIDS Hum Retro. 1996; 12: 495-499PubMed Google Scholar To investigate this difference, we used a cross-sectional method for measuring HIV-1 seroconversion. This method is based on the preantibody, p24 antigenaemic period of HIV-1 infection, the mean duration of this period, and the proportion of HIV-1-antibody negatives with p24 antigenaemia.5Brookmeyer R Quinn T Shepherd M Mehendale S Rodrigues J Bollinger B AIDS Epidemic in India: a new method for estimating current human immunodeficiency virus (HIV) incidence rates.Am J Epidemiol. 1995; 142: 709-713PubMed Google Scholar We found the apparent decline in HIV-1 incidence among a cohort of female sex workers, and another cohort of male sex workers, to be explained as an artifact of differential loss to follow-up among individuals at higher risk.4Beyrer C Brookmeyer R Natpratan C Ruckpaophunt S Nelson KE Khamboonruang C Measuring HIV-1 incidence in northern Thailand; prospective cohort results and estimates based on early diagnostic tests.J AIDS Hum Retro. 1996; 12: 495-499PubMed Google Scholar This effect was not trivial; among female commercial sex workers in the first 6 months we found an HIV-1 incidence of 23·8 per 100 person-years. 18 months later, the incidence was 11·8 per 100 personyears, which is about half. However, using the cross-sectional method, we estimated the rate to be 25·4% per year (95% CI 14·8–40·6), suggesting that the observed decline did not represent a true decline in HIV-1 acquisition rates. Brookmeyer and colleagues,5Brookmeyer R Quinn T Shepherd M Mehendale S Rodrigues J Bollinger B AIDS Epidemic in India: a new method for estimating current human immunodeficiency virus (HIV) incidence rates.Am J Epidemiol. 1995; 142: 709-713PubMed Google Scholar reported a similar finding among female sex workers in India; their conclusion was that loss to follow-up largely accounted for a sharp observed decline in HIV-1 seroconversions. These data suggest that loss to follow-up effects should be considered when investigating observed HIV-1 incidence declines before other explanations are proposed.