Effect of Different Human Papillomavirus Serological and DNA Criteria on Vaccine Efficacy Estimates

Gardasil (Merck & Co., Inc., West Point, Pennsylvania) and Cervarix (GlaxoSmithKline Vaccines, Rixensart, Belgium) are 2 highly efficacious prophylactic human papillomavirus (HPV) vaccines. Gardasil is a quadrivalent vaccine containing the recombinant L1 major capsid proteins of oncogenic HPV16 and HPV18 and low-risk HPV6 and HPV11, and Cervarix is a bivalent vaccine containing the recombinant L1 major capsid proteins of HPV16 and HPV18 (1). HPV vaccination is recommended for young adolescents before the initiation of sexual activity and exposure to the virus (2). However, the phase III clinical trials that led to vaccine licensure were conducted among young adult women, many of whom were sexually active and potentially HPV exposed. A combination of HPV DNA and serological testing was used to define a subcohort of women likely to be HPV naive before vaccination to estimate vaccine efficacy (VE) in an HPV-unexposed target population. Published estimates of VE against cervical intraepithelial neoplasia grade 2 or worse (CIN2+), irrespective of HPV DNA typing, were 42.7% (95% confidence interval (CI): 23.7%, 57.3%) for Gardasil and 64.9% (95% CI: 52.7%, 74.2%) for Cervarix in the naive-population subanalyses (3, 4). If the observed difference in estimated VE among HPV-naive recipients of Gardasil and Cervarix reflects vaccine performance, it may have important public health implications. Alternatively, these differences in VE estimates might reflect differences in the approaches used to simulate an HPV-unexposed group in the 2 trials—Females United to Unilaterally Reduce Endo/Ectovervical Disease (FUTURE I/II) and the Papilloma Trial Against Cancer in Young Adults (PATRICIA). In both trials, the criteria used to define HPV-unexposed groups included negative HPV DNA tests of cervical cells and negative serological results for HPV-related antibodies. Specifically, in both trials, subjects who tested HPV DNA positive for any of 12 oncogenic HPV types (type 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, or 59) were excluded from the HPV-naive group. PATRICIA additionally excluded subjects on the basis of HPV DNA positivity for HPV types 66 and 68/73, whereas FUTURE I/II excluded subjects on the basis of HPV DNA positivity for HPV types 6 and 11 (3, 4). For the serological analyses, PATRICIA used a viruslike particle (VLP)–based direct enzyme-linked immunosorbent assay (ELISA), which detects the presence of both neutralizing and nonneutralizing HPV antibodies (4). In contrast, a VLP-based competitive Luminex immunoassay (cLIA) was used in FUTURE I/II, which is designed to detect neutralizing HPV antibodies specifically directed against the V5 epitope of HPV16 and the J4 epitope of HPV18 (3). Previous analyses have demonstrated that ELISA is more sensitive than cLIA (5). Thus, the use of cLIA may have resulted in the inclusion of more HPV-exposed women within the naive subanalysis in FUTURE I/II compared with the use of the ELISA in PATRICIA. The objective of the present analysis was to investigate the extent to which methodological differences in estimating VE within HPV-naive subcohorts contributes to differences in VE estimates. To address our objective, we applied the FUTURE I/II and PATRICIA criteria within the Costa Rica HPV16/18 Vaccine Trial (CVT) to approximate the 2 naive subcohorts and to estimate VE for the following 3 endpoints: 1) CIN2+ regardless of type of HPV infection, 2) HPV type–specific CIN2+, and 3) HPV type–specific 12-month persistence.