Performance of Human Papillomavirus Attribution Algorithms to Predict Causative Genotypes in Anal High-Grade Lesions

Attribution algorithms have been successfully used to assign human papillomavirus (HPV) genotype risk in the cervix for the past 20 years. We show that these algorithms have limited use for defining HPV genotype risk in anal disease for populations with a high rate of multiple HPV infections. Background Gay and bisexual men (GBM) are at increased risk of human papillomavirus (HPV)-associated anal high-grade squamous intraepithelial lesions (HSILs). Understanding the fractions of HSILs attributable to HPV genotypes is important to inform potential impacts of screening and vaccination strategies. However, multiple infections are common, making attribution of causative types difficult. Algorithms developed for predicting HSIL-causative genotype fractions have never been compared with a reference standard in GBM. Method Samples were from the Study of the Prevention of Anal Cancer. Baseline HPV genotypes detected in anal swab samples (160 participants) were compared with HPV genotypes in anal HSILs (222 lesions) determined by laser capture microdissection (LCM). Five algorithms were compared: proportional, hierarchical, maximum, minimum, and maximum likelihood estimation. Results All algorithms predicted HPV-16 as the most common HSIL-causative genotype, and proportions differed from LCM detection (37.8%) by algorithm (with differences of -6.1%, +20.9%, -20.4%, +2.9%, and +2.2% respectively). Fractions predicted using the proportional method showed a strong positive correlation with LCM, overall (R = 0.73 and P = .002), and by human immunodeficiency virus (HIV) status (HIV positive, R = 0.74 and P = .001; HIV-negative, R = 0.68 and P = .005). Conclusions Algorithms produced a range of inaccurate estimates of HSIL attribution, with the proportional algorithm performing best. The high occurrence of multiple HPV infections means that these algorithms may be of limited use in GBM.