Multiple imputation and clinico-serological models to predict human papillomavirus status in oropharyngeal carcinoma: An alternative when tissue is unavailable.

ABSTRACT

In cancer epidemiological studies, determination of human papillomavirus (HPV) in oropharyngeal squamous cell carcinoma (OPSCC) typically depends on the availability of tumor tissue testing, and/or tumor tissue access. Identifying alternative methods for estimating HPV status can improve the quality of such studies when tissue is unavailable. We developed multiple predictive models for tumor HPV status and prognosis by combining both clinico-epidemiological variables and either serological multiplex assays of HPV or multiple imputation of HPV status (HPVmi ). Sensitivity, specificity and accuracy of these methods compared to either p16 immunostaining (p16 IHC) or survival were assessed. When compared to a reference of tumor tissue p16 IHC in 783 OPSCC patients, the clinic-HPVsero model incorporating a composite of 20 HPV serological antibodies (HPVsero ) and 4 clinical factors (c-index 0.96) performed better than using HPVsero (c-index 0.92) or HPVmi (c-index 0.76) alone. However, the model that contained a single HPV16 E6 antibody combined with four clinical variables, performed extremely well (clinic-s1-16E6; c-index 0.95). When defining HPV status by HPVsero , s1-16E6, HPVmi or through p16 IHC, each of these definitions demonstrated improved overall and disease-free survival in HPV-positive OPSCC patients, when compared to HPV-negative patients (adjusted hazard ratios between 0.25 and 0.63). Our study demonstrates that when blood samples are available, a model that utilizes a single s1-16E6 antibody combined with several clinical features has excellent test performance characteristics to estimate HPV status and prognosis. When neither blood nor tumor tissue is available, multiple imputation, calibrated on local population characteristics, remains a viable, but suboptimal option.