Competing-risks regression models in analysis of biomarkers as predictors of high-risk human papillomavirus (HPV) infection outcomes and incident CIN in the LAMS cohort

Abstract

To assess the prediction potential of a 5-biomarker panel for detecting high-risk human papillomavirus (HR-HPV) infections and/or cervical intraepithelial neoplasia (CIN) progression. Five biomarkers, lipocalin, plasminogen activator inhibitor-2, p300, interleukin-10, and stratifin, were assessed in cervical biopsies from 225 women of the Latin American Screening Study. Competing-risks regression models were constructed to assess their predictive power for (i) HR-HPV outcomes (negative, transient, or persistent infection) and (ii) CIN outcomes (no progression, incident CIN1, CIN2, or CIN3). p300, LCN2, stratifin were significantly associated with prevalent HR-HPV but lost their significance in multivariate analysis. In the multivariate model, only p300 was an independent predictor of CIN3 (odds ratio=2.63; 95% confidence interval, 1.05-6.61; P=0.039). In univariate competing-risks regression, lipocalin predicted permanent HR-HPV-negative status, but in the multivariate model, IL-10 emerged as a independent predictor of HPV-negative status (subhazard ratio=4.04; 95% confidence interval, 1.81-9.01; P=0.001). The clinical value of the panel in predicting longitudinal outcomes of HR-HPV infection and/or incident CIN is limited.