Abstract LB-165: Risk factor prediction models for triage of HR-HPV positive women: The Estampa study

Abstract

Abstract Background: Cervical cancer remains an important public health problem worldwide, being the 4th most common cancer in women globally. HPV testing has been proposed for primary cervical cancer screening, given its higher sensitivity compared to cytology to detect cervical cancer precursors. However, HPV testing may lead to costly and unnecessary increases in follow-ups and/or treatments for transiently HPV positive women. Thus, triage methods are needed for HR-HPV positive women to determine optimal management. This study aims to construct and validate a risk prediction model for cervical pre-cancer based on demographics, clinical, and lifestyle characteristics among HR-HPV positive women. Methods: This analysis from the Multicentric study of cervical cancer screening and triage with HPV testing (ESTAMPA) included HR-HPV positive women aged 30-65 recruited on nine countries in Latin America from 2012 to 2019 (n=4565). The outcomes were: (1) cervical intraepithelial neoplasia grade 2 or more severe disease (CIN2+), and (2) CIN grade 3 or more severe disease (CIN3+) before final pathology review. Covariates included demographic characteristics and cervical cancer risk factors. Univariate logistic regression models were used to assess potential predictors of CIN2+ and CIN3+, controlling for the effect of the country of residence, using a multilevel approach. A multivariate logistic regression model was used to predict outcomes. After computing the probability of CIN2+ and CIN3+ among participants, we categorized these probabilities in deciles, to set the scale boundaries. Receiver-operating curves (ROC) and the Area under the Curves (AUC) were constructed. Sensitivity, specificity and predictive values were used to summarize the performance of the scoring system. Results: The prevalence of CIN2+ and CIN3+ in the study population were 14.7% and 9.7%, respectively. Age at recruitment, history of last Pap test, number of pregnancies, use of oral contraceptives, cigarette use and age at first sexual intercourse were significantly associated (p<0.05) with having CIN2+ and CIN3+. The AUC for the risk factor prediction model was 0.66 for CIN2+ and 0.69 for CIN3+. The optimal cutpoint selected by the Youden's statistic was the 5th decile for CIN2+ and 6th decile for CIN3+. This optimal point was for women with probabilities of CIN2+ greater than 13% and greater than 10% for CIN3+. Thus, in these deciles the number of referrals to colposcopy decreased from 100% to 60% for CIN2+ (100% to 49% for CIN3+), and the sensitivity and specificity obtained in this decile was 80% and 43% for CIN2+, respectively (73% and 54% for CIN3+). Conclusions: Our risk prediction model demonstrated a close to acceptable discrimination. This risk prediction model has the potential to reduce the number of HR-HPV positive women who are referred to colposcopy or treatment unnecessarily, reducing the burden for the patient and the healthcare system. Further analyses after pathology review are underway. Citation Format: Marievelisse Soto-Salgado, Maribel Almonte, Ana Patricia Ortiz, Erick L. Suarez, Luis R. Pericchi, Armando Baena, Rolando Herrero. Risk factor prediction models for triage of HR-HPV positive women: The Estampa study [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr LB-165.