# Development and Validation of a Score-Based Model for Estimating Esophageal Squamous Cell Carcinoma and Precancerous Lesions Risk in an Opportunistic Screening Population

## TL;DR

A new score-based model helps identify people at high risk for esophageal cancer and precancerous lesions during opportunistic screenings, reducing the need for endoscopies.

## Contribution

The first score-based model for estimating ESCC and precancerous lesion risk in opportunistic screening populations.

## Key findings

- The model detected 70.0% of high-grade lesions, 81.3% of early ESCC, and 81.1% of advanced ESCC with 76.4% specificity.
- Using the model could reduce the number of individuals needing endoscopy by 75.6%.

## Abstract

Currently, there are no score-based models for estimating esophageal squamous cell carcinoma (ESCC) and precancerous lesions risk in an opportunistic population. The present study developed and validated a score-based risk prediction model for opportunistic screening for ESCC for the first time, comprising 8 variables on a 21-point scale. The model could detect 70.0%, 81.3%, and 81.1% of high-grade intraepithelial neoplasia, early ESCC, and advanced ESCC, respectively, with a specificity of 76.4%. Additionally, the score-based model could result in 75.6% fewer individuals subjected to endoscopy. The utilization of the score-based model enables risk stratification and individual self-assessment of ESCC during opportunistic screening.

Background: Opportunistic screening is one major screening approach for esophageal squamous cell carcinoma (ESCC). We aimed to develop a score-based risk stratification model to assess the risk of ESCC and precancerous lesions in opportunistic screening and to validate it in an external population. Methods: The study was a secondary analysis of a published esophageal cancer screening trial. The trial was conducted in 39 secondary or tertiary hospitals in China, with 14,597 individuals including 71 high-grade intraepithelial neoplasia (HGIN) and 182 ESCC, enrolled for opportunistic screening. Additionally, questionnaires and endoscopy were performed. The primary outcome was histology-confirmed high-grade esophageal lesions, including HGIN and ESCC. The predictors were selected using univariable and multivariable logistic regression. Model performance was primarily measured with the area under the receiver operating characteristic curve (AUROC). Results: The score-based prediction model contained 8 variables on a 21-point scale. The model demonstrated an AUROC of 0.833 (95% CI, 0.803–0.862) and 0.828 (95% CI, 0.793–0.864) for detecting high-grade lesions in the training and validation cohorts, respectively. Using the cut-off score determined in the training cohort (≥9), the sensitivity reached 70.0% (95% CI, 50.6–85.3%), 81.3% (95% CI, 63.6–92.8%), and 81.1% (95% CI, 64.9–92.0%) in the validation cohort for detecting HGIN, early ESCC, and advanced ESCC, respectively, at a specificity of 76.4% (95%CI, 75.4–77.4%). The score-based model exhibited satisfactory calibration in the calibration plots. The model could result in 75.6% fewer individuals subjected to endoscopy. Conclusions: This score-based model demonstrated superior discrimination for esophageal high-grade lesions. It has the potential to inform referral decisions in an opportunistic screening setting.