What question did this study set out to answer?

This research identifies predictors of STIs risk in British men and women and explores gender differences in behaviors contributing to STIs risk.

February 28, 2026Open Access

Findings from a transformer-based prediction model: social norms and confidentiality are associated with STIs risk in British men and women

Key Points

This research identifies predictors of STIs risk in British men and women and explores gender differences in behaviors contributing to STIs risk.
Analyzed data from the UK National Survey of Sexual Attitudes and Lifestyles with 15,162 participants.
Selected 119 features using the socio-ecological model.
Applied deep learning and traditional machine learning algorithms to predict STIs risk.
Evaluated model performance using metrics like AUC, accuracy, precision, and recall.
FT-transformer model achieved an AUC of 0.843 for males and 0.879 for females.
Accuracy was 87.0% for males and 87.5% for females.
Perceived social norms were the most influential factor for males, while confidentiality was most influential for females.

Abstract

Women were disproportionately affected by sexually transmitted infections (STIs), both physically and psychologically. The underlying reasons are likely multifaceted, ranging from individual behavior to relationship power, gender norms, and economic inequities. This study aims to identify predictors of STIs risk in women and men and to explore gender differences in the behavioral patterns that may contribute to STIs risk. We analyzed the data from the third United Kingdom National Survey of Sexual Attitudes and Lifestyles, including 15,162 participants aged 16–74 from 2010 to 2012. According to the five levels of individual, interpersonal, community, institutional, and structural in the socio-ecological model (SEM), we selected 119 features from the dataset. Then we applied three deep learning algorithms and two traditional machine learning algorithms to address the influential factors of STIs. To evaluate the performance, we computed the metrics such as the area under the receiver operating characteristic curve (AUC), accuracy, precision, and recall. The results interpretation of the best model is based on feature importance analysis within the context of SEM. The tabular transformer model, FT-transformer, demonstrated excellent performance in predicting STIs risk in British males (AUC = 0.843, Accuracy = 87.0%) and females (AUC = 0.879, Accuracy = 87.5%) among five models. The top 10 influential factors to predict STIs risk for British males and females are different. The most influential factor for males is perceived social norms, and for females is guaranteed confidentiality. The high accuracy of the transformer model in predicting STIs risk highlights the need to use multi-level factors to identify gender-specific risk factors, which could be used in the future to formulate gender-tailored interventions in STIs prevention, diagnosis, and treatment.

Demander à l'IA

Bookmark

View Full Paper