Applying Deep Learning for Predicting Retention in PrEP Care and Effective PrEP Use among Key Populations at Risk for HIV in Thailand

项目总结/摘要 尽管在艾滋病毒预防和治疗方面取得了很大进展， 特别是对关键人群，包括男男性行为者和变性妇女 （TGW）。暴露前预防（PrEP）已被证明有效地减少艾滋病毒感染， 在作为综合预防战略的一部分实施时，然而，PrEP的有效性 随着次优的保持和粘附而降低。虽然为评估遵守情况作出了许多努力， 对于PrEP及其与艾滋病毒预防有效性的关联，需要更多的研究来深化我们的研究。 了解个人层面的促进者和障碍，以保持在照顾和坚持PrEP。机器 学习有希望有效地解决这些问题，因为它有能力模拟复杂的非线性关系 在许多相互作用的因素，而不依赖于建模假设，以及最近的进展，在深 学习已经为各种临床预测应用带来了令人兴奋的结果。尽管机器 学习已经被应用于识别潜在的PrEP候选者，在探索机器学习方面做得很少， 特别是先进的深度学习技术，以评估保留PrEP护理的预测因素， 有效使用PrEP。 为了缩小知识差距，拟议的研究旨在探索先进的机器学习技术， 确定保留PrEP护理的保护性和风险因素，并在关键人群中有效使用PrEP， 泰国我们将进行描述性统计分析，以表征MSM和 TGW（目标1）;开发深度学习模型，以预测PrEP护理和有效PrEP使用中的随访损失 (Aim 2）;设计一个可解释的风险评分系统，以确定中断服务的高风险客户， 非有效的PrEP使用，具有可解释的推理逻辑和相关的人口统计学，行为，社会和 临床因素（目标3）。 这项研究是响应NIMH在艾滋病毒预防和战略目标3.2方面的优先研究， 调整现有干预措施以优化成果的战略。这项研究的结果和 基于预测模型的评分系统将为定制干预提供信息，以优化PrEP参与， 促进差异化的PrEP服务提供，为使用PrEP精确预防艾滋病毒奠定坚实基础 作为一个有效的战略。

Project Summary/Abstract HIV remains a major cause of morbidity and mortality despite great progress in HIV prevention and treatment, especially for key populations (KPs), including men who have sex with men (MSM) and transgender women (TGW). Pre-exposure prophylaxis (PrEP) has been shown effective in reducing HIV acquisition among different populations when implemented as part of a combination prevention strategy. However, effectiveness of PrEP decreases with suboptimal retention and adherence. While many efforts have been made to assess adherence to PrEP and its associations with HIV prevention effectiveness, more research is needed to deepen our understanding of individual-level facilitators and barriers to retention in care and adherence to PrEP. Machine learning holds promise to address those effectively due to its ability to model complex non-linear relationships among many interacting factors without relying on modeling assumptions, and recent advances in deep learning have resulted in exciting results for a variety of clinical prediction applications. Although machine learning has been applied to identify potential PrEP candidates, little is done in exploring machine learning, especially advanced deep learning techniques, to assess predictive factors for retention in PrEP care and effective PrEP use. To close gaps in knowledge, the proposed study aims to explore advanced machine learning techniques to identify protective and risk factors for retention in PrEP care and effective PrEP use among key populations in Thailand. We will perform descriptive statistical analysis to characterize PrEP use patterns among MSM and TGW (Aim 1); develop deep learning models to predict loss to follow up in PrEP care and effective PrEP use (Aim 2); and design an explainable risk scoring system for identifying clients at high risk of discontinuation and non-effective PrEP use, with interpretable reasoning logic and associated demographic, behavioral, social, and clinical factors (Aim 3). This study is responsive to NIMH’s priority research in HIV prevention and strategic goal 3.2 to develop strategies for tailoring existing interventions to optimize outcomes. The findings from this study and the prediction-model based scoring system will inform tailored interventions to optimize PrEP engagement and facilitate differentiated PrEP service delivery, paving a solid foundation for precise HIV prevention using PrEP as an effective strategy.