Scholarlink

Background <p>Eating disorders (EDs) are a growing global health concern and are increasingly associated with negative body-image perceptions linked to social-media exposure, particularly among adolescents. Early identification and screening of individuals at risk can influence effective patient management and care. AI models, such as machine learning, natural language processing, or wearables, have shown early promise as potential screening tools, offering early prediction, improved accuracy, reduced reporting bias, and better user compliance than conventional methods.</p> Methods <p>Literature from PubMed, Scopus, and Web of Science (2015–July 2025) produced 606 records after applying Boolean operators “OR” for synonyms and “AND” to link core concepts. 23 studies were included using the PICOT framework. Data extraction examined predictors, algorithms, metrics, and limitations.</p> Results <p>The findings on predicting EDs highlight six major domains: traditional machine-learning (ML) models, sensor-based systems, mobile and wearable technologies, natural language processing (NLP), questionnaire-based models, and integrated risk prediction frameworks. Among all the domains, sensor-based and wearable systems, along with NLP from social-media data, demonstrated the highest accuracy. While traditional questionnaire-based models integrating validated tools with AI/ML have shown moderate performance, they remain limited by small sample sizes, standardised settings, and a lack of validation in low-resource settings, such as India.</p> Conclusions <p>AI-based predictive models demonstrate promising early potential for scalable, likely less biased, and non-stigmatising detection of EDs. Despite the promising performance demonstrated across multiple AI-based approaches, certain challenges remain, including generalizability, ethical considerations, and real-world implications. Although most AI-based eating-disorder research has been conducted in high-income countries, evidence from developing and culturally diverse settings remains limited, where cultural variations and social stigma co-exist. In such contexts, the use of multimodal systems can support early prediction of eating disorders. Future research should be culturally diverse, integrate multimodal approaches with social and peer factors, and use validated frameworks for effective practice.</p> <p><i>Level of Evidence</i>: Level V. Opinions of respected authorities, based on descriptive studies, narrative reviews, clinical experience, or reports of expert committees.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI and machine learning for early identification of eating disorders: a narrative review and Indian contextual insights

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月25日

查看全文
  • Yutika Shirgaonkar,
  • Devaki Gokhale

摘要

Background

Eating disorders (EDs) are a growing global health concern and are increasingly associated with negative body-image perceptions linked to social-media exposure, particularly among adolescents. Early identification and screening of individuals at risk can influence effective patient management and care. AI models, such as machine learning, natural language processing, or wearables, have shown early promise as potential screening tools, offering early prediction, improved accuracy, reduced reporting bias, and better user compliance than conventional methods.

Methods

Literature from PubMed, Scopus, and Web of Science (2015–July 2025) produced 606 records after applying Boolean operators “OR” for synonyms and “AND” to link core concepts. 23 studies were included using the PICOT framework. Data extraction examined predictors, algorithms, metrics, and limitations.

Results

The findings on predicting EDs highlight six major domains: traditional machine-learning (ML) models, sensor-based systems, mobile and wearable technologies, natural language processing (NLP), questionnaire-based models, and integrated risk prediction frameworks. Among all the domains, sensor-based and wearable systems, along with NLP from social-media data, demonstrated the highest accuracy. While traditional questionnaire-based models integrating validated tools with AI/ML have shown moderate performance, they remain limited by small sample sizes, standardised settings, and a lack of validation in low-resource settings, such as India.

Conclusions

AI-based predictive models demonstrate promising early potential for scalable, likely less biased, and non-stigmatising detection of EDs. Despite the promising performance demonstrated across multiple AI-based approaches, certain challenges remain, including generalizability, ethical considerations, and real-world implications. Although most AI-based eating-disorder research has been conducted in high-income countries, evidence from developing and culturally diverse settings remains limited, where cultural variations and social stigma co-exist. In such contexts, the use of multimodal systems can support early prediction of eating disorders. Future research should be culturally diverse, integrate multimodal approaches with social and peer factors, and use validated frameworks for effective practice.

Level of Evidence: Level V. Opinions of respected authorities, based on descriptive studies, narrative reviews, clinical experience, or reports of expert committees.