AI Aids Housing Instability Detection, Human Oversight Essential

Researchers at the Institute for Systems Biology (ISB), in collaboration with Providence, have examined the capabilities of artificial intelligence (AI) in identifying social determinants of health, focusing on housing instability through clinical notes. The study highlights both the potential and the limitations of large language models (LLMs) in processing electronic health records.

The research analyzed over 25,000 clinical notes from 795 pregnant women using two versions of generative pre-trained transformers (GPT-4 and GPT-3.5), alongside a named entity recognition model, regular expressions, and human review. The goal was to assess AI’s ability to detect housing challenges, differentiate between current and past housing instability, and extract direct evidence from clinical documentation.

GPT-4 emerged as the most effective tool, surpassing human reviewers in identifying instances of housing instability. However, human reviewers outperformed AI in understanding when patients did not experience housing challenges, showcasing greater precision and accuracy in extracting evidence from clinical notes.

“These results show that LLMs present a scalable, cost-effective solution for an initial search for patients who may benefit from outreach,” said Jennifer Hadlock, Associate Professor, ISB.

Unlike earlier applications of AI, GPT-4 did not generate false evidence when properly instructed to extract verbatim text from notes. Still, researchers noted cases where AI misinterpreted clinical information, potentially leading to misleading conclusions—a significant concern given the overlap between housing instability and sensitive issues like domestic abuse.

“When a healthcare professional decides whether and how to reach out to offer help, they take great care to consider patient safety. Our results illustrate that it would still be essential to have a human read the actual text in the chart, not just the LLM summary,” added Hadlock.

The research underscores the value of AI as a tool to streamline initial reviews of health records, offering scalability and cost efficiency. However, human oversight remains indispensable for ensuring accuracy and safeguarding patient safety.

“Training healthcare professionals in the use and understanding of AI is fundamental, and healthtech companies offer training and continuous education programs that help integrate AI effectively into daily practice,” wrote Jesús Hernández, President of the Healthtech Mexico Association for MBN.