Understanding your medical documents - Is ChatGPT the answer?
The rise of patient portals and the ubiquity of 'Dr. Google' have brought many benefits but also some notable challenges. Patients now have easier access to their medical records, which empowers them to take ownership of their health journey. However, the complexity of medical terminology often poses a barrier to those without a medical background. What happens then is that patients try to figure it out themselves using the resources available: Dr. Google and some friend of a friend who has heard something about this topic. Wouldn't it be significantly better if there was a way to translate these complex medical terms into language that's more accessible to the general public? We're fortunate to live in an era of remarkable technological advancements, and as hinted in the title, there’s an ingenious solution to this challenge.
Large Language Models (LLMs) are unsurprisingly excelling in language-related tasks. So, it makes sense to use them for turning complex texts, like radiology reports, into something understandable by a crucial, yet often overlooked, group in healthcare: the patient. Rushabh Doshi and team from Yale have recently published some very promising results in the field of Radiology.
👩⚕️🖥️ Study design: Simplify 750 reports for better understanding
- In this study 750 diverse radiology reports from the MIMIC-IV database were analyzed, including CT, MRI, US, Mammography, and Radiographs (150 of each).
- The researchers utilized 4 different LLMs: ChatGPT-3.5 and -4, Bard (now known as Gemini), and Bing.
- They employed and compared 3 distinct prompts for simplification of the impression:
- “Simplify this radiology report.”
- “I am a patient. Simplify this radiology report.”
- “Simplify this radiology report at the 7th-grade level.”
- Simplification was evaluated using the average of four established readability indexes.
🎯 Results: Impressive Performance by LLMs
- All LLMs effectively simplified the radiology reports.
- Context-rich prompts, like “simplify at the 7th-grade level,” yielded better results, with ChatGPT-3.5 and -4 excelling in these scenarios.
🌍Real World Impact:
- LLMs can provide patient-friendly reports using very little resources
- This could definitely boost patient empowerment with all its advantages
- Integrating these tools into routine workflows requires fine-tuning each LLM's output to ensure optimal simplification without compromising medical accuracy.
🤔This made me curious, so I tried it myself
- “Simplify this radiology report.”
- “I am a patient. Simplify this radiology report.”
- “Simplify this radiology report at the 7th-grade level.”
✌️Personal takeaway:
LLMs have profoundly impacted the world, showcasing the vast potential and strengths of AI to a broad audience. This has helped to clear up many misconceptions about what this subgroup of AI can do. They might not predict lottery numbers accurately, but their value in crafting customized, patient-friendly medical documents is undeniable. The potential of these technologies to enhance patient empowerment is significant and should not be underestimated. This applicability extends to many areas in medicine and beyond. Imagine a future where AI chatbots primarily provide first-line support. This scenario offers a win-win: individuals in need receive instant, round-the-clock responses, and knowledge workers, including many doctors, can reap multiple benefits:
- Reduced workloads
- Fewer interruptions, leading to more deep work (or call it focused work, (or call it focused work, but I am a big fan of Cal Newport's Deep Work concept)
- More efficient consultations with well-informed patients
- A stronger foundation for shared decision-making
Of course, there are challenges, like the well-discussed hallucinations of LLMs and the risk of misinformation. As of now, these issues cannot be completely eliminated from my perspective.
We still have a long way to go to ensure LLMs effectively support healthcare providers and patients. However, they are a powerful tool for patient empowerment and have tremendous potential to improve resource allocation in our healthcare system.