Does AI Improve Doctors’ Diagnoses? Study Finds Out

https://newsroom.uvahealth.com/2024/11/13/does-ai-improve-doctors-diagnoses-study-finds-out/ ^ | 11/13/2024 | No Attribution

[RoosterRedux]

With hospitals already deploying artificial intelligence (AI) to improve patient care, a new study has found that using Chat GPT Plus does not significantly improve the accuracy of doctors’ diagnoses when compared with the use of usual resources.

The study, from UVA Health’s Andrew S. Parsons, MD, MPH and colleagues, enlisted 50 physicians in family medicine, internal medicine and emergency medicine to put Chat GPT Plus to the test. Half were randomly assigned to use Chat GPT Plus to diagnose complex cases, while the other half relied on conventional methods such as medical reference sites (for example, UpToDate©) and Google. The researchers then compared the resulting diagnoses, finding that the accuracy across the two groups was similar.

That said, Chat GPT alone outperformed both groups, suggesting that it still holds promise for improving patient care. Physicians, however, will need more training and experience with the emerging technology to capitalize on its potential, the researchers conclude.

...

The median diagnostic accuracy for the docs using Chat GPT Plus was 76.3%, while the results for the physicians using conventional approaches was 73.7%. The Chat GPT group members reached their diagnoses slightly more quickly overall – 519 seconds compared with 565 seconds.

The researchers were surprised at how well Chat GPT Plus alone performed, with a median diagnostic accuracy of more than 92%. They say this may reflect the prompts used in the study, suggesting that physicians likely will benefit from training on how to use prompts effectively. Alternately, they say, healthcare organizations could purchase predefined prompts to implement in clinical workflow and documentation.

[RoosterRedux]

You would think the headline would have been, "ChatGPT alone beats doctors significantly in diagnosing illnesses--study."

I guess the authors of the article didn't want to insult doctors who read it.;-)

[Jonty30]

I expect this to be available, I. Conjunction with a smart watch.

[9YearLurker]

So both performed poorly.

[RoosterRedux]

Both sets of doctors performed poorly compared to ChatGPT Plus.

[9YearLurker]

Put in specific-enough prompts and you almost can’t help but getting back what you want.

[Envisioning]

I was recently referred to another doctor by my primary care doctor . An AI ‘assistant’ texting me couldn’t even get me the correct phone number for the doctor I was referred to, had to look it up myself. AI? No thanks.

[mmichaels1970]

It writes code better than me. Takes it a while to get it where I want it to be. And sometimes when it’s wrong it gets caught in a “try doing this, try doing that” loop. But it’s taught be quite a bit about coding and I’ve been doing this over 30 years. Over 40 if you bring my childhood and my apple iie and BASIC into play.

[xp38]

I follow the precepts of Dr. Savage

[RoosterRedux]

I use ChatGPT every day and am always very careful to avoid leading questions (prompts) to make sure that I get unbiased responses. The doctors in this study may not have been adequately trained on creating good prompts.

I put the following question to ChatGPT about the article:

Me:

Why would the doctors using AI perform worse than AI alone?

ChatGPT:

Why Doctors Using AI Performed Worse Than AI Alone

The fact that doctors using AI performed slightly worse than the AI alone could stem from several factors related to how humans interact with technology, particularly in high-stakes and complex scenarios like medical diagnosis. Here are potential explanations:

1. Confirmation Bias

• Doctors may approach AI recommendations with preconceived notions about a diagnosis, leading them to interpret AI outputs in a way that confirms their initial ideas rather than reassessing the case objectively.
• This bias could cause them to misinterpret or disregard accurate suggestions from the AI.

2. Misuse or Misunderstanding of AI

• Physicians might lack sufficient training on how to effectively use AI tools, such as crafting appropriate prompts or interpreting AI-generated information.
• Without understanding the AI’s strengths and limitations, doctors might either over-rely on or underutilize its suggestions, leading to suboptimal outcomes.

3. Distrust of AI

• Physicians may not fully trust AI recommendations and could override correct suggestions with their own (sometimes incorrect) clinical judgments.
• Skepticism toward AI, especially when its outputs conflict with their experience or intuition, might cause errors.

4. Over-reliance on AI

• Conversely, doctors might defer too much to the AI and fail to critically assess its outputs, missing nuances that a human doctor might otherwise catch.
• AI is not infallible, and over-reliance can lead to a loss of critical thinking in clinical decision-making.

5. Cognitive Load and Decision Fatigue

• Introducing AI into the diagnostic process might increase cognitive load. Instead of simplifying the task, integrating AI suggestions with a doctor’s own reasoning could create additional mental steps.
• This can lead to mistakes or slower decision-making, especially in time-sensitive cases.

6. Mismatch Between AI and Human Diagnostic Processes

• AI and human physicians may approach problem-solving differently. For example, AI relies on patterns and data, while physicians often consider contextual factors and experience.
• These differences might lead to conflicts in decision-making when physicians misinterpret AI outputs or struggle to align them with the broader clinical picture.

7. AI Output Complexity

• The way AI presents its findings could impact usability. If the AI’s suggestions are too detailed, vague, or poorly prioritized, physicians might find it harder to integrate them effectively into their workflow.
• Physicians may also spend time validating or second-guessing AI outputs, reducing diagnostic accuracy.

8. Study Design Factors

• Artificial Setting: The controlled nature of clinical vignettes used in the study might not reflect the real-life dynamics of physician-AI collaboration.
• AI-Optimized Prompts: The AI performed better in isolation likely because the study used optimized prompts, whereas physicians might not yet know how to craft prompts that yield the best results.

Conclusion

The results highlight a need for training and better integration of AI into clinical workflows. Physicians must learn to effectively interact with AI, using it as a complementary tool while maintaining their critical reasoning skills. Additionally, healthcare systems need to design interfaces and workflows that facilitate seamless and accurate human-AI collaboration. Until these gaps are addressed, human interaction with AI might inadvertently reduce its effectiveness.

[RoosterRedux]

The AI used by companies and professionals for customer, client, and patient services seems, from my experience, to be very subpar.

[9YearLurker]

Anyone really relying on genAI results at this point really needs to ask for sources — and then check them out. Does it save time over a regular Google search? Perhaps. Right now to me it is the difference between the Android and iPhone interfaces. I lean more toward the DIY approach, but the plan is to take those options away from us. Wasn’t it Eric S. who gushed about the perfect search providing just one response?

[RoosterRedux]

That’s been my experience also. For coding, I have only used it for VBA (for Excel) and a little bit of Python, but it does in seconds what it takes me almost a week to do. But there is sometimes that “try this, try that” loop you mentioned.

[RoosterRedux]

I agree. I check and double-check source material to verify accuracy.

I think of it like a very smart but error-prone intern or associate-level helper.

[trebb]

I can imagine some of the AI answers:

Plain English - “The idiot put a lightbulb up his anus and it broke, causing serious bleeding and a complicated procedure to remove the shards.”

AI - “I cannot engage in this conversation”

[Sarcazmo]

You remember what the security experts said about STUXnet when they got a piece of that code to look at?

This was what, 15 years ago?

[antidemoncrat]

Ney, if that’s the case I might as well just get on line and diagnose my symptoms first as I’m sure that won’t cost my insurance $200 for a doctor visit.

[telescope115]

So what would this mean for the medical profession? Are med students going to com out of medical school without the extensive knowledge that is required of them? It seems to me that AI might become a crutch for doctors who lack critical thinking skills and experience.

One thing I’ve learned is that patient-to-doctor communication is paramount. I don’t know if I’m ready to trust my healthcare to a computer program.

[RoosterRedux]

I always diagnose my symptoms first with AI.

AI isn't always correct, but it almost always gives me ideas for expanding my search. I usually check the sources carefully before I rely on any AI response (and even then I consider that the sources may be wrong).

[RoosterRedux]

BTW, when I have serious questions for AI, I usually ask ChatGPT and Perplexity and then compare the answers.

ChatGPT is conversational and might give me answers it thinks I want. For this reason, I am always careful not to ask leading questions or questions that show a hint of bias.

Perplexity on the other hand can be blunt and, at times, almost argumentative.