A.I. Chatbots Defeated Doctors at Diagnosing Illness

A small study found ChatGPT outdid human physicians when assessing medical case histories, even when those doctors were using a chatbot.

 View from a hallway into an exam room of a health care center.

In an experiment, doctors who were given ChatGPT to diagnose illness did only slightly better than doctors who did not. But the chatbot alone outperformed all the doctors. Credit...Michelle Gustafson for The New York Times

By Gina Kolata

Nov. 17, 2024

View from a hallway into an exam room of a health care center.

Adam Rodman, an expert in internal medicine at Beth Israel Deaconess Medical Center in Boston, confidently expected that chatbots built to use artificial intelligence would help doctors diagnose illnesses.

He was wrong.

Instead, in a study Dr. Rodman helped design, doctors who were given ChatGPT-4 along with conventional resources did only slightly better than doctors who did not have access to the bot. And, to the researchers’ surprise, ChatGPT alone outperformed the doctors.

“I was shocked,” Dr. Rodman said.

The chatbot, from the company OpenAI, scored an average of 90 percent when diagnosing a medical condition from a case report and explaining its reasoning. Doctors randomly assigned to use the chatbot got an average score of 76 percent. Those randomly assigned not to use it had an average score of 74 percent.

The study showed more than just the chatbot’s superior performance.

It unveiled doctors’ sometimes unwavering belief in a diagnosis they made, even when a chatbot potentially suggests a better one.

The study illustrated that while doctors are being exposed to the tools of artificial intelligence for their work, few know how to exploit the abilities of chatbots. As a result, they failed to take advantage of A.I. systems’ ability to solve complex diagnostic problems and offer explanations for their diagnoses.

A.I. systems should be “doctor extenders,” Dr. Rodman said, offering valuable second opinions on diagnoses.

But it looks as if there is a way to go before that potential is realized.

Case History, Case Future

The experiment involved 50 doctors, a mix of residents, and attending physicians recruited through a few large American hospital systems, and was published last month in the journal JAMA Network Open.

The test subjects were given six case histories and were graded on their ability to suggest diagnoses and explain why they favored or ruled them out. Their grades also included getting the final diagnosis right.

The graders were medical experts who saw only the participants’ answers, without knowing whether they were from a doctor with ChatGPT, a doctor without it or from ChatGPT by itself.

The case histories used in the study were based on real patients and are part of a set of 105 cases that has been used by researchers since the 1990s. The cases intentionally have never been published so that medical students and others could be tested on them without any foreknowledge. That also meant that ChatGPT could not have been trained on them.

But, to illustrate what the study involved, the investigators published one of the six cases the doctors were tested on, along with answers to the test questions on that case from a doctor who scored high and from one whose score was low with the blood thinner heparin for 48 hours after the procedure.

The man complained that he felt feverish and tired. His cardiologist had done lab studies that indicated a new onset of anemia and a buildup of nitrogen and other kidney waste products in his blood. The man had had bypass surgery for heart disease a decade earlier.

The case vignette continued to include details of the man’s physical exam and then provided his lab test results.

The correct diagnosis was cholesterol embolism — a condition in which shards of cholesterol break off from plaque in arteries and block blood vessels.

Participants were asked for three possible diagnoses, with supporting evidence for each. They also were asked to provide, for each possible diagnosis, findings that do not support it or that were expected but not present.

The participants also were asked to provide a final diagnosis. Then they were to name up to three additional steps they would take in their diagnostic process.

Like the diagnosis for the published case, the diagnoses for the other five cases in the study were not easy to figure out. But neither were they so rare as to be almost unheard of. Yet the doctors on average did worse than the chatbot.

What, the researchers asked, was going on?

The answer seems to hinge on questions of how doctors settle on a diagnosis, and how they use a tool like artificial intelligence.

The key ingredient is how to use the prompts to query ChatGPT.

Unpacking FDA Guidance on Digital Health Technologies in Clinical Trials

Digital Health Technologies (DHTs) use in clinical research is no longer a futuristic concept—it’s quickly becoming the norm. Recognizing this shift, the FDA issued comprehensive guidance in 2023 on Digital Health Technologies for Remote Data Acquisition in Clinical Investigations. This guidance sets critical standards for validating, deploying, and monitoring these tools in clinical trials. The guidance stresses the importance of reliability, data quality, patient safety, and privacy—key factors as trials become more decentralized and reliant on technology.

The use of artificial intelligence will enable clinical trials to be evaluated quicker than present methods allowing FDA statistics to be calculated much quicker than present-day standards.  As a result, studies will be less labor intensive. Pipelines for drug delivery will be optimized.

What Are Digital Health Technologies?

According to the FDA, Digital Health Technologies encompass many tools designed to collect health data remotely. These include wearables, sensors, mobile applications, and software transmitting health-related data in real time. The goal of the FDA’s guidance here is to clarify what qualifies as a DHT and, more importantly, to establish how these technologies can contribute to clinical trials.

The FDA recognizes that data collected in traditional clinical settings can be limited. In-person visits provide only a snapshot of a patient’s condition, often missing the broader picture of their daily health patterns. This is where DHTs, like Aktiia’s continuous BP monitoring bracelet, come in. By enabling constant, real-world monitoring, Digital Health Technologies allow researchers to gather data on how a patient’s health fluctuates in their natural environment. This continuous data provides richer insights into conditions that require hypertension management, such as cardiovascular disease, diabetes, and chronic kidney disease, which often benefit from precise, long-term blood pressure tracking to inform clinical outcomes better.

The FDA emphasizes DHTs because they offer a more holistic view of patient health, moving beyond the confines of the clinic. However, with this new capability comes the responsibility to ensure that the data collected remotely is just as reliable and accurate as that collected in a controlled environment.

Key Considerations for Selecting DHTs in Clinical Trials

A significant point of emphasis in the FDA’s guidance is the concept of fit-for-purpose—ensuring that the selected DHTs are appropriate for the specific trial’s objectives and the patient population involved. The FDA doesn’t just want sponsors to use technology for technology’s sake; it wants to ensure that these tools effectively serve the trial’s intended goals.

The term “fit-for-purpose” is central to the FDA’s guidance, and it essentially means that a DHT should be appropriately validated and capable of addressing the specific objectives of a clinical trial. The technology must be well-suited for the patient population, the studied condition, and the data required for meaningful analysis. When the FDA urges sponsors to consider whether a DHT is “fit-for-purpose,” it’s pushing for a thoughtful selection process. This involves ensuring that the chosen technology can reliably collect data relevant to the trial’s endpoints, functions appropriately in the intended setting, and fits within the broader clinical context. Validation for the specific population also ensures that the DHT can accurately measure the health outcomes of interest. 

For instance, Aktiia’s BP monitor is fit for purpose in trials where continuous blood pressure monitoring is essential, such as in chronic kidney disease or hypertension-related studies. It has been carefully evaluated using an extended ISO81060-2 protocol adapted for a cuffless wrist device 

However, in trials focused on mental health or sleep disorders, where different metrics are crucial (e.g., brain activity or sleep patterns), a DHT designed for cardiovascular monitoring wouldn’t be fit for purpose unless adapted to the specific needs of those studies. A specific device (pulse-oximeter, continuous glucose monitor CGM} would have to be tested to meet ISO 15197:2013. while a pulse oximeter must meet ISO 80601-2-61:2017

As the FDA opens the door to more remote data collection, the agency remains aware of the risks associated with collecting data outside controlled environments where inconsistencies or errors could arise. Aktiia, however, has undergone rigorous testing and validation to meet these stringent requirements. Its blood pressure monitoring technology has been proven to track blood pressure accurately across different patient demographics, environmental conditions, and health statuses. This includes testing in real-world scenarios where patients may have comorbid conditions or varying physical states, such as during physical activity or periods of rest. Aktiia’s system still reliably collects accurate blood pressure measurements under these varied conditions.

Furthermore, the FDA emphasizes that trial endpoints must be directly tied to the data collected, and data integrity is paramount. In the case of Aktiia’s device, the company has implemented robust mechanisms to reduce noise and ensure the clarity of the data it collects. Aktiia’s technology effectively filters out potential data inconsistencies—such as those caused by patient movement or environmental factors—so that the blood pressure readings are accurate and precise.

Future Implications of the Guidance

Finally, the FDA’s guidance on Digital Health Technologies points to a future where clinical trials are increasingly decentralized and patient-centered. The agency sees DHTs as a way to make clinical trials more inclusive and reflective of real-world conditions. Trials that rely solely on clinic-based visits may exclude certain populations or fail to capture how treatments work in everyday life. By encouraging the use of Digital Health Technologies, the FDA is promoting a model of clinical research that is more flexible, accessible, and capable of capturing richer data and more reflective of patient experiences. The FDA’s forward-looking guidance sets the stage for a future where real-world evidence plays a more significant role in clinical research, ultimately speeding up drug development and improving patient outcomes.

Today's technology only allows for continuous monitoring of BP, glucose, and Oxygen saturation.  Other devices will come to market to measure electrolytes and blood chemistry. 

The FDA’s 2023 guidance on Digital Health Technologies for Remote Data Acquisition offers a comprehensive roadmap for integrating cutting-edge tools into clinical research. By focusing on validation, data integrity, risk management, and regulatory compliance, the FDA ensures that DHTs can enhance trial quality without sacrificing patient safety or data reliability. This guidance supports the safe and effective use of DHTs and pushes the boundaries of clinical research toward a more decentralized and patient-focused future, where real-time data offers deeper insights into how treatments impact lives.

Unpacking FDA Guidance on Digital Health Technologies in Clinical Trials

8 Top Pharma Trends In The Digital Health and AI Era - The Medical Futurist

8 Top Pharma Trends In The Digital Health and AI Era - The Medical Futurist

One area of. AI adoption is in PHARMA. The reasons are simple

1. Large corporations have significant resources.

2. Intense competition.

3. Regulatory effects from CMS, and FDA

Amid the digital health era, the pharmaceutical industry has been experiencing a rapid evolution. Thanks to the new paradigms that novel healthcare delivery approaches and digital health technologies have brought about, pharma companies need to adopt new strategies and consider new investment opportunities to stay relevant in the ever-evolving healthcare landscape.

In this article, we cover 8 major trends for pharma companies to consider in the digital health era. We rank them based on their return on investment (ROI) and return on vision (ROV). The latter is particularly relevant in this era considering its rapid rate of evolution. 

To better appreciate the importance of ROV, we can consider the Apollo program. While it was focused on space exploration, the endeavour has resulted in a multitude of spin-off products across various fields. In the healthcare sector, those spin-offs range from medical imaging to innovations in dental care. 

As such, current investments can have wide-ranging applications in the future. From AI in drug development to digital therapeutics, a host of new technologies and trends hold promising ROI and ROV for pharma companies.

We share a collection of 8 major ones below, ordered from the most practical and promising to the least

The niche is rich in  applications

1. Artificial Intelligence for drug research and development

The process of drug research and development has traditionally been a time-consuming and labor-intensive one. This has involved considerable trial-and-error research before a drug can proceed to further developmental stages. This process can be made more time- and cost-efficient with the assistance of artificial intelligence (AI).

AI models, such as those developed by Benevolent AI, can analyze significant amounts of datasets from scientific literature, clinical records, and chemical databases in a more time-efficient manner than humans can. From this information, they can precisely identify targets and how potential drugs will interact with them.

Companies like Schrödinger and Google DeepMind have used AI for drug formulation. Their software predicts the behavior of drug candidates and assesses their safety and effectiveness.

2. New reimbursement models

Pharma companies can tap into the new healthcare experience that patients can have in the digital health era to offer more than just medication. By combining medication and technology packages, they can offer more enticing reimbursement models for both payers and providers.

There have been several examples of such innovative models in the past that combine pharmaceuticals with technology. GSK has worked with Propeller Health on smart inhalers. Partners Healthcare Center and Japanese drug maker Daichii-Sankyo teamed up to bring a connected wearable for patients with atrial fibrillation.

Digital tools have been shown to improve health outcomes while minimizing financial costs. With such offerings, pharma companies can make their products stand out while being beneficial for both patients and insurance providers.

3. Large language models for improved workflow and customer service

Large language models (LLMs) have been popularised by tools such as ChatGPT and Google Gemini. Beyond the hype, the technology is a practical trend in the pharma industry.  LLMs can boost a company’s efficiency by optimizing internal operations and customer service.

Roche’s internal LLM tool, Roche GPT, assists the pharma company’s team in optimizing repetitive tasks and sharing knowledge. The tool further supports their business by automating structured data extraction about therapies and patients from scientific articles and clinical test results. Pfizer has also deployed a similar tool to help with its marketing efforts.

LLMs could further be used to improve customer service. With an LLM-powered chatbot, patients can get answers to their queries such as medication side effects in their native language

4. Automation in the supply chain

The pharma industry’s supply chain stands to gain a lot by embracing automation in its midst. For example, by integrating AI, drug shortages can be averted. By analyzing data from various sources, AI software can forecast potential disruptions and suggest adequate measures to ensure a steady supply of essential medication.

Automation in the supply chain does not only involve AI software but robotics is also part of the picture. Denso Robotics’ robots are capable of automating tasks in the manufacturing process. Exoskeletons are another example of robotics assistance. While not fully automating tasks, they augment manual factory worker’s ability to carry heavy loads and work in uncomfortable positions. In the future, we can even expect automated drone deliveries to be carried out within manufacturing sites and beyond.

5. Digital therapeutics 

Using software as treatment might have sounded like a science fiction concept a decade or so ago, but this prospect is very real and promising with the advent of digital therapeutics (DTx). DTx can be described as evidence-based software applications designed to prevent, manage, or treat medical conditions.

The accessibility, privacy, and minimal side effects that DTx provides have enticed pharma companies to invest in this trend. Pfizer has teamed with Sidekick Health to launch a DTx solution for atopic dermatitis. Eli Lilly also partnered with Sidekick Health to develop apps to support breast cancer treatment. 

Other companies like RelieVRx or HelloBetter integrate cognitive behavioral therapy principles in their apps to ease chronic pain. We share more promising DTx examples in a dedicated article.

6. in silico clinical trials

in silico clinical trials promise to enable the conduction of experiments wholly via computer simulation, without the need for animal or human testing. By running drug trials on computer simulations of organs, this approach can be both time and cost-effective while circumventing the side effects on live participants. 

While this promise has yet to be fully realized, progress has been made towards it. The Wyss Institute has developed organs-on-a-chip to emulate the complex structures and functions of living human organs. Their technology has been leveraged by Emulate Inc. for efficient drug development. The mathematical model of human physiology created by HumMod has been used in several research projects. Further envisioning a future trending towards in silico trials, the Virtual Physiological Human Institute has been set up to encourage the effective adoption of in silico medical research.

For further reading please click on this link

Machine Learning–Based Prediction for Incident Hypertension Based on Regular Health Checkup Data:

The World Health Organization (WHO) has identified cardiovascular diseases (CVDs) as the leading cause of mortality worldwide, with a staggering 17.9 million deaths recorded in 2019 [1]. This number is projected to rise to approximately 23 million by 2030. Of the multitude of CVDs, specific conditions such as myocardial infarction and ischemic stroke account for more than 85% of these CVD-related deaths [2]. The US Centers for Disease Control and Prevention (CDC) have highlighted that CVDs caused over US $216 billion in overall health care expenses and resulted in US $147 billion lost due to increased workplace absenteeism and corresponding productivity in the United States. As a result, CVDs impose a significant burden on the nation’s economy [3].

Given the acknowledged biological and economic risks associated with CVDs, it is widely recognized that hypertension plays a significant role in these health complications, including myocardial infarction and stroke [4]. Predicting hypertension onset is notably challenging due to the disease’s multifactorial origins, encompassing a wide range of genetic, environmental, and lifestyle factors. The subtle and often interrelated effects of these factors contribute to the complexity of early detection. For example, genetic predispositions may interact with lifestyle choices such as diet, exercise, and smoking habits, in ways that are not fully understood [5]. Environmental influences, including socioeconomic status and access to health care, further complicate the picture by affecting both the risk of developing hypertension and the ability to manage risk factors effectively [5,6]. Additionally, the asymptomatic nature of hypertension in its early stages means that it often goes unnoticed until more serious health issues arise, making timely and accurate prediction all the more difficult [7]. These challenges underscore the need for sophisticated predictive models that can integrate and analyze the myriad of contributing factors to identify individuals at risk of developing hypertension early in its progression. Considering the severe societal implications of hypertension across all nations, early diagnosis is crucial to mitigate its potential hazards. In this study, we propose a novel approach to predict the onset of hypertension using the population’s regular health checkup and demographic factors. In recent years, machine learning models have emerged as powerful tools across many fields, particularly in medical applications [8]. Their ability to analyze complex patterns and make accurate predictions has revolutionized how we approach health care challenges.

However, ensuring this methodology’s replicability and broad applicability in real-world settings presents an intricate challenge. To bolster the reliability of our hypertension projections, we conducted additional independent validation using distinct cohorts. This study investigated various machine learning approaches to strengthen the method’s robustness, replicability, and real-world practicality. We delved into the hypertension landscape across Asian populations through machine learning optics, firmly anchoring our methodology within the burgeoning realm of artificial intelligence (AI)–driven disciplines. This research endeavors to amplify our comprehension of global hypertension trends by channeling multifaceted machine learning analyses, thereby catalyzing more timely and precise diagnostic efforts.

This is only one area where MACHINE LEARNING IS BEING USED

Machine learning (ML) is a dynamic field at the forefront of artificial intelligence (AI), employing algorithms and data to enable machines to learn autonomously or semi-autonomously. In 2024, ML continues to evolve, characterized by several key trends and innovations that are shaping various sectors, from healthcare to finance.AI is set to significantly transform healthcare in several key areas:

Diagnostics and Imaging**

Enhanced Imaging Analysis:** AI algorithms can analyze medical images (X-rays, MRIs, CT scans) with high accuracy, often identifying conditions earlier than traditional methods.

Predictive Analytics:** AI can predict disease outbreaks and individual health risks based on data patterns.

Personalized Medicine**

Genomic Data Analysis:** AI will aid in interpreting genomic data to tailor treatments specific to individual genetic profiles.

Treatment Recommendations:** Machine learning models can suggest personalized treatment plans based on patient history and outcomes.

Patient Monitoring and Management**

Wearable Devices:** AI will enhance data from wearables for continuous health monitoring and early intervention.

Remote Patient Monitoring:** AI systems can analyze data from remote monitoring devices to alert healthcare providers of potential issues.

### 4. **Administrative Efficiency**

Streamlining Operations:** AI can automate administrative tasks such as scheduling, billing, and patient data management, reducing costs and improving efficiency.

Chatbots and Virtual Assistants:** These can handle patient inquiries, appointment scheduling, and follow-up care, freeing up healthcare professionals for more complex tasks.

Drug Discovery and Development**

Accelerating Research:** AI can analyze vast datasets to identify potential drug candidates and predict their effectiveness, significantly speeding up the research process.

Clinical Trials Optimization:** AI can help in designing and running clinical trials more efficiently by identifying suitable candidates and monitoring results in real-time.

Mental Health Support**

AI Therapy and Chatbots:** AI-driven tools can provide mental health support through chatbots or apps that offer cognitive behavioral therapy techniques.

Predictive Tools:** AI can help identify individuals at risk of mental health crises based on data patterns.

Ethics and Bias Mitigation**

Addressing Inequities:** As AI develops, there will be a focus on ensuring algorithms are trained on diverse datasets to minimize bias and improve health equity.

Conclusion

The integration of AI in healthcare promises to enhance patient outcomes, improve operational efficiency, and drive innovation in treatment and care delivery. However, ongoing ethical considerations and data privacy will be crucial as these technologies evolve.

JOURNAL OF MEDICAL INFORMATION