The headline suggests a rather bleak and disheartening outlook. Yet within lies a glimmer of hope: AI and GenAI offer innovative ways to deal with unintended human error in medicine.
- In 2016, a globally recognized US study found that medical errors by doctors are the third leading cause of death, following heart disease and cancer.
- The acclaimed study of the Johns Hopkins University School of Medicine estimated that nearly 40% of?those errors in diagnosing are linked to five conditions: stroke, sepsis, pneumonia, venous thromboembolism, and lung cancer.
- By 2023, a new study suggested that every 3rd Americans become permanently disabled or die annually across care settings because dangerous diseases are misdiagnosed.
- Diagnostic errors are a major concern globally. In Europe the same cause of incidences amount to 465 000, with medical human error equally representing the third leading cause of death.
- This reality may become a major opportunity for an increased application of AI and GenAI in medical diagnosis and treatment. A new survey?found that one in five British General Practitioners are using artificial intelligence, such as ChatGPT, in their clinical practice including treatment options. Beyond ethical concerns and risks of using AI in clinical practice, there benefits of such applications could become life-saving.
So what: Unintended human error is an opportunity
Clearly, unintended human error, occurring not only on the doctor's side but also on the patient's side, can lead to harm. The crucial fact, however, is that close to half of the errors in diagnosing are linked to few conditions only. In fact, since the Covid-19 pandemic, the ‘Big Three’ dangerous disease cases are vascular events, infections, and cancers.
The concentration around frequent and repetitive ocurrences makes the related source of errors not only more tractable, parsable for data processing, and screenable and manageable for hospitals, doctors and, indirectly, insurers - more than we ever imagined in the "pre-GenAI era".
Real life: How human error entered my own medical path
Over the last few months I had my own medical path (in a rather robust healthcare system such as Switzerland) changed my perspective on AI and GenAI in healthcare. The outlook and opportunity that AI applications will enhance one day diagnostic accuracy and support clinical decision-making and possibly even reduce human errors, gained a slightly different meaning than just one year ago:
- Ten months ago, I underwent medical surgery following the detection of an issue during a routine check-up. I seeked opinions from three qualified radiologist based on MRI imaging, yet none of them was able to accurately diagnose the severity of the issue. The advise was three times to "wait a few months" to see how things develop. My intuition kept telling me, something is seriously wrong. Which effectively was the case, ultimately necessitating a second surgery two months later. You may have had similar experiences of a miseleading medical expert judgement . Will future AI-based MRI analysis find the cause fast and prevent (the costs and strain of) a second surgery.
- Next up, I was prescribed medication at the doctor's office. While my online records correctly documented the doctor's prescription of a 500mg dosage, the assistant handed over the counter a pack of 1000mg units. In fact, at that moment she did the right thing. The doctor had correctly recorded the 500 dosage, the error was the inaccurate manual transfer of data into the prescription (as so often, due to lack of time). Had I not asked, I would have followed the wrong treatment plan. Can simple AI-powered validation tools eradicate such errors in the future.
Next up: How GenAI can be applied to diagnoses
By now, I could not be more positive and optimistic, that GenAI applications offer promising new avenues to eradicate scenarios as the two self-reported above in future. Already today, AI has made significant strides in the medical field, enhancing diagnostic accuracy, and supporting clinical decision-making.
I see three major areas of application, from which both patients and doctors should benefit:
- Predictive Diagnostics: predict potential health issues, aiding doctors in early diagnosis. Today, AI can identify subtle patterns and markers in imaging data that may be overlooked by the human eye, enabling earlier detection of diseases like cancer, Alzheimer's, and cardiovascular conditions.
- Early Detection: analyze medical images to detect anomalies that might be missed by human practitioners. Today, GenAI models, especially deep learning algorithms like convolutional neural networks (CNNs), are employed to analyze medical images such as X-rays, MRIs, CT scans, and ultrasounds. These models can detect anomalies—including tumors, fractures, lesions, and hemorrhages—with a high degree of accuracy. For instance, advanced AI models are used to detect and classify ovarian cancer from CT scan images and clinical biomarkers. These models employ deep learning techniques to analyze complex patterns in medical images, enabling earlier and more accurate diagnosis, which is crucial for conditions like cancer where early treatment can significantly improve outcomes.
- Personalized Treatments: enhance diagnostic accuracy, and support healthcare specialist in designing the optimal treatment path providing evidence-based recommendations, in particular in more complex cases. Today, Generative AI models like Med-PaLM 2 are being fine-tuned for healthcare applications to assist clinicians with decision-making. These models can process large amounts of clinical data, including patient histories, lab results, and genomic information, to provide insights and support complex diagnostic reasoning. This includes generating clinical notes, summarizing patient information, and answering complex medical queries, thereby streamlining clinical workflows and reducing administrative burden.
It matters: GenAI's potential to reduce cost of insurance in health
The use of AI tools could significantly impact the insurance industry, particularly in healthcare and malpractice coverage. It has two sides of a coin:
As AI assists with diagnoses, treatment suggestions, and documentation, there’s an increased risk of medical errors or data breaches, raising the current level of concern about liability and patient confidentiality. Insurers need to reassess policies. Potentially, premiums will need to be raised and new new clauses to be introduced to account for AI-related risks.
On the flipside, AI and GenAI can be applied for better diagnosis outcomes when deployed in an effective and ethically controlled manner. This includes ensuring the accuracy and reliability of the AI algorithms, maintaining patient privacy and data security, promoting transparency in AI decision-making, and integrating GenAI into existing processes to enhance human expertise rather than replace it.
In my mind, the power of increasingly larger and enriched medical history data sets is a game changer. For insurance, this has an immediate impact and is the single most likely lever in health insurance to transform sickcare into healthcare the fastest.
Insurers will benefit from GenAI in three particular ways:
- Risk Assessment Models: GenAI analyzes large datasets to predict a patient's risk of developing specific conditions such as sepsis, heart attacks, or strokes. These predictive models can assess risk factors and can transform product design to be fit for the customer's needs. This will enable insurers to develop personalized insurance products that are fairer to consumers and greater ability to contain rising claims costs for insurers. Personalized products are likely to become more affordable compared to standard product ranges.
- Personalized Risk Assessment and Product Development: By analyzing extensive datasets, including customer demographics and health records, GenAI can predict high-cost claims with greater accuracy. It can rapidly identify new emerging patterns and facilitate the implementation of preventive measures. Furthermore, GenAI can enable insurers to simulate the impact of treatments over time and assess the effectiveness of healthcare providers, thereby enhancing operational decision-making - currently, such decisions are often reliant on a "second opinion" covered by insurance.
Ultimately, AI can expedite the transition toward more personalized medicine. By tailoring diagnostic and treatment recommendations to the individual patient level, unique genetics and personal health history are considered more widely for the first time. If you follow this train of thought, AI and GenAI could have the potential to transform sickcare into healthcare, make prevention accessible more widely and, potentially, save lives.
Mirjam Bamberger is a member of the Management Committee of AXA's European Markets & Health. Previously, she served as the CEO of AXA Luxembourg, CEO of AXA Wealth Europe, and held various board member roles at AXA Switzerland. With over 25 years of experience, Mirjam has lived and worked in 9 different locations across the US, UK, China, Latin America, and Europe, holding executive and non-executive director positions within the financial services and high-tech sectors. She holds an Executive MBA with honors from IMD Lausanne, a master’s degree from the University of Cologne, and diplomas from University St. Gallen/Stanford, Cornell, DUW Berlin, and ICMA. Additionally, she is a certified director of the Swiss Board School with NED positions across Europe.
