Working at Oracle is the best opportunity I ’ ve ever had to impact this at scale because our core competency around infrastructure and data can make this vision a reality. We can improve the provider-patient relationships and people ’ s situations. And we can use real-life data to help build healthier communities. In Qatar, for example, they are using our electronic health records ( EHR ) and, more specifically, pediatric weights, to curb pediatric obesity. Based on this data, they decide whether a fast-food restaurant or a gymnasium should open in a certain area. Data can be powerful if people use it in the right way to make decisions. How has big data impacted patients already? I ’ m not sure that it has yet. Digitizing the health record was important. We fixed practical problems like not being able to read someone ’ s handwriting or missing charts because someone took them home over the weekend. Ultimately, digitizing the existing workflow isn ’ t enough. It has been helpful, but it ’ s not the real promise of tech in health care. Now, we ’ re incorporating disparate data sets that go far beyond the EHR. With the latest tech, we can use data to better understand who you are, who other patients like you are, the best way to engage with you and how to ensure that we ’ re improving, not worsening, health equity.
What are the promises and pitfalls of using big data and AI? When I was at Google, for example, we built a model for the U.K. National Health Service ( NHS ) around acute kidney injury. Typically, the rapid response team took four hours to diagnose this condition. Then, using our phone app with good user experience design that utilized creatinine levels we were able to decrease the diagnostic time to 14 minutes and reduce cardiac arrests by 30% and overall costs by 17%. This is just one example of how tools can assist caregivers to deliver better care. Next, we utilized even more data with AI to predict the chances of acute kidney injury: 600,000 variables per patient in a data set of 70,000 patients. With 90% accuracy, the computer predicted that a person would be on dialysis two days before any signs or symptoms appeared. That ’ s anticipatory medicine. That is the power of AI.
Dr. Feinberg speaking at the Oracle Cerner Health Conference in October 2022. Oracle acquired Cerner, a leading provider of digital information systems used in healthcare settings (including electronic health records) in June 2022, leading to the creation of Oracle Health
However, there is a catch. Our training data set was from a research project with Veterans Affairs and was 93% male, so the model didn ’ t work on females. So, what if, in 10 years, after we ’ ve incorporated AI and made major developments across our healthcare systems, we ’ ve made situations worse for underrepresented groups along the way? That scares me. My biggest worry is that we will hardwire the bad parts of health care. All of our large language models that use EHRs involve patients who have seen a doctor. They exclude anyone who ’ s never gone due to lack of access or financial ability, for instance. This missing subset can therefore affect and skew our models. Bias in health care can also influence our training models, and we have to be cognizant of that as well. People also talk about hallucinations in AI models. One example in health care is using ambient listening during a patient visit and then the computer writes a note. The patient was never weighed but the computer puts in a BMI, which shouldn ’ t happen. But in this case, the BMI is close to accurate. This is a hallucination but a helpful one because the doctor should have weighed the patient. You know the computer is going to be right a thousand times in a row, but the one time it misses, how will we catch that? It is crucial to hone the human-computer interface, know when to use technology and understand that the computer can get it wrong if we train it on the wrong models. We ’ re still untangling the nuances here. How will emerging technologies influence clinical care? Whether it is the mom caring for her family, a medical provider, a case manager in the emergency room, a person running a healthcare system or just you caring for yourself, technology shouldn ’ t get in the way of care. But, if you look at current EHRs, it often does. Providers look at the computer, not the patient, straining the patient-physician connection. Ideally, tech should augment and optimize care to easily recall medical history, summarize patient data and incorporate other factors to aid in clinical decision-making and allow for a stronger connection with the patient. Humans are best at connecting with people, and we should rely on technology to do things that humans aren ’ t so good at. Computer vision is a great example of this. Computers can read diabetic retinopathy scans, moles and lab pathology better “ Health care is about people caring for people. It ’ s built on trust. ”
3
Made with FlippingBook flipbook maker