My 68-year-old mom called me recently with a medical question. A week prior, she was hauling logs on her farm, fell on a pile of branches, and staked her calf on a branch. She pulled out a two-inch piece of wood, but was worried there was still part of the branch in her leg. The next morning, she went to the ER. After a normal X-ray and tetanus shot, they sent her home on antibiotics.
Any doctor reading this is probably cringing right now. When I worked with residents, this is what I call a “teachable moment.” Can you use an X-ray to rule out a wooden foreign body? No, because wood doesn’t always show up on an X-ray. This cognitive error led to multiple unnecessary visits, infection, surgery, and a hospital stay a week later.
This is just one personal example, but it’s illustrative of the opportunity for real-time technology assistance in medicine. No pilot in their right mind would choose to fly without their instruments and/or autopilot. We have lane assist and automatic braking in our cars, but the current state of medical technology doesn’t incorporate anything but the most basic of decision support. Too often, healthcare providers essentially fly blind. The good news is that this latest AI boom has unlocked the ability to add real-time cognitive support in a way that was pure science fiction a few years ago.
The U.S. spends more as a percentage of GDP than any other nation, and yet our life expectancy ranks 34th in the world (and is worsening), and our maternal mortality rate is abysmal. The reasons behind these figures are multifactorial. Pre-pandemic, there was a physician shortage. This shortage is accentuated post-pandemic. To meet this need, more midlevel providers have been joining the workforce.
While these practitioners can be extremely capable, upon entry into the healthcare workforce, they typically have less experience and education than their physician counterparts. Add this to the explosion of medical information we hold healthcare practitioners responsible for understanding and recalling during a brief appointment–often while sleep-deprived and overworked–this is a setup for failure. Even without the systemic issues in the U.S. healthcare system, it’s no shock that our health outcomes are subpar.
This is where the advancements in AI to support providers can make a difference. Carbon Health founder and CEO Eren Bali has compared this support to autonomous driving levels, with parallels between autonomous driving and autonomous medical practice. For reference, Level 0 has no driving support, Level 3 has conditional driving support (limited driving in controlled environments), and Level 5 is fully autonomous (no steering wheel). Common current features such as lane assist and adaptive cruise control are considered Level 2. When we apply this framework to medicine, blank paper charts are Level 0, basic clinical decision support is Level 1, and full sci-fi automated diagnosis and treatment is Level 5.
Market-leading EHRs offer Level 1 support. Algorithms can suggest the correct antibiotic after a physician has entered a diagnosis. The medication service checks for cross-reactivity and allergies. However, this current wave of AI and medical technology can support autonomous practice up to Level 3. With the right training, this could be similar to a good resident. This level will be skilled at first-order thinking and single-pathology issues, but less capable when caring for complex or critical patients. Levels 4 and 5 will need underlying technology advancement to unlock these capabilities, but we are so much closer to this than we ever were before.
Imagine a real-time AI assistant that advised an ultrasound or CT scan to rule out a deep foreign body when my mom initially presented to her rural ER. Imagine a bot capable of analyzing her prior culture data and antibiotic tolerances to suggest the best antibiotic in seconds. Initially, AI can augment a provider’s expertise and make the painful parts of practicing medicine effortless. The note will be written, the medical coding and billing will be automated, and the patient education will be tailored to the healthcare literacy level and language of the patient.
With evolving technology, providers can leverage lane assist and cruise control-type support, knowing they could always turn it off if they didn’t want the assistance, and patients can rest assured their providers have access to all tools. This is the future of healthcare.