A holiday wish list for healthcare AI
Because with a little holiday magic, maybe we can make it happen
One of the best parts of being a kid is being pleasantly surprised with the world. You find a penny on the ground? Amazing! There’s a rock that looks like an animal? So awesome! Maintaining that sense of joy and wonder at the world is hard as an adult, so it’s helpful to think about how far we’ve come in healthcare and technology in the last 10/100/1000 years. Antibiotics! Clean water! A chatbot that can tell you what to cook for a week!
But because I’m an adult and can’t be satisfied with what we have, I still want more. As we wrap up the holiday season, I wanted to close with a list of ideas or products that I’d love to see get more attention in the coming year.
This is my wishlist for healthcare AI, though many are foundational to the entire concept of healthcare AI. I’m sure there’s a brilliant entrepreneur out there who can figure out how to develop a business case for some of these ideas and help propel the field forward safely and responsibly.
More attention to data management to actually get insights out of all the data we collect
The amount of detail that’s captured in a medical image has dramatically increased with better microscopes and imaging techniques. This improvement has been a boon to AI training sets and has played a part in breakthroughs in AI-assisted diagnosis. However, health systems were never built to be data storage companies. Unlike other data-heavy companies like Facebook or Twitter, they don’t (usually) sell their patients’ data. The US government also hasn’t invested in its centralized data storage like countries with socialized medicine models. Therefore each health system is stuck with a huge amount of potentially life-saving information that is:
Expensive to store
Generally not cleaned or formatted to be easily used
They are legally required to keep for a specified period of years
In part because hospitals aren’t actually reimbursed for any data management functions, their data talent has been short-staffed for years and they’re unable to pay salaries anywhere close to the tech giants. So they’re also missing the people they need to be effective.
I believe one of the main reasons physicians are frustrated with EHRs is that we don’t see any benefit for our patients. If we could actually use all the data that’s stored in the EHR to help our patients, there would be more acceptance of the EHR. Currently it feels like the data just disappears into a bottomless pit, which makes the whole endeavor of putting a lot of information into the EHR feel inherently pointless.
Integrating data from wearables etc to inform decision-making
I have patients with an enormous amount of data on their wrists that their physician will never see or act on. There are exciting innovations in the Internet of Things field like nanomaterials and smart textiles and even inside cell phones that could transform the snapshot that physicians get to the summary of a TV series. Smart sensors are already collecting information; I’d love to see that data harnessed in a meaningful way for physicians and other clinicians.
Focus on validation and monitoring of AI tools
I’d like to see more studies that evaluate commercially available AI tools in clinical practice, especially that compare tools and that describe the best and worst ways to use them. Validation will likely need to be done on a local basis, and is challenging but necessary to ensure the safe use of AI platforms.
Education of physicians about how to use AI tools
There’s an ethical and practical imperative to give physicians relevant, useable information about these tools. The Center for Healthcare AI proposes a Model Facts card that describes important details about the information the tool was trained on, the kinds of patients and settings to use it in, and warnings. As I’ve said before, doctors are great at using dangerous tools if they understand their boundaries. We can’t expect physicians to use these tools well if we don’t give them the crucial information they need to make decisions.
Additionally, I’d like to see more focus on education practicing physicians on safe and effective use of AI platforms. All the tech-savvy medical educators I know are developing AI curricula for students and trainees, but there’s still relatively little out there for the practicing physicians who will make up the large bulk of users. Medical societies and others need to take a more active role in providing skills to ensure these new tools actually benefit our patients.
AI in public and global health
There hasn’t been much movement in public health using AI. There are patient privacy, cybersecurity, and consent issues that need to be addressed. The most exciting uses I’ve seen so far are wastewater screening for communicable diseases, weather prediction, and screening for mental health issues. The CDC is also actively developing AI/ML approaches for screening chest X-rays for TB, for example. Two especially interesting focus areas include:
“Use more of the data we have:
Identify opioid-related terms on death certificates, even if they’re misspelled
Impute missing data from surveys, or fix sparsity in geographical sampling
Use non-traditional data sources, including images, audio, social media, and data not specifically collected for public health analysis, such as electronic health records”
However, there’s still so much possibility for identifying disease patterns, evaluating effectiveness of public health interventions and providing AI readings of images when a radiologist or pathologist is not available. The global health community is working toward using AI effectively but I’d love to see more money and attention to this area.
Personalized medicine democratized
Much of the personalized medicine that’s been rolled out has been focused on biohacking and on a small subsegment of the population - both rich people able to buy fancy biohacking kits and people with very specific diseases. I’d love to see broader applications of personalization in medicine in the form of risk calculators, medication and nutrition suggestions, and care plans. We still do a lot of “try it and see” medicine that I’m hopeful could be minimized in the future.
Until I can recapture a childlike sense of wonder about the amazing parts of medicine, I’ll hope that we can move the field forward on these important topics. I’d love to know what would be on your wishlist - please share!