Machine Learning Will Transform Healthcare
20 Ideas in 20 Days — Day 3
I know we’re at peak hype for AI and machine learning, but when you shine the possibilities of the technology on a messy, broken industry like healthcare, the opportunity is massive.
I’ve seen a number of recent items in the media about machine learning improving diagnostic outcomes. But I believe that machine learning technology will be the innovation that transforms healthcare into a patient centered approach with the EHR (electronic health record) at the heart.
The move to electronic health records, in addition to advances in the ability to analyze unstructured data will provide rich population datasets that will greatly increase positive patient outcomes and decrease the cost and time it takes for patients to get the proper diagnoses and treatment.
I find it useful to bucket the potential for machine learning into prediction, diagnostic support, therapeutic support and patient engagement.
Prediction
I think this is the most exciting area because of the long term potential to allow doctors to focus on keeping patients healthy vs. trying to fix what is already wrong. Meshing personal genomic data with broad and preferably global patient population data will offer early prediction for cancer, heart disease, obesity, mental disorders, chronic stress and the list goes on.
Armed with this knowledge, the physician can work with the patient to determine the best course of action depending on the nature of the risk. I’m going to save my thoughts on patient engagement for tomorrow’s post, but I believe machine learning will further help us identify which tools and regimens work best, not just for a population bound by a clinical trial, but for patients that are similar in age, sex, economics, education and perhaps and maybe most importantly personality and temperament.
Diagnostic and Therapeutic Support
Applying machine learning is already being researched and in some cases implemented in multiple specialties.
IBM’s Watson Health is working together with the Mayo clinic to determine if oncology patients are a good match for clinical trials. In radiology, machine learning can quickly identify reference images and increase accuracy saving the doctor search and consult time and producing related savings. And there was a recent article in The New Yorker about how machine learning is being applied to teach computers to differentiate between cancerous and benign lesions in dermatology.
Clearly patients are more than their data. And there’s no question that we’ll still need doctors for the foreseeable future. But no single doctor can be all-knowing. And as much as we’d like them to be omnipotent, doctors can have off-days too.
To save docs the time of searching for the reference image, tracking down the clinical trial that will accept the patient who’s run out of alternatives and verifying that the lesion is or isn’t a melanoma (preferably without the cost and discomfort of a biopsy just to be sure) is an incredible opportunity.
There are challenges to be sure. Patient privacy is a big one, but I believe it’s possible for scrubbed data to be fed into the population data without violating doctor/patient confidentiality. Malpractice is another one. Much like the dilemma with self-driving cars — if the machine/doctor pair is wrong about the diagnoses, who is at fault? Certainly the complexity of our current regulatory and payer systems will play into the adoption rate in health care as well.
However, I believe that adding machine learning to the health care matrix will allow us to become a healthier population at a lower cost by giving doctors the information they need about who’s at risk, possible diagnoses, the cost/benefit of all possible therapies and, most importantly, the outcome information associated with the possible prevention/treatment paths.
