Americans encounter some form of artificial intelligence and machine learning technologies in nearly every aspect of daily life: We accept Netflix’s recommendations on what movie we should stream next, enjoy Spotify’s curated playlists and take a detour when Waze tells us we can shave eight minutes off of our commute.
And it turns out that we’re fairly comfortable with this new normal: A survey released last year by Innovative Technology Solutions found that, on a scale of 1 to 10, Americans give their GPS systems an 8.1 “trust and satisfaction” score, followed closely by a 7.5 for TV and movie streaming services.
But when it comes to higher stakes, we’re not so trusting. When asked about whether they trust an AI doctor diagnosing or treating a medical issue, respondents scored it just a 5.4.
Overall skepticism about medical AI and ML is nothing new. In 2012, we were told that IBM’s AI-powered Watson was being trained to recommend treatments for cancer patients. There were claims that the advanced technology could make medicine personalized and tailored to millions of people living with cancer. But in 2018, reports surfaced that indicated the research and technology had fallen short of expectations, leaving users to speculate the accuracy of Watson’s predictive analytics.
Patients have been reluctant to trust medical AI and ML out of fear that the technology would not offer a unique or personalized recommendation based on individual needs. A piece in Harvard Business Review in 2019 referenced a survey in which 200 business students were asked to take a free health assessment to perform a diagnosis—40% of students signed up for the assessment when told their doctor would perform the diagnosis, while only 26% signed up when told a computer would perform the diagnosis.
These concerns are not without basis. Many of the AI and ML approaches that are being used in healthcare today—due to simplicity and ease of implementation—strive for performance at the population-level by fitting to the characteristics most common among patients. They look to do well in the “general” case, failing to serve large groups of patients and individuals with unique health needs. However, this limitation of how AI and ML is being applied is not a limitation of the technology.
If anything, what makes AI and ML exceptional—if done right—is its ability to process huge sets of data comprising a diversity of patients, providers, diseases and outcomes and model the fine-grained trends that could potentially have a lasting impact on a patient’s diagnosis or treatment options. This ability to use data in the large for representative populations and to obtain inferences in the small for individual-level decision support is the promise of AI and ML. The whole process might sound impersonal or cookie-cutter, but the reality is that the advancements in precision medicine and delivery will make care decisions more data-driven and thus more exact.
Consider a patient choosing a specialist. It’s anything but data-driven: They’ll search for a provider in-network or maybe one that is conveniently located, without understanding potential health outcomes as a result of their choice. The issue is that patients lack the proper data and information they need to make these informed choices.
That’s where machine intelligence comes into play—an AI/ML model that is able to accurately predict the right treatment, at the right time, by the right provider for a patient, which could drastically help reduce the rate of hospitalizations and emergency room visits.
As an example, research published last month in AJMC looked at claims data from 2 million Medicare beneficiaries between 2017 and 2019 to evaluate the utility of ML in the management of severe respiratory infections in community and post-acute settings. The researchers found that machine intelligence for precision navigation could be used to mitigate infection rates in the post-acute care setting.
Specifically, at-risk individuals who received care at skilled nursing facilities (SNFs) that the technology predicted would be the best choice for them had a relative reduction of 37% for emergent care and 36% for inpatient hospitalizations due to respiratory infections compared to those who received care at non-recommended SNFs.
This advanced technology has the ability to comb through and analyze an individual’s treatment needs and medical history so that the most accurate recommendations can be made based on that individual’s personalized needs and the doctors or facilities available to them. In turn, matching a patient to the optimal provider has the ability to drastically improve health outcomes while also lowering the cost of care.
We now have the technology where we can use machine intelligence to optimize some of the most important decisions in healthcare. The data show results we can trust.
Zeeshan Syed is the CEO and Zahoor Elahi is the COO of Health at Scale.