Can AI & Machine Learning Make Healthcare Cheaper?
By Kumar Venkatesiah
The average cost of colonoscopy in the U.S. is over $5,500. This is a cost borne either by the consumer or by the insurer depending on their insurance status. Depending on your unique condition, the procedure can cost several times more.
What’s bothersome, however, is the poor success rate in detecting polyps and tumors. Healthcare experts agree a doctor’s independent skill in determining polyps play a defining role in the success of the procedure. Some studies estimate the success rate of detecting benign polyps at 20 percent and adenoma at 12 percent.
This is a poor success rate for a procedure that can cost several thousands of dollars. So why do these tests cost so much with such low success rate? A typical colonoscopy test takes between 20 minutes to an hour. Even discounting the pre-test procedures, a doctor may only be able to perform 8-10 tests in a day at max efficiency.
As for the cost; In addition to the doctor’s fee, patients also incur significant costs to use the hospital facilities, take pre-test medications, and consultation. In short, a good chunk of money is charged because of the limited number of patients a hospital can take in a day, not for the actual costs of performing the test.
This is why the advances in artificial intelligence (AI) and machine learning are exciting from a healthcare perspective. AI can help computers scan thousands of images and identify patterns at a fraction of time. Add machine learning to this, and we are looking at technology that can dramatically improve sensitivity and accuracy over time. Leveraging the cloud, these technologies would be capable of comparing every scanned image against the millions of data points gathered from hospitals across the world.
An initial test of such a procedure was performed by a team of doctors at Showa University in Yokohama, Japan. In the test, the team scanned the colon for several hundred different features that typically define a polyp. The study’s results are stunning as doctors were able to deliver 86 percent accuracy with AI. The test also delivered 79 percent specificity in its results and 94 percent sensitivity.
All these results point to an advance in healthcare that can deliver a vastly improved form of cancer screening. Not only this, AI-assisted colonoscopy technologies are much quicker to scan and interpret results compared to humans. Typically, such a system can look more than 300 features of polyps in less than a second. This brings down the time to perform a procedure from an hour to a matter of minutes. It was also found results delivered from an AI-assisted colonoscopy test were independent of the endoscopists’ individual skill.
In other words, AI can bring down the cost of a cancer screening test in two ways: it reduces the time it takes to perform a screening operation on one person (thereby increasing the number of procedures that can be conducted in a day), and also brings down the doctor’s fee since highly skilled endoscopists may no longer be required to perform screening tests.
Some observers in the insurance sector however worry technology and resources required to deploy such advanced image scanning machines may inflate scanning costs and thereby offset any savings made from a drop in an endoscopists’ fee. But as another experiment conducted by scientists at the Stanford University shows, these fears may be unfounded.
To perform this experiment, the scientists at Stanford University made use of a public API provided by Google for machine learning. This is the same algorithm that Google uses to identify objects by their image. For the experiment, the scientists fed the algorithm with over 130,000 images of skin lesions that cover over 2000 diseases. The results? A whopping 91 percent accuracy when it came to matching new diagnosis for skin cancer classification.
The cost of conducting such a test may not really be high. Google’s machine learning algorithm costs just a few pennies for every hour on the job and there is not a lot of hardware equipment to deploy considering this infrastructure is based in the cloud.
Cloud, AI, and machine learning could dramatically alter the way medical tests are conducted and priced. The lower cost of performing such advanced tests could also push doctors into prescribing such tests at an earlier stage. This could help detect and stop deadly diseases including cancer much before they are detected today.
From an insurer’s perspective, these advances in healthcare are profitable for two reasons. One, they bring down the cost of performing tests. But more importantly, the results are highly accurate. This helps insurance agencies avoid claims from patients undergoing further tests in detecting ailments that are accurately diagnosed with AI.
Truth be told, it will be at least a decade for such technologies to be available to hospitals across the world. But given the impact that these technologies have on healthcare, the onus is on patients and insurance companies to drive hospitals to adopt these cheaper, highly accurate technology systems.
About The Author
Kumar Venkatesiah is an eLearning consultant from India with more than nine years of experience. He works with telemedicine startups to help onboard new clients to the digital platform. You may reach him at kumarv2012@yahoo.co.in.