Find out how medical algorithms have the potential to change healthcare.
Twenty-one years ago, Garry Kasparov, one of the world’s greatest chess players, sat hunched over a chessboard as he tried to calculate his next move in a tournament televised around the globe. His opponent was calculating moves too, at a rate of 200 million per second. It was the IBM supercomputer, Deep Blue, which went on to beat Kasparov, prompting people to wonder if machines would surpass humans not only in games but also in jobs and many other functions in life.
Today, doctors are wondering the same thing as computer- and analytics-driven medical algorithms come into wider use in medical practices and hospitals. Will they eventually replace doctors, completely changing healthcare and the way patients are diagnosed and treated?
It is an area that is evolving quickly. To answer the question, first it helps to define algorithms. In their most fundamental form, they are a step-by-step set of instructions for solving a problem or performing a task. For example, the directions someone gives you to find their favorite hotdog stand can be considered an algorithm. So is the manual for putting together a bicycle. Digital algorithms make all kinds of modern conveniences function today, including online searches, smartphones and global positioning systems (GPS). They help us choose movies on Netflix and make purchase recommendations based on our online shopping history.
In healthcare, medical algorithms can aid in diagnosis, treatment, short- and long-term prognosis, and care management. They can help remove uncertainty and improve the efficiency, accuracy and speed of decisions that doctors and other healthcare professionals make.
Medical algorithms can be developed by high-tech companies, software firms and medical groups themselves. Medical Algorithms Company, known as Medal, is one of the largest providers of algorithms to doctors and medical practices, selling more than 24,000 algorithmic products that are used via the web, through iOS and Android apps and by application programming interfaces. These products allow users to access data across various systems and devices. The information that makes algorithms run is drawn from numerous sources, including patients’ electronic medical records (EMRs), data inputted from medical and scientific research and surveys, big data analysis of medical histories and outcomes, and proprietary research done by high-tech and software firms, universities and hospitals.
Dr. Dana Edelson, of the University of Chicago Medical Center, developed an algorithmic analytics tool she calls eCart, which stands for “electronic cardiac arrest risk triage.” It uses more than 30 variables from the health records of hospitalized patients to calculate in real time their probability of having an adverse event such as cardiac arrest. “ECart can predict death in the next 24 hours with an accuracy of 94 percent,” Edelson told Crain’s Chicago Business. Doctors can follow patients’ eCart readouts on a computer dashboard and receive alerts about high-risk patients via text message or advisory notices in patients’ EMRs. Doctors and hospitals across the country now use eCart, including the Chicago area’s NorthShore University Health System and Alexian Brothers Medical Center.
A Diagnosis Tool as Useful as a Stethoscope
Indeed, some have called medical algorithms the 21st century stethoscope. While these algorithms, which are a type of artificial intelligence (AI), are generally still in their early stages of development for healthcare, they have the potential to dramatically improve the accuracy and speed of diagnosis, serving as another tool for doctors, much like stethoscopes, x-rays and MRIs. In fact, algorithms may soon outperform doctors and radiologists in quickly spotting disease or injury on scanned images. The University of California, Irvine, has completed a study using an algorithm that can detect cerebral hemorrhages (bleeding in the brain) much faster than radiologists can review patient scans, allowing doctors to send victims of strokes or trauma to lifesaving surgery much sooner.
In another example, doctor-researchers at Google used an algorithm to detect symptoms of diabetic retinopathy (DR), a complication found in nearly half of diabetics that can lead to blindness. The algorithm detected DR as successfully as highly trained ophthalmologists. The hope is that using such an algorithm will help screen patients in need, particularly in parts of the country or the world that lack medical specialists capable of detecting the disease.
In the United Kingdom, Oxford University has teamed up with high-tech company Ultromics to develop algorithms that improve the diagnosis of coronary heart disease by more than 90 percent. The algorithm-driven software extracts more than 80,000 data points from an echocardiogram (the traditional way to image the heart using sound waves) to produce new, objective and thorough metrics that vastly improve diagnosis. Ultromics expects to launch the product for cardiologists’ use in 2019.
Increasing Safety and Easing Doctors’ Burdens
Not only could algorithms help doctors diagnose more accurately in general, they could also help avoid misdiagnosis which, common in the United States, is dangerous and can result in malpractice lawsuits. The Washington Post reports that 20 percent of patients who sought a second opinion at the prestigious Mayo Clinic in Rochester, Minn., were misdiagnosed by their primary care physicians. With 10,000 diseases and just 200 to 300 symptoms, accurate diagnosis is difficult, according to Mark L. Graber, a senior fellow at the research institute RTI International and founder of the Society to Improve Diagnosis in Medicine. Five percent of adults seeking outpatient care—12 million people—are misdiagnosed annually, a trend that is expected to worsen, potentially jeopardizing patient safety and system efficacy.
Medical algorithms can help doctors reverse that trend by providing them with accurate diagnostic information and doing it speedily for faster treatment.
Algorithms also can ease doctors’ burdens in other ways:
- Manage higher workloads—Faced with a growing population, aging of baby boomers, and retirement of practicing doctors, the United States could have a shortage of 100,000 doctors by 2030. Those still practicing will most likely grapple with heavier patient loads and a nonstop daily work pace. Using algorithms would help them diagnose that much quicker and more accurately.
- Reduce administrative work and increase patient time—The advent of EMRs has made less efficient paper-based patient files largely obsolete, but it has also meant that doctors now use half their time to complete EMR documentation requirements to fulfill state regulations, insurance demands and quality control initiatives. Algorithm-driven bots can handle repetitive administrative tasks like maintaining EMRs, freeing doctors to focus on face-to-face patient interaction instead of the computer screen.
- Access to research and clinical trials—Doctors are lifelong learners and keeping up with the latest literature is a professional requirement. But there are 28 million biomedical papers on the National Institutes of Health Pubmed.com alone, making it extremely challenging and time-consuming to find all the pertinent information for a case. Algorithms can comb through such databases in seconds. They also can search through thousands of clinical trials to locate the one that could offer a cutting-edge experimental treatment for a patient much faster than a doctor could by searching the web or making phone calls.
Algorithms as a Team Member
IBM didn’t stop working on algorithms 20 years ago when Deep Blue triumphed over Kasparov. The company moved on from chess to the TV game show Jeopardy!, showing that its latest AI iteration, Watson, could understand natural language and process algorithms well enough to beat the show’s best human champions. Now IBM has rechristened Watson as “Doctor” Watson with its aim to improve healthcare.
Among the things that IBM Watson can now do for doctors and hospitals is identify appropriate cancer treatments when time is of the essence. In one study, IBM Watson took just 10 minutes to analyze a brain-cancer patient’s full genome and recommend a treatment plan. It took human experts 160 hours to make a similar plan. Watson used its language-processing abilities to analyze 23 million medical-journal articles, government listings of clinical trials and other data as part of the process. Both Watson and the clinicians identified gene mutations that suggested potentially beneficial drugs and trials. The patient’s team, which included an oncologist, a neuro-oncologist and bioinformaticians, however, identified mutations in two genes that led them to recommend a trial that targeted both with a combinatorial drug therapy, according to IEEE Spectrum. Watson failed to recommend that trial. It would have been the patient’s best chance for survival had he lived. So, even though Watson’s solution came far more rapidly, it was not the best.
Rather than viewing this example as doctor versus machine, it shows how algorithms complement doctors’ work. Ultimately, it appears that IBM’s Watson and other algorithm-driven technology can be valuable tools for doctors, instead of their replacements. What’s more, there is no algorithm for bedside manner. In the world of medicine, empathy and understanding are uniquely human traits—ones that calm and reassure a patient faced with a daunting diagnosis, no matter how accurate or speedily obtained it is.