Machine learning coupled with the explosion of data offers the very real possibility of addressing the most intractable problems in healthcare. For the first time, data is helping to answer healthcare’s toughest questions.
Like Google’s Research Director Peter Norvig, I too believe it’s not better algorithms that are fueling our advancement. Instead, it’s the surge in data sources and the innate ability of machine learning to automatically apply complex calculations to vast stores of data and derive rules that help us to understand the correlations, patterns and predictions within the data.
Take health costs, for example.
In 2015, our country spent $3.2 trillion a year on healthcare, or $9,990 per person, and the actuaries at the Centers for Medicare & Medicaid Services project the total to rise to nearly $5.6 trillion in 2025. Research by the Kaiser Family Foundation has shown that nearly half of our nation’s healthcare spending – $1.6 trillion – is driven by the top five percent of the population with the highest spending. More than 20 percent of healthcare costs are attributed to the top one percent.
There are many important ways machine learning plus the proliferation of data can help address health costs, and even more on the horizon.
Hospital readmissions have always been costly to the health system and to patients. The passage of the Affordable Care Act made them even more expensive by imposing financial penalties on hospitals with relatively higher rates of Medicare readmissions.
The industry responded by applying machine learning to diverse and plentiful data sources to identify the factors most likely to influence a Medicare patient’s hospital readmission risk – and therefore where to focus money and care management resources to prevent readmission. As many as 100 or more variables can contribute to a patient’s risk of readmission, and machine learning allows us to identify the variables that have the most weight and are most easily addressed. For example, patients who do not speak the same language as their healthcare provider have a much greater risk of readmission as do those who live more than 10 miles from a pharmacy.
Machine Learning Remedies ‘Anything and Everything’
Returning to the one percenters – the individuals who account for 20+ percent of total healthcare costs. These people are among the sickest people in the healthcare system, and many of them are likely to die within the next 12 months. At that stage of illness, typically many types of treatment are tried.
Earlier this month, advocates of federal ‘Right to Try’ legislation met with Vice President Mike Pence and earned White House support for giving terminally ill patients the right to try investigational medicines that have not yet received full approval by the Federal Drug Administration. It’s easy to understand why there is support for this legislation. Any parent would want our health system to try anything and everything rather than see his or her child die. Yet, this kind of federal legislation could lead to even more spending on the top one percent, and very well may not improve health outcomes or patient satisfaction.
Machine Learning Improves Effectiveness in Marketing & Healthcare
Rather than ‘trying anything and everything’ and hoping for a good outcome, machine learning plus data helps us identify which therapies work for which people and which therapies don’t, just as it’s done for marketing and advertising.
Department store magnate John Wannaker (1838-1922) is considered by some to be a pioneer in marketing, opening one of the first and most successful department stores in the United States. He is perhaps best known for his quip, “Half the money I spend on advertising is wasted; the trouble is I don’t know which half.” Today, Google can tell you by using machine learning.
For 40 years, tamoxifen has been used to treat breast cancers that are hormone-receptor positive. Research has shown tamoxifen lowers the risk of breast cancer recurrence, breast cancer in the opposite breast, and death from breast cancer. As you might expect, there are side effects, including menopausal symptoms, and more rarely, endometrial cancer and blood clots in the lungs.
Of all the breast cancer patients taking tamoxifen, we know it works for approximately 70 percent; for the other 30 percent, they experience costly side effects without the therapeutic benefits. A combination of various genetic factors, family history and personal characteristics explain the drug’s ineffectiveness.
Machine Learning Leads to Personalized Medicine
Machine learning offers us the possibility to provide breast cancer patients personalized medicine in the form of determining in advance who will benefit from treatment with tamoxifen. The healthcare provider would have this information upfront and could prescribe the appropriate treatment immediately, saving money and providing more effective care.
The foundation of personalized medicine – medicine that’s specifically tailored to address a patient’s condition while minimizing costly and unintended side effects – is machine learning. Even more exciting to me is the very real potential for machine learning to identify in advance a person at risk and prescribe how to prevent the person from becoming ill. I am confident that we have just begun to scratch the surface of personalized care, and the future is very bright.