Where Industry Meets Innovation

  • Contact Us
  • sign in Sign In
  • Sign in with certificate
mit campus

Resources

Search News

  • View All
  • ILP News
  • MIT Research News
  • MIT Sloan Management Review
  • Technology Review
  • Startup Exchange

ILP Institute Insider

August 7, 2017

Transforming Decision Making with Analytics

Dimitris Bertsimas, a professor of operations research and an applied mathematician, has combined his work in analytics with computational power to make predictive and prescriptive algorithms. With a particular focus on healthcare, his modeling offers more personalized treatment and cost-saving measures.

Steve Calechman

Data holds the potential to predict the future. But without a framework, even the best information is merely quantity. Dimitris Bertsimas is a professor of operations research and an applied mathematician. His work is in analytics, taking old methods of quantification, like statistics, and combining them with computational power to make predictive and prescriptive algorithms. With a particular focus on healthcare, his modeling offers more personalized treatment and cost-saving measures. But the desire to make better decisions and maximize resources isn’t the stranglehold of medicine. It applies across industries like finance, transportation, aviation, energy, anything which involves human activity. “That’s the beauty of analytics,” he says.

The patient in 3D
Healthcare is a natural for analytics, but Bertsimas had particular motivation to target the field. His parents died in 2009, his mother from diabetes, his father from gastric cancer, and, with both, he saw the limitations of treatment. His mother had been dealing with her disease since Bertsimas was young, and he says that medicine had remained static. Doctors could offer good, generic advice about diet and exercise, but specifics were rare.

Out of that need, Bertsimas says that he developed a program, Life Analytics, which would know individuals’ body measurements and blood levels, along with lifestyle preferences. The database could decide what and when to eat, and when and how much to exercise, all based on personal makeup, since, as he says, two people, for example, metabolize the same banana differently. Delivered through a smart phone, the program also identifies restaurants in a person’s vicinity and recommends food from the menu to order.



Dimitris Bertsimas
Professor of Operations Research
Co-director,
Operations Research Center (ORC)

His father was diagnosed with Stage IV gastric cancer in 2007. Since it was inoperable, Bertsimas and his family visited various major-city hospitals, searching for chemotherapy advice and proposals. Seemingly, he says, each hospital only had experience with specific drugs; no overall therapeutic picture or means of comparison was available.

Bertsimas says that he designed a rudimentary, two-dimensional system, with toxicity on one axis, survivability on the other, offering the ability to compare benefits and tradeoffs. He then compiled, with his research group, a more extensive database of various cancers, looking at clinical trials for specific populations and the associated survivability rates and toxicity levels. It didn’t provide analysis, just data presented in a visual and accessible way. “I call this descriptive analytics,” he says.

Moving into the future
The next step was designing predictive analytics. The reality of drug trials is that most involve ones that have already been used individually or in different combinations – new ones receive little attention, Bertsimas says. With a data footprint that went back to the 1990s, he created a system that could look at new drugs and predict outcomes with high accuracy. This modeling would allow pharmaceutical companies to identify promising trials, while avoiding significant early investment costs, with savings ranging from $10-50 million, he says.


The final iteration has been developing presumptive analytics, taking what is already known, combining it with modeling, and being able to predict and propose the next 10 drug trials. The result is better delivery of care. In one example, Bertsimas says that his methods can add 5 months of life to a patient with gastric cancer. It’s not a cure, he says, but with an initial diagnosis of 10 months, a 50 percent improvement is significant.

Bertsimas says that from his work, his dream is to see more personalized care. In the current system, often two people with the same symptoms, but of different ages, will receive the same treatment course. His designed algorithms can take into account a person’s history and propose specific diagnoses, treatment and possible outcomes. Over the last 15 years, he’s used the approach in a Boston hospital with 1 millions patients. The initial results show “significant promise,” he says, and, over the next decade, he wants to see the medical industry incorporate algorithms that continually learn from data in order to propose more targeted recommendations.

Along with care, Bertsimas says medicine’s business side can be optimized. As of 2014, in the United States, healthcare costs approximately $3 trillion, 17.5 percent of the economy. But for the outlay, services and outcomes don’t match up, he says. Employers primarily pay most of the bill, and for a company of 5,000 employees, that could mean 12,000-13,000 people covered. Bertsimas says that his predictive models can help determine future costs, and his prescriptive models can design policies that maximize benefits within a budget. Companies he’s worked with have seen a 10-15 percent savings.

It’s not just with healthcare, and it’s not just savings that a company can experience. Bertsimas says that he partnered with a European entertainment company that sold music and movies. His analytics monitored sales and popularity of artists and titles and could then tailor the merchandise in specific stores. The result was a double-digit increase in sales. “That’s an example of where data can be a weapon,” he says.

In a clinical description, Bertsimas says his work uses data to build abstract models that lead to better decisions. In a field that’s still growing, his job, he says, is also to expand knowledge. He does that with his research, but he says that he mainly judges himself by the ability to teach his concepts. He works with doctoral students, and, as the next generation of scientists, he tries to convince them that more than impressing their colleagues, or even himself, they should work on projects that have an impact.

It’s a fundamental attitude of MIT, to make concepts work, which are widely employed and which advance what has been. With data, the previous model used statistics and human knowledge to give answers. It worked for a particular time, but the shortcoming was that disciplines acted on their own. Bertsimas has looked to take fields that were viewed separately, combine them, and add machine learning to increase speed and insight. “When you do that, new things happen,” he says.