Georgia Tech Professor Eva Lee discusses Personalized medicine

How OR/MS is helping a revolutionary improvement in healthcare.

Personalized Medicine

By Douglas A. Samuelson

Imagine a way to deliver medical care that takes into account genetics, personal lifestyles, different effects of medications on different people, medication interactions, nutrition and compliance with medical directives. Such a system could radically decrease preventable deaths in the medical system, most of which result from unusual reactions to properly prescribed and administered medications or from interactions of multiple medications [4, 12]. It could also make patients’ lives more pleasant by reducing uncomfortable side effects and guiding them to more effective rehabilitation. Wouldn’t that be exciting? And wouldn’t it be even more exciting if OR/MS could help make this happen?

Personalized Medicine Evea Lee

Eva Lee

Well, it’s actually here. It’s called “personalized medicine,” and several researchers are testing the method and the computer software in clinical settings. One of the analysts in the forefront of this revolution is Eva Lee [2], a university professor and a prominent and often-honored member of INFORMS. Lee won an Edelman Award in 2007, along with Memorial Sloan Kettering Cancer Institute, for her work on better placement, in real time, of brachytherapy (embedded radiation sources) seeds to treat prostate cancer. She was a finalist/laureate for another Edelman Award in 2012, with the Centers for Disease Control and Prevention (CDC), for work on public health and emergency preparedness, and in 2015, with Grady Health System and Emory University School of Medicine for transforming the emergency department workflow and patient care. Lee serves on the White House panel that recommends responses to pandemics and other health emergencies. In short, she’s a recognized leader in applying OR/MS to healthcare.

Recently, Lee, a professor in the Stewart School of Industrial and Systems Engineering and director of the Center for Operations Research in Medicine and HealthCare at Georgia Tech, has been working on software to track all aspects of a patient’s data and develop integrated medical recommendations. She now has preliminary results from several studies applying her software in clinical settings.

Promising Early Results

In a study with Columbia University, following 2,000 women with high genetic risk of breast cancer, lifestyle changes greatly reduced the risk. Only 11 of the study subjects actually developed cancer during the 10-year study. “We can find genomic, physiological and behavioral factors that act as triggers,” she says, “and those turn out to make a big difference.

“We include all the biology of the cancer,” she adds. “Knowing how it is triggered, how it spreads and what biochemicals in the body support or resist its spread can provide vital guidance. And the biology is precisely about the individual.”

For instance, in prostate adenoma, the most common form of prostate cancer, “the usual treatment plan is specifically designed for the state of the tumor but assumes the prostate is homogeneous,” she says. “But the tumor is distributed unevenly. We can zoom into key places and avoid the bladder if the tumor is not close to it, concentrating more radiation (or drug) where the cancer cells are.”

Cervical cancers are typically about 35 percent fatal once they spread into surrounding tissue. In a clinical trial using personalized medicine, patients had “excellent outcomes and survival rate,” Lee says, “and we might have 5 percent failure in the longer term. We couldn’t achieve this without the personalized approach” [8, 9].

“Chronic diseases are typically not entirely genetic,” Lee continues. “We’re looking at diabetes. The problem is that there are multiple comorbidities – obesity, hypertension, heart disease. The numerous types of drugs taken make patients more prone to adverse reactions to medications. And with diabetes and obesity, we really don’t know what’s in the foods we eat. There are lots of medications and hormones in meat.”

It should also be noted that high fructose corn syrup and artificial sweeteners are present in many foods, not just desserts, with numerous, poorly identified effects.

As far as obesity is concerned, “We’ve identified a gene or two, but we don’t know which came first, the gene or the other factors in the body,” Lee says.

For instance, there are known changes in the composition of the large population of bacteria in the intestine that are essential to digesting food. Foods appear to influence the growth of some strains of these bacteria. At least one virus, Adenovirus 36, is apparently associated with obesity [12]. And then there are behavioral factors.

“I have a colleague who lost 45 pounds just by walking 10 miles a day, with no change of diet,” Lee notes. “It’s sometimes quite easy to produce a big effect by a change of lifestyle.”

Infectious Diseases

Of course, personalized medicine applies to acute infectious diseases. “We worked with a major pharmaceutical manufacturer on a malaria vaccine,” Lee recounts. “Unfortunately, it was only effective for about 23 percent of the subjects. I wouldn’t say it’s useless, just highly personalized, and we need to know more about the other factors.”

Analyzing genomic information, Lee and her vaccinology colleagues identified which patients would benefit from it and how it would benefit them. The result is very important to world-wide trials set for next year [1, 5]. Now, Lee and her team are trying to design next-generation medications tailored to individuals [11].

“We had better luck advising on containing Zika,” Lee reports. “My team has had pretty good predictions for Puerto Rico on disease trend after sufficient cases were reported. We recommended some containment strategies, and they worked pretty well through the summer. But local public health declared success prematurely. It wasn’t totally contained when they thought it was.  We did have a lot of pregnant women getting tested, and those numbers are reliable. It is a serious health burden.”

Just having people paying attention to the threat helped, but, according to Lee, the virus maybe making a comeback around Florida and possibly Houston. “It’s hard to control because of the amount of summer travel,” Lee says.

And that was before hurricanes Harvey, Irma and Maria devastated the Houston area, southern Florida, and Puerto Rico and other Caribbean islands, respectively. Refugee movements and relocations to temporary shelter, contamination of water and breakdowns of public services are highly likely to multiply the difficulties.

“The flood and the extreme damp and polluted environment make for the perfect breeding conditions.” Lee says of the virus. “We worked with Harris County (Houston area) since May 2016. They did an excellent job. Now, this is a true challenge as they race to eliminate the mosquitos and puddle of water where the eggs are laid.”

Puerto Rico suffered unprecedented destruction in terms of flooding, sewage and chemical waste in the wake of Hurricane Maria. “It worries me greatly about the level of health issues, short term and long term,” Lee states.

Allergies greatly complicate predictions of response to infection, Lee explains, adding that people have very different capabilities and responses, which makes diagnosis difficult. Is it an allergy? A cold? Flu? Early anthrax symptom? Allergies affect the ability of the immune system to repel infections. Sensitivities to medication vary widely. Is there a genetic component to reactions to infection? For example, the tendency to have a depressive reaction to flu is well known, and it may be because it had survival value during historic flu epidemics [12].

Mental Health

Personalized Medicine

Research indicates that 20 percent to 30 percent of college students are depressed. Image © Cathy Yeulet |

Professor Lee describes mental health as “the most dangerous diseases we’re facing,” adding that 20 percent to 30 percent of college students are depressed. “These are educated, capable people,” she says. “Some of them may be able to get medical help and receive prescribed medication to control their mood. There aren’t enough psychiatrists in universities to help them. We don’t call it endemic because of social stigma, and we don’t call it pandemic because we don’t know how to treat it.

“We know depression is highly heritable,” Lee continues. “The mother, in particular, affects the child both genetically and behaviorally. But the strong stigma makes it hard to get people to seek treatment. If kids can get help, they can be helped to feel better. This is why patient privacy is so important; people won’t talk about it, so they don’t get the right treatment.”

Social media appears to aggravate the problem. “People make comparisons to others and feel disappointed,” Lee observes. “It’s natural for us to be affected by what others say and what they have. It is human-nature. It’s like an advertisement. The majority of people, perhaps 70 percent, respond to ads. Maybe as few as 1 percent are never affected. What makes them resilient? We don’t know. We need to. This will be very useful to mitigate the social influences on depression.”

Mental illness is also related to terrorism, as many of the attacks are “lone wolf” events, often perpetrated by people who have recently discontinued strong psychotropic medications without anyone monitoring them. Clearly, social media plays a role here, too, as many of the events involve little communication that can be detected by law enforcement. Other contributing factors are stigma and the necessity of having care available quickly for those who need it – and not just for mental illness.

As detailed in Homeland Security Presidential Directive-21, community resilience and individual resilience are important elements of national health security, and, according to Lee, the topic of resilience deserves more coverage in public debate and media commentary about healthcare.

Speaking of resilience, Professor Lee says she and her colleagues have had some successful results with Alzheimer’s, making comparisons of brain physical, chemistry and cognitive tasks performance between patients and people without the disease. “The content of the information they [patients] process seems to play an important role,” she says [3,7].

The Analytics Component

How revolutionary is personalized medicine? “We can’t keep saying we have missing data problems,” Lee says. “Missing data remains a challenge, but we can minimize its effects by tailoring it to the specific problems at hand. Our models make good enough use of whatever data we do have to yield good predictions that indicate good or bad courses of treatment. Now the data problem is that everyone’s measurement is different. People are not homogeneous.”

According to Lee, we’re still learning how to take differences among people into account in assessing which effects of treatment and lifestyle have significant effects on outcomes. “Imputation is no longer all that useful,” she says. “And in electronic medical records, we can’t worry so much about what’s missing, because assessing the quality and completeness of the information we have is all part of the data capture routine.”

That is, the absence of some data can be quite informative, such as blood tests not performed because the patient was unconscious and could not give consent. Unconsciousness is very different than choosing not to comply.

Professor Lee and her team are now testing out their first “personalized” models integrating pharmacokinetics (the way in which medications travel through the body and affect it) and pharmacodynamics (the way the body affects the medications). In medical-speak short form, such models are called “PK-PD” models.

“This model is really powerful in terms of prediction. It connects the individual’s response to the actual dosage of drug taken,” Lee explains. “Testing out behavior and compliance or noncompliance alone is not sufficient; the model tells us so much more. The response (of the body) is actually reflecting and capturing the behavior of how the patient takes his or her drug. The model is forgiving and gives good estimates.”

Using the PK-PD relationship she establishes for each patient, Lee optimizes the treatment plan specifically to the individual. The results, implemented at Grady Health Systems based in Atlanta, are remarkable. “Doctors spend less time experimenting with which drug type and dosing should be used,” Lee says. “The method also enables doctors to use higher doses early without great concern about adverse effects, thereby getting better outcomes, shortening the course of treatment and lowering the overall dose of medication. The patients maintain good, healthy blood glucose level. As an extra bonus, the cost and amount of drug used is lower. Cost is not in our equation, but we always use less resource when we optimize.”

She adds that, as far as she and her colleagues know, this was the first successful incorporation of personalized patient data into a mathematical predictive model directing treatment [10]. “This kind of predictive analytics, integrating PK and PD on a patient-by-patient level, has been an aspiration of medical practice for nearly a century,” she stresses. “But now we’re actually doing it, with real patients, with good results. This is a major breakthrough.”

As their work continues, Lee and her colleagues are eagerly awaiting the response of the medical practice community.

“OR/MS supports this personalized medicine approach because we’re used to taking advantage of whatever data streams we have,” Lee adds. “O.R. is not the same as data science. O.R. people are not used to collecting data. Some of us do, but we don’t just collect data. Collecting data and data management are part of data science, but we do more. We build (mathematical) models from data and reach solutions. We influence decisions, policies, strategies, many layers of impact. The key is to implement O.R. on actual patients in the clinical setting, with the O.R. analysts embedded within the medical care team.”


Personalized medicine, using OR/MS models to tailor medical guidance and treatment to highly specific individual characteristics of patients, is a revolutionary improvement in medical and preventive care. Models can now integrate effects of medications, interactions, patients’ behavior, genetics, diet and other factors to develop recommendations for the most effective course of action. There is a tremendous potential benefit here for all of mankind, and a major opportunity for OR/MS analysts who can work closely within medical teams in clinical settings.

Douglas A. Samuelson, D.Sc., is president and chief scientist of InfoLogix, Inc., a small R&D and consulting company in Annandale, Va. In recent years, he has specialized mostly in healthcare outcomes and policy, emergency preparedness and response, and various defense and national security issues, especially cybersecurity. He is a contributing editor and columnist for OR/MS Today.


  1. Jen Christensen, CNN report, 2017, “First malaria vaccine to be widely tested in Africa next year,”, April 24.
  2. Peter Horner, 2015, “The Most Interesting Person in the (O.R.) World,” OR/MS Today, April.
  3. W.T. Hu, K. Watts, P. Tailor, T. Nguyen, J. Howell, R. Lee, N. Seyfried, M. Gearing, C, Hales, A. Levey, J. Lah and E.K. Lee, 2016,” CSF complement 3 and factor H are staging biomarkers in Alzheimer’s disease,” Alzheimer’s Disease Neuro-Imaging Initiative, Acta Neuropathologica Communications, Vol. 4, No. 1:14. doi: 10.1186/s40478-016-0277-8.
  4. Brent James and Douglas A. Samuelson, 2013, “Change We Can Live With: Building the Data Capabilities and Analytics to Make Critical Improvements in Patient Safety and Wellness,” OR/MS Today, June.
  5. Dmitri Kazmin, Helder I. Nakaya, Eva K. Lee, et al., 2017, “Systems analysis of protective immune responses to RTS, S malaria vaccination in humans,” Proceedings of the National Academy of Sciences, Vol. 114, No.9, pp. 2,425–2,430.
  6. Eva K. Lee, Amanda F. Mejia, Tal Senior and James Jose, 2010, “Improving patient safety through medical alert management: an automated decision tool to reduce alert fatigue,” American Medical Informatics Association Annual Symposium Proceedings, pp. 417-421.
  7. Eva K. Lee, T.L. Wu, F. Goldstein and A. Levey, 2012, “Predictive model for early detection of mild cognitive impairment and Alzheimer’s disease,” Optimization and Data Analysis in Biomedical Informatics, Fields Institute Communications, Vol. 63, pp. 83-97.
  8. Eva K. Lee, Fan Yuan, Alistair Templeton, Rui Yao, Krystyna Kiel and James C.H. Chu, 2013, “Biological planning for high-dose rate brachytherapy: Application to cervical cancer treatment,” Interfaces, The Daniel H. Wagner Prize for Excellence in Operations Research Practice, No. 43, No. 5, pp. 462-476.
  9. Eva K. Lee, Yu Cao, Fan Yuan, Alistair Templeton, Rui Yao, James C.H. Chu, 2016, “Optimizing tumor control probability in PET-guided radiation therapy treatment – Application to HDR cervical cancer,” 58th annual meeting of the American Association of Physicists in Medicine.
  10. Eva K. Lee, Xin Wei, Fran Baker, Michael Wright and Alexander Quarshie, 2017, “Outcome-Driven Personalized Treatment Design for Managing Diabetes,” to appear in Interfaces Wagner Prize Issue 2018.
  11. Helder I. Nakaya, Thomas Hagan, M. Kwissad, E.K. Lee, et al., 2015, “Systems analysis of immunity to influenza vaccination across multiple years and in diverse populations reveals shared molecular signatures,” Immunity, Vol. 43, No. 6, pp.1186-1198, December.
  12. Douglas A. Samuelson, 2008, “Can We Detect ‘The Coming Plague’?: How Emerging Health Threats Are Sneaking Up on Us,” OR/MS Today, June.
  13., 2007.