The Promise and Challenges of Health Analytics

On February 24 I attended a workshop in MIT on the Future of Health Analytics.  The event was sponsored by MIT Connection Science, a recently organized research initiative aimed at leveraging data science to quantify and analyze human behaviors, and to leverage the new insights thus obtained in key societal applications, including healthcare, transportation and finance.  Connection Science, - with which I’m affiliated as a Fellow, - was founded by Media Lab professor Alex “Sandy” Pentland.  He’s the author of several books, including the recently published Social Physics: How Good Ideas Spread.  

I’ve worked with Pentland for the past several years, and have previously written about his research on Reinventing Society in the Wake of Big Data, as well as his work with the World Economic Forum and others on the creation of trust frameworks for the sharing and protection of personal data.  In his opening remarks at the workshop, Pentland talked about Big Data and Health.  The little data breadcrumbs that we leave behind as we move around in the world can now be reality mined to help us better understand our behaviors and thus improve our lives and health.  He discussed several applications of big data to health, based on research at his MIT Human Dynamics Lab, as well as startups he’s involved in.

Following Pentland, Dr. Dennis Ausiello talked about Quantitative Human Phenotyping and the opportunities it offers to transform the practice of medicine.  Ausiello is professor of medicine at the Harvard Medical School, Chief Emeritus of Medicine at Massachusetts General Hospital (MGH), and director and co-founder of the Center for Assessment Technology and Continuous Health (CATCH), a joint MGH-MIT initiative aimed at finding new ways of measuring the human condition, i.e., phenotypes.

What are phenotypes?  According to the National Human Genome Research Institute: “A phenotype is an individual’s observable traits, such as height, eye color, and blood type.  The genetic contribution to the phenotype is called the genotype.  Some traits are largely determined by the genotype, while other traits are largely determined by environmental factors.”  Wikipedia offers a broader definition:  “A phenotype is the composite of an organism’s observable characteristics or traits,… [it] results from the expression of an organism’s genes as well as the influence of environmental factors and the interactions between the two.”  It further adds that whereas the genotype is an organism’s full hereditary information, the phenotype is what that heredity actually produces.

In a recent paper, Ausiello notes that health analytics has the potential to become the next frontier in medicine, driven by the confluence of three key revolutions: 

• The digital revolution, including all the various mobile devices and associated software and apps that enable us to collect huge amounts of information about an individual’s actual behavior.
• The genetic revolution, which has identified very large numbers of genetic variations that contribute to human traits and to the risk of disease.
• The data revolution, which is enabling us to collect, store and analyze extremely large and disparate data sets relevant to human health, yielding insights on individual patients as well as entire populations.

But, achieving this next frontier requires major changes in how medicine is generally practiced.  “Our current system of delivering health care is episodic and reactive.  That is, patients see their physicians largely at regularly scheduled intervals (typically 1 year) and/or when symptoms appear or worsen… During their episodic appointments, the methods physicians use to assess disease in our patients have largely remained the same for decades.  The typical office visit will document the patient’s medical history and symptoms since the last visit (usually several months ago); parameters such as weight, heart rate, blood pressure, and respiratory rate; a physical examination; and perhaps standard blood tests such as general chemistry values and a lipid panel.  Whereas specialized blood diagnostics and imaging studies are used to investigate specific diagnoses, the most commonly used measures reflect an uneasy balance between cost, the time constraints of an office visit, and the ability to detect significant changes in health status.” 

In addition, despite dramatic advances in genomics, the use of genetic information to improve medical diagnoses and treatments remains a challenge.  As an example, Ausiello cited predicting the risks of developing cardiovascular disease over a 10 year period.  Medicine can predict the risks of getting cardiovascular disease with a .787 probability using traditional risks factors, - including age, gender, smoking, cholesterol, and family history.  Adding a genetics-based risk score barely improves the probability to .788. 

How can that be?  How can we both talk about unprecedented advances in human genetics, yet their clinical translation to the diagnosis, prevention and treatment of something as common as cardiovascular disease lags so far behind?   “This is likely due to several factors,” explains Ausiello, “including the sheer number of genetic loci - [that is, the specific location of a gene, DNA sequence, or position on a chromosome] - that can contribute to individual risk, and perhaps most critically, the importance of largely unmeasured environmental and behavioral factors that influence risk of important conditions such as obesity, type 2 diabetes, cardiovascular disease, and cancer.” 

For chronic diseases, around 60% of the risk factors might be due to the kind of behavioral and environmental everyday factors that our digital revolution now enables us to measure.   CATCH is investigating several such new types of phenotype measurements: 

Continuous measurements:  “[M]illions of implanted cardiac devices such as pacemakers and implantable cardioverter-defibrillators provide ready access to beat-by-beat heart rate data.  Continuous glucose monitors (typically accessed via a small sensor in the interstitial space) have long been used primarily to guide dosing of automated insulin infusion pumps, but may yield insights into the dynamics of glucose regulation.” 

Behavioral measurements:  “Behaviors such as exercise, diet, and medication adherence make significant contributions to several diseases ranging from cancer to diabetes and cardiovascular disease.  Data from wearable devices such as activity monitors (such as digital pedometers), and wrist-based monitors (eg, devices that measure skin galvanic response as a reflection of stress) could provide insight into individual behaviors as well as facilitate feedback.” 

Environment-responsive measurements:  These include diet and inhaled or ingested toxins.  “Exposures are typically accessed on rare occasions through survey instruments based on recall or blood or urine assays.  Although continuous measurement of environmental exposures may not be necessary (or feasible), enabling more facile, passive quantification of environmental exposures will create an important new data resource that can be integrated with genetic and clinical information.” 

Pathway-inspired measurements: “[A] wide variety of inflammatory cells and pathways are being studied in auto-inflammatory disease as well as common diseases such as type 2 diabetes and cardiovascular disease… Shifting pathway-based measurements from the province of research studies to point-of-care devices for use in physician offices or the home would represent a significant leap in making these measurements more continuous.” 

Integrative Analyses:  The human microbiota, for example, has been shown to be an important contributor to many chronic diseases.  “The community of approximately 10¹⁴ bacterial, archaeal, fungal, and viral cells or particles that reside on each individual constitutes the human microbiota…  Recent studies in normal volunteers and disease cohorts (such as those with inflammatory bowel disease, obesity, diabetes, or cardiovascular disease) are revealing how all of these genetic and environmental factors can shape what types of microbes are present, and their aggregate influence on human metabolism and immunity (and in some animal models, even behavior).” 

Realizing this next frontier of medicine requires important changes in the culture of both medical research and patient care.  “Scientific collaborations will need to convene a wider range of expertise than are traditionally sought, including device engineers, front-line physicians, geneticists, and experts in sociology and behavior.  Collection of these novel data types will require new approaches to data ownership and security that appropriately balance an individual’s control over use of their data with a permission and trust framework for secondary use of data in specific contexts.  Analyses must be focused on actionable insights and rendered visually to allow patients and their caregivers to understand the medical implications.”

And, perhaps most importantly, patients must be part of this new healthcare ecosystem.  “The traditional barriers between clinical care and clinical research must be replaced by a new model in which patients are at the center as fully informed participants, and individual wellness is pursued hand-in-hand with a spirit of inquiry.”

Comments

[Bill Tetzlaff]

David Patterson started some related wotk two years ago, with a NIH grant, to merge genomic data, with clinical recoreds system history, and cancer treatment options used, and data mine it. The idea is to better understand Cancer, and the treatment regeim success, based on personal history, medical history, and genomic data. I find it very intregueing. He is on a sabatical this year, so I have not run into him at UC Berkeley, and don't know what he is up to during the sabatical. I agree that this is an area with great potential.

As a Computer Scientist, I am embarassed, that all the doctors I talk to find that medical records systems are slowing them down. Partly, and I know because I watch them, they are poorly designed. Better HCI would help. The real value will come when data mining has been used to evaluate alternatives, and doctors can be given suggestions for treatment, and back them up with statistics on alternatives used on similar parients.

[Pasquale Di Cesare]

I think the real challenge is to get “we”, as educated individuals and society, possibly empowered by the analytics tools which technology will provide, realizing the need to really take ownership of our health, either because we want it or because the entire “sick-care” system is likely to hit a wall in terms of sustainability and even solvency. The system looks “sick” today, prevention is poorly supported, medical education is incredibly lacking fundamentals such as nutrition and lifestyle and costs are exploding. When we will realize the importance of being our own health advocates the pressure on the system to improve will be higher and we will stop delegating our health to pharmaceutical industry interests, lobbyists and too many individuals keeping the “status-quo” even between the best and most dedicated MD’s. Maybe only then the promises of the next frontier in medicine will be met. As often the case, it is rather a question of people rather than technology.