Imagine a world where you can walk into a hospital, submit a urine and blood sample, and be told 20 minutes later that you not only have a particular type of ear infection, but also a 50% chance of developing diabetes in the next ten years. Such is the promise of “personalized medicine,” in which the development of molecular diagnostics and therapies enables medical care to be tailored to individual bodies.
This vision of personalized medicine is predicated upon the collection and analysis of “big data”: the use of computerized tools and databases to find hidden patterns within large data sets. Data, proponents of personalized medicine say, will lead to more precise and objective ways of diagnosing and treating disease, and will help to redefine heterogeneous categories like cancer and obesity.
But to some extent, medical care has always been personalized. Throughout the 20th century, physicians practiced patient-centered care and used the “art” of clinical judgment in order to avoid one-size-fits-all approaches to medicine (see Richard Tutton’s paper in Social Science & Medicine). What then, is new about modern forms of personalized medicine, and why should we pay attention to them?
In the quest for personalized medicine, the post-genomic or “omics” fields such as genomics, proteomics, and metabolomics take center stage. Researchers work with gigabyte upon gigabyte of data, sitting in front of computers to analyze the patterns and statistical relationships contained within molecular information.
But alongside this focus on data, alternative narratives and ways of engaging with biology are emerging. Life, for the post-genomic sciences, is made up of systems and networks (see, for example, an article on network medicine). Linear notions of disease causation, as well as the reductionist metaphors of the informational code of life, are out. Notions of complexity, probability, variability, and dynamic systems are in.
Here metabolomics, the post-genomic study of the molecules and processes that make up metabolism, appears as a potentially radical new way of examining biology and disease. By analyzing the molecular composition of urine, blood, and tissue samples, metabolomics explores how metabolism changes in response to disease, diet, and environment.
To study metabolism, researchers use a variety of biochemical and statistical technologies to develop “biomarkers,” quantifiable biological entities which can be used to understand biological processes or diagnose diseases like cancer or diabetes. Metabolomics, they claim, provides a compelling alternative to genomics, in that it shows the dynamic outcome of the interactions between genes, metabolic pathways, and the environment.
My research for the past four years has centered on this emerging field of metabolomics, and on how it signals the changing practices and objects of biomedicine. Consequently, this post is concerned with how metabolomics—as a case study for the dynamics of big data—is being used to develop “personalized medicine.” Moreover, it explores what this means for the kinds of persons and populations that are implicated in or “made up” (see Ian Hacking’s essay “Making Up People”) in personalized medicine research.
Finding persons in metabolomics research
For now, let’s leave these questions and turn to a recent meeting I had with Professor Jeremy Nicholson, the head of the Department of Surgery and Cancer at Imperial College London, and also the founding father of one of the leading metabolomics[i] laboratories in the world. We are meeting in the sixth-floor space of the Computational and Systems Medicine laboratory, a sprawling set of offices and experimental facilities housed primarily within Imperial’s South Kensington campus.
It’s been a while since I’ve been back here, and I’m blown away by the changes that have occurred since the laboratory began developing the “Phenome Centre,” a multi-million pound private-public partnership to carry out broad-scale metabolic phenotyping and epidemiological studies. The Centre, which opened at Hammersmith hospital in mid-2013, is a “legacy” of the London 2012 Olympic Games.
As the story goes, Jeremy Nicholson helped to transform the biochemical machinery used to test for illegal performance-enhancing drugs into a cutting edge laboratory. The resulting Centre—a collaboration between a number of UK universities, the analytical technology companies Waters Corporation and Bruker Biospin, and the government-run Medical Research Council (MRC) and National Institute for Health Research (NIHR)—is a testament to the increasing size and strength of the field of metabolomics in the UK and beyond[ii]. Following from these developments, the sixth-floor space has been transformed from a slightly haphazard arrangement of desks and machines, to a modern facility neatly partitioned by glass walls and digital displays.
Professor Nicholson warmly welcomes me into his office, and when I complement him on the new space, he jokingly says, “It’s because of all of this darn publicity, we had to make it look good. I miss the old lab. It felt more like a place where you could actually do science.” Exuding energy and excitement, he begins to tell me about the range of developments occurring across his laboratory.
There are efforts to improve histology with the “iKnife,” a surgical knife that can analyse the biochemical signature of the smoke generated during electrocautery. There are clinical trials on diagnostic methods for breast and colon cancer. There are pilot projects to develop biomarkers for Alzheimer’s disease, and even to improve intensive care treatments for hypoxia by sending teams of intensive care practitioners up Mount Everest, where they conduct experiments on themselves and other volunteers at high altitude.
In the back of my mind, I am aware that such developments are enveloped in the hype and promise of post-genomic technologies. Even so, I can’t help but become caught up in this vision of the technological future of biomedicine. These projects are all housed under the theoretical umbrella of personalized medicine, in which metabolomics will provide more accurate, specific, and cost-effective biomedical solutions. Professor Nicholson has managed to unite a diverse array of medical professionals and research specialties around one cutting-edge, post-genomic vision.
Central to this vision of personalized medicine is the notion of the “patient journey,” where metabolomics technologies are integrated into hospital settings to model and monitor patients as they move through various parts of medical care. Also central is the notion of “molecular epidemiology,” where metabolomics technologies enable longitudinal studies to molecularly characterize states of health and disease within populations. Together, metabolomics researchers claim, the combined notions of the patient journey and molecular phenotyping are “two sides of the same coin” that will revolutionize biomedicine.
However, as I listen to Professor Nicholson, I ask myself: What are the end goals of this research? How are researchers using new practices and forms of knowledge to investigate personalized and population medicine? Moreover, what types of persons are we dealing with here, in this search for personalized medicine?
The person in anthropology
Since the 1930s, anthropology—spurred by the work of Marcel Mauss—has been interested in notions of the person and personhood, as well as related notions such as the individual, the self, and the body. Early in this history, as British social anthropology investigated the cultures of colonized tribes and peoples, it defined the person as a nexus of social positions or relationships, rather than as an amalgamation of individual characteristics. But when anthropologists began to look at different cultural contexts, the notion of the person as occupying a social position began to break down. In more recent anthropological thought—espoused, for example, in Marilyn Strathern’s work in Papua New Guinea or Viveiros de Castro’s work in the Amazon—persons emerged as constantly contested, negotiated, and changing entities.
But persons, as such work also teaches us, are shaped by the shifting dynamics of knowledge and power. As Michel Foucault’s foundational work on techniques of power demonstrates, the rise of modern institutions and technologies for data collection promote the exercise of power and control at the level of populations (through processes of “governmentality”) and individuals (through “technologies of the self”). Such techniques of power have the fundamental ability to alter subjectivities—as scholars have noted with the concept of “biological citizenship”—and to influence how individuals and collectives live and experience their lives.
This brief digression into anthropological notions of persons provides an important starting point for thinking through the efforts—as embodied by the Phenome Centre—to develop personalized medicine with metabolomics. Biomedicine and data-intensive research are establishing new kinds and quantities of relationships between persons and things. Increasingly, for example, nature and culture are intertwined: notions like kinship, life and death, and time are redefined through technologies like assisted reproduction, organ transplantation, and cell culture, respectively. Perhaps, then, with changes in the practices and technologies of biology, notions of persons are also being conceived of and “made up” differently.
Persons as multiple, molecular, and…relational?
With this in mind, I want to turn back briefly to some of the happenings within the metabolomics laboratory. During my fieldwork several years prior to the aforementioned meeting with Professor Nicholson, I observed a PhD student carrying out a metabolomics experiment to determine more “personalized” ways of diagnosing pancreatitis. Pancreatitis, which often manifests as a sudden and serious inflammation of the pancreas, is easily confused with other conditions that give rise to abdominal pain. This student’s work sought to determine metabolic markers that could detect whether a given patient had pancreatitis upon hospital admission.
One day, I observed a meeting between the student and her supervisor, in which she discussed the difficulties she was having analyzing her data. The data had come from the urine and blood of patients hospitalized for pancreatitis. The student explained, with frustration, that her experiments did not show a clear difference between patients diagnosed with pancreatitis and patients diagnosed with other conditions. Instead, her results showed three separate clusters of patients, each of which contained a mixture of pancreatitis and non-pancreatitis patients.
As the student discussed the perplexing result with her supervisor, it became clear that each of the three clusters corresponded to the location within the hospital in which each patient had been treated. One cluster was from patients in the intensive care unit (ITU). Another was from patients in the accident and emergency (A&E). And the last was from patients admitted into the gastroenterology ward.
Seeing these results, the student and her supervisor acknowledged that the experiment had been impacted by differences in sampling and experimental conditions. But the results also inspired the supervisor to begin to rethink the category of pancreatitis, and what it meant for the health of the patients in the experiment. He said, “[The data] reinforces the fact that…diseases are not homogenous or easy to separate…And it shows that people behave differently when they get really sick. And actually, we can see that, we can measure it.”
Together, the student and the supervisor used this moment of uncertainty to rethink diseases as molecularly diverse entities. According to their reasoning, the similarities and differences between patients were not due to clinical symptoms, but rather to a series of complex molecular characteristics reflecting the severity of the patients’ disease, the hospital environment in which they were being treated, the drugs they were taking or receiving, the composition of their gut bacteria, and their overall state of well-being.
Observing two metabolomics researchers reasoning through their work, it became clear to me that the persons (patients) in this experiment were multiple and molecular entities. They were defined according to groups of biochemicals or sets of statistical relationships, which reflected the body’s response to the environment in space and time.
Ultimately, though metabolomics researchers draw on the standard demographic categories like sex, gender, ethnicity, and age in their research, their subsequent understandings and definitions of persons do not necessary correspond to these same categories. As metabolomics research looks for more personalized approaches to disease diagnosis and treatment, it makes up persons as metabolic patterns and relationships. This is reminiscent, as a colleague described, to the modern possibility of tracking down internet users not by their IP addresses, but by their “personalized footprint” generated by website visits and Google searches. Persons, then, are not only constituted by information that they contain, but also by the emergent patterns of relationships that they embody. What, then, does this mean for how persons are understood and treated with personalized medicine?
As metabolomics researchers seek to collect information and maximize their knowledge of disease, they recognize that it is impossible to measure and model everything about the world. Despite this, they try to capture the interactions between many things—genes, gut bacteria, environment—in order to provide the best possible approximation. There is a seductive element of complexity and holism here, one that captivated me during my meeting with Professor Nicholson. With the ability to measure life—with precision, at the molecular level, as the outcome of the organism in its environment—the possibilities for metabolomics seem limitless.
In conclusion, I want to return to the original premise for this post: what it means to deliver personalized medicine with metabolomics research, and what types of persons emerge in the process. Perhaps, as metabolomics research shows, persons are defined relationally, but according to statistical relationships rather than social relationships. Perhaps personalized medicine is as much about making up new persons, as it is about attributing ever finer metabolic resolution to individuals. Perhaps in the quest to define persons, new elements of personhood are foregrounded, while others retreat from view.
What is at stake, I want to suggest, is the very nature of measurement in metabolomics research. What types of information or relationships is metabolomics capable of capturing? Or, put otherwise, what types of information or relationships is metabolomics not capable of measuring?
Here, I am reminded of a conversation I had once with a clinician doing a short research project in the metabolomics laboratory, while still working shifts in the intensive care unit. The clinician described a challenging case, where a woman suffering from a drug overdose was in need of a liver transplant. In such cases, clinicians relied on a statistical test called Model of End-Stage Liver Disease (MELD)[iii], which calculated a patient’s probability of dying over a certain period of time without a liver transplant. Though the clinician and his team had suspected that the woman’s prognosis was poor, her MELD score was not high enough to qualify her for a liver transplant. Contrary to the results of the test, the clinician drew on what he referred to as “gut feeling”–tacit knowledge of the woman’s skin color, pallor, quality of respiration, and general change over time—to sense that the woman was more sick than her MELD score indicated.
A few weeks later, as the clinician and his team predicted, the woman’s health took a turn for the worse. Subsequently, she scored high enough on the statistical test (MELD) to qualify for and receive a life-saving liver transplant. This case, said the clinician, typified how medical practitioners often knew that patients were in need of a transplant before the statistical models indicated so. They carried out “their own pattern recognition” based on their judgments and impressions of patients, and by applying statistical models repeatedly to patients until they displayed a result that would qualify them for a liver transplant. What struck me about this case was not only that the patient’s care was inherently individualized, but also that her treatment relied on a series of factors that were beyond quantification with standard clinical methods. This type of care, though not based on data or post-genomic technologies, was highly “personalized.” It entailed interpretive care attuned to the individual status and needs of the patient. For me, this raised key questions about whether metabolomics’ experiments, as they configured patients into a series of measurable and seemingly objective variables, were able to capture those elements of health—or personhood—that facilitated the best medical care for the individual.
As this and the other cases I have discussed show, metabolomics entails shifts not only in the practices for carrying out personalized medicine, but also in the definitions and categories of persons being treated. What is still unclear—and what I am grappling with in my own work—is what such persons might be, and what consequences they might have for the provision of medical care in the 21st century.
Ultimately, is the type of personalized heralded by metabolomics addressing the type of person we tend to think of anthropologically, as existing relationally within society? Can this type of medicine, with its basis in data and statistics, ever be personalized in the way we want it to be, or in a way that leads to more attentive and higher quality care for patients?
Nadine Levin is a Research Fellow at the University of Exeter, where she is exploring how Open Access and Open Data policies affect the practice of post-genomic research. She completed her DPhil in 2013 in Anthropology at Oxford University, with a dissertation that explored how researchers in the field of metabolomics create, analyze, and use data to make claims about metabolism and health. Her current research explores how metabolism is being configured in relation to big data, and what consequences this has for biomedicine.
[i] The terms “metabolomics” and “metabonomics” are used interchangeably to describe the field, though both terms have a slightly different history. While metabolomics is said to focus on the characterization of metabolism at the cellular or organ level, and metabonomics is said to focus on the combined effects of the environment, disease processes, and gut bacteria at the organismal level, both terms entail overlapping sets of practices and ideas. The term metabolomics is attributed to researchers working on model plants and organisms at the University of Manchester, while the term metabonomics is attributed to researchers working on nuclear magnetic resonance (NMR) and bodily fluids at Imperial College London.
[ii] The past decade has seen a surge of developments in metabolomics research. Though the field was formally inaugurated into the scientific literature in the late 1990s, it has grown significantly in the last five years with the establishment of several major funding initiatives and industrial partnerships worldwide. In the United States, the National Institutes of Health (NIH) recently initiated an investment of more than $50 million USD in metabolomics research through an NIH Common Fund. In parallel, a number of research groups with expertise in metabolomics have emerged throughout the United Kingdom, resulting in the development not only of the MRC-NIHR Phenome Centre, but also of the comprehensive MetaboLights database for metabolomics experiments and associated information at the European Bioinformatics Institute (EBI). In Canada, developments in metabolomics have been supported through investments in the Metabolomics Innovation Centre, a nationally funded core facility supporting metabolomics activities across a range of Canadian universities, as well as the Human Metabolome Project, an attempt to catalogue the range of metabolites present in human beings.
[iii] Model of End-Stage Liver Disease (MELD) is a linear regression on a combination of clinical measures of serum bilirubin, creatinine, and prothrombin time. It was implemented in the National Health System (NHS) in the United Kingdom in 2002 to replace the older system for assessing need for transplantation, which entailed a combination of the Child-Turcotte-Pugh (CTP) score and overall waiting time for a liver transplant. As a clinician explained, “Basically everyone just does it on a website now, you just type in your results and you get a score. And the higher the score, with a max of forty, the more likely a patient is to die.”