Topic of Panel: Should We Have A Unique Individual Identifier For Health Care, And What Are The Alternatives For Such An Identifier?

DR. CHUTE: My name is Chris Chute. I am professor of medical informatics and associate professor of epidemiology at the Mayo Foundation.

.......

Data security, which is fundamentally the issue to address appropriate confidentiality concerns, I believe is possible when it is targeted against patient data rather than the identifier and is clearly required. I should say without hesitation that the Mayo Foundation is profoundly and unwaveringly committed to the notions of patient confidentiality and data security.

However, if we think from the perspective of a patient, and we are all patients at some point in our lives, when we engage the health care process, the issues that we want to know is, is there anything wrong with me or my loved ones, and if so, what does it mean, these findings, and finally, what can we do about it.

If we pause for a moment, the answer to these and related questions derives from a body of experience with patients in the past. Historically, this has been something akin to anecdote and folklore, and it is in the modern era that we are beginning to overcome these ad hoc mechanisms of understanding how we can take better care of patients, by analyzing repositories of patient information appropriately linked.

Nevertheless, American health care, the way it is practiced today, is profoundly fragmented. This is not fundamentally bad, but it does emphasize a decentralization of specialties and services. A given patient can engage health services at a number of providers even within a single episode. Furthermore, patients, as they move throughout society, are highly mobile and can encounter coverage options with multiple providers at different points in their careers and lives.

Information transfer between and among these phases of health care are somewhat inefficient in our current mechanism. Furthermore, clinical decisions and research, which are premised upon the linkage of information, is often significantly incomplete and potentially biased and might lead to wrong conclusions.

Consider the delivery of services in laboratories, x-rays or other clinical studies. These can be generated from external facilities and resources to a major health care provider that had their own mechanisms of enumeration and numbering. That record or information may or may not be linked back to the right patient at the right time in the right context. Worse, that information may be merged due to insufficient linkage and insufficient identification with patients with a similar name. Anecdotally, I can attest that when I was a medical student, there was a large volume of patient findings and reports that pertained to my father in my own medical record. Obviously, nobody acted on the fact that I might have had prostate cancer at 19, but it was a serious problem.

An example at the broader level of research illustrates a fragmentation of data not only within single patients but across populations of patients. If we take the relationships, which is only recently understood, between papilloma virus and subsequent cervical cancer, these events are often and findings are often separated by decades. They are often identified from multiple providers with different geographic history, and the detailed information is rarely transferred.

The natural history of these chronic conditions then can be significantly misunderstood for a lack of appropriate data linkage. Similarly, our understanding of the incidence and prevalence of disease can be significantly underreported either in a public health context or in an academic context when we try to understand the impact of disease in our society and efforts to improve their treatment and management.

The population-based research, as it is engaged today, offers an opportunity to overcome incomplete and biased experiences that result from single episode hospital-based studies. In a hospital series which tries to elucidate a natural history or an impact of disease, they are often reliant exclusively on immediate episode related information. They are often confronted with a very skewed representation of patients that are the consequence of socioeconomic referral to a given health care facility and are fundamentally non-representative of the population at large if only by the virtue of their being sick.

An alternative, of course, is population-based research which is not reliant upon hospitalization. An example of that that I will expand upon in our own experience is the 30-year old Rochester Epidemiology Project which is fundamentally dependent upon linkage of information across different providers that pertains over time to a single patient. The Rochester Epidemiology Project was begun in 1966. It has generated over 1,000 peer reviewed publications which we hope are widely regarded as a contribution to our understanding of health outcomes and disease. It integrates the health experience across providers for persons in Olmstead County, Minnesota.

The logistics of the Rochester Epidemiology Project, for which I am somewhat directly responsible, have de facto surrounded the master patient index approach. We are intimately familiar with the shortcomings and the inaccuracies that are associated with trying to link patient data from various providers without the common basis. Reliance on names, date of birth, sundry identification numbers including the Social Security number are, in our experience, fraught with inaccuracy and error. We engage in this process so that we can develop a master diagnostic and procedure index of patients in the population-based area, and we maintain that information in a highly secure data format. We have severely restricted access to that information, and any communication of the data associated with those indices are encrypted.

The benefits of such a study allow us to recognize notions of hospital biases. We were the first to demonstrate that the natural history of disease can differ profoundly in a hospital series from that in a community. We were among those to first recognize that hospital series patients tend to be sicker, a famous Berkson(?) bias as it is called in the epidemiologic textbooks.

We were the first to recognize, as an example, that multiple sclerosis as a disease, which is information that we published in the 1950s, had double the prevalence in populations than had been previously anticipated and had a vastly improved prognosis than that which was expected, with persons and population based gainfully employed decades after the onset of the diagnosis, which was in contradistinction to the hospital experience where obviously ill patients were seen disproportionately. We were the first resource to unequivocally demonstrate distance referral bias where patients referred from distant sites, by and large, tend to be more healthy than patients seen within the community. Again, this is a function of data linkage, and it can subsequently distort the quality of care metrics that are presented as they are associated with given medical centers.

One of the questions that I think we might consider is patient data as a valued resource. Mayo Foundation and the Mayo Clinic, we believe, are widely regarded as outstanding health care institutions. Why is that? Why did we emerge from the cornfields of southern Minnesota, which is fundamentally where we are? We have a heritage of organizing, preserving and linking patient data. This dates back to a common medical record structure in 1907 that was fundamentally enhanced by the introduction of a common identifier to facilitate linkage of all patient events within the activities of the foundation over, I might add, a century.

We have a significant and longstanding commitment and resource expenditure on indexing this information so that we can learn more about disease and natural history and reincorporate that, as was illustrated by Codman(?) at the turn of the century, into our health care practice to continuously improve what we do and how we take care of patients. Our record during this 91 years of continuous usage of indexing and linking identifiers on 5.5 million patients, wherein we conduct more than 4,000 studies per year overall, is that we have had no breach of confidentiality traceable to the use of patient information in research.

We recognize and do not dismiss the special concern associated with confidentiality. Clearly, we have, as I said, an unwavering commitment to maintain and preserve the confidentiality of all our patients. However, our overriding concern is the welfare of our patients, and to the extent that we engage in research to look at outcomes and management, we do use patient information and patient record materials to study this process.

The question of whether anonymity in research databases is an adequate response to the identifier problem is an interesting one. Fundamentally, researchers can and should deal with information that is anonymous, that has no patient identifiers within it. The real issue is how is that dataset generated for the researcher. At some point, to make the research database credible and unbiased, a linkage of patient events that precede an episode under study is required. That very linkage implies the existence even if deep within a computer system, of some sort of consistent patient identifier. That identifier can be discarded once the linkage has been undertaken, but the absence of an identifier absolutely would preclude that linkage in the first place and would essentially make impossible the practical conduct of outcomes research, disease natural history, and treatment response analysis........