Stat counter

View My Stats

Sunday, December 13, 2015

Predicting the future has it limits

The IOM has issued it's report on diagnostic errors which I believe highlights highlights major issues in the diagnostic realm. I have not had a chance to review it completely but I have been involved with the Society for Improved Diagnosis in Medicine (SIDM). I have heard much of discussions leading up to the report and have devoted much thought to problems in diagnosis.

One colleague in Pathology whose thinking and writings have influenced me greatly in this realm is Dr. Elliott Foucar who for more than 20 years highlighted the limitations of anatomic pathology is the diagnostic sphere. One key concept that he highlighted is that pathology made a transition in the 20th century from being a taxonomic activity to being engaged in the assessment of risk. I now have come to realize that the remainder of medicine has followed in the transition without having grasped that the transition has occurred. The SIDM and the IOM have delved into this world without explicitly recognizing the relevance of this concept to the act of making a diagnosis.

What Dr. Foucar has written about is how anatomic pathology at it origins focused on the autopsy with the sole purpose of identifying characteristics of disease which caused someone's death. An autopsy of a woman with a breast mass might identify a tumor with a specific histology and evidence of tumors elsewhere. This information was subsequently used to make predictions about live individuals who shared certain characteristics with someone already dead. For example, a live woman might present with a ulcerated 5 cm breast mass which when samples demonstrated a tumor with great similarity to what had already been observed in patients on the autopsy table. In contrast, another live woman might present with a similar mass which was showed a cyst and inflammatory cells. Histology was an amazingly powerful tool which allowed for clear separation of these distinct entities. What anatomic histology functioned as here is as a taxonomic tool, separating a being inflammatory process from a clearly malignant one. Diagnosing a ulcerated breast cancer as an infected cyst would be a clear diagnostic error, misidentifying one entity with a almost certain malignant course as a distinct entity with a benign clinical course.

Fast forward to the 21st century. Now instead of attempting to distinguish distinct clinical entities with distinct pathologies, anatomic pathology tries to separate infinite grades of dysplasias from forms of actual cancer. There is no clear line which can separate low grade dysplasias from high grade dysplasias; scoring systems are riddled with subjectivity and inconsistencies, whether the tissues are breasts, prostates, or pigmented lesions of skin, The life expectancy of those diagnosed with dysplastic lesions tend to be identical to those diagnosed with malignancies.

The transition which has occurred is one where the pathologist has moved from being a taxonomist to being an actuary. Actuaries use information to identify people or groups who are at risk for certain outcomes. Their functions are essential for certain financial entities such as insurance companies which need to put money aside to cover costs which might occur when particular events occur such as floods, fire, or illness.  They assign specific odds for such events in particular populations.

In many respects, diagnoses are always actuarial statements. Someone who presents with chest pain, shortness of breath, an abnormal EKG, and positive cardiac muscle enzymes almost certainly is having an acute myocardial infarction. This carries with it the immediate morbidity but the diagnosis has a prediction about the future built into it as well. Those affected are at near immediate risk for death. This is a prediction one does not need to wait around for months or years to assess. The acute MI diagnosis is an actuarial statement of short term outcomes.

A host of tools within the cardiovascular realm have purported to extend the ability to predict future outcomes much further into the future. We can image the coronary circulation using a variety of tools. However, moving the time frame for the predictions creates all sorts of problems, both practical and conceptually. For a patient who has clear cardiac symptoms and functional impairment short of what qualifies as an acute MI, imaging tools may allow for predictions in the non-acute time frame. However, there are efforts to push out the time windows for predictions out into years or even decades into the future. Conceptually, do "abnormal" findings observed from studies of asymptomatic individuals represent distinct diseases or diagnoses or do they represent risk factors? What represents an abnormality if the finding is common within the population studied?

Our ability to peer inside of people now is unprecedented. Whether we peer using ultrasound, x-rays, CAT scans, MRI scans, or PET scans, we can literally visualize and see what could not previously be seen unless someone was cut open on the operating table. A recent story published in the New York Times discussed new observations regarding the use of scans looking for calcification of coronary arteries. The absence of calcium seem to predictive of much lower risk of cardiac events in a 10 year time line than what conventional risk calculators might suggest. However, questions were raised because of the nearly 5% incidental findings on scan. The reality is we don't know what the range of normal findings might look like. Similarly, the ability to do ultrasound on thyroids has led to an epidemic of thyroid biopsies. We are about to embark on widespread CAT scans to detect lung cancers, most of which are likely to be indolent.

As we peer at and into into people with tools of increasing resolution and sensitivity, we will see things. Whether the things we observe will represent variations with little or not relevance to future outcomes, or whether they will represent abnormalities which place a given patient at risk for a bad outcomes will be asked. The temptation will always be there to view anything out of the ordinary as a reason to act. As the tools get more sensitive, the opportunities to observe variations will become overwhelming. When will we recommend acting and when will we write variations off as simply variations? Like screenings at the airport and those working at homeland security, we will have to deal with rare signals within a sea of noise. Ideally, we can become like actuaries, if we can accumulate enough solid data to allow us to make data driven decisions. Until then, we will make our decisions driven by fear and complacency.

What the electronic medical record has revealed about health care delivery

I have and am living through the deployment of the electronic medical record (EHR) within the health systems I am employed by. It is something that will continue to evolve for the remainder of my practice lifetime and beyond. Stories regarding EHR deployment have tended to focus either on the pain involving end users or the promise of implementation. There is no question whatsoever that EHR deployment has been disruptive and has had unintended consequences on the delivery of health care services.  I would also suggest that it has been revealing of aspects and elements of health care delivery of which we had very limited awareness.

That the movement of operations from paper based to electronic information systems ends up revealing novel aspects of the business of health care should not come as a surprise. Virtually all industries which have made this transition have undergone similar realizations. Archaic legacy processes which were hidden and not measurable become revealed and quantified. Value added activities become separable from non-value added actions. Management, which was previously driven by intuitive decision making can move toward data driven decisions.

Medicine is not one industry and the habit we have to lump all health care activities into one bucket has really obscured the nature of the challenges we face, especially relating to the movement to digital information systems.  My perspective is by its very nature narrow, colored by the fact that I am physician and a sub specialty doctor to boot. However, even within the narrow universe of physician perspectives, the differences within even this narrow scope dwarf the similarities. Some physicians are primarily technicians who view their jobs are doing very narrowly defined tasks. These might include replacement of a joint, removal of a specific tumor, insertion of a stent, or repair of a hernia in a given person at a given time. For a physician who performs this type of service, the information which comes out of the EHR or for them to enter into the EHR is very focused on their specific value added service.

In contrast, other physicians manage chronic conditions over extended periods of time. Some of these doctors (or APPs) may be generalists while some may be specialists. The role of the EHR under these circumstances is completely different from the role it plays for the technician providers, both from the perspective of what any given physician needs to extract to make effective decisions and from the perspective of what needs to be entered into EHRs at any given moment of time. If a patient has multiple chronic medical problems which all need to be simultaneously managed, the quality of the inputs and the ability to digest information over time ultimately determines the quality of the care delivered and the outcomes achieved.

If one asks practitioners who primary focus is acute interventions what value is added by the EHR and how this can be improved, you will get very different answers than if you ask physicians whose primary focus is long term care, including disease prevention and chronic disease management. The answers will be so different that you can't help but conclude that these groups are engaged if distinctly different industries. That is my point. We are not deploying the EHR in one industry, but in reality multiple different industries lumped inappropriately under one roof. To believe the EHR needs of an orthopedic surgeon who primarily replaces knees will overlap significantly with an allergist who primarily manages chronic asthma, or a primary care physician who manages many problems over time is delusional. The EHR is ideally an information management tool and both the types and perhaps more importantly the flows of information required by different practitioners is completely different. Both airline pilots and diesel mechanics are involved in the transportation industry but they require very different information sets to do their jobs. Similarly, whether one is dealing with an acute with an acute issue or a chronic issue in health care delivery will mean that what information one needs to address a particular problem or problems will be very different.

Obscuring this reality has been the dramatic change in medicine which has occurred in the past 50-75 years. When my parents were growing up, for most people one went to a physician when one was very sick and the the measure of success was whether you were dead within days to at most weeks. Information systems were not essential. There were not many drugs that patients took chronically and the outcome measurement was relatively simple; your were either dead or not dead. Decisions regarding optimal management were primarily intuitive and not dependent upon sophisticated information systems.

Fast forward to the year 2015 and much had changed. While the healthcare delivery system still deals with acute and life threatening issues, many if not most encounters with health systems and providers deal with chronic issues managed over long term time lines of years to decades. However, we are still locked into a very similar delivery model as was used 75 years ago. Furthermore, while we pay lip service to evidence based medicine, we still operate primarily on an intuitive level. We have no choice since within the outpatient delivery realm, our information systems are so dismal that we are left with few other options. We measure money in and out and don't do this so well.

The culture where we value intuitive master clinicians creates another challenge in that when software engineers were developing the current first generation EHR tools. There were few clinicians who actually had substantial understanding of what information they needed to make decisions. Little effort was made to engage front line clinicians in the design of EHRs but when it was made, few front line personnel had any idea of what to tell EHR developers. In the absence of a significant cadre of clinicians who practiced in an information driven manner, there was no one to effectively guide the EHR development process away from where it went. We created billing justification tools because that is where we could define where information made a difference.

For the most part, measurements of things which reflect activities that might add value to patient's lives were few and far between. Efforts are being made to figure out how to measure what is important and measure whether resources invested to deliver specific services are worth the cost. However, the going is slow. Standardization and measurement is hard work and meets resistance for a host of reasons. One particular problem is that activities which are the financial foundations of practice may in fact generate very limited value to patients. What we as physicians may view as valuable may not be viewed so positively by the recipients of our care. We are very likely to experience surprises as physician centric universe moves more and more toward a patient centric one.