Ningishzida - lord of the good tree

The Healthcare Analytics Blog

This Blog is where I write about my favorite kind of data analytics, Clinical and Healthcare analytics. Of course, all kinds of data and information analysis are fair game for discussion in keeping with my growing fascinations with Mathematics, Statistics and Data Mining.

You are on Blog page 1, You can go Forward to page two.

A "New" Career in Analytics

The inaugural contribution, where I talk about the things that brought my focus and energies onto the problem of clinical information over-load and the uncontrollable accumulation of semi-structured and inaccessible data.

I spent a lot of my clinical career collecting information that was never going to be looked at again. There are many reasons for that exercise in futility, and my take on it is this.

Historically, the information was merely to prompt the provider's memory of what was done at the previous patient visit. By browsing many medical records made decades before I started practice, I have seen many patient visits reduced to a half a line of text. Many times, these were not even complete sentences. My observations included many examples that I think show a more obsessive data collection for the memory function has not always improved patient care.

I recall many patients’ complaints, upon transferring to my practice, that caregivers they had previously consulted rarely knew why their patient was there or, clearly, didn't know where they had left off on the last visit. Rechecks were often (for the patient at least) confused dialogues sometimes culminating in the recheck actually being completed satisfactorily and sometimes not.

Now, I would be the first to admit that more than a few patients have the impression that putting down another physician to their new provider is an effective way of getting into the good graces of the a provider. However, having found myself at times demonstrating the same doctor-patient forgetfulness testifies to some truth being involved during the buttering up of the new doc. The sad reality is that it is not uncommon for providers to also stand on the bones of other providers to be elevated in the esteem of their patients. I blame the competitive underpinnings of the business of medicine for the professional climate. All that said, the likelihood is that the medical record is not overly effective beyond it’s current role of self-serving evidence for the provider and proof to payers that the billed service did take place.

The allusion to payers introduces the reader to the charting requirements of the Medicare Administrators for Evaluation and Management (E & M) documentation as a standard. This is the requirement for record keeping where medicare and medicaid patients but has evolved into the defacto standard for all clinical records. Though records have more structure and are more complete than ever it is difficult to show greater usefulness to the clinical process or any increased likelihood to be reviewed by the provider for subsequent visits. It is, in my assessment, the rare physician who regularly revisits and updates a care plan outside of academic settings. This includes the planned review of medications and therapeutics for continued appropriateness. The presures of time and production make this habit more a more good theory than standard practice.

Attempting to Make the Medical Record More Useful

I had the great fortune a number of years ago to attend a keynote address by Dr. Larry Weed at an informatics conference in Burlington VT. Prof. Weed is renowned as the developer of the concept of the problem oriented medical record on which I and most practitioners cut my clinical teeth. The presentation he gave was a revisiting of the old needs and philosophy which was the driving force behind the original project and the introduction of new paradigms of medicine existing at the end of the twentieth century. The new need was to make the Medical Record more useful by making it more relevant and accessible.

The retrospective was a critical analysis (in a frank voice) of evolutionary course of the record. Two central issues were obvious. Firstly the medical profession continues to propagate the myth that all needed knowledge can be held in the competent physicians head. This issue lays the foundation for the second issue. That the legal profession considers their role to police this impossible task and their biggest tool is our own records which are rarely consulted or cross-checked a year after it is "committed to paper". In setting the bar at an impossible height for mere mortals to hurdle in the face of the huge body of learning which is twenty first century medicine, we lay the foundation for our own failure.

Specialization has been a solution to the knowledge gap but is also on weak underpinnings with the poor communication of ideas and knowledge on specific cases amongst the sundry caregivers. Should care be found to span the borders between care systems or hospital domains, the communication problem is multiplied. A study by the Institute of Medicine (NIH) in 1998 summarized the crisis. A patient admitted to hospital in the U.S. had a greater chance of dying of a disease of complications not directly related to the reason for admission, than the admission disease or disorder itself. Further this risk was unrelated to provider competence, where they trained or even the number of conferences they attended. It related to poor communication and records access.

As Professor Weed stated, not only is a better way of keeping records and a better way of making them accessible long overdue, but the tools which are at hand to supplement our human memories also have in then a promise of even more effectiveness in making connections about disease and the human condition. It used to be that the greats of medicine were distinguished by their ability to make the connections the average physician misses. Computational power can turn us all into Oslers and Hughlings-Jacksons if we make room in our Science and our Art for these new tools.

This is the place of Analytics and the Networks of Medicine. We can't afford to let another person perish because of parochial thinking and proprietary impediments to the flow of information. The Profession of Medicine must set itself above the mere commerce of healthcare which masquerades as a competitive market which offers patient choices. This kind competition breeds a drive to segment the market to the goal of branding over substance. It used to be called snake oil. We all need to communicate about our patients and to learn.

Do We Simply Need More Experts?

(Added August 2014)
Having finished my rant in true Dennis Miller fashion, it behooves me to be constructive now. When I first had contact with Professor Weed, he was beginning a project and start-up company called Problem-Knowledge Couplers (PKC)® now part of Sharecare. Professor Weed has influenced the thinking of many students and practitioners over his more than 45 years of teaching. I have been no exception, but consider that, notwithstanding the standard continuing medical education required of us all, it’s easy to get buried in practice with little opportunity for deep thought about how one is doing what one is doing.

Much has been written over the years since my first exposure my first exposure to these ideas, and the tools of PKC are becoming part of a logical extension to the electronic health or medical record (EMR/HER). See for example “The Use of Problem-Knowledge Couplers in a Primary Care Practice” Perm J. 2010 Spring; 14(1): 47–50. and "WebMD Founder, Dr. Oz Buy Clinical Decision Support Company" .

The idea begins with a conceptual list called in the trade so to speak differential diagnosis. The computer provides a medium to review the possible options and interactions on the list among its members in the context of the online accessible world medical care and research experience. In doing so an opportunity to derive relative probabilities can be carried out. Further, the real-time access to the world literature allows the most current context for decision support.

Where should all this all logically lead? Well, let’s survey what else is out there in medical decision support and where it hasn’t gotten to yet. Wikipedia which is my favorite first go to source refers to medical decision support systems as expert systems.. I do take some exception with this Wikipedia author’s referring to the tool in this way as I think it is not only inaccurate but misses an important implied distinction about whom is doing the thinking. The article redeems itself with the quote of Robert Hayward [of the Centre for Health Evidence], "Clinical Decision Support systems link health observations with health knowledge to influence health choices by clinicians for improved health care". This I think is a better characterization of the genre.

Considering the place of clinical decision support tools, a great stumbling block to adoption is one of the responsibility for the decision. When a medical text (read: “authority”) is consulted does the text make the decision? Better still, one of the most ground breaking decision tools introduced over the past 30 years I believe is the Cochrane Library home for the Cochrane Reviews, but it remained (and still does) that the decision maker is the professional who is weighing the evidence and not the evidence itself no matter what form it is presented in. I reserve the right to digress on this later, but suffice to say that this was the message that I heard from Professor Weed in 1996.

With that in mind, I will proceed with my, non-exhaustive, list of the decision support offerings (for later comment).

All existing EHR software include a basic set of what I call aid to memory tools. It is important for me to point out I am not minimizing this basic facilities. This are now no longer “nice to haves” but, rather, absolute minimum requirements of any system. These would include, a drug interaction check, a risk factor reminder list, individual patient allergy and adverse reaction list, medication and problem lists (active and inactive) and follow-up reminder list. What these as yet are not routinely, fully integrated among each other and especially the patients, own, symptom and diagnoses inventory – again active and inactive. This will very soon be an absolute requirement for an electronic medical record which will set it above a mere specialized word processor database.

The next generation decision support tools are those which derive information and “tertiary measures” from facts about the individual’s condition or state of health and well being. These should be things usually done for providers not usually necessarily by providers. Ad hoc real-time epidemiologic assessments of a patients disease context and statistical analysis of signs and symptoms by their clustering is time as well as space.

The various acuity and trauma scores (APACHE etc) were developed so that an arriving patient’s state could be immediately portrayed and followed as interventions process. These sets of information start to allude to the information overload which Professor Weed held up for our appreciation. Add to this the inherent weakness of the individual human brain as well as the team of brains we are a part of, in communicating and keeping at the fore relevant information real-time. We all obviously need a dedicated and learned assistant constantly looking over each of our shoulders.

Consider Wikipedia’s “lay-list” of, See Also:

I find this list interesting in that support for the making of a diagnosis is rather over-represented here. It is often not well enough appreciated that diagnosis is an evolving process which only begins with a differential diagnostic list subject to a constant revision process. Often the search for better decision tools is a quest to eliminate all uncertainty. This is, by the nature of the universe and of clinical problem solving, unlikely to be realized. That is why the clinician needs constant reminder to review the clinical picture.

See also the following, for a picture of the current state of the art:

Remembering the Zebras while Keeping Track of the Horses

I had a teacher in medical school who gave to my clinical group a two part caution which has remained with me ever since. “Isn’t it so that if you hear hoof beats outside your window, you think first of a horse, not of a zebra. However, never forget the zebra!”

I am constantly exploring how to remember to figure the zebra into the equation as well as keeping track of my horses. More to come. The media (and particularly television) leaves us with the sense that "Real" doctors find zebras (like House does) whereas the challenge most commonly (and probably most importantly) faced by clinicians, is keeping track of the horses.

Creating 'Derived Information' from Patient Data'

(Added September 30 2014)
Data is only useful when it becomes information. Patient data can be information in many ways. The clinical note should describe how or what the patient feels (subjective data) and whithout further processing, is information about the patient when interpreted by the clinician. For some data to be useful is must be compared in a descriptive simple time analysis and interpreted by the clinician and for other data it is desirable to apply extensive transformation and statistical analysis to provide the best information in context, about a patient. This later information is based on what I call derived data as opposed to raw data or simple measurements like blood pressure. The majority of quantitative data in a clinical record is raw unprocessed data. It is productive to consider all the ways we can use the data in medical records to create new information about specific diseases, about the care process, about a patients r esponse compared to some peer group, about a provider and his practice and so much more.

Payer and care networks collect and aggregate data about provider effectiveness and compliance with established standards. The provider should be given all the tools to generate this kind of information and more, on demand. Every provider should have the means to find outliers and trends within her own practice and to compare them to her peers by demography, geography, genetics and culture. And, she should be enabled to ask why. The more the record can be smart enough to extract this information easily the smarter the providers become and the more they can think of their aggregate data, and that of their peers doing for them and their patients. Knowledge feeds on itself.