Thursday, July 28, 2011

Genesis (Of The Statistical Medievalist)

It may seem strange to the reader that this post details the genesis of The Statistical Medievalist but I take as my model for this unorthodox temporal sequencing, as well as explicitly stealing the posts title from Have Gun Will Travel, my all time favorite TV series. In Episode 1 of the 6th and final season entitled Genesis, it is revealed how Paladin was transformed from a man with 'feet of clay' into a heroic champion for hire.

So along those lines, and in keeping with the spirit of the 'Knight Without Armour' it seems to be the right time to describe the genesis of The Statistical Medievalist.

Having been trained in medicine at the end of the last century, at a time when the medieval Guild system was in authority and in near complete control of the education of medical professionals, I am very much intellectually rooted in the times of the mage, the maven, and the master guildsman. However, in the pursuit of the science of medicine, and as someone with an undergraduate major in Mathematics, I have always been drawn to the meta-theoretic areas of medicine, i.e., the theory of medical practice including as I have said in introduction to the blog, biostatistics.

These then have been the two enduring lines of thought and exploration that I have traversed over my entire medical career. In this way, going back to my idolization of Paladin, I know what is too live in two worlds, such as he does in San Francisco (Dandified and Elegant Man of Letters) or the Gun For Hire (Solitary Warrior in pursuit of justice for his client).

In my own case, those worlds are the intellectual pursuit of mathematical truth, insight, or rigor in decision making existing in dynamic tension with the need for immediate and often dramatic action in pursuit of the best outcome for my 'client' usually with action being impelled with incomplete knowledge and understanding.

Thus, while I treasure the medieval values and power of the Medical Education Guild system, I find I must embrace fully and without contradiction the necessity to balance the medieval logical world view best epitomized by Aristotelian Syllogistic logic in contradistinction to the multi-valued 'fuzzy' logical systems of thought inherent in the modern quantum views of reality and implicit in the use of statistical methodologies for analysis of medical therapies.

Furthermore in another but related sense, I am attracted to the medieval view of medicine as I  live through the transformation of my profession. No longer am I a Master Guildsman who owns a unique job shop, I am being transformed into a well compensated technician laboring in an implicit medical assembly line in which statistical defined outputs and control limits are valued more than the unique outcomes of the individual patient. To better define and understand these transformational forces I would reference you gentle reader to the following graphic which illustrates the so-called Product-Process Matrix, a fundamental tool for understanding Process Architectures.

I had the experience of working as a Consulting Cardiologist at the now defunct Internal Medicine Branch of the Aeromedical Consultation Service at the USAF School of Aerospace Medicine, a job in which on a busy week, I would exhaustively evaluate, consul, and more often than not perform heart catheterizations on 4 active duty military aviators. Thus,  I understand at a visceral level the medieval job-shop practice environment. These 'clients' were all worth up to or more than $12 million dollars per copy to the USA military with respect to costs to replace them at the most important time in their careers, i.e., Major to Colonel ranks. Therefore, a highly individualized process was appropriate to their needs as well as the needs of the US military.

However, I also had the contemporaneous experience of working in an Urgent Care Clinic, in which on the heaviest day of my practice in that environment I evaluated, counseled, and treated (together with my 'team') 80 patient's in 16 hours of operations. This aberrancy was fueled by the need of the clinic to meet throughput demands driven by cash flow considerations. The 'on the hoof' value of the 'clients' was hard to judge, but certainly no one had invested multi-10's of millions of dollars in their training and upkeep. Their issues were usually more mundane than that of my military aviators, but honestly was there any humanistic justification for an almost 200:1 disparity in face time with your doctor?

It is in this sense of Medieval that I find my truest sense of despair with the transformation of my guild into a humanistically irrelevant set of processes designed to separate 'clients' or their 'payors' from the maximum amount of wealth with the minimal cost or thought possible.

And so it is in this other, darker, blacker, but more realistic sense that I am the Statistical Medievalist, an obsolescent guild master fighting a delaying but ultimately hopeless rear-guard action against relentless opponents wielding a metaphorical mace, statistical control theory.

The awful truth of statistical control theory, is that perfection lies at the 'golden' mean.  In this sense, mean-ing the greatest return for the least investment of gold. Immodestly as someone who believes he is operating well above the upper control limit, I know that those process improvement forces designed to move practitioners from below the lower control limit into the mean, must by the laws of statistics and reality also force me to drop below the upper control limit into the range of the average (meaning in my world view, thoughtless) practitioner.

So in conclusion and to be amplified at a later date, recall the talismanic power of  TANSTAAFL! 


Tuesday, July 26, 2011

The Case For Case Based Reasoning

Case-based reasoning has been formalized for purposes of computer reasoning as a four-step process[1]:
  1. Retrieve: Given a target problem, retrieve cases from memory that are relevant to solving it. A case consists of a problem, its solution, and, typically, annotations about how the solution was derived.  
  2. Reuse: Map the solution from the previous case to the target problem. This may involve adapting the solution as needed to fit the new situation.
  3. Revise: Having mapped the previous solution to the target situation, test the new solution in the real world (or a simulation) and, if necessary, revise.  
  4. Retain: After the solution has been successfully adapted to the target problem, store the resulting experience as a new case in memory.  
The complexities associated with programming and implementation of a knowledge management system based on case histories is both non-obvious and difficult, but ironically this is the actual process that an expert physician uses in his day to day clinical work.

As always I am struck by the central ironies of the practice of 'scientific medicine' in the 1st decades of the 21st Century. The central thrust of the evidence based medical movement, a movement that I am sympathetic too with respect to its articulated central goal of improving clinical practice, involves the notoriously difficult challenge of adopting and adapting clinical trial data to the individual case. However, too often the proponents of 'EBM' in their enthusiasm and zeal for the movements global goals prefer mindless execution of the guidelines if or if not the individual patient before you conforms to the patient characteristics of the clinical trial from which the 'evidence' is derived.

This is not a trivial problem. No one has captured the complexities of medical practice, and the dynamic tension that exists between data and experience, reflection and action, tincture of time vs the need to act with incomplete knowledge than the first and greatest of all scientific physicians Hippocrates.

Recall the 1st aphorism of Hippocrates, recall it and burn it into your brain and your heart.....

"Life is short, the art is long, opportunity fleeting, experience delusive, judgment difficult."

The first aphorism of Hippocrates eloquently captures the crux and flux of modern medical practice as precisely today as it did in classical Greece. In spite of 2000 years of continuous and relentless advances Medicine remains more art than science. The complexity of human biology, an intractably unique human psychology, complex and poorly understood or characterized social forces, the enigmatic dimension of spirit, and the variegated interplay of human cultural forces and beliefs act in concert to make each ‘patient’ bewilderingly unique.

The statistical analysis of groups is an inductive process; i.e., reasoning from specific cases to a
general case or rule. Conversely, logical analysis of the individual is deductive; reasoning from general cases to a specific instance. Thus, at the most fundamental level, there are operational and philosophical impediments to the application of fundamentally inductive conclusions in solving the problems of the individual case of human disease.

The power of the randomized controlled trial, is the extraction of the essential commonalities between large numbers of cases and the experiences of individual patients. The power of the method lies in teasing out beneficial effects between competing treatments. However, in that extraction of treatment effects, the individual case is sacrificed for the sake of the power of the data extraction.

However, often lost in the blind pursuit of the RCT, are the serious methodological limitations and caveats related to its use:
(i) The study of rare diseases is difficult or impossible since studies may not have sufficient statistical power to detect clinically significant therapeutic benefits.
(ii) Referral bias is endemic to this form of clinical research and results are often not broadly generalizable to nonacademic centers.
(iii) Large multi-center trials of common illnesses can demonstrate statistically significant differences
between treatment groups, independent of any genuine operational treatment differences.
(iv) The process of clinical trial design involves conscious and implicit constraint of the patient population. So- called exclusion criteria act to select for the ‘purest cases’ of any given illness. This explicit attempt to limit patient heterogeneity is scientifically justifiable in order to increase the confidence that variations in dependent variables are statistically attributable to the study treatment rather than be due to intrinsic biological variability of the study population. However, this same process also seriously undermines the ability to generalize study results to the more diverse and unselected natural patient
populations. Ironically, generalization of experimental results is the true purpose of all clinical experiments.
(v) Placebo controlled double blind studies are increasingly prohibitively expensive, and increasingly funded by 'industry' rather than having government funding. This sometimes makes it difficult to separate science from marketing. Furthermore, the ever expanding role of industry in study design and funding constrains the universe of potential studies and potential problems being studied.
(vi) Double blind studies are seriously constrained by resource and time constraints and important even vital clinical questions may not be approached with this methodology. Many perhaps most important clinical questions require time frames of experimentation that are not practical for clinical scientists to study. For example basic science research in pirion diseases was originally unusually slow due to the experimental models available and the necessity to wait for 5-6 years to see evidence of infection in the animal models used in their study.
(vii) Knowledge and technology is not static, even in those cases where long term follow up is available, technological changes in diagnosis and therapeutics threatens to make results irrelevant even at the time
of their publication. For example the original BARI trial concluded that PCI was an inferior strategy to CABG for treatment of multi vessel coronary artery disease in patients with diabetes treated with oral agents. However, over the 5.4 years of average follow up accomplished in the trial, STENT technology advanced from bare metal to various drug eluting STENT's making the conclusions of the trial irrelevant to the 'state of the art' at the time publication of the trial results.

The power of Case Based Reasoning lies in the practical realization that each case presents the expression of a common etiological factor(s) in interaction with a biologically complex and often times unique individual. Therefore, the method has the ability to extract commonality, i.e., how one case resembles another case, when that 'resemblance' is that of a family resemblance in its sense as articulated by Ludwig Wittgenstein in the Philosophical Investigations,  Familien√§hnlichkeit, that is a resemblance more subliminal than precisely defined. Just as we recognize Johnny as vaguely similar to his third youngest brother Jack, we might well be stymied to detail precisely how they actually seemed to have a common family origin. In the same way, one case of psittacosis may not be identical, but reminiscent to another case of our own experience or existing within the literature. Therefore, central to the method and fundamental to the recognition of the case, is the 5th and ultimately the 1st case-based R-step, recognition of the case as a case of 'x.' The case based method uses analogy of solution successes to determine the best solution set from a database of cases. But fundamental to this method, is the recognition of the commonality of cases. Again not a trivial issue.

However, for me the real philosophical attraction of case based reasoning, is the realization that with the passage of time, while the knowledge inherent in RCT methodologies can become irrelevant or superceded by time and events, case based knowledge always builds upon its previous foundation becoming deeper, richer, and more relevant with the passage of time. To the mature practitioner this is a comfort and a validation of the life time of learning involved in becoming an expert practitioner of medicine.

While not precisely congruent with respect to the issues I am raising in this essay, the following recent Perspective Article from the New England Journal Of Medicine makes an analogous point, that an ugly case makes an enormous impact on our clinical practice, and promotes clinical wisdom. (Level IV Evidence — Adverse Anecdote and Clinical Practice. Alison M. Stuebe, M.D. N Engl J Med 2011; 365:8-9July 7, 2011-  The central message of this thoughtful essay can be summarized by what I call the 1st Law of Surgery: Good surgical judgement comes from experience, and experience comes from bad surgical judgement.    

Saturday, July 16, 2011

The New Galenism

The term Galenism in this essay has the usual pejorative sense of a pernicious influence, which grants that while Galen transformed medicine in late antiquity, he yet so dominated its thought that subsequent thinking and innovation was stifled by his authority.

In the defense of Galen, his own personal ideal of the physician is one that I can enthusiastically subscribe to, and is best captured by the title of his treatise "That the best Doctor is also a Philosopher" (Galen On The Ideal Physician. P Brain. South African Medical Journal, 52:936-938, 1977)(Original Source - Claudii Galeni Opera Omnia. ed. C.G. Kuhn, Lepzig Cnobloch. 1821-1833. Reprinted, 1964: Hildesheim, Olms. The work translated is in vol I, pp. 53-63.) 

However, the genesis of Galenism owed less to his own habits of thought and investigative methodologies than it did to the mindless acceptance of his successors of his results, and who then embraced him as the ultimate authority in medical practice.  

One can see the current infatuation with practice guidelines as the reintroduction of slavish conformity to 'expert' opinion akin to the forces that kept the western practice of medicine mired in antiquity until the renaissance. 

Take as an example a recent case referred to me by the "Performance Improvement" department in one of the locum tenens hospital sites I am currently working at. The patient was an elderly female who presented with atrial fibrillation with rapid ventricular rate, mild chest pain, focal and severe ST segment changes in the anterior lateral leads of her EKG. She had a troponin I elevation of a significant degree occurring several hours after her initial assessment. Once her heart rate was controlled and once she was back in normal sinus rhythm her chest pain resolved, but her troponin I elevation remained mildly elevated. She had no EKG evidence of transmural infarction, and once she had returned to normal sinus rhythm her focal EKG changes remained but were attenuated compared to those seen initially. 

By the Universal Definition of Acute Myocardial Infarction, (, this patient had a so-called Type II AMI , i.e., rather than the AMI being a consequence of plaque rupture and thrombotic occlusion occurring within the vessel causing focal myocardial injury, the pathophysiological etiology for the myocardial injury signal is due to increased myocardial oxygen demand in combination with inadequate myocardial supply of oxygen and nutrients. Common causes for this pathophysiological state are anemia, arrhythmia (as in this case), hyper- or hypotension, vasoconstriction or arterial spasm. 

I am referred the case, as I did not begin this young lady (75 years of age) on Statin drug therapy since it was a protocol item. The entire implication of a Type II AMI is that is not a consequence of unstable plaque occluding or severely obstructing an epicardial coronary artery. And as such, since it is not the consequence of unstable cholesterol plaque, it is not obvious to a clinical expert in cardiology (that is myself) that statin drug therapy is indicated in this case. That is simply put, anti-cholesterol therapy is not, nor should it be a protocol item for post MI care in the Type II infarct. However, the change in nomenclature and the emphasis on correct pathophysiological characterization of the mechanism of an acute myocardial infarction now threatens the mindless execution of ‘best practices.’

The effect of acute statin therapy on plaque stabilization and prevention of Type I AMI is clear cut and supported by decades of biologically congruent results with powerful clinical data endpoints. Furthermore, because the effects of statins are so immediate in the setting of plaque stabilization, a significant effect of their use is probably related to their anti-inflammatory effects with respect to the vascular system. Thus, the absolute risk:benefit ratio is so highly weighted to treatment that the decision to treat epicardial lesions is virtually axiomatic across all genders and ages of patients. The survival advantage for statin drug therapy with respect to plaque stabilization is immediate and profound and occurs within many studies as soon as 6 months post event.

While the thoughtful use of guidelines to enhance patient care and improve medical outcomes is highly desirable, the mindless pursuit of ‘core measure’ statistics is anathema to the thoughtful physician. By not treating this patient with statin drug therapy I have invoked in my hospital a ream of paperwork designed to ‘improve my practice.’

However, if I had simply followed the guideline my performance would have been ‘perfect’ but my practice thoughtless, flawed, and ‘Galen-istic.’ While statin drugs are for the most part among the safest of all drugs and drug classes there still remain real potential risks with their use.

Since in the patient under discussion the guideline does not apply as the patient has no evidence of unstable plaque as her pathophysiology, could an argument be made for statin drug therapy based on ATP III guidelines?

This particular patient had a lipid profile in which her HDL cholesterol was 54 mg/dl, HDL/Cholesterol ratio was 3.4, and her LDL cholesterol was 119  mg/dl. According to the National Cholesterol Education Program, this patient has a 10 year risk of developing symptomatic CVD (due to epicardial disease - not supply/demand mismatch) of <9%. Furthermore her ATP III LDL goal is <130 mg/dl. Thus, no argument for statin drug therapy can be made based on population normative data and conventional risk assessment.

The PROSPER study (The PROspective Study of Pravastatin in the Elderly at Risk) (Randomized trial of 5,804 seniors age 70-82 years of age) demonstrated an actual cardiovascular death rate of 3.3% among patient’s on statins vs a 4.2% in the placebo group, a 0.9% absolute difference in mortality rate in this age group, with an absolute all-cause mortality rate difference in treatment versus placebo group of 0.2%. As a consequence because of the absence of a clinically significant difference in mortality between treated groups and placebo groups, I find little data to support a cholesterol intervention in this patient based on survival advantage within her age group.

Furthermore, the current patient had advanced COPD, was on home oxygen and is essentially bed-ridden due to her need for continuous oxygen therapy. Echocardiography demonstrates severe pulmonary artery hypertension due to the magnitude of her COPD. Again, we see a patient who is not manifestly obviously directly comparable to or appropriate to treat with guidelines designed to mitigate the effects of elevated cholesterol levels on the survival of predominately middle aged males.

So, in summary, a patient without evidence of unstable coronary plaque, without indication for drug therapy related to a population derived risk assessment, in an age group controversial with respect to absolute survival advantages of statin drug therapy, and with severe co-morbid pulmonary conditions was not treated with statin drug therapy AND the ghost of Galen has pronounced my practice inadequate and deserving of ‘practice improvement.

What is going on here? It seems to me that the more thoughtful your practice, the more headaches will be brought to that practice by those who inexpertly and without art-fullness drive that practice to the 'best practices' if or if not that practice applies to the precise patient in front of you. I am tempted to draw an analogy between the effect of standardization in education methodologies to the attempted standardization in medical treatments. But I will leave that rumination to a future post.