Diagnosis and outcome

Medical knowledge, all scientific knowledge, is statistical. Even in the simplest, most straightforward experiment, there is some level of imprecision, some degree of unavoidable ambiguity ( error). All results are expressed some margin of error.

Medical data often has a very wide margin of error. Reasons for this include the heterogeneity of the population studied. Often, what we call a “diagnosis” can be resolved into more than one disease entity. Until we knew better, no distinction was made between estrogen receptor positive estrogen receptor negative breast cancer.Now, these different entities are often treated in different ways. sometimes, a new treatment defines a new subgroup. tratuzamab makes Her2 expression important in breast and stomach cancers, Rituximab makes the CD 20 a critical factor in the treatment of lymphoma.
In my opinion, the true value of a diagnosis is how it defines treatment. The diagnosis and treatment outcome are inseparable. If a diagnosis gives a 30% response rate to treatment, it is a poor diagnosis. A diagnosis of lung cancer, in the absence of molecular definition,  is hardly more than a guess, regardless of the details appended. A diagnosis of CML, with a bcr-abl translocation is a strong statement that leads to a near 100% probability of response.

Response is not the point. People living better and longer is the point. Response is a useful measure that hints that a treatment is on the right track. The documentation and publication of a response is a public service, but, aside from the psychological lift, it does not necessarily benefit the patient.

There are unfounded beliefs in the practice of ooncology Probably the most destructive of these is the idea that efficacy is a function of toxicity. The more toxic the therapy, the more powerful it is, the more cancer it kills. This is an unfounded belief, not evidence based and not scientific.
It is true that some cancers can be cured by very toxic treatment, and at this time, extremely dangerous and toxic treatment is the only known way to reliably treat many leukemias, certain lymphomas, etc. But oral, low side effective  tyrosine kinase inhibitors, like imatinib, are more effective in the treatment of chronic myelogenous leukemia than a life endangering, life altering bone marrow transplant.

A related, unfounded belief is that the microscopic appearance of a tumor, and the pathologist’s impression of its aggressiveness should determine the toxicity of the therapy; The most primitive of cancer, testicular cancer, is among the most reliably curable. And it is the most normal appearing elements of that cancer ( teratoma) that is the most difficult to eliminate. The microscopic “aggressiveness” estimate of the pathologist is an aggregate of the number of cells in mitosis ( the act of division), how bizarre the cells look, how different from normal they are, etc. These parameters do not correlate with outcome. A high mitotic rate could confer increased susceptibility to treatment ( even treatment that is not directed at mitosis). Bizarre appearing cells may be programmed for death under certain conditiona

Traditional Forces in Medical Decisions

Medicine is both the beneficiary and victim of traditions.

The benefit comes from the credibility afforded to medicine, translated into the respect afforded  doctors.  There is a tradition of professional excellence, enforced by a licensing mechanism and a legal system that monitors quality.

Traditional  forces are  conservative,  They encourage the continuation of "standard practice." Usually that is a good thing.  Standard practice is usually the best available course. But the tradition is a force that retards change, and sometimes change means improvement.

All traditions, medical or otherwise, are the products of history, and as such, preserve a wisdom and perpetuate myths.

A core tradition in medical  oncology is that more toxic treatment is more effective treatment.  This stems form the history of the field (  see: The Emperor  of all Maladies) .  The first  successes in treatment of cancer  were seen when therapy was brought to an intensity that  carried a significant chance of toxic death, death from the treatment itself     ( not the disease).  That  tradition carries through to the present.  When cancer  cases are discussed in conference,  the question of dose   is always raised.

But the notion that higher dose yields better results, has often  not  been  tested in most circumstances.  The dosage of drugs is established in  Phase I clinical trials, These trials include desperate, heavily pre-treated patients with clearly heterogeneous primary and secondary problems.  These trials usually assume the optimal dose is  at or near the maximum tolerated dose .  Once that dose is established, it becomes cowardly  to   reduce the dose, unless forced to do so by toxicity. I am not sure that the approach is globally  valid.

The classification and  treatment of cancers based upon microscpic appearance,  That  is a tradition.  The appearance of cells in the microscope and the impression of analogy to aberrant normal tissue dates from Virchow  (1821-1902). This approach organizes the treatment of cancer, but it has other consequences.

There is currently a battle between the microscope and molecular biology for the orimacy  in diagnosis. The  molecular biology sometimes identifies mutations that may be fundamental to the cancer process and these  mutated gene products can be the targets of therapy.  Frequently, neither of these  happens. The mutations are of unknown significance and the targeted agents work poorly, or not at all ( at $10,000 per month).  But even when clear targets emerge, they struggle against the tradition of the microscope  for recognition.

Sometimes, treatment does not change despite a  better understanding of the disease process.  Lymphocyte predominant Hodgkin disease is an example.  Hodgkin Diseases is a set of lymphomas that are caused by a very distinctive appearing cell: the Reed Sternberg cell.  Cells with a similar appearance are  characteristic of a disease called Lymphocyte Predominance Hodgkin Disease, but it has been known for 30 years that this is a  non-Hodgkin lymphoma.

The traditional treatment for Hodgkin  is ABVD, a mixture of toxic  medicines that can damage the heart, lungs, nerves and immune system.  This combination is not standard for Non Hodgkin lymphoma....except for lymphocyte predominant Hodgkin Disease, because that is the tradition.  ( I do not follow this tradition). Continuing to use a regimen that has worked in the past  is a very reasonable approach, but the Hodgkin Disease regimen has considerable long term toxicity.

Tradition is a reasonable fall back position, especially when it has a good  track record.   It should not stand in the way of progress. There is value in tracing  how we come to treat certain diseases in certain ways. Often, there is no basis at all.

Tradition has a role, but it must be subject to scrutiny.

Epistemology of Medical Practice

How does the physician make a medical decision?   A spectrum of deductive and inductive techniques  are used in this process. They contain economic, social, scientific, traditional, among other, influences. I want to  examine this process,

The published medical literature that constitutes the effable basis for medical decisions comes in two forms: models (basic science) andclinical trial data. The two work off of one another.  Both are intrinsically flawed by the same problem: heterogeneity. The purist doctor ( devoid of extraneous influences like money and trying to look good) has the problem of balancing the inputs and resolving  the internal inconsistencies.

 Models, often expresses as molecular mechanisms,  describe how things work,  They  are  useful, simplified, ways of thinking about  how things happen. There is no claim that these models are complete.  They are a framework that can be tested, to an extent,  by laboratory research and clinical trials

The clinical trial is very different.  It  tests an intervention. It looks at the percentage of people who benefited or were harmed ( or both) by the intervention   This is strong, hard to obtain, human experimentation evidence.

How these two interact constitutes the conceptual basis of medical practice.  The correlation between clinical trail data and the model is often  not very good. What does that mean for the model? What does that mean for the clinical study?

Models

Models  depend upon the interpretation of laboratory experiments that are subject to limitations of  currently available technique. They are often appealing.  At the time of presentation, they seem to account for much of  the available  data.

Models are  very fragile,  and  they should be.  Contradictory laboratory or clinical evidence should promptly collapse the model and clear the space for the generation of a new model. The checking of results is limited  because of the expense and the esoteric nature of the work. For practitioners who are not directly involved in the research endeavor, the model is an act of faith, a belief based upon its beauty and trust in the investigators

Models are supported by a scaffold of economic and social interactions.  Over time, careers and corporations are built on these models.  The models generate communities, that came to their visions based upon underlying common beliefs, Some of these believes are  stated, others are long-standing assumptions that are hidden from view,  There is a  “common knowledge,” a set of propositions that evolve into postulates. Based upon what they “know” the community   generates a structure that protects the model.

 Clincal Trials

The outcomes of clinical trials are usually  expressed in statistical terms.The trial asks if there is  a difference between two  ( or more) courses of action. But differences occur by chance. A difference that would be predicted by a Gaussian distribution to occur by chance  less than 5% of the time  ( P<.05)   is considered significant:  That is the  arbitrary Holy Grail.

The tested population in a clinical trial is always heterogeneous. That is a recognized issue. That is the reason that larger trials are “more powerful”. Large numbers are believed to compensate for heterogeneity, “random” variations are canceled,

This approach puts the  clinical trial in the role of testing  an idea, not necessarily looking for truth. The heterogeneity in clinical trials is  an asset to the trial, not something to be canceled,  it is  a truth  to be discovered. It is a way to identify responders, victims of toxicity, and  new diseases. But the heterogeneity stands in the way  of the ( P<0.05) declaration of significance.

The  result of the clinical trial is correct.  If the trial is well designed, it predicts what would happen in a similar population treated in an identical manner.  This information is  useful  for insurance companies, drug manufacturers, epidemiologists, health care planners. etc.  It is  an accurate portrayal  of the selected outcomes in a heterogeneous population.

For the treating  physician theses results  should be  far less useful.  It is rare ( perhaps, impossible) for the patient that needs treatment  to match the “average” patient in the trial. Worse,  the information about the treatment and outcomes of those patients , included in the trial,  that most closely match the physician’s actual  patient, are obscured in the statistical smoke.  The data on  those patients has been sacrificed to  statistical “significance,”

The trialists look for sources of heterogeneity, but the raw data is generally not accessible to the practitioner, the consumer of the trial data. S/he only has a statistical summary of the outcomes and cannot search the trial data for the closest match to the patient at hand.

Making the Decision

How should  the physician behave?  The current dogma  is that practice should remain the same until an adequately powered clinical trial has  demonstrated the need for change. This is the  practice  of “evidence based medicine”. Discoveries in basic science and  changes in models,  are not recognized.  Among the consequence of this approach:  insurance companies are allowed to save money by avoiding payment for expensive new therapies; and there is  stability of the market for drug manufacturers.

We would hope that a clinical decision  weighs  all the factors. Clinical trials apply to  populations. An individual is never a perfect representative of a population. Hence, the application of a clinical trial is always flawed.The models are always tenuous,  but they are the best we have at the moment. As models account for increasing amounts of data, they become more credible, they become  a way to summarize and check the applicability of  the clinical trial data.  Which patients in the clinical trial fit the model?

Clinical trials should turn to identifying heterogeneity instead of burying it. Populations that vary in outcome should be investigated for corresponding variations on biology  The raw data should be available for analysis, so that the practitioner can examine the outcomes  of  the best matches to the patient needing treatment.

Models should  both predict,  and be tested by,  looking at  more completely defined populations.

It is time to start using the technology we already have.

.