“The best model” the dealer said, “is this one. It rarely breaks down, in fact over 10 years only 16% have to be replaced or have major repairs.”
“That’s one in six” my friend said. The dealer replied that he could buy (expensive) insurance to cover the period not under warranty. “Best we can do” he said.
Each time I use the Framingham calculator I’m reminded of the similar imprecision of modern medicine.
As I fiddle with it today, entering a total cholesterol near the top of the range, a ‘good’ cholesterol in the middle of the range and a normal blood pressure yields a hypothetical me with a 16% chance of a heart attack over the next 10 years. While this is prognostically 4 times higher than if my total cholesterol were the lowest on the scale and my good cholesterol highest, it still means that even with these excessive cholesterol levels my chances of having a heart attack is still far below even 50/50. Most men my age with bad cholesterol values don’t have a heart attack or die. Indeed if there were 6 men similar to me only one of us would have a heart attack in the following 10 years.
Why do some get heart attacks and others don’t? This is the irritating imprecision of medicine. While all cholesterols (good and bad) are the same, each of the 6 men have different ‘other factors’ that play into a complex yet unknown equation that somehow leads one of us to suffer the bad outcome.
When I see a patient with a bad condition I always remember my first patient with cancer of the lung. Mr. Walker was 59, and other than pain in his shins (an infrequent but known secondary effect in a small number of patients with lung cancer) he was well. Yet there was no hope. Removal of the lung would not save Mr. Walker - we had good epidemiological studies showing dismal prognosis and as most older patients could not survive with a single lung, high operative mortality rates. I presented his case to our professor, a man older than the patient. To my recommendation that Mr. Walker be sent home without having his lung removed, my professor replied that a few of his lung cancer patients who had surgery survived for very long periods, so why deny this one the chance. The professor recommended that I see if my patient could walk up a flight of stairs without stopping. If he passed this ‘stress test’ he could survive with one lung, said the professor. He was old enough to have seen enough lung cancer to know that not all were alike.
Wouldn’t it be fine if our physician knew which of the 6 of us was going to suffer the bad outcome? Not everyone who smokes a pack a day for 40 years will get lung cancer. In fact, most won’t. We all know committed smokers who believe that inhaled smoke gives the lung tissues a protective coating. Perhaps they are right.
Of 100 men with prostate cancer limited to the prostate 75 will not have any evidence of metastases 10 years later. (For those having a prostatectomy only slightly more, 85, do not develop metastases over 10 years. Although the disease phenotypes are the same - cell type adenocarcinoma, anatomic location within the prostate at the time of diagnosis - only about 25% of men go on to develop metastases. While this might look like the play of chance, science insists there must be a reason. Greater diagnostic precision will reduce prognostic imprecision and lead some men to avoid debilitating and unnecessary prostatectomy.
Physicians make diagnoses on the basis of anatomical location of symptoms, physical signs (anatomical and physiological - e.g. blood pressure) and laboratory tests that measure specific body systems for homeostasis such as oxygen transport, blood sugars, blood, immune response, the presence of pathogenic organisms and toxins, and so on.
A series of papers over the past decade have demonstrated that we need to rethink our reductionist conceptions of diseases as being a set of distinct entities, like congestive heart failure, or sickle cell disease or AIDS and begin to understand that these illnesses are each complex involving multiple genetic, environmental and social factors. As we begin to understand these factors and their relationships, our conceptions of disease will change, diagnostic and prognostic labels and estimates will be altered, and new systems of understanding human physiology and pathobiology will enter the language of diagnosis and the practice of medicine. These discoveries have arisen in the emerging field of network analysis.
The genetic component can be thought of as a limited but as yet incompletely identified set of genes that control human development and cellular function throughout life. Some genes are known to be involved in specific areas of human development and pathophysiological response because mutations in these genes result in specific diseases, such as sickle cell anemia, familial cardiomyopathy, pulmonary arterial hypertension, diabetes, various cancers and so on. Some of these are monogenetic in that a single mutation is common to all who have the defect. In others seemingly identical diseases can result from mutations in more than one gene or can arrive from different mutations in the same gene - for example the clinical disease - familial pulmonary arterial hypertension - can arise from over 50 different mutations.
It seems helpful to conceptualize two categories of genes, those that have a specific role with a particular type of cell and those that are more generic. The specific-role or primary genes become apparent when mutations arise. Sickle cell anemia derives from a mutation in a single gene that results in the substitution of valine for glutamic acid at a specific position in the molecule that makes up the beta-chain of hemoglobin. Under hypoxic (or other) conditions this results in the formation of hemoglobin polymers which cause the erythrocyte to assume a sickle shape. Genes involved in various malignancies probably function in a similar but more complex way.
The other category of disease modifying genes have broader effects that serve to modify threats to cells from the specific disease-related genes and from environmental threats such as temperature, radiation, hydration and tonicity, oxygen, micro and macro nutrients, infective agents and toxins. The ability of an individual to accommodate these genetic and environmental threats is also part of the genomic makeup of that individual. The resultant pathology depends on the interaction of the gene with the environment.
Conceptually, it might look like this:
From the same paper here is the disease network for sickle cell anemia, a disease that can present with many pathophenotypes:
The figure is copied from the article by Loscalzo, Kohane and Barabasi <1>
The primary genetic abnormality, hemoglobin S (red) can be affected by other genetic abnormalities, if they are present (grey). The various clinical presentations of sickle cell anemia (in blue) are thus the result of a network of cellular and sub-cellular events, aided and modified by environmental agents (green) and the genomic elements that control the bodies generic reactions (yellow).
On a general level none of this is new or surprising. We know that our genetic complement and our lifetime interaction with environmental factors are somehow responsible for our particular pathophenotype experience - the disease we have. What is perhaps new is the notion that although the phenotypes may look identical, they are not: What is different about them is how they are produced - their underlying networks of causation and damage control. Understanding these networks will add precision to diagnosis and will be helpful in developing pharmacologic interventions that disrupt or disconnect the disease causing elements of the network.
While the primary nodes of causality - primary and secondary genomes and environmental factors (physical and social) are limited, the secondary networks of each primary node are more complex and the number of possible interactions between them large. Thus a single anatomic or cellular or molecular pathophenotype might have been the product of a very large number of possible interactions involving the primary nodes of causation.
Even in this simple schematic model, there are an exponential set of of possible interactions each of which might produce a different presentation of the ‘same’ anatomic or functional disease or pathophenotype. Thus some adenocarcinomas of the breast are affected by primary genomes such as the BRACA genes, secondary genomes such as estrogen receptors, and are likely further modulated by environmental causes, and perhaps environmental factors that play a role in prognosis and the role of the secondary genome in control of proliferation, immune response, apoptosis/necrosis and so on.
Although logically unassailable the new model of disease would be mathematically unusable due to the exponential number of possible combinations of nodes and sub-nodes (if all possible connections are considered to be random events).
Fortunately they are not. The pathways are part of non-random networks between primary and secondary nodes of influence, themselves interconnected. Network analysis is changing the way we look at biological, social, economic, electronic and other networks.
Practical applications are in the future, perhaps decades from now. But we should pause when considering a diagnosis to hand on to a patient, for each diagnosis comes with a prognosis and a set of care burdens and, unfortunately, with considerable imprecision about likely outcomes. Recent publications on aggressive control of blood sugars are only the latest to confirm how little we know about the natural history of the diseases we diagnose. (see next post
References
1. Loscalzo J, Kohane I, Barabasi A. Human disease classification in the post-genomic era: A complex systems approach to human pathobiology.
No comments:
Post a Comment