Lose the forest for the disease: Part II
Epidemiology has encroached on our perception of multifactorial diseases (see previous blog entry), and it has also distorted the way in which medicine is practiced and taught. Now that the medical system is encumbered with complex diseases – heart disease, diabetes, obesity, etc. – we are attempting to identify causative factors for these expensive illnesses. Yet, whether we examine coarse characteristics like race or fine-grained genetic features, the best we can do is to identify risk factors, not causes. Risk factors – like age, gender, or genetic mutations – are identified based on population-level trends in disease incidence. The use of the phrase “risk factors” originated in The Framingham Heart Study (1948), an epidemiologic study designed to examine how cardiovascular disease is linked to lifestyle, and, based on the 5,000 or so men and women studied in Framingham, MA, they have published a widely used risk calculator. Can anyone’s risk of developing cardiovascular disease be boiled down to these 7 characteristics? If you live in Framingham, MA, it can, but if you live anywhere else (e.g. the UK, Ref. 1), your risk may be significantly different. This is the first problem that arises with the use of epidemiology to make claims about disease: how similar are you – or your patient – to the individuals followed in a particular study? Are broad categories like age, gender, and race sufficient to constitute similarity? (Given the study in the UK, probably not)
While this suggests that we should be designing more nuanced studies – where we assemble cohorts of patients that resemble real and varied communities (e.g. African Americans in Cleveland, or 2nd generation children of Jewish immigrants in NY) – current research, particularly in genomics (Ref. 2), is moving in the opposite direction, amassing thousands of individuals grouped together by coarse features like race, gender, or age. The impressive size of these studies effectively allows us to study how a handful of features vary with disease, and the astronomical number of individuals is necessary to wash-out the large amount of individual-to-individual variation that confounds complex diseases. The high degree of heterogeneity in the study population weakens the correlation to disease incidence, making the widespread application of the calculated risks tenuous. Often, the few genetic polymorphisms associated with a disease can only explain a fraction of the disease incidence (Refs. 3,4). For instance, variation in two well-publicized genes linked to obesity accounts for less than 2% of the variation in BMI (Ref. 3). The individual variation seen with complex illnesses – from the severity/features of the illness, to the lifestyle surrounding it – and the large numbers of people required for research studies suggests that there may be underlying latent variables or population stratification (read about Simpson’s paradox), and identifying the latent variables may paint a more nuanced picture of the factors that influence a person’s disease risk.
Another issue lies in bringing epidemiology to the clinic, for it influences how we view our patients. Translating epidemiology into the practice of medicine begins in medical school, and the influence of epidemiology is apparent from the problem cases that comprise our board exams. These cases often sound like “a 44 year old African American male with a history of diabetes and hypertension presents with complaints of chest pain…”, where every word clues us in to their risk factors: age, gender, race, diabetes, and hypertension. As a result, medical students are trained to view patients as nothing more than a grocery list of risks – a dehumanizing process (for both us and the patients) that may have limited utility. For both the patient and physician, the care of health operates at the level of the individual, not the population. Rather than understanding what is “normal” for the patient in front of us, we are trained to define “normal” relative to the population. Ideas like population-based normality and risk factors are tools for regulating public health and, as such, serve as a wonderful resources for institutions – the CDC, the WHO, governments, cities, etc. – because they can use these population-based averages to implement large-scale health interventions. If, for example, African Americans living in a particular zip code have higher rates of colon cancer and diabetes, the city government might team up with civil engineers and city planners to reduce the number of fast food chains and increase the number of grocery stores (read more on food deserts).
There is no doubt that epidemiology continues to improve our understanding of health and disease. Indeed, the health of entire countries has benefited from the lessons learned from Framingham (Ref. 5). Nonetheless, we have to be cautious in (1) making the ecological fallacy in assuming that an individual behaves like the average and (2) viewing patients as nothing more than an inventory of risk factors.
References:
- Brindle, P. Predictive Accuracy in the Framingham coronary risk score in British men: prospective cohort study. British Medical Journal, 327 (2003).
- Ahmed, S. Newly Discovered Breast Cancer Susceptibility Loci on 3p24 and 17q23.2. Nature Genetics, 41 (2009).
- Bogardus, C. Missing Heritability and GWAS Utility. Obesity. 17 (2), 2009.
- Kraft P, Hunter DJ. Genetic risk prediction—Are we there yet? N Engl J Med 360:1701–1703 (2009)
- Puska, P. From Framingham to North Karelia: From Descriptive Epidemiology to Public Health Action. Progress in Cardiovascular Disease, 53 (2010).
