A post by Jason Shafrin has been sitting in my Google Reader starred items for too long. It’s time to explain to myself why it is there. It’s about the ways in which patients differ from one another–patient heterogeneity.
Why do I care about this? I could list all the possible ways it is relevant to comparative effectiveness research, the distribution of health care costs, the problem of allocating the risk of those costs across potential payers, and so forth. But the day-to-day practical reason I care is because the dimensions of patient heterogeneity provide a way to organize control variables in multivariate analyses, which is basically what I do for a living.
The paper Shafrin summarized appeared in a 2010 issue of Medical Care. Kaplan et al. organize patient characteristics into six categories. Shafrin explains them as:
Immutable characteristics: These are intrinsic factors that the patient cannot change such as demographic characteristics or genetic factors. […]
Health Profile. This category represents the patient’s current health condition. This can be represented by factors such as a patient’s disease burden, mental/physicial functioning and other measures. […]
Personality Profile Measures. A patient’s personality can affect health outcomes as well. […]
Behavioral Profile. This includes disease management skills and health habits (e.g., smoking, diet, exercise).
Medical/Treatment Context. The category measures differences across patients in the relationship with their physician, the setting where care is given, and continuity of care.
Life Context. The goal of this final category is to measure whether the patient experience a stressful life event or whether they have a significant social support system.
Measures for each of these categories are not always available in the data I work with. However, in characterizing the work I am doing or proposing, it is important to acknowledge that all these areas are important and relevant (this is often done with a theoretical or conceptual model). Omitted variables in these categories can bias results, which is why I so often use instrumental variables (IV) techniques. If instruments are valid, the resulting estimates are free of omitted variable bias. (Oh how I wish more reviewers of my grant proposals understood that fact. Oh how I wish so many others who do observational studies understood why they need to consider IV approaches.)
