Original articlesUses of Ecologic Studies in the Assessment of Intended Treatment Effects
Introduction
Ecologic studies have generally been considered poor substitutes for individual-level studies in epidemiologic research, not only because they tend to be confounded by differences in aspects other than the factors being studied (frequently referred to as the “ecologic fallacy”), but also because the confounding factors in ecologic data are usually difficult to detect and adjust for by standard statistical modelling 1, 2, 3, 4, 5. More recently, however, some epidemiologists have described situations in which ecologic studies may be more appropriate than individual-level studies. These include studies of “group” effects that cannot be measured at the individual level 6, 7, 8 and those in which biases arising from sampling and individualized data collection are large and intractable 9, 10.
In this article, we offer a rationale for ecologic studies in the assessment of intended treatment effects. One particular goal of this article is to illustrate that ecologic studies of intended treatment effects, especially if based on geographic area variations, may sometimes be preferable to observational studies at the individual level, because they are relatively immune to confounding by indication.
Section snippets
Randomized Controlled Trials Are Not Always Possible
Whenever possible, the efficacy of a treatment should be assessed by randomized controlled trials (RCTs). In many situations, however, RCTs are either unethical or infeasible 11, 12, 13, 14. In an effort to quantitate the need for postmarketing studies of the intended beneficial effects of drugs, Strom et al. surveyed the 100 most common drug uses (comprising <100 drugs, since a single drug can have multiple uses) in 1978 and found that 31 of the 100 uses had not been approved by the FDA at the
Scenario 1
A treatment for a hypothetical disease is proposed to alter a life-threatening pathophysiological process, with expected improvements in survival. It turns out, however, that the treatment has no impact on survival. Death from the hypothetical disease is closely associated with the severity of the disease, and treatment is also positively related to disease severity, so that disease severity confounds the effect of treatment on mortality. Suppose the use of this treatment is influenced by
Discussion
Our hypothetical examples demonstrate that confounding by disease severity can constitute an important threat to the validity of analysis at the individual level. Although we have assumed that the treatment does not cause death, it is associated with a doubling of the mortality rate because both mortality and the treatment are closely related to disease severity. On the other hand, despite substantial variation in the frequency of treatment among patients residing in the six geographic areas
Acknowledgements
We are grateful to Qing Zhang for assistance in computer programming, K.S. Joseph for thoughtful comments on a previous draft of the manuscript, and Catherine McCourt for her encouragement and support of this work.
References (32)
- et al.
Postmarketing studies of drug efficacyWhy?
Am J Med
(1985) The evolution of clinical practice and time trends in drug effects
J Clin Epidemiol
(1994)- et al.
Adapting a clinical comorbidity index for use with ICD-9-CM administrative database
J Clin Epidemiol
(1992) - et al.
A new method of classifying prognostic comorbidity in longitudinal studiesDevelopment and validation
J Chron Dis
(1987) - et al.
Postmarketing studies of drug efficacyHow?
Am J Med
(1984) - et al.
Professional uncertainty and the problem of supplier-induced demand
Soc Sci Med
(1982) - et al.
Comparison of relative risks obtained in ecological and individual studiesSome methodological considerations
Int J Epidemiol
(1987) - et al.
The ecological fallacy
Am J Epidemiol
(1988) - et al.
Ecological bias, confounding, and effect modification
Int J Epidemiol
(1989) - et al.
Invited commentaryEcologic studies-biases, misconceptions, and counterexamples
Am J Epidemiol
(1994)