Health Risk Science - Population-based studies

The term population-based is traditionally used to describe a study that involved a defined “general population”, as opposed to hospital-based or occupation-based populations. Epidemiologic studies have a tacit need to be based in populations, and as such, most epidemiologic studies can be loosely considered as population-based. 

The etymology of the word epidemiology arises from epi-demos, meaning upon the people.  By definition, epidemiology has as one of its fundamental concepts, the population.  But conflicting definitions in epidemiology are common where the names of study designs and concepts are concerned.  Readers of epidemiologic literature should be aware that several terms are used idiosyncratically by epidemiologists.  The first section below describes population-based designs, and following this is a short section on the concept of population.

Population-based designs
For many epidemiologists, a famous population-based study is the Framingham Heart Study (Dawber, 1951), which began by enumerating a sampled cohort of 5209 men from the city of Framingham, Massachusetts in 1949 and following them over the last decades.  This is an example of a cohort study design, and it (along with the case-control study design) is taught as one of the main approaches in epidemiology to studying the etiology of illness.  Given that the study population was sampled from the residents of a defined location, such a study is often considered “population-based.”

Traditionally, epidemiologic studies were often labeled by various methodological descriptors, indicating the origins of the source population being exploited for the study.  For example, hospital-based studies and industry-based studies obviously try to accrue patients and workers, respectively.  But these are not universally referred to as population-based approaches.  The common usage of the term implies sampling of individuals from the general population, one that is defined by geopolitical borders.  Sometimes even case-control studies (another epidemiologic design) are not considered population-based because the sampling is often perceived as outcome-based.  This, however, is at odds with other usage. 

The use of the term population-based is a misnomer.  What is actually being referred to is a study that uses a directly defined population (as opposed to indirectly defined).  For example, a case-control study can begin by defining a source population and then directly sampling from that population to form the control series.  Take, as one example, this text concerning a case-control study in Montreal (Parent et al. 2006):

The present paper describes associations between diesel and gasoline engine emissions and lung cancer, as evidenced in a 1979-1985 population-based case-control study in Montreal, Canada. Cases were 857 male lung cancer patients. Controls were 533 population controls and 1,349 patients with other cancer types.

This study actually serves to illustrate a more modern explanation of the meaning of ‘population-based’ and how epidemiologic studies have the tacit reliance on a population.  This Montreal study, in theory, provides two equivalent samples of the source population (metropolitan Montreal residents during 1979 to 1985).  One, a directly defined population, was formed (533 controls) by sampling from electoral lists of Montreal.  The authors refer to this series as population controls.  Different lists were used over the course of the study, relating to the dynamic population membership of those moving into or out of Montreal over the years. 

But there was an alternative approach used in the Montreal study that secondarily defined its population.  That is, the source population was defined indirectly via the method of identifying the cases of lung cancer (diagnoses in pathology departments of various hospitals).  The second control group (the 1349 “cancer controls”) were identified as men with other cancer diagnoses, found in the same hospitals.  In this fashion, the source population is actually being defined as the catchment populations of the hospitals, but since the Montreal study restricted itself to Montreal residents, both “population controls” and “cancer controls” are (in theory) representative of the same Montreal source population.

Populations
One of the essential concepts in epidemiology concerns that of the population.  Definitions of different types of populations are used by the various study designs and statistical theories to explore aspects of the frequency of occurrence of illnesses. In epidemiology, in referring to a population of individuals, the concept of time is essential.  By and large, populations come in two flavours:  cohort-type (closed) and dynamic-type (open).  Cohorts are populations that are defined by an event and are static.  Examples include admission to a study, having been diagnosed with an illness, or having worked for a particular employer.  Once someone is admitted to that population it is final (though this is quite independent of whether follow-up over time is incomplete).  In contrast to this, dynamic populations are defined by a transient state and have turnover.  Examples include residence in a particular city, temporary exposure to an occupational substance, or the use of a prescription for a certain amount of time.  Membership to the population is temporary and only while the state exists.

References:
Dawber TR, Meadors GF, Moore FEJ (1951) Epidemiological approaches to heart disease: the Framingham Study. American Journal Public Health 41:279-286
Parent ME, Rousseau MC, Boffetta P, Cohen A, and Siemiatycki J (2006) Exposure to Diesel and Gasoline Engine Emissions and the Risk of Lung Cancer. American Journal of Epidemiology 165: 53-62.

Further reading:
Miettinen OS. Theoretical Epidemiology: Principles of Occurrence Research in Medicine. Wiley, 1985.
Rothman KJ and Greenland S.  Modern Epidemiology, Second Edition, Lippincott-Raven, 1998.