When individual linked data on pharmaceutical use and health outcomes are not available we can only assess associations between (1) population data on pharmaceutical use and (2) population health outcomes such as mortality or morbidity attributable to a specific disease [5]. The analysis of aggregate data on medicine use and health outcome comprises a type of "ecological analysis" that uses data on groups to make inferences about the health of individuals [39,40]. If we assume, in the absence of good reasons for so doing, that individual level relationships can be inferred from aggregate level relationships, then we are said to have committed the "ecological fallacy" [41,42].
The dominant view in the epidemiological literature is that ecological studies should only be conducted when individual level data are unavailable. Even then they are only seen as providing, at best, inexpensive and relatively efficient ways of generating hypotheses that need to be tested in analyses of relationships between these variables measured in individuals (e.g. [5,10,39,43]). According to the approach adopted here, we need to identify the major threats to the validity of inferences from aggregate to individuals and then either design our studies to avoid them or to analyze the data in ways that minimize these errors.
Assessing changes for the better in a health outcome
Vital statistics on population mortality and treated morbidity that are collected as a standard part of public health surveillance in most developed countries can be used to monitor trends in population health [44,45]. Disease case registers (e.g. of cancer mortality or incidence, cardiovascular disease, diabetes) provide trend data on disease incidence and prevalence for a population using standardised criteria for defining cases. Registry data improve upon vital statistics by producing time series data on disease incidence or prevalence that use consistent diagnostic criteria.
Trends in the prevalence of self-reported morbidity, degree of disability and quality of life can be estimated from periodic cross-sectional surveys of large representative samples of the population. These data provide cross-sectional estimates of the prevalence of self-reported health status. Sometimes they also collect data on self-reported use of medicines [45].
Measuring medicine use in the population
Trends in medicine use can be inferred from aggregate pharmaceutical data. These may comprise data on the sales by volume or formulation of a specific drug or a drug class. Sales data do not provide any information on who is being prescribed the medicine (unless they are only prescribed to people within a narrow age range or to a single sex). They also do not tell us who is having the drugs dispensed and who is complying with the recommended use of the medicines.
Data on the number of scripts that are written or dispensed are closer to medicine use than sales data but they usually do not include information on patient diagnoses or on patient compliance. Even limited information on the characteristics of those who are prescribed a drug (such as age and sex) improves on aggregate prescription data because it increases our capacity to study covariations between population medicine exposure and population health outcomes.
Answers to all your Biology Questions
