We can assess the plausibility of alternative explanations by conducting further analyses of data and by citing supportive data from other studies. If we have measures of population exposure to other medical interventions, or aggregate data on population trends in risk factors, we can assess the plausibility of these explanations by regression analyses (e.g. [54,55]). Sensitivity analyses can also be conducted using the study data to evaluate the plausibility of alternative explanations [16,21].
We can also look for supporting evidence from other studies. This evidence may include the absence of similar trends in population health outcomes in populations that have not introduced the medicine. The presence of a similar relationship occurring at different times in countries that have introduced the medicine at different dates, and its absence in countries that have not done so, reduces the plausibility of alternative hypotheses (e.g. [47]). Our confidence in the impact of medicine use on population health outcomes will increase with replications of the relationship between medicine use and health outcomes in different countries which introduce the medicine at different times.
The role of epidemiological modelling
Epidemiological models of a disease implemented in computer programs can be used to assess the plausibility of alternative explanations of improvements in population health outcomes. Macrolevel epidemiological models model the mortality of hypothetical populations (e.g. 1000 adults aged 35–39 years) from a starting age a year at a time until a specified age (e.g. 75 years). Empirical probabilities of experiencing different health states are derived from population data on disease incidence and mortality. Data from clinical trials or observational studies on the effectiveness of interventions are used to predict the effects of interventions on mortality and morbidity in the cohort. Markov models are used to model transitions between different disease states over the lifetime of the cohort of individuals [56].
Microlevel models attempt to simulate the lifetime mortality experience of large samples of individuals who vary in characteristics that predict the risk of developing a disease, dying from the disease or another cause, and responding to interventions. These models sum the estimated effects on individuals to predict the effects of interventions on population health outcomes [56]. Such models require more knowledge of the natural history of the disease, the effects of different interventions on its natural history, and relationships between individual characteristics and disease risk and treatment outcome. They can also be computationally complex.
Sensitivity analyses are used to assess the robustness of the results of both types of modelling. In macro-level models, sensitivity analyses evaluate the impact of uncertainty about key model parameters on the model's predictions about population health outcomes [56]. In the case of microlevel models, sensitivity analyses assess the effects of different starting assumptions on the results of modelling.
Epidemiological models can be used to assess the likelihood that observed trends in disease incidence or mortality can be attributed to an intervention. A good example of the utility of this type of epidemiological modelling is provided by studies that assessed the plausibility of the hypothesis that a decline in prostate cancer mortality in the USA in the early 1990s was due to the widespread adoption of prostate specific antigen (PSA) screening in that country in the late 1980s (e.g. [57,58]).
Answers to all your Biology Questions
