Scope of epidemiology/causality

Question 1

Cause vs association

Answer 1

Cause: Antecedent event/characteristic/condition that is necessary for a disease to occur, given everything else is fixed. Importantly, cause implied a direction. X causes Y.

Association: statistical dependence between an event/characteristic/condition and an outcome of interest. Associations are undirected (symmetric) with regard to time. If X is associated with Y, then Y is associated with X.

Associations are directly observable, causation is not. Intuitively we believe that associations are the result of causal factors (though alternative explanations may be selection bias, confounding and/or chance). We must develop a plausible explanation for the observed association.

1. X causes Y
2. Y causes X
3. X and Y have a common cause which we have missed (confounding)
4. Selected on the premise of having both X and Y (collider bias vis a vis Berkson’s bias)

Question 2

Causal models

Answer 2

1. Sufficient-component models: - deal with sufficient vs necessary causes
2. Graphical models - causal webs, path models
3. Potential-outcome models

Question 3

Potential-outcome (counterfactual) model

Answer 3

Concerned with comparing frequency of an outcome in populations that were exposed to a putative “causal factor” vs those that weren’t exposed (reference or counterfactual group). These histories (exposed/unexposed) represent alternative “potential outcomes” for a population. Many of the standard measures used on epidemiology (e.g. risk difference, relative risk) are based on this model of causation.

Example: Among a group of smokers, we want to be able to predict what would have happened if these individuals didn’t smoke (over the same time period). Since we can’t observe the latter, we compare the group of smokers to a different group of non-smokers. We subsitute measures of effect with measures of association.

Question 4

Sufficient-component cause model - sufficient cause, compontent cause, necessary cause

Answer 4

1. Sufficient cause: minimum set of conditions that are needed for outcome to occur. Note that a condition may have many sufficient causes
2. Within a sufficient cause each condition constitutes a “component cause.” Each component cause acts simultaneously or at different times to cause the disease. Within a sufficent cause, a disease will not occur if one component cause is missing.
3. Necessary cause: component cause which must be present in order for disease to occur

Question 5

Graphical models

Answer 5

Visually represent investigators assumptions about causal relationships among exposure, outcome and covariates (directed acyclic graphs, DAGs). Simple graphical model is typical triangle depicting confounding.

Question 6

Bradford-Hill criteria (5)

Answer 6

1. Time: cause must precede effect
2. Strength of association: if factor is causal, then there will be a strong positive association between the factor and the disease
3. Biological gradient: if there is a dose-response relationship between factor and disease then the plausibility of a factor being causal are increased
4. Consistency: if association exists in a variety of different circumstances (different places, different samples) then it is more plausible that a causal association exists
5. Plausibility: more likely that causal relationship exists if a biological mechanism has been identified (though limited by current understanding)

Question 7

Henle-Koch postulates - posulates (4), limitations (2)

Answer 7

1. Agent can be consistently recovered from animals with disease
2. Agent cannot be isolated from animals without the disease
3. Once isolated and grown in culture, agent can induces disease in experimentally infected animals
4. Agent can be recovered from experimentally infected animal

Limitations:

• focus on bacteriology methods (what if agent can’t be grown?);
• ignores multifactorial nature of many diseases (agent may not induce disease unless other conditions are present e.g. immunocompromise)

Question 8

Objectives of epidemiology (5)

Answer 8

1. Determine of the origin/cause of disease whose cause is unknown
2. Describe ecology and natural history of disease
3. Investigate and control of a disease whose cause is unknown or poorly understood
4. Plan, monitor and assess disease control programs
5. Assess economic effects of disease and associated control programs

Question 9

Types of epidemiological investigation

Answer 9

1. Descriptive epidemiology: observing and recording diseases and possible risk factors
2. Analytic epidemiology: assessing link between possible risk factors and observations
3. Experimental epidemiology: observations from groups of animals that can be selected and modified
4. Theoretical epidemiology: mathematical modeling and simulation

Question 10

Epidemiology sub-disciplines

Answer 10

1. Clinical epidemiology - basis for EBM 2. Computational epidemiology 3. Genetic epidemiology - inherited defects 4. Field epidemiology 5. Participatory epidemiology - local knowledge 6. Molecular epidemiology