Lecture 11- causality

association

statistical dependence between two events, characteristics or other variables, doesn’t mean causation

risk factor

an exposure that increases the chance or probability of getting the outcome, not necessarily a cause of disease, could be a surrogate for underlying causes

causation

potential for changing an outcome by changing the exposure 

criteria for causality- strength 

strong associations may likely be causal (later refuted), RR or OR of 10 is more likely to be a causal association than an RR or OR of 2

criteria for causality- consistency 

an association has been observed repeatedly, it is likely a causal association

criteria for causality- specificity

association is constrained to a particular exposure-outcome relationship, in a specific association, a given disease results from a given exposure and not from other types of exposures

criteria for causality- temporality

the cause must be observed before the effect, only criterion that holds true in all cases

criteria for causality- biological gradient

dose-response curve, shows a linear trend in the association between exposure and disease

criteria for causality- plausibility

the association must be biologically plausible from the standpoint of contemporary biological knowledge, does the association make sense?, lab-based studies are helpful

criteria for causality- coherence 

the cause-and-effect interpretation of our data should not seriously conflict with the generally known facts of the natural history of the disease

criteria for causality- experiment

evidence from experiments can help support the existence of a causal relationship

criteria for causality- analogy

relates to the correspondence between known associations and one that is being evaluated for causality

deterministic causality

involves a deterministic model, claims that a cause is invariably followed by an effect, leaves nothing to chance, employs “necessary” and “sufficient” causes

necessary cause

a factor whose presence is required for the occurrence of the effect, every case of the disease will have this necessary cause as the exposure factor

sufficient cause

a cause that is sufficient by itself to produce the effect, set of factors together present as sufficient

sufficient component cause model (causal pie model)

constituted from a group of component causes, which can be diagrammed as a pie, emphasizes multi-causality and a necessary cause

causes can be- necessary and sufficient

very uncommon in epidemiology

causes can be- sufficient but not necessary

smoking and lung cancer, not all smoker may get lung cancer

causes can be- necessary but not sufficient

exposure to the influenza virus and getting the flu, you won’t get it if you’re vaccinated

causes can be- neither necessary or sufficient 

mostly seen in chronic diseases that have multiple contributing causes

probabilistic causality

involves a probabilistic model also called a stochastic model

stochastic process

incorporates some element of randomness, chance, probability, a cause is associated with the increased probability that an effect will happen

role of chance in associations

assessed through hypothesis testing

hypothesis

any conjecture cast in a form that will allow it to be tested and refuted

null hypothesis

there is no association between the exposure and outcome, observed association is by chance

alternate hypothesis 

there is an association between the exposure and outcome, observed association is not by chance

alpha level

significance level, probability of making wrong decision about null hypothesis, probability of making a mistake, usually set at 0.05 in statistical tests, α = .05 means 5% probability of incorrectly rejecting true null

P value

most commonly p < 0.05 is the critical value for statistical significance

P value interpretation 

there is <5% probability that the observed association is explained by chance alone, statistically significant associations when p<0.05 (the alpha is set at 0.05), may not mean they are all clinically/biologically significant and meaningful

confidence interval

the range of values within which the true value will fall, 95% corresponds to the alpha level of 0.05, not statistically significant if the null value of the measure of association falls in the CI

null value

value that denotes no association between exposure and outcome, for OR and RR, if 1.0 (null value) is not in 95% CI, then statistically significant