causal evaluation (7)

Question 1

what is correlation

Answer 1

2 events occurring at the same time or in a temporal sequence

does not imply causation, try infer causation from correlation/associations

Question 2

reverse causation

Answer 2

depends on which stance you take
sugar intake –> diabetes
diabetets –> sugar intake

Question 3

DAG - directed acyclic graph

Answer 3

nodes are exposures, outcomes, covariates
arrows indicate causal relationship
acyclic, no directed loops as factor can’t cause itself

Question 4

ruling out reverse causation

Answer 4

establish temporal relationship, exposure comes before outcome

Question 5

bias

upwards an downwards

Answer 5

bias is systematic error, associations could be deviated from the truth
upward is bias against the null
downward is bias towards the null

Question 6

what is a confounder

Answer 6

common causes or related to both of exposure and outcome could cause up or down bias

Question 7

ruling out confounder bias

Answer 7

difficult in observational studies due to unobserved or unknown confounders
residual confounding

Question 8

residual confounding

Answer 8

confounding still there after a variable has not been adequately adjusted for
eg smoking yes or no without dosage

Question 9

what to do for know or observed confounders

Answer 9

adjustment
restriction
matching
stratification
standardisation

Question 10

matching

Answer 10

matching exposed/unexposed (control vs case) by the confounder
eg income in a study for expensive supplements and diabetes

Question 11

restriction (easiest)

Answer 11

only restrict comparison in certain confounder subgroup

eg only recruit people with income greater than 80k

Question 12

stratification

Answer 12

recruit everyone but compare within each confounder subgroup

eg compare diabetic risk for each income category not as a whole

Question 13

standardisation

Answer 13

applies a standard population to the stratified RRs

reduces confounding and background factors when comparing incidence/ prevalence between populations

Question 14

direct standardisation

Answer 14

observed age specific rates in the study population are applied to a standard (sometimes artificial) population structure

Question 15

direct standardisation for death rate

Answer 15

(deaths/pop) x standard pop = standardised death rate

Question 16

direct standardisation method

Answer 16

each age group calculate specific rate, apply each rate to standardised pop.
add standardised rates to give overall age standardised rate

Question 17

indirect standardisation

Answer 17

use age specific rates for own pop and pop comparing with

shows what we would expect to see in our pop if occurrences where equivalent to the comparator

Question 18

direct pros cons

Answer 18

pop can be compared with each other
events with small numbers may be unstable
can compare over time

Question 19

indirect pros cons

Answer 19

compare to standards not each other

better for small pop./rare events because age specific deaths not needed

Question 20

standardisation for adjustment

Answer 20

by categorical variables eg age/sex

standardised rate can be used in regression

Question 21

mediators

Answer 21

intermediate variable between exposure and outcome
caused by exposure
causes outcome
obesity->diabetes->heart attack
obesity->heart attack
but diabetes is mediator

Question 22

mediated and unmediated associations

Answer 22

can split association between obesity and heart attack to association via diabetes (mediated)
or not via diabetes (unmediated)

Question 23

over adjustment bias

Answer 23

don’t adjust for mediators when looking for total association between exposure and outcome
adjusting for mediators = bias towards the null