Causal Inference

Question 1

What was the debate regarding red meat consumption

Answer 1

• WHO and other organizations did a series of reviews on red meat/processed meat as probably carcinogenic
• this study done by Russ and Dena suggested that individuals do not change their red meat consumption
• controversial conclusion

Question 2

What is the optimal study design for making causal inferences?

Answer 2

• RCT
• assign participants at random to interventions
• ensures balance in known and unknown prognostic factors
• any differences in the groups can then be attributed to the intervention

Question 3

What is the revised hierarchy of evidence?

Answer 3

• recognizes that some study designs lower on the hierarchy can sometimes be more useful if they are rigoursly conducted (rigorous case control may be better than poorly conducted RCT)
• also removes systematic review/meta analysis from the top because it recognizes that the quality of these is highly dependent on the quality of the primary studies

Question 4

What is the first step to investigate a causal relationship?

Answer 4

• look at evidence from randomized control trials
• in red meat example, did systematic review of RCTs randomizing participants to lower and higher amounts of red meat
• found that evidence from the RCTs does not support a causal relationship between red meat intake and adverse health outcomes
• there were problems with these RCTs (design limitations) so moved onto cohort studies

Question 5

How are cohort studies normally conducted in nutrition?

Answer 5

• recruit participants and measure intake of various nutritional exposures through questionnaires or biomarkers and follow up over years
• correlate exposures with health outcomes

Question 6

What did the review of cohort studies suggest with red meat consumption?

Answer 6

-suggested positive association between consumption of red meat and cardiovascular mortality, MI, cancer mortality, etc.

Question 7

Why might we question the results found in the cohort studies?

Answer 7

• confounding bias (most important)
• selection bias
• bias in classification of the exposure
• bias due to missing data
• bias in measurement of the outcome
• bias in selection of the reported results

Question 8

What biases can also affect RCTs?

Answer 8

• selection bias
• bias in classification of the exposure
• bias due to missing data
• bias in measurement of the outcome
• bias in selection of the reported results

Question 9

What is confounding bias?

Answer 9

• distortion of the association between an exposure and outcome due to the association of the exposure with other prognostic factors that also influence the outcome
• confounding bias is not a concern with RCTs because in an RCT you randomize people to conditions so you get a balance of the confounders

Question 10

What is selection bias?

Answer 10

• bias in the estimated effect of the exposure on an outcome that arises from the procedures used to select individuals into the study or analysis
• RCTs and observational studies

Question 11

Can misclassification of exposure happen in RCT?

Answer 11

• no, RCTs people are assigned an exposure
• no bias related to measuring the exposure

Question 12

When does bias from missing data occur?

Answer 12

-participants lost to follow up, study participants experiencing deteriorating help so they can’t make study visits, exposure associated with deteriorating health (could make exposure seem less harmful than it is because the people who are getting sick from it are dropping out of study)

Question 13

Can bias in the measurement of the outcome affect RCTs?

Answer 13

• yes, can affect RCTs and cohort studies
• differential and non-differential

Question 14

What criteria are in place that help determine if a relationship is causal?

Answer 14

-temporal relationship

-strength of the association

-dose-response relationship

-replication of the findings

-cessation of exposure

• biologic plausability
• consistency with other knowledge
• specificity of the association
• consideration of alternate explanations

*not all of these have to be met for a relationship to be causal

Question 15

Describe temporal relationship

Answer 15

• if a factor causes a health outcome, the exposure to the factor should precede the health outcome
• why case control is not ideal to establish causal relationship

Question 16

Describe strength of the association

Answer 16

• the greater the magnitude of the association between exposure and outcome, the higher the certainty of evidence supporting a causal relationship
• RR greater than 2 or less than 0.5 is considered to be large
• compelled to believe stronger associations because it is unlikely that the entire association could be explained away by other factors or biases

Question 17

Describe dose-response association

Answer 17

• a relationship in which increasing levels of exposure, either in the frequency, magnitude, or duration, is associated with either increasing or decreasing levels of the outcome of interest
• more confident because the biases are unlikely to produce a dose response association (may produce a one off effect but unlikely to see a pattern if it is due to biases)

Question 18

Describe replication of findings

Answer 18

• the effect is consistently observed across different studies and populations
• effect could be attributed to a bias particular to one study design or a confounder that acts only in one population or due to chance

Question 19

Describe cessation of exposure

Answer 19

• the effects of an exposure should decline or be eliminated completely once the exposure is removed
• a potential problem: only select situations where it can be applied- pathogenic pathway for the disease may have already been initiated so stopping the exposure may not reduce your risk of the outcome

Question 20

Describe biologic plausability

Answer 20

• the consistency of the observed association with biological knowledge
• provides weak evidence for causal relationship because it is subjective

Question 21

Describe consistency with other knowledge

Answer 21

• the consistency of the observed association with forms of evidence
• weak
• evidence coming from study designs lower on the hierarchy have higher risk of bias so if there are inconsistencies, they could just be due to other biases

Question 22

Describe specificity

Answer 22

• an association is specific when the exposure of interest is associated with only one disease and when that disease is only caused by the exposure of interest
• many exposures associated with adverse health outcomes so in real life this doesn’t really play out

Question 23

Describe consideration of alternate explanations

Answer 23

• an association is more likely to be causal when there are no alternate explanations for the observed association
• not very useful

Question 24

Which criteria were fulfilled for the association between red meat and adverse health outcomes?

Answer 24

• temporality because most of evidence came from prospective cohort studies which assign exposure which precedes outcome
• strength of association NOT met (small RR)
• dose response relationship observed (nutritional exposures are hard to study- food exposures are very densely correlated with each other so dose response association makes us less confident because it may be that other exposures correlated with the exposure of interest are confounding the association)
• consistent effect across different studies (minority reported negative (in asian pop) or no associations- suggests that some positive associations observed in other studies may be due to confounding factors not present in Asian populations)
• cessation of exposure; don’t know yet

Question 25

What methods are used to facilitate causal inference?

Answer 25

• covariate adjustment: analytic method to adjust for the effects of confounding factors (including confounding variables as covariates which removes their confounding effects- can only do this with confounders we know about and have measured in the study)
• matching: matches participants according to confounding factors and eliminates the effects of confounders

Question 26

What is the GRADE system? When is data rated highly versus poorly? **on exam

Answer 26

• can be used to evaluate the certainty of evidence supporting causal relationships across different studies (looking at TOTALITY of evidence not a specific study)
• RCTs start at high quality/certainty evidence
• observational studies start at low quality/certainty evidence
• certainty may be downgraded for risk of bias, inconsistency (different studies produce different results), indirectness (question addressed by studies is different from the question of interest), imprecision (studies are too small to draw a reliable conclusion), publication bias
• certainty may be rated up if there is a large effect, dose-response, and suspected biases act against observed direction of effect