Twin Designs and Causal Inference PDF

Twin designs and causal inference 12 November 2023 21:03 Main Ideas Notes What are the concerns? ○ Causal effect of exposure A on outcome B ○ Spurious relationship between exposure A and outcome B Notes ○ Matching ▪ Match each treated individual with one control who is the same age (or close) and el...

Twin designs and causal inference 12 November 2023 21:03 Main Ideas Notes What are the concerns? ○ Causal effect of exposure A on outcome B ○ Spurious relationship between exposure A and outcome B Notes ○ Matching ▪ Match each treated individual with one control who is the same age (or close) and eliminate the others. ▪ The new treatment effect estimation is unbiased by age. ▪ Genetic conformation and causality (Pingault et al., 2018) Summary PSYC0036 Genes and Behaviour Page 1 Triangulation ○ Different designs off ▪ The effect of fa ○ Different strengths a ▪ CoT assumes th ○ No one design gives a findings (e.g. smokin ○ ○ But what happens if we change A? ○ Causal inference is essential in psychopathology ▪ Can be used to identify causes of diseases as targets for interventions ○ There is a difference between correlation and cause ○ Genetic confounding: This occurs when the exposure and the outcome are both associated with a common genetic factor. ▪ For example, if smoking is associated with both lung cancer and a certain genetic variant, then that genetic variant could confound the relationship between smoking and lung cancer. ○ Misidentification: This occurs when the exposure is measured incorrectly. ▪ For example, if self-reported smoking is used as a measure of exposure, but some people misreport their smoking status, then this could misrepresent the true relationship between smoking and the outcome. ○ Active gene-environment correlation: This occurs when the exposure is influenced by the genotype. ▪ For example, people with a certain genotype may be more likely to smoke. ▪ In this case, the genotype is both a cause of the exposure and a cause of the outcome (lung disease), which can confound the relationship between the exposure and the outcome. ○ Passive gene-environment correlation: This occurs when the exposure is influenced by the genotype of another person. ▪ For example, a child raised by a smoking parent is exposed to second-hand smoke, even if the child does not smoke themselves. ▪ In this case, the genotype of the parent is a cause of the child's exposure, which can confound the relationship between the child's exposure and the outcome. ○ Reverse causation: This occurs when the outcome causes the exposure, rather than the other way around. ▪ For example, people with lung cancer may be more likely to smoke. ▪ In this case, lung cancer is a cause of smoking, rather than smoking being a cause of lung cancer. Counterfactual ○ Counterfactual approach to theorize causal inference: ▪ The fact is the administered treatment (e.g. anti-depressant) ▪ The “counterfactual” would be the same person, a “clone” without the treatment ▪ But mutually exclusive: same person cannot be administered and not administered the treatment at the same time ○ Theoretical situation: ▪ Only difference between the treatment group and the counterfactual group is being administered or not the treatment ▪ Any difference between the 2 groups is caused by the treatment Notes Using RCT Design as an approximation of the counterfactual model Twin and co-twin as counterfactuals ○ Co-twin: as close to a clone as it gets? ▪ Share 100% of their genetic material ▪ Share common environmental influences ▪ Perfect match ○ However, one twin can be exposed to a risk factor whilst the other is not (they are discordant): ▪ e.g. smoking and lung cancer ▪ e.g. peer victimization and depression □ => estimate causal effect by comparing exposed vs. non-exposed twins Stressful life-event and depression (Kendler et al., 2010) ○ Question: Do dependent stressful life-events have a causal effect on Major Depression (MD). ▪ A large proportion of the observed effect is due to environmental and genetic confounding. ▪ Still seems that the link is causal as shown by the discordant MZ design and propensity score matching estimates. ○ Study shows that the effect of dSLEs on MD is stronger in female twins than in male twins. ▪ This suggests that women may be more vulnerable to the effects of stressful life events ○ odds ratio for MD is significantly higher in people who experience a dSLE than in those who do not, even after controlling fo r genetic and environmental confounding ○ Children of twins design (CoT) (D'Onofrio, 2013) ○ Aim: examine the nature of the relationships between parental risk factors and children outcomes ○ Participants: ▪ Adult MZ and DZ pairs ▪ And their children ○ Similar or different phenotypes can be examined in parents & children: ▪ Parental and child IQ ▪ Smoking during pregnancy and child birthweight (D'Onofrio, 2013) ○ D'Onofrio (2013) findings ▪ Pattern A: Direct environmental influence causes intergenerational relationship. □ Children of affected parents are more likely to be affected than children of unaffected parents. □ There is no difference in the rates of disorder between children of MZ twins and children of DZ twins. □ Shows that the trait is environmentally based ▪ Pattern B: Shared genotypic factors cause intergenerational relationship. □ The rates of disorder are the same for children of MZ twins, regardless of whether or not the parent is affected. □ Children of unaffected DZ twins are still affected at half the rate of children of MZ twins. □ Higher rates of trait in children of MZ but not children of DZ suggests genes account for at least some of the intergenerational transmission ▪ Pattern C: Effectively shared environment causes intergenerational relationship. □ All children are affected at the same rate, regardless of the affection status of the parent or whether the twins are MZ or DZ. □ Suggests shared environmental factors are the most important Estimation of the CoT model ○ The estimate of interest is p : ▪ is there a direct phenotypic relationship between parental phenotype (T1) and child's phenotype (C1), whilst controlling for intergenerational genetic (g) and shared environmental influences (s) er different possibilities amily-level variables (e.g. socioeconomic status) can be examined with CoT but not with discordant MZ and limitations he absence of assortative mating, which is not an issue in discordant MZ twins a definite answer and the convergence of results from different designs – i.e. triangulation – provides more confidence in the g during pregnancy and birthweight). PSYC0036 Genes and Behaviour Page 2