Table 3 Sensitivity of meta-analyzed candidate predictors to unmeasured confounding

Variable	E-values^a
Variable	Effect estimate^b	95% CI^c
Relationship with mother when growing up
Ref: Very bad/somewhat bad	–	–
Very good/somewhat good	1.09	1.00
Relationship with father when growing up
Ref: Very bad/somewhat bad	–	–
Very good/somewhat good	1.25	1.14
Family structure around age 12
Ref: Parents were married	–	–
Parents were divorced	1.32	1.17
Parents were never married	1.16	1.00
One or both parents had died	1.20	1.00
Subjective financial status of the family around age 12
Ref: Got by	–	–
Lived comfortably	1.31	1.23
Found it difficult	1.29	1.18
Found it very difficult	1.30	1.14
Experienced abuse when growing up
Ref: No	–	–
Yes	1.86	1.72
Felt like an outsider in the family when growing up
Ref: No	–	–
Yes	1.68	1.54
Self-rated health when growing up
Ref: Good	–	–
Excellent	1.65	1.44
Very good	1.46	1.34
Fair	1.39	1.28
Poor	1.54	1.33
Immigration status
Ref: Born in this country	–	–
Born in another country	1.24	1.00
Frequency of religious service attendance around age 12
Ref: Never	–	–
≥1/week	1.20	1.00
1–3/month	1.28	1.06
<1/month	1.10	1.00
Year of birth
Ref: 1998–2005 (current age: 18–24 years)	–	–
1993–1998 (current age: 25–29 years)	1.11	1.00
1983–1993 (current age: 30–39 years)	1.10	1.00
1973–1983 (current age: 40–49 years)	1.10	1.00
1963–1973 (current age: 50–59 years)	1.27	1.00
1953–1963 (current age: 60–69 years)	1.24	1.00
1943–1953 (current age: 70–79 years)	1.34	1.00
1943 or earlier (current age: 80+ years)	2.41	1.00
Gender
Ref: Male	–	–
Female	1.37	1.25
Other	4.50	1.00

CI confidence interval.
^aThe formula for calculating E-values can be found in VanderWeele and Ding⁴⁹.
^bThe E-value for the effect estimate is the minimum strength of association (on the risk ratio scale) that an unmeasured confounder would need to have with both the predictor and the outcome to entirely explain away the observed association between them, conditional on the measured covariates.
^cThe E-value for the limit of the 95% confidence interval closest to the null denotes the minimum strength of association (on the risk ratio scale) that an unmeasured confounder would need to have with both the predictor and the outcome to shift the confidence interval to include the null value, conditional on the measured covariates.

Search