Skip to main content

Table 5 Elements of implementation for causal mediation analysis

From: Implementation and reporting of causal mediation analysis in 2015: a systematic review in epidemiological studies

References

Sample sizea

Power and sample size calculation for mediation analysis

Exposure-mediator interaction in the model

Power and sample size calculation for interaction analysis

Lack of power mentioned as a non-causal explanation of findings

Studies with negative findings

 Freeman et al. [28]

N = 51 for adjusted and n = 54 for unadjusted analysis

80 % power to detect large indirect effects (Fritz M, Mackinnon DP. Required sample size to detect the mediated effect)

Not reported

Not applicable

Limited power to detect whether mediated effects were statistically significant

 Jackson et al. [29]

N = 26,197

Not reported

✓

No

No

 Kositsawat et al. [30]

N = 1765

Not reported

Not reported

Not applicable

No

 Louwies et al. [31]

N = 55

Not reported

Not reported

Not applicable

Called for cautious interpretation given small sample size

 Mendola et al. [33]

Varied by outcomes (i.e. 1 study site did not report infant apnea)

Not reported

✓

No

No

Studies with positive findings

 Banack et al. [26]

N = 7212

Not reported

✓

No

No

 D’Amelio et al. [27]

N = 37

80 % power, two-sided significance level of 0.05, to detect differences in uOC greater than 1.71 (t test on log-scale)

✓

No

Acknowledged small sample size as a limitation

 Lu et al. [32]

N = 58,322 for metabolic

N = 19,572 for fibrinogen

Not reported

✓

No

No

 Messerlian et al. [34]

All singleton births (n = 18,147); only first births (n = 8651)

Not reported

Not reported

Not applicable

Limited power mentioned; limited sample size to evaluate spontaneous and induced preterm birth separately

 Raghavan et al. [35]

Varied by mediators (n = 2159 for diet score; and n = 2098 for physical activity index)

Not reported

Not reported

Not applicable

Lack of power to examine the association in the mediation framework

 Rao et al. [36]

N = 433

Not reported

X

No

No

 Song et al. [37]

Varied by mediators (biomarkers)

Not reported

✓

No

Despite some significant mediation by several biomarkers, the sample size may still not be large enough to provide more precise estimates or to detect mediation by other potential factors with smaller mediation effects

 Xie et al. [38]

N = 167

Not reported

X

No

Small sample size acknowledged

  1. aSmallest sample size used in the causal mediation analysis