Skip to main content

Table 2 Characteristics of mediation analyses with time-to-event outcome in healthcare research, n = 149

From: Mediation analysis with a time-to-event outcome: a review of use and reporting in healthcare research

Included study characteristic

Result

Funding source, n (%)

 

 Government

113 (76)

 Foundation

37 (25)

 Hospital

6 (4)

 Industry

6 (4)

 University

4 (3)

 Professional association

1 (< 1)

 None stated

13 (9)

Study design, n (%)

 Cohort

131 (88)

 Randomised Controlled Trial

8 (5)

 Case-cohort

5 (3)

 Case control

4 (3)

 Cross-sectional

1 (< 1)

Type of analysis, n (%)

 

 Confirmatory/Hypothesis-based

72 (48)

 Exploratory

74 (50)

 Not able to infer

3 (2)

Mediation analysis is primary aim of study, n (%)

69 (46)

Multiple mediators tested, n (%)

76 (51)

Type of mediator, n (%)

 

 Continuous

60 (40)

 Binary

56 (38)

 Categorical

25 (17)

 Interval/Ordinal

25 (17)

 Latent

8 (5)

Most common content of mediatora, n (%)

 

 Physiologic (e.g. blood pressure, heart rate, weight)

34 (23)

 Psychological/psychiatric

32 (21)

 Lifestyle (e.g. alcohol, smoking, nutrition, exercise, sleep)

31 (21)

 Biomarker (blood test results)

24 (16)

 Health

17 (11)

 Comorbidity

13 (9)

 Treatment

8 (5)

 Functioning

8 (5)

 Socioeconomic

8 (5)

 Environment

6 (4)

 Reproductive

2 (1)

Most common outcomes, n (%)

 

 New medical condition or exacerbation of an existing condition

68 (46)

 All-cause mortality

48 (32)

 Cause-specific mortality

21 (14)

 Disability or sick leave

6 (4)

Causal diagram included, n (%)

 

 Causal steps/change in coefficient (n = 87)

22 (25)

 Counterfactuals (n = 32)

16 (50)

 SEM/path (n = 23)

18 (78)

 Product of coefficients (n = 6)

3 (50)

 Cannot infer (n = 1)

0 (0)

Sample size, median (IQR)

3345 (637–16,061)

Power/sample size, n (%)

 

 Calculation

1 (< 1)

 Consideration

3 (2)

Method of mediation analysis, n (%)

 

 Causal steps, including Baron-Kenny

41 (28)

 Change in coefficient in a single regression

46 (31)

 Counterfactuals

32 (21)

 SEM/path

23 (15)

 Product of coefficients

6 (4)

 Cannot infer

1 (< 1)

Statistical tests for no mediation/indirect effect, n (%)

 

 Sobel

7 (5)

 Other product test

14 (9)

 Difference test

2 (1)

 Z-test of mediated proportion

1 (< 1)

 Joint significance test

1 (< 1)

 Olaf & Finn test

1 (< 1)

Type of time-to-event model, n (%)

 

 Cox proportional hazard

114 (77)

 Additive hazard

10 (7)

 Linear

7 (5)

 Discrete time survival model

6 (4)

 Failure time/parametric survival

5 (3)

 Marginal structural model

3 (2)

 Log linear Poisson

1 (< 1)

 Quantile regression

1 (< 1)

 Cannot infer

5 (3)

Specific mediation software mentioned, n

 

 Causal steps/change in coefficient

 

  SAS “mediate” macro

2

  PRODCLIN

1

 Counterfactuals

 

  R

8

  R “mediation”

2

  SAS

1

  SAS “mediate” macro

1

  STATA “medeff”

1

 SEM/path

 

  Mplus

13

  SAS

1

  STATA mediation package

1

  LISREL

1

Competing risks considered, n (%)

4 (3)

If clustering of data, was this addressed in the analysis? n (%)

 

 Not multilevel

114

 Yes

19 (54)

 No

8 (23)

 Cannot determine

8 (23)

Cox models, outcome frequencyb, n (%)

 

 > or equal to 5%

74 (65)

 > or equal to 10%

55 (48)

Rare outcome limitation for Cox model mentionedb, n (%)

8 (7)

Temporal separation clearly defined, n (%)

 

 Yes

37 (25)

 Overlap exposure and mediator

89 (60)

 Overlap mediator/outcome

7 (5)

 Cannot determine

19 (13)

 Acknowledged as a limitation

20 (13)

Mediation assumptions (or limitation) stated, n (%)

 

 No unmeasured confounding of exposure/outcome

29 (19)

 No unmeasured confounding of mediator/outcome

29 (19)

 No unmeasured confounding of exposure/mediator

22 (15)

 No exposure-dependent confounding of mediator-outcome

17 (11)

 Accurate measurement of mediator

31 (21)

Interaction between exposure and mediator considered/tested, n (%)

46 (31)

Method to address confounding of exposure (more than one can be used), n (%)

 

 Regression/modelling

137 (92)

 Stratification/restriction

14 (9)

 Randomisation

6 (4)

 None

9 (6)

Method to address confounding of mediator (more than one can be used), n (%)

 

 Regression/modelling

138 (93)

 Weighting

13 (9)

 Stratification/restriction

13 (9)

 Matching

1 (< 1)

 None

10 (7)

Sensitivity analysis related to mediation analysis, n (%)

 

 Any

25 (17)

 Confounding

8 (5)

 Accurate measurement/specification of mediator

7 (5)

 Temporal sequence assumption

6 (4)

 Testing a combined mediator or all mediators in same model

5 (3)

 Interaction/moderation

2 (1)

Measures of mediation reported, n (%)

 Causal steps/change in coefficient method (n = 87)

 

  Indirect effect

7 (8)

  Proportion mediated

52 (60)

 Counterfactuals (n = 32)

 

  Indirect effect

29 (91)

  Proportion mediated

22 (69)

 SEM/path (n = 23)

 

  Indirect effect

16 (70)

  Proportion mediated

5 (22)

 Other (n = 7)

 

  Indirect effect

3

  Proportion mediated

4

Measures of precision reported, n (%)

 Causal steps/change in coefficient (n = 87)

 

  Indirect effect confidence interval

6 (7)

  Proportion mediated confidence interval

17 (20)

  Statistical test p-value or equivalent

10 (11)

 Counterfactuals (n = 32)

 

  Indirect effect confidence interval

29 (91)

  Proportion mediated confidence interval

14 (44)

 SEM/path (n = 23)

 

  Indirect effect confidence interval

15 (65)

  Proportion mediated confidence interval

2 (9)

  Statistical test p-value or equivalent

4 (17)

 Other (n = 7)

 

  Indirect effect confidence interval

3

  Proportion mediated confidence interval

3

  Statistical test p-value or equivalent

2

  1. aTotal exceeds 100% because of multiple mediators in many studies
  2. bDenominator is 114