From: Mediation analysis methods used in observational research: a scoping review and recommendations
Methodological characteristics | Overall (n = 174) | Causal steps, change-in-coefficient, and joint significance (n = 28) | Traditional mediation analysis (n = 123) | Causal mediation analysis (n = 23) |
---|---|---|---|---|
Year published, n (%)a | ||||
 2015 | 21 (12.1) | 4 (14.3) | 16 (13.0) | 1 (4.3) |
 2016 | 29 (16.7) | 8 (28.6) | 16 (13.0) | 5 (21.7) |
 2017 | 27 (15.5) | 2 (7.1) | 21 (17.1) | 4 (17.4) |
 2018 | 47 (27.0) | 6 (21.4) | 34 (27.6) | 7 (30.4) |
 2019 | 50 (28.7) | 8 (28.6) | 36 (29.3) | 6 (26.1) |
Study design, n (%)a | ||||
 Cross-sectional | 84 (48.3) | 15 (53.6) | 64 (52.0) | 5 (21.7) |
 Case-control | 7 (4.0) | 1 (3.6) | 3 (2.4) | 3 (13.0) |
 Prospective cohort | 78 (44.8) | 12 (42.9) | 53 (43.1) | 13 (56.5) |
 Retrospective cohort | 5 (2.9) | 0 (0.0) | 3 (2.4) | 2 (8.7) |
 Analytical sample size, median (IQR) | 428.5 (157.5–2026.0) | 443.5 (146.5–1442.3) | 326.0 (147.0–935.0) | 2249.0 (703.0–9484.0) |
Software program used, n (%)a | ||||
 SAS | 21 (12.1) | 4 (14.3) | 9 (7.3) | 8 (34.8) |
 Stata | 27 (15.5) | 5 (17.9) | 18 (14.6) | 4 (17.4) |
 SPSS | 67 (38.5) | 13 (46.4) | 54 (43.9) | 0 (0.0) |
 Mplus | 26 (14.9) | 1 (3.6) | 25 (20.3) | 0 (0.0) |
 R | 14 (8.0) | 0 (0.0) | 5 (4.1) | 9 (39.1) |
 LISREL | 1 (0.6) | 0 (0.0) | 1 (0.8) | 0 (0.0) |
 Unclear, multiple programs used | 5 (2.9) | 1 (3.6) | 4 (3.3) | 0 (0.0) |
 Not mentioned | 13 (7.5) | 4 (14.3) | 7 (5.7) | 2 (8.7) |
Number of exposures, n (%)a | ||||
 1 | 116 (66.7) | 19 (67.9) | 81 (65.9) | 16 (69.6) |
 2 | 35 (20.1) | 3 (10.7) | 26 (21.1) | 6 (26.1) |
 3 or more (max. 14) | 23 (13.2) | 6 (21.4) | 16 (13.0) | 1 (4.3) |
Type of exposure, n (% of studies)b | ||||
 Continuous | 105 (60.3) | 15 (53.6) | 84 (68.3) | 6 (26.1) |
 Binary | 51 (29.3) | 8 (28.6) | 28 (22.8) | 15 (65.2) |
 Categorical | 20 (11.5) | 5 (17.9) | 11 (8.9) | 4 (17.4) |
 Latent | 10 (5.7) | 2 (7.1) | 8 (34.8) | 0 (0.0) |
Number of mediators, n (%)a | ||||
 1 | 86 (49.4) | 14 (50.0) | 59 (48.0) | 13 (56.5) |
 2 | 35 (20.1) | 5 (17.9) | 28 (22.8) | 2 (8.7) |
 3 or more (max. 19) | 53 (30.5) | 9 (32.1) | 36 (29.3) | 8 (34.8) |
Type of mediator, n (% of studies)b | ||||
 Continuous | 136 (78.2) | 19 (67.9) | 103 (83.7) | 14 (60.9) |
 Binary | 31 (17.8) | 10 (35.7) | 13 (10.6) | 8 (34.8) |
 Categorical | 9 (5.2) | 3 (10.7) | 3 (2.4) | 3 (13.0) |
 Count | 1 (0.6) | 0 (0.0) | 0 (0.0) | 1 (4.3) |
 Time-to-event | 2 (1.1) | 0 (0.0) | 2 (1.6) | 0 (0.0) |
 Latent | 10 (5.7) | 3 (10.7) | 7 (5.7) | 0 (0.0) |
Number of outcomes, n (%)a | ||||
 1 | 126 (72.4) | 22 (78.6) | 88 (71.5) | 16 (69.6) |
 2 | 26 (15.0) | 2 (7.1) | 22 (17.9) | 2 (8.7) |
 3 or more (max. 8) | 22 (12.6) | 4 (14.3) | 13 (10.6) | 5 (21.6) |
Type of outcome, n (% of studies)b | ||||
 Continuous | 110 (63.2) | 18 (64.3) | 86 (69.9) | 6 (26.1) |
 Binary | 47 (27.0) | 6 (21.4) | 27 (22.0) | 14 (60.9) |
 Categorical | 6 (3.4) | 2 (7.1) | 3 (2.4) | 1 (4.3) |
 Count | 4 (2.3) | 0 (0.0) | 2 (1.6) | 2 (8.7) |
 Time-to-event | 8 (4.6) | 3 (10.7) | 3 (2.4) | 2 (8.7) |
 Latent | 10 (5.7) | 3 (10.7) | 7 (5.7) | 0 (0.0) |
Diagram of model included, n (%)a | ||||
 Yes | 130 (74.7) | 14 (50.0) | 106 (86.2) | 10 (43.5) |
 No | 44 (25.3) | 14 (50.0) | 17 (13.8) | 13 (56.5) |
Repeated measurements, n (%)a | ||||
 Yes | 41 (23.6) | 6 (21.4) | 28 (22.8) | 7 (30.4) |
 No | 132 (75.9) | 22 (78.6) | 94 (76.4) | 16 (69.6) |
 Unclear | 1 (0.6) | 0 (0.0) | 1 (0.8) | 0 (0.0) |
Type of mediation model, n (%)a | ||||
 Single mediator | 114 (65.5) | 20 (71.4) | 74 (60.2) | 20 (87.0) |
 Multiple mediator | 41 (23.6) | 5 (17.9) | 34 (27.6) | 2 (8.7) |
 Both single and multiple mediator | 16 (9.2) | 3 (10.7) | 12 (9.8) | 1 (4.3) |
 Unclear | 3 (1.7) | 0 (0.0) | 3 (2.4) | 0 (0.0) |
Regression for the mediator equation, n (% of studies)b | ||||
 Linear | 122 (70.1) | 14 (50.0) | 98 (79.7) | 10 (43.4) |
 Logistic | 15 (8.6) | 2 (7.1) | 7 (5.7) | 6 (26.1) |
 Poisson | 1 (0.6) | 0 (0.0) | 0 (0.0) | 1 (4.3) |
 Cox proportional hazards | 2 (1.1) | 0 (0.0) | 2 (1.6) | 0 (0.0) |
 Multilevel linear | 7 (4.0) | 0 (0.0) | 7 (5.7) | 0 (0.0) |
 Multilevel logistic | 1 (0.6) | 1 (3.6) | 0 (0.0) | 0 (0.0) |
 Unclear | 16 (9.2) | 2 (7.1) | 8 (6.5) | 6 (26.1) |
 Not estimated | 12 (6.9) | 9 (32.1) | 3 (2.4) | 0 (0.0) |
Regression for the outcome equation, n (% of studies)b | ||||
 Linear | 109 (62.6) | 17 (60.7) | 87 (70.7) | 5 (21.7) |
 Logistic | 39 (22.4) | 5 (17.9) | 23 (18.7) | 11 (47.8) |
 Probit | 1 (0.6) | 0 (0.0) | 1 (0.8) | 0 (0.0) |
 Log-linear | 1 (0.6) | 0 (0.0) | 0 (0.0) | 1 (4.3) |
 Multinomial logistic | 1 (0.6) | 1 (3.6) | 2 (1.6) | 0 (0.0) |
 Poisson | 3 (1.7) | 0 (0.0) | 0 (0.0) | 3 (13.0) |
 Negative binomial | 3 (1.7) | 0 (0.0) | 1 (0.8) | 2 (8.7) |
 Cox proportional hazards | 10 (5.7) | 4 (14.3) | 4 (3.3) | 1 (4.3) |
 Additive hazards model | 1 (0.6) | 0 (0.0) | 0 (0.0) | 1 (4.3) |
 Multilevel linear | 9 (5.2) | 2 (7.1) | 7 (5.7) | 0 (0.0) |
 Multilevel logistic | 1 (0.6) | 0 (0.0) | 1 (0.8) | 0 (0.0) |
 Multilevel log-linear | 1 (0.6) | 0 (0.0) | 1 (0.8) | 0 (0.0) |
 Unclear | 3 (1.7) | 1 (3.6) | 1 (0.8) | 1 (4.3) |
Type of confidence interval for the indirect effect, n (%)a | ||||
 Normal-based | 7 (4.0) | 0 (0.0) | 6 (4.9) | 1 (4.3) |
 Percentile bootstrap | 7 (4.0) | 0 (0.0) | 5 (4.1) | 2 (8.7) |
 Bias-corrected bootstrap | 35 (20.1) | 0 (0.0) | 35 (28.5) | 0 (0.0) |
 Bias-corrected and accelerated bootstrap | 2 (1.1) | 0 (0.0) | 2 (1.6) | 0 (0.0) |
 Percentile and bias-corrected and accelerated bootstrap | 1 (0.6) | 0 (0.0) | 1 (0.8) | 0 (0.0) |
 Bootstrap, not specified | 36 (20.7) | 1 (3.6) | 28 (22.8) | 7 (30.4) |
 Distribution of the product | 3 (1.7) | 0 (0.0) | 3 (2.4) | 0 (0.0) |
 Monte Carlo | 2 (1.1) | 0 (0.0) | 2 (1.6) | 0 (0.0) |
 Bayesian credible intervals | 3 (1.7) | 0 (0.0) | 3 (2.4) | 0 (0.0) |
 Unclear | 18 (10.3) | 0 (0.0) | 8 (6.5) | 10 (43.5) |
 Not reported | 60 (34.5) | 27 (96.4) | 30 (24.4) | 3 (13.0) |
Standard error for indirect effect, n (%)a | ||||
 Yes | 37 (21.3) | 0 (0.0) | 35 (28.5) | 2 (8.7) |
 No | 137 (78.7) | 28 (100.0) | 88 (71.5) | 21 (91.3) |
Statistical test for indirect effect, n (%)a | ||||
 Yes | 62 (35.6) | 7 (25.0) | 46 (37.4) | 9 (39.1) |
 No | 112 (64.4) | 21 (75.0) | 77 (62.6) | 14 (58.3) |
Effect size measure, n (%)a | ||||
 Proportion mediated | 65 (37.4) | 3 (10.7) | 43 (35.0) | 19 (82.6) |
 Standardized effect | 6 (3.4) | 0 (0.0) | 6 (4.9) | 0 (0.0) |
 No effect size measure | 103 (59.2) | 25 (89.3) | 74 (60.2) | 4 (17.4) |
Inclusion of confounders, n (%)a | ||||
 Yes | 125 (71.8) | 19 (67.9) | 88 (71.5) | 18 (78.3) |
 No | 41 (23.6) | 8 (28.6) | 31 (25.2) | 2 (8.7) |
 Unclear | 8 (4.6) | 1 (3.6) | 4 (3.3) | 3 (13.0) |
Sensitivity analyses for unmeasured confounders, n (%)a | ||||
 Yes | 3 (1.7) | 0 (0.0) | 0 (0.0) | 3 (13.0) |
 No | 170 (97.7) | 28 (100.0) | 123 (0.0) | 19 (82.6) |
 No, but discussed as unnecessary | 1 (0.6) | 0 (0.0) | 0 (0.0) | 1 (4.3) |
Effect modifiers considered, n (%)a | ||||
 Yes, a priori stratification | 10 (5.7) | 3 (10.7) | 5 (4.1) | 2 (8.7) |
 Yes, through an interaction term | 28 (16.1) | 6 (21.4) | 19 (15.4) | 3 (13.0) |
 No | 136 (78.2) | 19 (67.9) | 99 (80.5) | 18 (78.3) |
Assessment of exposure-mediator interaction, n (%)a | ||||
 Yes | 17 (9.8) | 3 (10.7) | 5 (4.1) | 9 (39.1) |
 No | 157 (90.2) | 25 (89.3) | 118 (95.9) | 14 (60.9) |