Skip to main content


Table 1 General description of epidemiological studies that were eligible for the systematic review

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

References Country, population, and sample size of studya Directed acyclic graph (DAG) included? Specific mediation hypothesis specified? Exposure Primary outcome Mediator Confoundersb
Randomized controlled trials
 D’Amelio et al. [27] Italy
Non-diabetic women with postmenopausal osteoporosis
(n = 46)
No Biologic mechanisms discussed All treated with calcium 1200 mg/day and cholecalciferol 800 UI/day
Randomized to with PTH 1–84 100 μg/day subcutaneous
Without PTH 1–84 100 μg/day subcutaneous (binary)
Glucose metabolism, (continuous, log scale) Total osteocalcin (OC) undercarboxylated (uOC)
Biomarkers that were unbalanced between the two treatment groups at baseline including uOC and serum tartrate resistant acid phosphatase 5B (TRAP5b)
 Freeman et al. [28] England
Patients with persecutory delusions from 6 mental health sites (n = 59)
No Guided by cognitive model of persecutory delusions Randomized to street exposure in areas of relative deprivation during busy mid-day
A neutral control condition which included sitting in a room watching mildly humorous television clips for 10 min (binary)
(1) State Paranoia using six visual analog scales (VAS)
(2) State social paranoia scale
(3) Schizotypal Symptoms Inventory—Paranoia (continuous)
Hallucinations VAS
Distress VAS
Anxiety VAS
Depression VAS
Brief core schema scales (BCSS)
Threat anticipation
Interpretation bias (continuous)
Reasoning measures
Jumping to conclusions
Possibility of being mistaken
Alternative explanations
Hypothetical contradiction (binary)
Probability of being mistaken (continuous)
Baseline measures of paranoia, all of the mediators considered, and center
Cohort studies
 Banack et al. [26] United States
Nationally representative noninstitutionalized
Sample of adults aged 20 to 80 years in the U.S. (1988–2004) (n = 7212)
Yes Guided by previous research Obesity defined as body mass index ≥30 kg/m2 vs. 18.5–29.9 kg/m2(binary) All-cause mortality with follow-up through 2006 (binary) Self-reported acute cardiac event (e.g. stroke or myocardial infarction) (binary) Age, gender, race, education, smoking status, and cardiorespiratory fitness
 Jackson et al. [29] New Jersey and Pennsylvania, United States
Older adults dually enrolled in medicare and pharmacy assistance programs; “new users” (n = 26,197)
No Mediators selected based on previous literature New user of first generation antipsychotic versus new user of second generation antipsychotic (binary) Mortality with 180 days (binary) Medical events stroke, ventricular arrhythmia, acute myocardial infarction, venous thromboembolism, pneumonia, bacterial infection (besides pneumonia), and hip fracture) (binary) 70 different demographic characteristics, health service utilization and medication usage, co-existing medical and psychiatric illness, and indicators of functional impairment
 Kositsawat et al. [30] Memphis, Tennessee and Pittsburgh, Pennsylvania, United States
Black and white medicare eligible—community dwelling adults aged 70–79 years without diabetes at year 2 of the study (n = 2193)
No Rationale not clear Serum vitamin D levels (25-hydroxyvitamin D) <20 ng/mL (binary) A1c level ≥6.5 % at year 4 (binary) Diabetes status at year 4 (binary) Confounders considered in mediation analysis not reported
 Louwies et al. [31] Belgium
Working nurses aged between 22 and 59 years without cardiovascular diseases and diabetes (n = 55)
No Guided by previous literature Subchronic black carbon exposure (continuous) Diastolic blood pressure
Systolic blood pressure
(continuous )
Retinal microcirculation (continuous) Age, sex, body mass index, smoking, use of anti-hypertensive medication, γ-GT, A1c, distance to major road, clinic, and average weekly temperature
 Lu et al. [32] United States
Adults free of coronary heart disease who participated in 9 National Heart, Lung, and Blood Institute funded cohort studies with body mass index ≥ 20 kg/m2 (1954–2001)
(n = 58,322 for metabolic risk factors; n = 19,572 for fibrinogen analysis)
Yes Biologic mechanisms discussed Body mass index (categories ≥30 kg/m2, 25–<30 kg/m2, 20–25 kg/m2)
(categorical and continuous)
First fatal or non-fatal occurrence of ischemic heart disease, acute myocardial infarction, or angina pectoris (binary) Explored in data combined from nine cohort studies
Systolic blood pressure, total serum cholesterol, glucose
Explored in data combined from three cohort studies
Fibrinogen, high-sensitive C-reactive protein (continuous)
Age, sex, smoking, race/ethnicity, socioeconomic status, alcohol intake, physical activity, and dietary intake
 Mendola et al. [33] United States
Singleton newborns with ≥23 weeks of gestation (n = 210,610)
Yes Biologic mechanisms discussed Preeclampsia (binary) Ten neonatal outcomes (binary) Preterm birth (binary) Study site, maternal age, maternal race/ethnicity, insurance status, marital status, parity, pre-pregnancy body mass index, and chronic diseases during pregnancy
 Messerlian et al. [34] Montreal, Canada
Women aging 20–45 years without preexisting medical conditions potentially associated with both infertility and preterm birth and primary analysis was restricted to singleton pregnancies (n = 18,147)
Yes Noted that the biologic mechanisms are unclear Reason for infertility (ovulatory, endo-tubal, male factor, uterine abnormalities, unexplained, unspecified) (categorical) Preterm birth categorized as <32, <35, <37, ≥37 weeks) (ordinal) Any type of Infertility treatment
Maternal age, parity, education, smoking, and alcohol or substance use during pregnancy, and body mass index
 Raghavan et al. [35] Framingham, Massachusetts, United States
Participants without type 2 diabetes who had whole-genome, common variant genotyping and were followed for a median of 13 years at exam 5 (n = 2361)
Yes Informed by the literature Parental history of diabetes—none, one or two parents (ordinal) Incident type 2 diabetes in offspring (binary) Metabolic
corrected insulin response, HOMA-IR, metabolic syndrome, components score
genetic risk score Lifestyle
diabetogenic, diet score, physical activity index (continuous)
Age, sex and genetic risk score (for models not focused on genetic mediators)
Case control studies
 Rao et al. [36] Karnataka, India
Source population from which cases and controls were drawn included adults who were either patients or visitors at 4 major cancer hospitals (n = 452)
Yes Yes, critical period model guided the DAG construction Early life socioeconomic disadvantage (low/high) Cases
Diagnosed with oral and/or oropharyngeal cancer (ICD-10 codes C00-C10).
Visitors or those seeking medical care for medical conditions not related to tobacco or alcohol (binary)
Smoking, chewing quid and/or tobacco, alcohol (binary) Age, sex, adult socioeconomic measures and paternal alcohol drinking
 Song et al. [37] United States
Source population from which cases and controls were drawn included postmenopausal women at 40 clinical centers (n = 3049)
Yes Mediators selected based on previous literature Low birth weight (ordinal) Cases
Self-reported first-time use of medication for diabetes during the follow-up periods
For each incident case, controls were selected at random from women who remained free from cardiovascular diseases and/or diabetes at the diagnosed time in the case patient (binary)
Biomarkers of insulin resistance, leptin and its receptor, sex steroid hormones and their binding protein, inflammation, endothelial function, cellular ageing and blood pressure
Two sets of confounders were considered:
(1) Before birth: race/ethnicity and family history of diabetes
(2) After birth: age, smoking, alcohol consumption, physical exercise, dietary fiber intake, dietary glycaemic load, and BMI
 Xie et al. [38] Shanghai, China
Pre-pubertal and early pre-pubertal boys aged 8-15 years old (n = 167)
No Yes, biologic mechanisms discussed Total phthalates (continuous) Cases Diagnosis if constitutional delay of growth and puberty defined by bone age <1.75 years than chronological age
age and Tanner stage (1 or 2) matched (binary)
Serum testosterone level (continuous) Age and body mass index
  1. γ-GT gamma glutamyl transferase; HOMA-IR homeostatic model assessment for insulin resistance; ICD international classification of diseases; PTH parathyroid hormone
  2. aOverall sample size of the study
  3. bConfounders included in the causal mediation analysis
  4. cThe results of mediation analysis were graphically presented