implementation and reporting of causal mediation analysis ...studies apply and report the elements...

Liu et al. BMC Res Notes (2016) 9:354 DOI 10.1186/s13104-016-2163-7

RESEARCH ARTICLE

Implementation and reporting of causal mediation analysis in 2015: a systematic review in epidemiological studiesShao‑Hsien Liu1*, Christine M. Ulbricht2, Stavroula A. Chrysanthopoulou3 and Kate L. Lapane2

Abstract

Background: Causal mediation analysis is often used to understand the impact of variables along the causal path‑way of an occurrence relation. How well studies apply and report the elements of causal mediation analysis remains unknown.

Methods: We systematically reviewed epidemiological studies published in 2015 that employed causal media‑tion analysis to estimate direct and indirect effects of observed associations between an exposure on an outcome. We identified potential epidemiological studies through conducting a citation search within Web of Science and a keyword search within PubMed. Two reviewers independently screened studies for eligibility. For eligible studies, one reviewer performed data extraction, and a senior epidemiologist confirmed the extracted information. Empirical application and methodological details of the technique were extracted and summarized.

Results: Thirteen studies were eligible for data extraction. While the majority of studies reported and identified the effects of measures, most studies lacked sufficient details on the extent to which identifiability assumptions were satisfied. Although most studies addressed issues of unmeasured confounders either from empirical approaches or sensitivity analyses, the majority did not examine the potential bias arising from the measurement error of the media‑tor. Some studies allowed for exposure‑mediator interaction and only a few presented results from models both with and without interactions. Power calculations were scarce.

Conclusions: Reporting of causal mediation analysis is varied and suboptimal. Given that the application of causal mediation analysis will likely continue to increase, developing standards of reporting of causal mediation analysis in epidemiological research would be prudent.

Keywords: Causal mediation analysis, Systematic review, Causal inference, Causality

© 2016 The Author(s). This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

BackgroundCausal mediation analysis identifies potential pathways that could explain observed associations between an exposure and an outcome [1]. This approach also exam-ines how a third intermediate variable, the mediator, is related to the observed exposure-outcome relationship. Causal mediation analysis has been used to study genetic factors in disease causation [2, 3], pathways associated

with response to clinical treatments [4], and mechanisms impacting on public health interventions [5, 6]. There are two approaches for conducting causal mediation analysis. The first, primarily applied in the social sciences, involves the comparison between regression models with and without conditioning on the mediator [7]. The second approach uses the counterfactual framework [8, 9], which allows scientists to decompose the total effect into direct and indirect effects [8–13]. Using the counterfactual framework can help to address the potential bias arising from both incorrect statistical analysis and suboptimal study design [14–16].

Open Access

BMC Research Notes

*Correspondence: [email protected] 1 Clinical and Population Health Research Program, Graduate School of Biomedical Sciences, University of Massachusetts Medical School, 368 Plantation Street, Worcester, MA 01655, USAFull list of author information is available at the end of the article

http://creativecommons.org/licenses/by/4.0/

http://creativecommons.org/publicdomain/zero/1.0/

http://creativecommons.org/publicdomain/zero/1.0/

http://crossmark.crossref.org/dialog/?doi=10.1186/s13104-016-2163-7&domain=pdf

of 21Liu et al. BMC Res Notes (2016) 9:354

The field of causal mediation is relatively new and techniques emerge rapidly. With the rapid development of software packages [11–13, 17], the implementation and/or discussion of this methodology is increasing. In a preliminary search in PubMed, we identified 33 arti-cles in 2013, 59 in 2014, and 61 in 2015. While these software packages allow for estimation in a number of settings, limitations on automated procedures for con-ducting sensitivity analyses on unmeasured confounding or measurement errors remain. However, causal media-tion analysis requires careful implementation of the approach and appropriate evaluations for assumptions to derive valid estimates and the extent to which these studies apply and report the elements of causal mediation analysis remains unknown. Therefore, understanding how these methods have been applied to address issues of bias, how studies have implemented the approach, and how estimates are interpreted may provide useful guid-ance for future reporting.

The purpose of this review was to systematically review epidemiological studies in which causal mediation analy-sis was used to estimate direct and indirect effects. In this review, we will extract information on the elements critical to be reported and summarize our findings on how epide-miological studies have conducted and presented results from causal mediation analysis. We will also give recom-mendations for scientists considering to conduct studies applying causal mediation in the medical literature.

MethodsSelection of articlesOur aim was to identify original empirical epidemiologi-cal research published in 2015 that used causal media-tion analysis. Two search strategies were used to achieve this goal. First, we retrieved all published studies citing one of the seminal papers [8, 10, 12, 13] on causal media-tion analysis using the Web of Science database. One hundred and fifty-seven articles were identified with this approach. Second, we conducted a keyword search within PubMed through working with a research librar-ian at the University of Massachusetts Medical School. We developed the following keyword search algorithm: causal mediation analysis OR (“causal” AND “mediation analysis” AND “Mediat*”). This search term returned 61 unique records in PubMed dating from January 1, 2015 to December 31, 2015. We excluded the following types of publications or studies: (i) methodological or simula-tion studies without an empirical application; (ii) stud-ies without examining the effects on health outcomes, that is, studies not including mortality, morbidity, and diagnostic markers, for both mental and physical health; (iii) animal studies or genetic studies; (iv) letters, meet-ing abstracts, review articles, and editorials; (v) studies

without formal discussion of causal framework or using traditional approach, cross-sectional design, and studies using multilevel models or structural equation models approach.

We used the guidelines from the Preferred Reporting Items for Systematic Reviews and Meta-Analyses [18]. After excluding duplicate records, titles and abstracts of the remaining articles were assigned to two review-ers who independently evaluated each study to assess eligibility. Articles with titles and abstracts were then evaluated by two reviewers through full-text review. Any discrepancy in eligibility was discussed and resolved between reviewers. One reviewer (S-H L) performed data extraction, and two reviewers including a senior epi-demiologist (SC and KLL) confirmed the extracted infor-mation for all eligible studies.

Information abstractionWe considered several elements believed to be impor-tant for transparent and complete reporting of causal mediation analyses. These included: (1) motivation for applying causal mediation analysis, (2) evaluation of identifiability assumptions of effects identified, (3) use of sensitivity analyses for unmeasured confounding and/or measurement error of mediators, and (4) elements of implementing causal mediations analysis includ-ing power calculations, inclusion of exposure-mediator interactions, and bias analysis for interactions. A brief description and rationale for each element chosen is pro-vided in the following sections.

Rationale for causal mediation analysisExplanations of cause-effect associations may be enhanced through additional analyses of mediation and interaction. Mediation and interaction phenomena are not mutually exclusive [1]. Several theoretical and practi-cal considerations can also be the motivations to conduct empirical studies for these phenomena of causal effects. Empirically studying mediation can help to: (1) improve understanding; (2) confirm/refute theory; and (3) refine interventions [1]. In this review, we extracted informa-tion about whether studies reported (i) the reason for applying causal mediation analysis; (ii) the effect esti-mates calculated; and (iii) the motivation of the applica-tion presented.

Identification of effects and identifiability assumptionsIn a counterfactual framework, three measures are esti-mated: (1) natural direct effect; (2) natural indirect effect; and (3) controlled direct effect [8, 9]. The natural direct effect expresses how much the outcome (Y) would change if the exposure (A) was set to A = 1 compared to A = 0 (if binary) intervening to set the mediator (M)


to what it would have been if exposure had been A = 0 (defined by Y1M0

− Y0M0). The natural indirect effect

comparing fixing the mediator to M1 versus M0 if the exposure is set to level A = 1 (defined by Y1M1

− Y1M0 ).

The controlled direct effect expresses how much the outcome would change on average if the exposure were changed from A = 0 to A = 1 but the mediator were set to a fixed level in the population (defined by Y1M–Y0M).

For the mediation analysis to have a causal interpre-tation, we assume that adjustment for the four types of confounding has been addressed. The four types of con-founding are: (1) confounding of the exposure-outcome relationship; (2) confounding of the mediator-outcome relationship; (3) confounding of the exposure-mediator association; and (4) mediator-outcome confounders also affected by the exposure [19]. For controlled direct effect, assumptions (1) and (2) are required. For the identifica-tion of natural direct and indirect effects, assumptions (3) and (4) are also needed [13]. However, for studies with randomized treatments, assumptions (1) and (3) are satisfied and control only needed to be made for (2) and (4). We extracted information about what identifiability assumptions were acknowledged in relation to identified effects of estimates.

Sensitivity analysisIn addition to unmeasured confounding common in observational studies [8, 9, 20], measurement error of the mediator could potentially affect the regression coef-ficient from both the mediator and the outcome regres-sions and thus result in biased estimates for direct and indirect effects [21–23]. Furthermore, interaction anal-ysis could also be a part of research interests to under-stand how and why the effect occurs in an observed phenomena. If control has not been made for two sets of confounding factors for each of the exposures, the results from interaction analysis will be biased [1]. In causal mediation analysis, sensitivity analysis can be used as a technique to evaluate the extent to which the direct and indirect effects are robust to assumption violations [24, 25]. We abstracted information on bias analysis to assess: (i) whether sensitivity analysis was conducted or empirically analyzed for identification assumptions; (ii) which identification assumption was a concern and what approach was used for sensitivity analysis; (iii) whether the rationale and approach to conduct sensitivity analysis for measurement errors of the mediators was included; and (iv) whether bias analysis for the interaction was included.

Power calculationsStudies may be powered to detect a main effect, but may not be sufficiently powered to detect an interaction of a

certain magnitude. We hypothesized that many studies implementing causal mediation analyses may be under-powered. We extracted information about power calcu-lations for interaction from each study. However, further development and methodologic work regarding power calculations for direct and indirect effects is needed [1]. With this in mind, we extracted information regarding what authors reported on the issue of power calculations for causal mediation analysis without judgment regarding which formulas were appropriate.

Exposure‑mediator interactionsIn the traditional approach for mediation analysis, no interaction between the effects of the exposure and the mediator on the outcome is assumed [8, 9]. Causal medi-ation analysis, on the other hand, provides the decom-position of the direct and indirect effects that are valid even in the presence of interaction between the exposure and the mediator on the outcome and when non-linear models are needed [8, 9]. This gives rise to the question of when to include or exclude interactions in conducting causal mediation analysis. The decision to include inter-action terms is often driven by statistical findings which may be problematic if statistical power is lacking. As such, a recommended approach is to include exposure-mediator interactions in the outcome model by default and only exclude the interaction terms if the magnitude of interactions is small and the estimates of direct and indirect effects are not altered much in the presence of the interaction terms [1]. Leaving the interaction terms in the outcome model is suggested to avoid drawing incorrect causal conclusions, to help allow for additional model flexibility, and to understand the dynamics of mediation [1]. Therefore, we extracted information about whether or not studies allowed for interactions in the outcome model.

Effects of estimates and results from exposure‑mediator interactionIn this review, we assessed whether studies reported both estimates from allowing for exposure-mediator interac-tions in the outcome model in addition to the effect of estimates without interaction in the model. Moreover, we also extracted estimates from sensitivity analysis con-ducted for direct/indirect effects and interactions. We also extracted information about explanations of discrep-ancies when noted.

ResultsFigure 1 shows the process of identifying eligible arti-cles for the review. We retrieved 157 and 61 studies from citation search in Web of Science and keyword search in PubMed, respectively. After excluding duplicate studies


(n = 22), studies not focusing on the effects of health-related outcomes (n = 57), review articles (n = 6), methodological or simulation studies (n = 46), letters, meeting abstracts and brief reports (n = 10), animals studies (n = 2), studies not using causal mediation analy-sis (n = 9), genetic studies (n = 9) or studies using mul-tilevel models, structural equation models approach, and cross-sectional design (n = 27), and studies using tradi-tional approach or without formal discussion of formal causal framework (n = 17), we had 13 epidemiological studies that applied causal mediation analysis [26–38].

Summary of study design, primary exposure, outcomesTwo studies used randomized controlled trials; 8 were cohort studies; and 3 were case–control studies (Table 1). We found that studies were not clustered in one specific area (e.g. 3 studies evaluated risks from environmental exposures including environmental substances [31, 38] and changes in environments [28] and 3 studies evalu-ated parental conditions before [34, 35] and during [33] pregnancy). Regarding outcomes of interest, 4 studies

used the first occurrence of a pre-specified event [32, 35–37]. Other studies also examined levels of biomark-ers [27, 30, 31, 38], mortality [26, 29], or neonatal health outcomes [33, 34]. Nearly half of studies used biomark-ers as the primary mediator [27, 31, 32, 35, 37, 38]. Other studies used a pre-specified medical event [26, 29, 30], health behaviors [34, 36], psychological symptoms [28], and another a neonatal health outcome [33]. All studies provided information on the confounders in the causal mediation analysis and the majority of studies provided a hypothesized directed acyclic graph (DAG).

Motivation for applying causal mediation analysisThe reason for applying causal mediation analysis among all studies was to evaluate mediation (Table 2). With the exception of one study, most studies reported and iden-tified the measures of either direct/indirect effect or controlled direct effect. While the motivation for most studies was to improve understanding, one study used mediation analysis to confirm/refute theory, and one study did not report the motivation.

Fig. 1 Identification of epidemiological studies using causal mediation analysis in 2015


Tabl

e 1

Gen

eral

des

crip

tion

of e

pide

mio

logi

cal s

tudi

es th

at w

ere

elig

ible

for t

he s

yste

mat

ic re

view

Refe

renc

esCo

untr

y, p

opul

a‑tio

n, a

nd s

ampl

e si

ze

of s

tudy

a

Dire

cted

acy

clic

gr

aph

(DA

G)

incl

uded

?

Spec

ific

med

iatio

n hy

poth

esis

spe

cifie

d?Ex

posu

rePr

imar

y ou

tcom

eM

edia

tor

Conf

ound

ersb

Rand

omiz

ed c

ontr

olle

d tr

ials

D’A

mel

io e

t al.

[27]

Italy

Non

‑dia

betic

wom

en

with

pos

tmen

opau

sal

oste

opor

osis

(n =

46)

No

Biol

ogic

mec

hani

sms

disc

usse

dA

ll tr

eate

d w

ith c

al‑

cium

120

0 m

g/da

y an

d ch

olec

alci

fero

l 80

0 U

I/day

Rand

omiz

ed to

with

PT

H 1

–84

100

μg/

day

subc

utan

eous

Or

With

out P

TH 1

–84

100

μg/d

ay s

ubcu

ta‑

neou

s (b

inar

y)

Glu

cose

met

abol

ism

, (c

ontin

uous

, log

sc

ale)

Tota

l ost

eoca

lcin

(OC

) un

derc

arbo

xyla

ted

(uO

C)

(con

tinuo

us)

Biom

arke

rs th

at w

ere

unba

lanc

ed b

etw

een

the

two

trea

tmen

t gr

oups

at b

asel

ine

incl

udin

g uO

C a

nd

seru

m ta

rtra

te re

sist

‑an

t aci

d ph

osph

atas

e 5B

(TRA

P5b)

Fre

eman

et a

l. [2

8]En

glan

dPa

tient

s w

ith p

erse

cu‑

tory

del

usio

ns fr

om

6 m

enta

l hea

lth s

ites

(n =

59)

No

Gui

ded

by c

ogni

tive

mod

el o

f per

secu

‑to

ry d

elus

ions

Rand

omiz

ed to

str

eet

expo

sure

in a

reas

of

rela

tive

depr

ivat

ion

durin

g bu

sy m

id‑d

ayO

rA

neu

tral

con

trol

co

nditi

on w

hich

in

clud

ed s

ittin

g in

a

room

wat

chin

g m

ildly

hum

orou

s te

levi

sion

clip

s fo

r 10

min

(bin

ary)

(1) S

tate

Par

anoi

a us

ing

six

visu

al a

nalo

g sc

ales

(VA

S)(2

) Sta

te s

ocia

l par

a‑no

ia s

cale

(3) S

chiz

otyp

al S

ymp‑

tom

s In

vent

ory—

Para

noia

(con

tinu‑

ous)

Voic

esH

allu

cina

tions

VA

SD

istr

ess V

AS

Affec

tive

Anx

iety

VA

SD

epre

ssio

n VA

SBr

ief c

ore

sche

ma

scal

es (B

CSS

)Se

lf‑fo

cus

Thre

at a

ntic

ipat

ion

Inte

rpre

tatio

n bi

as

(con

tinuo

us)

Reas

onin

g m

easu

res

Jum

ping

to c

oncl

u‑si

ons

Poss

ibili

ty o

f bei

ng

mis

take

nA

ltern

ativ

e ex

plan

a‑tio

nsH

ypot

hetic

al c

ontr

adic

‑tio

n (b

inar

y)Pr

obab

ility

of b

eing

m

ista

ken

(con

tinu‑

ous)

Base

line

mea

sure

s of

pa

rano

ia, a

ll of

the

med

iato

rs c

onsi

dere

d,

and

cent

er


Tabl

e 1

cont

inue

d

Refe

renc

esCo

untr

y, p

opul

a‑tio

n, a

nd s

ampl

e si

ze

of s

tudy

a

Dire

cted

acy

clic

gr

aph

(DA

G)

incl

uded

?

Spec

ific

med

iatio

n hy

poth

esis

spe

cifie

d?Ex

posu

rePr

imar

y ou

tcom

eM

edia

tor

Conf

ound

ersb

Coho

rt s

tudi

es

Ban

ack

et a

l. [2

6]U

nite

d St

ates

Nat

iona

lly re

pres

enta

‑tiv

e no

nins

titut

iona

l‑iz

edSa

mpl

e of

adu

lts a

ged

20 to

80

year

s in

the

U.S

. (19

88–2

004)

(n

= 7

212)

Yes

Gui

ded

by p

revi

ous

rese

arch

Obe

sity

defi

ned

as

body

mas

s in

dex

≥30

kg/

m2 v

s.

18.5

–29.

9 kg

/m2

(bin

ary)

All‑

caus

e m

orta

l‑ity

with

follo

w‑u

p th

roug

h 20

06

(bin

ary)

Self‑

repo

rted

acu

te

card

iac

even

t (e.

g.

stro

ke o

r myo

card

ial

infa

rctio

n) (b

inar

y)

Age

, gen

der,

race

, edu

‑ca

tion,

sm

okin

g st

atus

, an

d ca

rdio

resp

irato

ry

fitne

ss

Jack

son

et a

l. [2

9]N

ew J

erse

y an

d Pe

nnsy

lvan

ia, U

nite

d St

ates

Old

er a

dults

dua

lly

enro

lled

in m

edic

are

and

phar

mac

y as

sis‑

tanc

e pr

ogra

ms;

“new

us

ers”

(n =

26,

197)

No

Med

iato

rs s

elec

ted

base

d on

pre

viou

s lit

erat

ure

New

use

r of fi

rst g

en‑

erat

ion

antip

sych

otic

ve

rsus

new

use

r of

sec

ond

gen‑

erat

ion

antip

sych

otic

(b

inar

y)

Mor

talit

y w

ith 1

80 d

ays

(bin

ary)

Med

ical

eve

nts

stro

ke,

vent

ricul

ar a

rrhy

th‑

mia

, acu

te m

yoca

rdia

l in

farc

tion,

ven

ous

thro

mbo

embo

lism

, pn

eum

onia

, bac

teria

l in

fect

ion

(bes

ides

pn

eum

onia

), an

d hi

p fra

ctur

e) (b

inar

y)

70 d

iffer

ent d

emo‑

grap

hic

char

acte

ristic

s, he

alth

ser

vice

util

iza‑

tion

and

med

icat

ion

usag

e, c

o‑ex

istin

g m

edic

al a

nd p

sych

iat‑

ric il

lnes

s, an

d in

dica

‑to

rs o

f fun

ctio

nal

impa

irmen

t

Kos

itsaw

at e

t al.

[30]

Mem

phis

, Ten

nes‑

see

and

Pitt

sbur

gh,

Penn

sylv

ania

, Uni

ted

Stat

esBl

ack

and

whi

te

med

icar

e el

igi‑

ble—

com

mun

ity

dwel

ling

adul

ts a

ged

70–7

9 ye

ars

with

out

diab

etes

at y

ear 2

of

the

stud

y (n

= 2

193)

No

Ratio

nale

not

cle

arSe

rum

vita

min

D le

vels

(2

5‑hy

drox

yvita

‑m

in D

) <20

ng/

mL

(bin

ary)

A1c

leve

l ≥6.

5 %

at

year

4 (b

inar

y)D

iabe

tes

stat

us a

t yea

r 4

(bin

ary)

Conf

ound

ers

cons

ider

ed

in m

edia

tion

anal

ysis

no

t rep

orte

d

Lou

wie

s et

al.

[31]

Belg

ium

Wor

king

nur

ses

aged

be

twee

n 22

and

59

yea

rs w

ithou

t car

‑di

ovas

cula

r dis

ease

s an

d di

abet

es (n

= 5

5)

No

Gui

ded

by p

revi

ous

liter

atur

eSu

bchr

onic

bla

ck

carb

on e

xpos

ure

(con

tinuo

us)

Dia

stol

ic b

lood

pre

s‑su

reSy

stol

ic b

lood

pre

ssur

e(c

ontin

uous

)

Retin

al m

icro

circ

ulat

ion

(con

tinuo

us)

Age

, sex

, bod

y m

ass

inde

x, s

mok

ing,

use

of

ant

i‑hyp

erte

nsiv

e m

edic

atio

n, γ

‑GT,

A1c

, di

stan

ce to

maj

or

road

, clin

ic, a

nd a

ver‑

age

wee

kly

tem

pera

‑tu

re


Tabl

e 1

cont

inue

d

Refe

renc

esCo

untr

y, p

opul

a‑tio

n, a

nd s

ampl

e si

ze

of s

tudy

a

Dire

cted

acy

clic

gr

aph

(DA

G)

incl

uded

?

Spec

ific

med

iatio

n hy

poth

esis

spe

cifie

d?Ex

posu

rePr

imar

y ou

tcom

eM

edia

tor

Conf

ound

ersb

Lu

et a

l. [3

2]U

nite

d St

ates

Adu

lts fr

ee o

f cor

onar

y he

art d

isea

se w

ho

part

icip

ated

in 9

N

atio

nal H

eart

, Lun

g,

and

Bloo

d In

stitu

te

fund

ed c

ohor

t stu

d‑ie

s w

ith b

ody

mas

s in

dex ≥

20

kg/m

2 (1

954–

2001

)(n

= 5

8,32

2 fo

r m

etab

olic

risk

fac‑

tors

; n =

19,

572

for

fibrin

ogen

ana

lysi

s)

Yes

Biol

ogic

mec

hani

sms

disc

usse

dBo

dy m

ass

inde

x (c

ateg

orie

s ≥30

kg/

m2 , 2

5–<

30 k

g/m

2 , 20

–25

kg/m

2 )(c

ateg

oric

al a

nd

cont

inuo

us)

Firs

t fat

al o

r non

‑fa

tal o

ccur

renc

e of

isch

emic

hea

rt

dise

ase,

acu

te m

yo‑

card

ial i

nfar

ctio

n,

or a

ngin

a pe

ctor

is

(bin

ary)

Expl

ored

in d

ata

com

bine

d fro

m n

ine

coho

rt st

udie

sSy

stol

ic b

lood

pre

ssur

e,

tota

l ser

um c

hole

s‑te

rol,

gluc

ose

Expl

ored

in d

ata

com

bine

d fro

m th

ree

coho

rt st

udie

sFi

brin

ogen

, hig

h‑se

nsiti

ve C

‑rea

ctiv

e pr

otei

n (c

ontin

uous

)

Age

, sex

, sm

okin

g, ra

ce/

ethn

icity

, soc

ioec

o‑no

mic

sta

tus,

alco

hol

inta

ke, p

hysi

cal a

ctiv

‑ity

, and

die

tary

inta

ke

Men

dola

et a

l. [3

3]U

nite

d St

ates

Sing

leto

n ne

wbo

rns

with

≥23

wee

ks

of g

esta

tion

(n =

210

,610

)

Yes

Biol

ogic

mec

hani

sms

disc

usse

dPr

eecl

amps

ia (b

inar

y)Te

n ne

onat

al o

ut‑

com

es (b

inar

y)Pr

eter

m b

irth

(bin

ary)

Stud

y si

te, m

ater

nal

age,

mat

erna

l rac

e/et

hnic

ity, i

nsur

ance

st

atus

, mar

ital s

tatu

s, pa

rity,

pre

‑pre

gnan

cy

body

mas

s in

dex,

and

ch

roni

c di

seas

es d

ur‑

ing

preg

nanc

y

Mes

serli

an e

t al.

[34]

Mon

trea

l, Ca

nada

Wom

en a

ging

20

–45

year

s w

ithou

t pr

eexi

stin

g m

edic

al

cond

ition

s po

tent

ially

as

soci

ated

with

bo

th in

fert

ility

and

pr

eter

m b

irth

and

prim

ary

anal

ysis

w

as re

stric

ted

to

sing

leto

n pr

egna

n‑ci

es (n

= 1

8,14

7)

Yes

Not

ed th

at th

e bi

o‑lo

gic

mec

hani

sms

are

uncl

ear

Reas

on fo

r inf

ertil

ity

(ovu

lato

ry, e

ndo‑

tuba

l, m

ale

fact

or,

uter

ine

abno

rmal

i‑tie

s, un

expl

aine

d,

unsp

ecifi

ed) (

cat‑

egor

ical

)

Pret

erm

birt

h ca

tego

‑riz

ed a

s <

32, <

35,

<37

, ≥37

wee

ks)

(ord

inal

)

Any

type

of I

nfer

tility

tr

eatm

ent

(bin

ary)

Mat

erna

l age

, par

ity,

educ

atio

n, s

mok

‑in

g, a

nd a

lcoh

ol o

r su

bsta

nce

use

durin

g pr

egna

ncy,

and

bod

y m

ass

inde

x

Rag

hava

n et

al.

[35]

Fram

ingh

am, M

as‑

sach

uset

ts, U

nite

d St

ates

Part

icip

ants

with

out

type

2 d

iabe

tes

who

ha

d w

hole

‑gen

ome,

co

mm

on v

aria

nt

geno

typi

ng a

nd w

ere

follo

wed

for a

med

ian

of 1

3 ye

ars

at e

xam

5

(n =

236

1)

Yes

Info

rmed

by

the

liter

atur

ePa

rent

al h

isto

ry o

f di

abet

es—

none

, on

e or

two

pare

nts

(ord

inal

)

Inci

dent

type

2

diab

etes

in o

ffspr

ing

(bin

ary)

Met

abol

icco

rrec

ted

insu

lin

resp

onse

, HO

MA

‑IR,

met

abol

ic s

yndr

ome,

co

mpo

nent

s sc

ore

Gen

etic

gene

tic ri

sk s

core

Li

fest

yle

diab

etog

enic

, die

t sc

ore,

phy

sica

l act

iv‑

ity in

dex

(con

tinuo

us)

Age

, sex

and

gen

etic

risk

sc

ore

(for m

odel

s no

t fo

cuse

d on

gen

etic

m

edia

tors

)


Tabl

e 1

cont

inue

d

Refe

renc

esCo

untr

y, p

opul

a‑tio

n, a

nd s

ampl

e si

ze

of s

tudy

a

Dire

cted

acy

clic

gr

aph

(DA

G)

incl

uded

?

Spec

ific

med

iatio

n hy

poth

esis

spe

cifie

d?Ex

posu

rePr

imar

y ou

tcom

eM

edia

tor

Conf

ound

ersb

Case

con

trol

stu

dies

Rao

et a

l. [3

6]Ka

rnat

aka,

Indi

aSo

urce

pop

ulat

ion

from

whi

ch c

ases

and

co

ntro

ls w

ere

draw

n in

clud

ed a

dults

who

w

ere

eith

er p

atie

nts

or v

isito

rs a

t 4 m

ajor

ca

ncer

hos

pita

ls

(n =

452

)

Yes

Yes,

criti

cal p

erio

d m

odel

gui

ded

the

DA

G c

onst

ruct

ion

Early

life

soc

ioec

o‑no

mic

dis

adva

ntag

e (lo

w/h

igh)

Case

sD

iagn

osed

with

ora

l an

d/or

oro

phar

yn‑

geal

can

cer (

ICD

‑10

code

s C

00‑C

10).

Cont

rols

Visi

tors

or t

hose

see

k‑in

g m

edic

al c

are

for

med

ical

con

di‑

tions

not

rela

ted

to

toba

cco

or a

lcoh

ol

(bin

ary)

Smok

ing,

che

win

g qu

id a

nd/o

r tob

acco

, al

coho

l (bi

nary

)

Age

, sex

, adu

lt so

cio‑

econ

omic

mea

sure

s an

d pa

tern

al a

lcoh

ol

drin

king

Son

g et

al.

[37]

Uni

ted

Stat

esSo

urce

pop

ulat

ion

from

whi

ch c

ases

and

co

ntro

ls w

ere

draw

n in

clud

ed p

ostm

eno‑

paus

al w

omen

at

40 c

linic

al c

ente

rs

(n =

304

9)

Yes

Med

iato

rs s

elec

ted

base

d on

pre

viou

s lit

erat

ure

Low

birt

h w

eigh

t (o

rdin

al)

Case

sSe

lf‑re

port

ed fi

rst‑

time

use

of m

edic

atio

n fo

r di

abet

es d

urin

g th

e fo

llow

‑up

perio

dsCo

ntro

lsFo

r eac

h in

cide

nt

case

, con

trol

s w

ere

sele

cted

at r

ando

m

from

wom

en w

ho

rem

aine

d fre

e fro

m

card

iova

scul

ar d

is‑

ease

s an

d/or

dia

be‑

tes

at th

e di

agno

sed

time

in th

e ca

se

patie

nt (b

inar

y)

Biom

arke

rs o

f ins

ulin

re

sist

ance

, lep

tin a

nd

its re

cept

or, s

ex s

ter‑

oid

horm

ones

and

th

eir b

indi

ng p

rote

in,

infla

mm

atio

n,

endo

thel

ial f

unct

ion,

ce

llula

r age

ing

and

bloo

d pr

essu

re(c

ontin

uous

)

Two

sets

of c

onfo

unde

rs

wer

e co

nsid

ered

:(1

) Bef

ore

birt

h: ra

ce/

ethn

icity

and

fam

ily

hist

ory

of d

iabe

tes

(2) A

fter

birt

h: a

ge,

smok

ing,

alc

ohol

co

nsum

ptio

n, p

hysi

cal

exer

cise

, die

tary

fibe

r in

take

, die

tary

gly

cae‑

mic

load

, and

BM

I

Xie

et a

l. [3

8]Sh

angh

ai, C

hina

Pre‑

pube

rtal

and

ear

ly

pre‑

pube

rtal

boy

s ag

ed 8

‑15

year

s ol

d (n

= 1

67)

No

Yes,

biol

ogic

mec

ha‑

nism

s di

scus

sed

Tota

l pht

hala

tes

(c

ontin

uous

)Ca

ses D

iagn

osis

if c

on‑

stitu

tiona

l del

ay o

f gr

owth

and

pub

erty

de

fined

by

bone

ag

e <

1.75

yea

rs th

an

chro

nolo

gica

l age

Cont

rols

age

and

Tann

er s

tage

(1

or 2

) mat

ched

(b

inar

y)

Seru

m te

stos

tero

ne

leve

l (co

ntin

uous

)A

ge a

nd b

ody

mas

s in

dex

γ-G

T ga

mm

a gl

utam

yl tr

ansf

eras

e; H

OM

A-IR

hom

eost

atic

mod

el a

sses

smen

t for

insu

lin re

sist

ance

; ICD

inte

rnat

iona

l cla

ssifi

catio

n of

dis

ease

s; P

TH p

arat

hyro

id h

orm

one

a Ove

rall

sam

ple

size

of t

he s

tudy

b Con

foun

ders

incl

uded

in th

e ca

usal

med

iatio

n an

alys

isc T

he re

sults

of m

edia

tion

anal

ysis

wer

e gr

aphi

cally

pre

sent

ed


Evaluation of identifiability assumptions and sensitivity analysesFour studies did not report identification assumptions for measures of effects identified (Table 3). With the excep-tion of two studies, the empirical approach or sensitivity analysis was used to address the issue of confounding. There were 9 studies addressing unmeasured confound-ing for the mediator-outcome relationship. Five studies provided the empirical approach and four studies used sensitivity analysis to address the concern. For measure-ment error or misclassification of mediators, 3 studies addressing this issue (Table 4). Two studies provided the rationale for doing sensitivity analysis for measurement error of mediators. Furthermore, they also noted that the

bias may result from misclassification of the mediator and robustness of findings was also discussed.

Elements for implementation of causal mediation analysisMost studies had a relatively large sample size (Table 5). Three studies had small size (n < 100) and this limitation was acknowledged. The majority of studies did not report whether the power or sample size calculation was calcu-lated. For exposure-mediator interaction, most studies did not report or did not have the exposure-mediator interaction in the model. Among those six studies allow-ing for exposure-mediator interaction in the model, none reported power or sample size calculation and bias analy-sis for the interaction.

Table 2 Rationale and measures of effect estimated and reported for Causal Mediation Analysis

a Reason for applying causal mediation analysis: Mediation, Interaction, or Interferenceb Motivation for each application of causal mediation analysis. For mediation (1) improve understanding; (2) confirm/refute theory; (3) intervention refinement. For interaction (1) help allocate resources better; (2) identifying groups in which treatments may be harmful or beneficial (qualitative or cross-over interactions); (3) understand mechanisms; (4) increase statistical power of main effect analysis, and (5) understand which mediator to intervene upon to eliminate most of the effect of primary exposure. For interference (1) quantify spillover effects for cost-effectiveness studies; (2) understand what proportion must be treated to attain population outcomes desired; (3) create knowledge for intervention development and refinementc “Natural” was not specifically used in the article but appeared to have counterfactual framework and appropriate references

References Reasona Measures discussed or reported Motivation for applicationb

Randomized controlled trials

D’Amelio et al. [27] Mediation Natural direct and natural indirect effectsEmphasized direct effect

Improve understanding to show that above and beyond how the treatment works through the mediator, there is an independent effect

Freeman et al. [28] Mediation Direct and indirect effectsc

Proportion mediated by various factorsImprove understanding of mechanisms

Cohort studies

Banack et al. [26] Mediation Similar to controlled direct effect (with caveat that no manipulation of obesity could actually occur)

Refute/confirm that selection bias drives the obesity para‑dox in cardiovascular disease

Jackson et al. [29] Mediation Natural direct and indirect effectsProportion mediated by each medical event

Improve understanding of mechanisms

Kositsawat et al. [30]

Mediation Not identified Not clear

Louwies et al. [31] Mediation Direct and indirect effectc Improve understanding of mechanisms

Lu et al. [32] Mediation Natural direct and natural indirect effectPercent excess risk mediatedNatural indirect effect emphasized


Mendola et al. [33] Mediation Controlled direct effect Improve understanding

Messerlian et al. [34]

Mediation Controlled direct effect Improve understanding

Raghavan et al. [35] Mediation Direct and indirect effects but only indirect effects reportedc

Proportion of risk mediated through genetic and meta‑bolic factors

Improve understanding of what mediators might be ripe for intervention

Case control studies

Rao et al. [36] Mediation Controlled direct effect Improve understanding

Song et al. [37] Mediation Effect not mediatedmediated effectc

Proportion mediated through various biomarkers


Xie et al. [38] Mediation Direct and indirect effectc

Proportion of effect mediated through testosteroneImprove understanding


Tabl

e 3

Exam

inat

ion

of Id

enti

fiabi

lity

Ass

umpt

ions

for C

ausa

l Med

iati

on A

naly

sis

Refe

renc

esN

o un

mea

sure

d ex

posu

re‑o

utco

me

con‑

foun

ders

No

unm

easu

red

med

iato

r‑ou

tcom

e co

nfou

nder

sN

o un

mea

sure

d ex

posu

re‑m

edia

tor

conf

ound

ers

No

med

iato

r‑ou

tcom

e co

nfou

nder

aff

ecte

d by

the

expo

sure

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

lyse

s or

se

nsiti

vity

ana

lyse

sA

ckno

wle

dged

as

sum

ptio

nEm

piri

cal a

naly

‑se

s or

sen

sitiv

ity

anal

yses

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Stud

ies

estim

atin

g co

ntro

lled

dire

ct e

ffect

s on

ly

Ban

ack

et a

l. [2

6]✓

Not

repo

rted

✓U

nmea

sure

d co

nfou

nder

ca

rdio

resp

irato

ry‑

fitne

ssEs

timat

es o

f the

di

rect

effe

ct o

f ca

rdio

resp

ira‑

tory

fitn

ess

on

mor

talit

y fro

m

wel

l‑est

ablis

hed

liter

atur

e. N

o lit

erat

ure

on e

sti‑

mat

es o

f pre

va‑

lenc

e di

ffere

nces

of

unm

easu

red

conf

ound

er—

so a

ra

nge

of 1

0–90

%

was

con

side

red

Not

app

licab

le

Men

dola

et a

l. [3

3]✓

Not

repo

rted

✓U

nmea

sure

d co

n-fo

unde

r mat

erna

l in

fect

ion

Estim

ates

of t

he

dire

ct e

ffect

of

mat

erna

l inf

ec‑

tion

on n

eona

tal

outc

ome

rang

ed

from

2 to

10.

Pre

v‑al

ence

diff

eren

ces

of u

nmea

sure

d co

nfou

nder

—so

a

rang

e of

1–9

9 %

w

as c

onsi

dere

d.

Whe

ther

this

was

do

ne b

ecau

se

no li

tera

ture

was

av

aila

ble

on w

hich

to

bas

e th

e se

nsi‑

tivity

ana

lyse

s w

as

not r

epor

ted

Not

app

licab

le


Tabl

e 3

cont

inue

d

Refe

renc

esN

o un

mea

sure

d ex

posu

re‑o

utco

me

con‑

foun

ders

No

unm

easu

red

med

iato

r‑ou

tcom

e co

nfou

nder

sN

o un

mea

sure

d ex

posu

re‑m

edia

tor

conf

ound

ers

No

med

iato

r‑ou

tcom

e co

nfou

nder

aff

ecte

d by

the

expo

sure

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

lyse

s or

se

nsiti

vity

ana

lyse

sA

ckno

wle

dged

as

sum

ptio

nEm

piri

cal a

naly

‑se

s or

sen

sitiv

ity

anal

yses

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Mes

serli

an e

t al.

[34]

✓It

is u

ncle

ar if

they

w

ere

addr

essi

ng th

is

conc

ern

alth

ough

ad

ditio

nal p

re‑

spec

ified

str

atum

‑ sp

ecifi

c w

ith d

iffer

ent

refe

renc

e ca

tego

ries

and

expo

sure

gro

ups

wer

e us

ed fo

r sen

si‑

tivity

ana

lyse

s

✓St

ratifi

ed a

naly

ses

“tria

ngul

ated

” th

ose

deriv

ed

from

mar

gina

l st

ruct

ural

mod

els.

It is

unc

lear

if th

ey

wer

e ad

dres

sing

th

is c

once

rn

Not

app

licab

le

Rao

et a

l. [3

6]✓

Unm

easu

red

con-

foun

der s

ituat

ion

that

unm

easu

red

conf

ound

ers

coul

d be

cor

rela

ted

with

ex

posu

re, m

edia

tor,

and

outc

ome

wer

e co

nsid

ered

. Usi

ng

para

met

ers,

such

as

γ (c

ondi

tiona

l inc

reas

e in

risk

for o

ral c

ance

r),

P1 (p

reva

lenc

e in

sm

oker

s/ch

ewer

s/dr

inke

rs),

and

P2

(pre

vale

nce

amon

g no

n‑sm

oker

s/no

n‑ch

ewer

s/no

n‑dr

ink‑

ers)

wer

e sp

ecifi

ed.

The

bias

intr

oduc

ed

by u

nmea

sure

d co

n‑fo

unde

rs th

at m

ay

entir

ely

inva

lidat

e th

e co

ntro

lled

dire

ct

effec

t was

cal

cula

ted

✓U

nmea

sure

d co

nfou

nder

co

nsid

ered

with

th

e ex

posu

re‑o

ut‑

com

e re

latio

nshi

p

Not

app

licab

le


Tabl

e 3

cont

inue

d

Refe

renc

esN

o un

mea

sure

d ex

posu

re‑o

utco

me

con‑

foun

ders

No

unm

easu

red

med

iato

r‑ou

tcom

e co

nfou

nder

sN

o un

mea

sure

d ex

posu

re‑m

edia

tor

conf

ound

ers

No

med

iato

r‑ou

tcom

e co

nfou

nder

aff

ecte

d by

the

expo

sure

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

lyse

s or

se

nsiti

vity

ana

lyse

sA

ckno

wle

dged

as

sum

ptio

nEm

piri

cal a

naly

‑se

s or

sen

sitiv

ity

anal

yses

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Stud

ies

estim

atin

g na

tura

l dire

ct a

nd in

dire

ct e

ffect

s

D’A

mel

io e

t al.

[27]

Rand

omiz

ed c

ontr

olle

d tr

ial‑n

ot a

pplic

able

✓a

Not

repo

rted

Rand

omiz

ed c

ontr

olle

d tr

ial‑n

ot a

ppli‑

cabl

e✓

aN

o se

nsiti

vity

an

alys

es, b

ut

adju

sted

for

biom

arke

rs th

at

wer

e un

bal‑

ance

d be

twee

n th

e tw

o tr

eat‑

men

t gro

ups

at

base

line

Fre

eman

et a

l. [2

8]Ra

ndom

ized

con

trol

led

tria

l‑not

app

licab

le✓

No

sens

itivi

ty a

naly

‑se

s, bu

t adj

uste

d fo

r bas

elin

e co

nfou

nder

s; ca

n’t

rule

out

Rand

omiz

ed c

ontr

olle

d tr

ial‑n

ot a

ppli‑

cabl

e✓

Not

repo

rted

Jack

son

et a

l. [2

9]✓

Show

ed ri

sk fa

ctor

s by

an

tipsy

chot

ic g

roup

✓N

o se

nsiti

vity

an

alys

es, b

ut

adju

sted

for m

any

risk

fact

ors;

cann

ot

rule

out

resi

dual

co

nfou

ndin

g

✓N

o se

nsiti

vity

ana

ly‑

sis,

but r

esid

ual

conf

ound

ing

(i.e.

de

liriu

m) a

t bas

e‑lin

e th

at c

ould

bi

as th

e to

tal a

nd

indi

rect

effe

cts

upw

ards

was

ac

know

ledg

ed

✓N

o se

nsiti

vity

an

alys

es, b

ut

cond

ucte

d st

rat‑

ified

ana

lyse

s by

med

iato

rs to

pr

ovid

e qu

alita

‑tiv

e ev

iden

ce fo

r w

heth

er o

r not

th

e as

soci

atio

n be

twee

n m

edia

‑to

r and

mor

tal‑

ity is

mod

ified

by

ant

ipsy

chot

ic

type


Tabl

e 3

cont

inue

d

Refe

renc

esN

o un

mea

sure

d ex

posu

re‑o

utco

me

con‑

foun

ders

No

unm

easu

red

med

iato

r‑ou

tcom

e co

nfou

nder

sN

o un

mea

sure

d ex

posu

re‑m

edia

tor

conf

ound

ers

No

med

iato

r‑ou

tcom

e co

nfou

nder

aff

ecte

d by

the

expo

sure

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

lyse

s or

se

nsiti

vity

ana

lyse

sA

ckno

wle

dged

as

sum

ptio

nEm

piri

cal a

naly

‑se

s or

sen

sitiv

ity

anal

yses

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Lou

wie

s et

al.

[31]

XN

o se

nsiti

vity

ana

lyse

s, bu

t adj

uste

d fo

r co

nfou

nder

s in

Ta

ble

1, e

xcep

t day

of

the

wee

k

XN

ot re

port

edX

Not

repo

rted

XN

ot re

port

ed

Lu

et a

l. [3

2]✓

Excl

uded

firs

t 3 y

ears

of

follo

w‑u

p to

redu

ce

the

influ

ence

of b

ase‑

line

conf

ound

ers

Rest

ricte

d th

e an

alys

is

to n

ever

‑sm

oker

s to

bet

ter c

ontr

ol

for c

onfo

undi

ng b

y sm

okin

g

✓U

nmea

sure

d co

n-fo

unde

rCo

mm

on c

ause

of

met

abol

ic

med

iato

rs a

nd

coro

nary

hea

rt

dise

ase

(e.g

. fam

ily

hist

ory,

gen

etic

fa

ctor

s, re

sidu

al

conf

ound

ing

due

to m

easu

rem

ent

erro

r in

diet

and

ph

ysic

al a

ctiv

ity).

Sens

itivi

ty a

naly

‑se

s do

ne w

ith

two

scen

ario

s: (1

) m

ild c

onfo

und‑

ing

(incr

ease

d ha

zard

ratio

by

fact

or o

f 1.1

an

d pr

eval

ence

20

% fo

r nor

mal

w

eigh

t/25

%

for o

verw

eigh

t/ob

ese)

; and

(2)

stro

ng c

onfo

und‑

ing

(incr

ease

d ha

zard

ratio

by

fact

or o

f 1.8

and

pr

eval

ence

of

45 %

for n

orm

al

wei

ght a

nd 4

0 %

fo

r ove

rwei

ght/

obes

e)

✓Re

stric

ted

the

anal

ysis

to

neve

r‑sm

oker

s to

be

tter

con

trol

for

conf

ound

ing

by

smok

ing

✓N

ot re

port

ed


Tabl

e 3

cont

inue

d

Refe

renc

esN

o un

mea

sure

d ex

posu

re‑o

utco

me

con‑

foun

ders

No

unm

easu

red

med

iato

r‑ou

tcom

e co

nfou

nder

sN

o un

mea

sure

d ex

posu

re‑m

edia

tor

conf

ound

ers

No

med

iato

r‑ou

tcom

e co

nfou

nder

aff

ecte

d by

the

expo

sure

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

lyse

s or

se

nsiti

vity

ana

lyse

sA

ckno

wle

dged

as

sum

ptio

nEm

piri

cal a

naly

‑se

s or

sen

sitiv

ity

anal

yses

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Ack

now

ledg

ed

assu

mpt

ion

Empi

rica

l ana

ly‑

ses

or s

ensi

tivit

y an

alys

es

Rag

hava

n et

al.

[35]

XN

ot re

port

edX

No

sens

itivi

ty a

naly

‑se

s, bu

t med

iatio

n an

alys

is w

as

cond

ucte

d w

ith

all t

hree

met

abol

ic

med

iato

rs(C

IR, H

OM

A‑IR

and

M

SS) t

oget

her

XN

o se

nsiti

vity

ana

ly‑

ses,

but m

edia

‑tio

n an

alys

is w

as

cond

ucte

d w

ith

all t

hree

met

a‑bo

lic m

edia

tors

(CIR

, HO

MA

‑IR a

nd

MSS

) tog

ethe

r

XN

ot re

port

ed

Son

g et

al.

[37]

✓N

o se

nsiti

vity

ana

lysi

s, bu

t inc

lude

d al

l th

e co

varia

tes

that

m

ay c

onfo

und

the

rela

tions

hip

✓N

o se

nsiti

vity

an

alys

is, b

ut

incl

uded

all

the

cova

riate

s th

at

may

con

foun

d th

e re

latio

nshi

p

✓N

o se

nsiti

vity

an

alys

is, b

ut

incl

uded

all

the

cova

riate

s th

at

may

con

foun

d th

e re

latio

nshi

p

✓Se

nsiti

vity

ana

lysi

s w

as c

ondu

cted

th

roug

h ex

clud

‑in

g BM

I, a

med

i‑at

or‑o

utco

me

conf

ound

er

that

is p

ossi

bly

affec

ted

by th

e ex

posu

re (l

ow

birt

h w

eigh

t)

Xie

et a

l. [3

8]X

Not

repo

rted

XN

ot re

port

edX

Not

repo

rted

XN

ot re

port

ed

Effec

ts n

ot id

entifi

ed

Kos

itsaw

at e

t al.

[30]

XN

ot re

port

edX

Not

repo

rted

XN

ot re

port

edX

Not

repo

rted

CIR

beta

cel

l cor

rect

ed in

sulin

resp

onse

; HO

MA-

IR h

omeo

stat

ic m

odel

ass

essm

ent f

or in

sulin

resi

stan

ce; M

SS m

etab

olic

syn

drom

e sc

ore

a Ide

ntifi

abili

ty a

ssum

ptio

ns w

ere

not s

peci

fical

ly m

entio

ned

in th

e ar

ticle

but

app

eare

d to

hav

e ap

prop

riate

refe

renc

es


Tabl

e 4

Sens

itiv

ity

anal

ysis

for m

easu

rem

ent e

rror

or m

iscl

assi

ficat

ion

of m

edia

tor i

n ca

usal

med

iati

on a

naly

sis

Refe

renc

esM

edia

tor

Ratio

nale

App

roac

hRe

sults

Jack

son

et a

l. [2

9]M

edic

al e

vent

s (b

inar

y)A

lgor

ithm

s w

ith h

igh

posi

tive‑

pred

ictiv

e va

lues

wer

e us

ed to

iden

tify

med

ical

eve

nts

durin

g fo

llow

up

Fals

e ne

gativ

es is

a c

once

rn u

nder

som

e sc

enar

ios

How

resu

lts w

ould

cha

nge

wer

e ex

amin

ed

give

n va

rious

sce

nario

s of

non

‑diff

eren

tial

and

diffe

rent

ial m

iscl

assi

ficat

ion

Perf

ect s

peci

ficity

for o

bser

ving

the

med

ical

ev

ent,

but v

arie

d th

e se

nsiti

vity

from

0.2

5 to

0.

75 s

epar

atel

y fo

r tho

se w

ho s

urvi

ved

and

for t

hose

who

die

d w

as a

ssum

edEa

ch s

cena

rio w

as a

ssum

ed th

at m

edia

tor

mis

clas

sific

atio

n w

as n

on‑d

iffer

entia

l with

re

spec

t to

antip

sych

otic

type

, cov

aria

tes,

and

othe

r med

iato

rs b

ut s

ome

scen

ario

s al

low

ed

for d

iffer

entia

l mis

clas

sific

atio

n w

ith re

spec

t to

dea

th. A

hyb

rid a

ppro

ach

was

als

o us

ed

The

prop

ortio

n m

edia

ted

was

hig

her t

han

the

naïv

e es

timat

ors

for s

ome

med

ical

eve

nts

and

grew

as

sens

itivi

ty d

ecre

ased

from

0.7

5 to

0.2

5. T

he s

ensi

tivity

am

ong

thos

e w

ho s

ur‑

vive

d, ra

ther

than

thos

e w

ho d

ied,

app

eare

d to

hav

e m

ore

influ

ence

on

thes

e re

sults

It w

as s

ugge

sted

that

15

to 4

5 %

of t

he m

orta

lity

diffe

renc

e m

ight

be

expl

aine

d by

som

e co

ndi‑

tions

giv

en s

cena

rios

assu

med

com

pare

d to

9

% u

sing

naï

ve a

ppro

ach

Auth

ors s

ugge

sted

to a

ddre

ss m

edia

tor m

is‑

clas

sific

atio

n w

hen

it is

sus

pect

ed, p

refe

r‑ab

ly th

roug

h va

lidat

ion

sub‑

stud

ies

or b

ias

anal

yses

Lu e

t al.

[32]

Biom

arke

rs (c

ontin

uous

)N

ot re

port

edTh

e im

pact

of m

easu

rem

ent e

rror

in th

e m

edia

tors

by

calib

ratin

g th

e re

gres

sion

coe

f‑fic

ient

s w

as a

sses

sed

Ass

umin

g th

at 1

‑tim

e m

easu

rem

ents

for e

ach

met

abol

ic ri

sk a

t bas

elin

e ex

plai

n on

ly 6

5 %

of

thei

r tru

e va

riabi

lity

(i.e.

35

% m

easu

re‑

men

t err

or)

Aft

er c

orre

ctin

g fo

r a p

resu

med

35

% m

easu

re‑

men

tEr

ror i

n ea

ch m

etab

olic

risk

fact

or in

crea

sed

the

over

all t

he p

erce

ntag

e of

exc

ess

rela

tive

risk

med

iate

d fro

m 4

7 %

(33–

63 %

) to

69 %

(5

2–87

%) f

or o

verw

eigh

t, an

d fro

m 5

2 %

(3

8–68

%) t

o 73

% (5

8–88

%) f

or o

besi

ty

Rao

et a

l. [3

6]Sm

okin

gC

hew

ing

quid

and

/or t

obac

coA

lcoh

ol (b

inar

y)

Dic

hoto

miz

atio

n of

med

iato

r var

iabl

e w

as

done

to s

impl

ify th

e an

alys

is b

ut th

e es

ti‑m

ates

from

the

anal

ysis

cou

ld b

e bi

ased

The

sens

itivi

ty a

naly

sis

for n

on‑d

iffer

entia

l m

iscl

assi

ficat

ion

erro

r of b

inar

y m

edia

tor

was

use

d

The

pred

ictiv

e va

lue

wei

ghtin

g es

timat

ors

for

outc

ome

regr

essi

on w

as u

sed

The

sens

itivi

ty a

naly

sis

was

car

ried

out w

ithou

t ac

coun

ting

for t

he c

lust

erin

g us

ing

the

plau

sibl

e se

nsiti

vity

val

ues

rang

ing

from

0.7

5 to

1.0

and

spe

cific

ity fr

om 0

.75

to 1

.0

In th

e ab

senc

e of

exp

osur

e m

edia

tor i

nter

ac‑

tion,

the

sens

itivi

ty a

naly

sis

indi

cate

d a

slig

ht

over

est

imat

ion

of th

e co

ntro

lled

dire

ct e

ffect

The

bias

see

med

to b

e la

rger

whe

n th

e se

nsiti

v‑ity

and

spe

cific

ity d

ecre

ased


Tabl

e 5

Elem

ents

of i

mpl

emen

tati

on fo

r cau

sal m

edia

tion

ana

lysi

s

a Sm

alle

st s

ampl

e si

ze u

sed

in th

e ca

usal

med

iatio

n an

alys

is

Refe

renc

esSa

mpl

e si

zea

Pow

er a

nd s

ampl

e si

ze c

alcu

la‑

tion

for m

edia

tion

anal

ysis

Expo

sure

‑med

iato

r int

erac

tion

in th

e m

odel

Pow

er a

nd s

ampl

e si

ze c

alcu

la‑

tion

for i

nter

actio

n an

alys

isLa

ck o

f pow

er m

entio

ned

as a

no

n‑ca

usal

exp

lana

tion

of fi

nd‑

ings

Stud

ies

with

neg

ativ

e fin

ding

s

Fre

eman

et a

l. [2

8]N

= 5

1 fo

r adj

uste

d an

d n =

54

for u

nadj

uste

d an

alys

is80

% p

ower

to d

etec

t lar

ge in

di‑

rect

effe

cts

(Frit

z M

, Mac

kinn

on

DP.

Requ

ired

sam

ple

size

to

dete

ct th

e m

edia

ted

effec

t)

Not

repo

rted

Not

app

licab

leLi

mite

d po

wer

to d

etec

t whe

ther

m

edia

ted

effec

ts w

ere

stat

isti‑

cally

sig

nific

ant

Jack

son

et a

l. [2

9]N

= 2

6,19

7N

ot re

port

ed✓

No

No

Kos

itsaw

at e

t al.

[30]

N =

176

5N

ot re

port

edN

ot re

port

edN

ot a

pplic

able

No

Lou

wie

s et

al.

[31]

N =

55

Not

repo

rted

Not

repo

rted

Not

app

licab

leCa

lled

for c

autio

us in

terp

reta

tion

give

n sm

all s

ampl

e si

ze

Men

dola

et a

l. [3

3]Va

ried

by o

utco

mes

(i.e

. 1 s

tudy

si

te d

id n

ot re

port

infa

nt a

pnea

)N

ot re

port

ed✓

No

No

Stud

ies

with

pos

itive

find

ings

Ban

ack

et a

l. [2

6]N

= 7

212

Not

repo

rted

✓N

oN

o

D’A

mel

io e

t al.

[27]

N =

37

80 %

pow

er, t

wo‑

side

d si

gnifi

‑ca

nce

leve

l of 0

.05,

to d

etec

t di

ffere

nces

in u

OC

gre

ater

than

1.

71 (t

test

on

log‑

scal

e)

✓N

oA

ckno

wle

dged

sm

all s

ampl

e si

ze

as a

lim

itatio

n

Lu

et a

l. [3

2]N

= 5

8,32

2 fo

r met

abol

icN

= 1

9,57

2 fo

r fibr

inog

enN

ot re

port

ed✓

No

No

Mes

serli

an e

t al.

[34]

All

sing

leto

n bi

rths

(n =

18,

147)

; on

ly fi

rst b

irths

(n =

865

1)N

ot re

port

edN

ot re

port

edN

ot a

pplic

able

Lim

ited

pow

er m

entio

ned;

lim

ited

sam

ple

size

to e

valu

ate

spon

‑ta

neou

s an

d in

duce

d pr

eter

m

birt

h se

para

tely

Rag

hava

n et

al.

[35]

Varie

d by

med

iato

rs (n

= 2

159

for d

iet s

core

; and

n =

209

8 fo

r ph

ysic

al a

ctiv

ity in

dex)

Not

repo

rted

Not

repo

rted

Not

app

licab

leLa

ck o

f pow

er to

exa

min

e th

e as

soci

atio

n in

the

med

iatio

n fra

mew

ork

Rao

et a

l. [3

6]N

= 4

33N

ot re

port

edX

No

No

Son

g et

al.

[37]

Varie

d by

med

iato

rs (b

iom

arke

rs)

Not

repo

rted

✓N

oD

espi

te s

ome

sign

ifica

nt m

edia

‑tio

n by

sev

eral

bio

mar

kers

, the

sa

mpl

e si

ze m

ay s

till n

ot b

e la

rge

enou

gh to

pro

vide

mor

e pr

ecis

e es

timat

es o

r to

dete

ct m

edia

tion

by o

ther

pot

entia

l fac

tors

with

sm

alle

r med

iatio

n eff

ects

Xie

et a

l. [3

8]N

= 1

67N

ot re

port

edX

No

Smal

l sam

ple

size

ack

now

ledg

ed


Table 6 Estimates of Direct and Indirect Effects With and/or Without Mediator-outcome Interaction

References Without exposure‑mediator interaction (95 % confidence interval)

With exposure‑mediator interaction (95 % confidence interval)

Discrepancy found with and with‑out exposure‑mediator interaction reason discussed

Banack et al. [26] Not reported Controlled direct effect with CVDRisk ratio: 0.62 (0.49, 0.78)Risk difference: −0.12 (−0.20, −0.04)Controlled direct effect without CVDRisk ratio: 1.30 (1.13, 1.49)Risk difference: 0.03 (0.01, 0.05)Total effectRisk ratio: 1.24 (1.11, 1.39)Risk difference: 0.03 (0.02, 0.05)

Not applicable

D’Amelio et al. [27] Effects of treatment on glucose level at 12 months mediated by OC at 6 months:

Natural direct effect: −0.033 (−0.186, 0.121)Natural indirect effect: −0.050 (−0.178,

0.078)Total effect: −0.082 (−0.174, 0.009)

Not reported Not applicable

Freeman et al. [28] Anxiety (Boot SE), P valueDirect effect: 0.21 (0.16), 0.19Indirect effect: 0.18 (0.11), 0.09Total effect: 0.39 (0.16), 0.01Depression (Boot SE), P valueDirect effect: 0.24 (0.14), 0.09Indirect effect: 0.15 (0.11), 0.18Total effect: 0.39 (0.16), 0.01BCSS—negative self (Boot SE), P valueDirect effect: 0.33 (0.18), 0.06Indirect effect: 0.06 (0.08), 0.48Total effect: 0.39 (0.16), 0.01BCSS—positive self (Boot SE), P valueDirect effect: 0.40 (0.17), 0.02Indirect effect: −0.01 (0.05), 0.92Total effect: 0.39 (0.16), 0.01BCSS—negative other (Boot SE), P valueDirect effect: 0.22 (0.16), 0.17Indirect effect: 0.18 (0.11), 0.13Total effect: 0.39 (0.16), 0.01

Not applicable Not applicable

Jackson et al. [29] StrokeDirect effect: 1.13 (1.05,1.22)Indirect effect: 1.005 (1.001,1.011)Total effect: 1.14 (1.06,1.22)

StrokeDirect effect: 1.13 (1.06,1.22)Indirect effect: 1.005 (1.001,1.011)Total effect: 1.14 (1.06,1.22)

NoNot applicable

Kositsawat et al. [30] Not reported Not applicable Not applicable

Louwies et al. [31] Systolic blood pressureDirect effect: 2.93 (CIs not reported)Indirect effect: −0.42 (−1.35 to 0.17)Total effect: not reportedDiastolic blood pressureDirect effect: 3.15 (CIs not reported)Indirect effect: −0.59 (−1.44 to 0.07)Total effect: not reported


Lu et al. [32] Overweight (formetabolic mediators)Blood pressureNatural direct effect: 1.16 (1.09–1.24)Natural indirect effect: 1.06 (1.03–1.08)Total effect: 1.22 (1.14–1.30)Obesity (formetabolic mediators)Blood pressureNatural direct effect: 1.28 (1.15–1.43)Natural indirect effect: 1.13 (1.07–1.19)Total effect: 1.42 (1.25–1.60)

Overweight (formetabolic mediators)Blood pressureNatural direct effect: 1.16 (1.09–1.24)Natural indirect effect: 1.05 (1.02–1.08)Total effect: 1.22 (1.14–1.30)Obesity (formetabolic mediators)Blood pressureNatural direct effect: 1.28 (1.15–1.42)Natural indirect effect: 1.10 (1.03–1.17)Total effect: 1.43 (1.25–1.62)

NoNot applicable


Effects of estimates and derived results from exposure‑mediator interactionTable 6 shows the estimates from causal mediation anal-ysis with and without interaction in the model for the associations between the primary study exposure and outcome listed in Table 1. While the majority of the stud-ies reported estimates from either with and or without interaction in the model, 3 studies did not report iden-tified estimates of effects. Among 6 studies allowing for exposure-mediator interaction, 2 studies presented results from both with and without interaction in the model and no substantial discrepancies were found.

DiscussionOur review shows that reporting of research on meth-ods using causal mediation analysis to better understand mechanisms of observed exposure-outcome relationship is varied and suboptimal in the field of epidemiology. After reviewing 13 epidemiological studies, we found that while the field of causal mediation analysis has made significant strides, majority of the studies lacked suffi-cient details on whether the identifiability assumptions were satisfied in relation to identified effect estimates.

Furthermore, despite most studies addressing the con-cern for unmeasured confounders either from empirical approaches or sensitivity analyses, we found that over half of studies did not examine the potential bias arising from the validity of the mediator. In addition, the major-ity of studies did not provide or comment information on the power calculation or issues of sample size. While some studies allowed for exposure-mediator interaction, only a few presented results from both with and without interaction in the model.

Although it was difficult to judge the adequacy of con-trol for confounding in the reviewed studies without increased knowledge of the specific datasets and subject areas, we found that most studies did not provide enough information on whether either the empirical approach or sensitivity analysis was conducted for identification assumptions in relation to effect estimates identified. It has been emphasized that controlling for mediator-out-come confounders is important when direct and indirect effects are examined [8, 9, 20]. When there is concern for uncontrolled confounding, sensitivity analyses have been recommended to quantify the extent to which the unmeasured confounding variable would have to be to

BCSS brief core schema scales; CI confidence interval; CVD cardiovascular disease; HOMA-IR homeostatic model assessment for insulin resistance; OC osteocalcin; SE standard errors; SHBG sex hormone-binding globulina Despite allowing for interaction, only models assuming no interaction were adopted due to no significant interaction between any of the exposures and mediators was observed

Table 6 continued

References Without exposure‑mediator interaction (95 % confidence interval)

With exposure‑mediator interaction (95 % confidence interval)

Discrepancy found with and with‑out exposure‑mediator interaction reason discussed

Mendola et al. [33] Not reported Peri‑ or intraventricular hemorrhage:Controlled direct effect: 3.2 (1.4–7.7)Total effect: 2.9 (2.4–3.4)

Not applicable

Messerlian et al. [34] Uterine: <35 weeksControlled direct effect: 2.43 (0.85, 6.93)Total effect: 2.27 (1.32, 3.89)


Raghavan et al. [35] Metabolic syndrome score:Direct effect: not reportedIndirect effect: 1.20 (1.07, 1.33)Total effect: not reported


Rao et al. [36] Smoking: low vs. highControlled direct effect: 1.5 (1.4, 1.5)Total effect: 1.6 (1.4, 1.9)


Song et al. [37] HOMA‑IR:Effect not mediated: 1.32 (0.95, 1.88)Mediated effect: 1.22 (1.02, 1.49)SHBG:Effect not mediated: 1.97 (1.25, 3.10)Mediated effect: 1.16 (1.03, 1.33)E‑selectin:Effect not mediated: 1.56 (1.10, 2.21)Mediated effect: 1.12 (1.02, 1.23)Systolic blood pressure:Effect not mediated: 1.61 (1.25, 2.16)Mediated effect: 1.03 (1.01, 1.07)

Not reporteda Not applicable

Xie et al. [38] Not reported Not applicable Not applicable


invalidate inferences about the direct and indirect effects [15, 24, 39]. Several approaches can be used to address unmeasured confounding [1]. For example, researchers can choose to report how large the effects of the con-founder variable would need to be to completely explain the effects of estimates. To improve reporting of causal mediation analyses in epidemiological literature, we rec-ommend the following. First, studies should be trans-parent on whether the empirical approach or sensitivity analyses were used to evaluate identifiability assump-tions. Second, studies must carefully consider the extent to which bias is present due to concerns regarding valid measurement of the mediator. Several approaches are available to address this issue [21, 22]. Third, if research-ers are concerned about the presence of multiple bias in the study, we recommend that researchers prioritize the approaches depending on the context to strengthen their findings.

We found that the majority of studies did not report whether the statistical power or sample size calculation was calculated or if the researchers believed that the sample size available was sufficient to estimate direct and indirect effects with sufficient precision. However, we recognize that approached for calculation power and sample size for direct and indirect effects is limited in the current literature, especially for the exposure-mediator interaction [1]. To understand what sample size is suffi-cient for mediation analysis, it is currently recommended that researchers use previously published tables for ade-quate power in single-mediator models [40]. In addition, we also recommend that studies should comment on whether lack of power or insufficient sample size was a likely non-causal explanation of findings especially for these with relatively small sample size.

It has been proven that under sequential ignorability and the additional no-interaction assumption, the esti-mate based on the product of coefficients method can be interpreted as a valid estimate of the causal mediation effect as long as the linearity assumption holds [41, 42]. However, in many studies it is unrealistic to assume that the exposure and mediator do not interact in their effects on the outcome. Carrying out mediation analysis incor-rectly assuming no interaction may result in invalid infer-ences [13]. Despite the progress of statistical methods in mediation analysis under settings with a binary mediator or count outcomes for exposure-mediator interactions [13], we found that most studies did not report whether there was exposure-mediator interaction in the model.. Although more assumptions are required for the decom-position of a total effect into direct and indirect effects even in models with interactions and non-linearity under the counterfactual framework, this decomposition of total effects allows investigators to assess whether most

of the effect is mediated through a particular intermedi-ate or the extent to which it is through other pathways. Therefore, we recommend that future studies include exposure-mediator interactions by default in the out-come model as suggested [1]. We recommend that exposure-mediator interactions only be excluded if the magnitude of interactions is small and do not change the estimates of direct and indirect effects very much.

Our review is subject to some limitations. First, we included only epidemiological studies published in 2015 and limited to those published in English. The findings may not be representative of all publications using causal mediation analysis. However, it is reasonable to give some time for the development and use of methods given that seminal articles for applications were mainly published in 2012 or 2013 and we are interested in a “snapshot” of cur-rent practices in reporting such complex methods from the most recent year. Second, the reporting practices of published studies may be influenced by journals’ require-ments. Authors may be reporting their approach and findings given word limitations from journals and thus may have limited space to provide details needed for the method. Nevertheless, with methods that require careful implementation of the approach, such reporting is neces-sary to evaluate the extent to which the method has been appropriately applied. Third, it is possible that we missed some relevant articles due to lack of standardized ter-minology or exchangeable jargons to describe the study design of causal mediation analysis. However, we believe that including papers which cited the seminal papers reduced the likelihood of this happening. Despite the limitations, this is the first review to examine how epide-miological studies have used causal mediation analysis, what appropriate procedures and analysis are needed to conduct such complex technique, and what elements are critical to report for the method, which is we believe is a strength of our review.

ConclusionsAlthough the application of causal mediation analysis is increasing in epidemiology, there is an opportunity for improving the quality and presentation of this methodol-ogy. We found that there is varied and suboptimal report-ing of this emerging approach in literature. We identified that the majority of studies addressing unmeasured con-founding for the mediator-outcome relationship. We recommend that future studies should: (1) provide suf-ficient details on whether either the empirical approach or sensitivity analysis was conducted for identifiability assumptions in relation to effect estimates identified, (2) comment on the bias that may arise from the validity of mediator, (3) discuss whether lack of statistical power or insufficient sample size issue was likely a non-causal


explanation of findings, and (4) allow the inclusion of exposure-mediator interaction in the model and present results derived from models with and without interaction terms. We hope that the development of best practices in reporting complex methods in epidemiological research and the adoption of such reporting standards may help quality assessment and interpretation of studies using causal mediation analysis.

Authors’ contributionsAll listed authors have made material contribution to the completion of this manuscript. Mr. Liu conceived and designed the study as well as performed data extraction. This study was conducted under the guidance of Drs. Lapane and Ulbricht with input from the remaining authors. Extracted information were confirmed by Drs. Lapane and Chrysanthopoulou for all eligible studies. Mr. Liu also wrote the first draft of the article, and all other authors revised the manuscript for important intellectual content. All authors read and approved the final manuscript.

Author details1 Clinical and Population Health Research Program, Graduate School of Bio‑medical Sciences, University of Massachusetts Medical School, 368 Planta‑tion Street, Worcester, MA 01655, USA. 2 Division of Epidemiology of Chronic Diseases and Vulnerable Populations, Department of Quantitative Health Sciences, University of Massachusetts Medical School, Worcester, MA 01605, USA. 3 Division of Biostatistics and Health Services Research, Department of Quantitative Health Sciences, University of Massachusetts Medical School, Worcester, MA 01605, USA.

Competing interestsThe authors declare that they have no competing interests.

FundingThis work was supported by the National Institute on Aging [grant number 1R21AG046839‑01 to Dr. Kate Lapane]; and the National Cancer Institute [Grant number 1R21CA198172 to Dr. Kate Lapane].

Received: 30 April 2016 Accepted: 14 July 2016

References 1. VanderWeele TJ. Explanation in causal inference: methods for mediation

and interaction. Oxford: Oxford University Press; 2015. 2. Vanderweele TJ, Asomaning K, Tchetgen Tchetgen EJ, Han Y, Spitz

MR, Shete S, et al. Genetic variants on 15q25.1, smoking, and lung cancer: an assessment of mediation and interaction. Am J Epidemiol. 2012;175:1013–20.

3. Huang Y‑T, Vanderweele TJ, Lin X. Joint analysis of SNP and gene expres‑sion data in genetic association studies of complex diseases. Ann Appl Stat. 2014;8:352–76.

4. VanderWeele TJ, Lauderdale DS, Lantos JD. Medically induced preterm birth and the associations between prenatal care and infant mortality. Ann Epidemiol. 2013;23:435–40.

5. Mumford SL, Schisterman EF, Siega‑Riz AM, Gaskins AJ, Wactawski‑Wende J, VanderWeele TJ. Effect of dietary fiber intake on lipoprotein cholesterol levels independent of estradiol in healthy premenopausal women. Am J Epidemiol. 2011;173:145–56.

6. Nandi A, Glymour MM, Kawachi I, VanderWeele TJ. Using marginal struc‑tural models to estimate the direct effect of adverse childhood social conditions on onset of heart disease, diabetes, and stroke. Epidemiology. 2012;23:223–32.

7. Baron RM, Kenny DA. The moderator‑mediator variable distinction in social psychological research: conceptual, strategic, and statistical consid‑erations. J Pers Soc Psychol. 1986;51:1173–82.

8. Robins JM, Greenland S. Identifiability and exchangeability for direct and indirect effects. Epidemiology. 1992;3:143–55.

9. Pearl J. Direct and indirect effects. In: Proceedings seventeenth confer‑ence uncertain Artificial Intelligence. 2001. p. 411–20.

10. Vanderweele TJ, Vansteelandt S. Conceptual issues concerning media‑tion, interventions and composition. Stat Interface. 2009;2:457–68.

11. Imai K, Keele L, Tingley D, Yamamoto T. Causal mediation analysis using R. In: Advances in social science research using R. Vol 196. 2010. p. 129–54.

12. Imai K, Keele L, Tingley D. A general approach to causal mediation analy‑sis. Psychol Methods. 2010;15:309–34.

13. Valeri L, Vanderweele TJ. Mediation analysis allowing for exposure‑mediator interactions and causal interpretation: theoretical assumptions and implementation with SAS and SPSS macros. Psychol Methods. 2013;18:137–50.

14. Preacher KJ, Rucker DD, Hayes AF. Addressing modearated mediation hypotheses: theory, methods, and prescriptions. Multivar Behav Res. 2007;42:185–227.

15. VanderWeele TJ, Vansteelandt S. Odds ratios for mediation analysis for a dichotomous outcome. Am J Epidemiol. 2010;172:1339–48.

16. Richiardi L, Bellocco R, Zugna D. Mediation analysis in epidemiology: methods, interpretation and bias. Int J Epidemiol. 2013;42:1511–9.

17. Emsley R, Liu H. PARAMED: Stata module to perform causal mediation analysis using parametric regression models. Statistical software compo‑nents. Boston: Boston College Department of Economics; 2013.

18. Moher D, Liberati A, Tetzlaff J, Altman DGTPG. Preferred reporting items for systematic reviews and meta‑analyses: the PRISMA statement. PLoS Med. 2009;2009(6):e1000097.

19. VanderWeele TJ. Mediation analysis: a practitioner’s guide. Ann Rev Publ Health. 2016;37:annurev—publhealth—032315–021402.

20. Cole SR, Hernán MA. Fallibility in estimating direct effects. Int J Epidemiol. 2002;31:163–5.

21. Valeri L, Lin X, VanderWeele TJ. Mediation analysis when a continuous mediator is measured with error and the outcome follows a generalized linear model. Stat Med Engl. 2014;33:4875–90.

22. le Cessie S, Debeij J, Rosendaal FR, Cannegieter SC, Vandenbroucke JP. Quantification of bias in direct effects estimates due to different types of measurement error in the mediator. Epidemiology. 2012;23:551–60.

23. Blakely T, McKenzie S, Carter K. Misclassification of the mediator mat‑ters when estimating indirect effects. J Epidemiol Community Health. 2013;67:458–66.

24. Hafeman DM. Confounding of indirect effects: a sensitivity analysis exploring the range of bias due to a cause common to both the media‑tor and the outcome. Am J Epidemiol. 2011;174:710–7.

25. VanderWeele TJ. Bias formulas for sensitivity analysis for direct and indi‑rect effects. Epidemiology. 2010;21:540–51.

26. Banack HR, Kaufman JS. Does selection bias explain the obesity para‑dox among individuals with cardiovascular disease? Ann Epidemiol. 2015;25:342–9.

27. D’Amelio P, Sassi F, Buondonno I, Spertino E, Tamone C, Piano S, et al. Effect of intermittent PTH treatment on plasma glucose in osteoporosis: a randomized trial. Bone. 2015;76:177–84.

28. Freeman D, Emsley R, Dunn G, Fowler D, Bebbington P, Kuipers E, et al. The stress of the street for patients with persecutory delusions: a test of the symptomatic and psychological effects of going outside into a busy urban area. Schizophr Bull. 2015;41:971–9.

29. Jackson JW, VanderWeele TJ, Blacker D, Schneeweiss S. Mediators of first‑versus second‑generation antipsychotic‑related mortality in older adults. Epidemiology. 2015;26:700–9.

30. Kositsawat J, Kuchel GA, Tooze JA, Houston DK, Cauley JA, Kritchevsky SB, et al. Vitamin D insufficiency and abnormal hemoglobin A1c in black and white older persons. J Gerontol Ser A Biol Sci Med Sci. 2015;70:525–31.

31. Louwies T, Nawrot T, Cox B, Dons E, Penders J, Provost E, et al. Blood pressure changes in association with black carbon exposure in a panel of healthy adults are independent of retinal microcirculation. Environ Int. 2015;75:81–6.

32. Lu Y, Hajifathalian K, Rimm EB, Ezzati M, Danaei G. Mediators of the effect of body mass index on coronary heart disease decomposing direct and indirect effects. Epidemiology. 2015;26:153–62.

33. Mendola P, Mumford SL, Mannisto TI, Holston A, Reddy UM, Laughon SK. Controlled direct effects of preeclampsia on neonatal health after accounting for mediation by preterm birth. Epidemiology. 2015;26:17–26.

34. Messerlian C, Platt RW, Ata B, Tan S‑L, Basso O. Do the causes of infertility play a direct role in the aetiology of preterm birth? Paediatr Perinat Epidemiol Engl. 2015;29:101–12.


• We accept pre-submission inquiries

• Our selector tool helps you to find the most relevant journal

• We provide round the clock customer support

• Convenient online submission

• Thorough peer review

• Inclusion in PubMed and all major indexing services

• Maximum visibility for your research

Submit your manuscript atwww.biomedcentral.com/submit

Submit your next manuscript to BioMed Central and we will help you at every step:

35. Raghavan S, Porneala B, McKeown N, Fox CS, Dupuis J, Meigs JB. Meta‑bolic factors and genetic risk mediate familial type 2 diabetes risk in the Framingham heart study. Diabetologia. 2015;58:988–96.

36. Rao SK, Mejia GC, Roberts‑Thomson K, Logan RM, Kamath V, Kulkarni M, et al. Estimating the effect of childhood socioeconomic disadvantage on oral cancer in india using marginal structural models. Epidemiology. 2015;26:509–17.

37. Song Y, Huang Y‑T, Song Y, Hevener AL, Ryckman KK, Qi L, et al. Birth‑weight, mediating biomarkers and the development of type 2 diabetes later in life: a prospective study of multi‑ethnic women. Diabetologia. 2015;58:1220–30.

38. Xie C, Zhao Y, Gao L, Chen J, Cai D, Zhang Y. Elevated phthalates’ exposure in children with constitutional delay of growth and puberty. Mol Cell Endocrinol. 2015;407:67–73.

39. Tchetgen Tchetgen EJ, Shpitser I, Tchetgen EJT, Shpitser I. Semipa‑rametric theory for causal mediation analysis: efficiency bounds, multiple robustness and sensitivity analysis. Ann Stat Inst Math Stat. 2012;40:1816–45.

40. Fritz MS, MacKinnon DP. Required sample size to detect the mediated effect. Psychol Sci. 2007;18:233–9.

41. Jo B. Causal inference in randomized experiments with mediational processes. Psychol Methods. 2008;13:314–36.

42. Imai K, Keele L, Yamamoto T. Identification, inference and sensitivity analysis for causal mediation effects. Stat Sci. 2010;25:51–71.

implementation and reporting of causal mediation analysis ...studies apply and report the elements...

Documents