meta-analysis of biofeedback for tension-type headache: efficacy, specificity, and treatment...

18

Upload: independent

Post on 14-Nov-2023

0 views

Category:

Documents


0 download

TRANSCRIPT

omitted results for measured outcome data), we imputed zeroeffects. This conservative method leads to lower limit effect sizes,thus taking into account the inaccuracies in reporting (Rosenthal,1994).

Integration of Effect Sizes

We integrated all outcome variables into three symptom cate-gories, indexing headache pain as the primary outcome variable aswell as physiological and psychological/behavioral variables assecondary outcome.

Integration of dependent effect sizes. Most studies used mul-tiple outcome measures, thereby producing dependent effect sizesas a function of intercorrelations between assessments. Indepen-dencies between effect sizes were also caused by the multiple useof a single CG to establish effect sizes of different active treatmentgroups. These dependencies were handled by averaging the mul-tiple outcome effect sizes within treatment groups and the multipletreatment effect sizes within studies. To obtain effect variances,the individual effect size variances were averaged with covarianceadjustment (Gleser & Olkin, 1994).3

Integration of independent effect sizes. We weighted theeffect sizes d by the inverse of their sampling variances (Hedges &Olkin, 1985) and calculated mean d�s by averaging the weightedeffects. To outline the general efficacy of biofeedback for TTH, wepooled the outcome on all available headache variables and addi-tionally presented average effect sizes for all specific headachevariables. Further separate integrations were carried out with re-spect to other outcome categories and time points. The homoge-neity statistic Q (Shadish & Haddock, 1994) was calculated todetermine whether each set of ds shared a common populationeffect size. Due to initial heterogeneity, all analyses were com-puted within the random effect model (REM). The proportion ofsystematic unexplained variance (�2) in the REM was calculatedfrom the expected total sum of squares of weighted effect sizes.

Moderator Analyses

Moderating variables were defined a priori based on character-istics of the investigated treatments and study participants. Hy-pothesized moderators were feedback modality, EMG training site,home training, treatment duration, diagnosis, age, and duration ofTTH. Each moderator was coded according to planned contrastsamong its levels and analyzed by random effects analysis ofvariance by using the Q statistic (Hedges & Olkin, 1985).Weighted average effects per level were calculated with themethod of moments. The statistic QB was used as an omnibus testfor differences between levels. QW was used to test for homoge-neity within levels. The presence of a moderator is indicated byheterogeneity between levels and homogeneity within levels. Thepercentage of reduction in the unexplained variance component(�2

explained) was calculated for each moderating variable. Since ourmethod of reconstructing effects sizes (i.e., directly calculatedeffects � 1 vs. reconstruction from significance bounds � 0) wascorrelated with headache effect sizes (rpbis � .52, p � .01), thecontrol variable “effect size reconstruction” was added to themodel.4

Sensitivity Analyses

Varying formulas for effect size calculation. Differences be-tween Hedges’ d and its pre-post equivalent have been described toinfluence meta-analytical results (Hartmann & Herzog, 1995). Toanalyze the impact of the method of effect size calculation, weseparately integrated and compared effect sizes and variances thatwere based on varying formulas (Ray & Shadish, 1996).

Publication bias. Results of a meta-analysis may be biaseddue to the fact that studies with nonsignificant results are lesslikely to be published than are those leading to significant results.Controlling for the association between effect sizes and theirsample sizes, we examined a potential publication bias graphicallywith a funnel plot analysis and numerically with the fail-safe Ncriterion (Rosenthal, 1979).

Intention-to-treat (ITT) analysis. Missing data from dropoutsrepresent a potential bias to treatment studies. ITT analysis is awidely recommended solution to it in primary research (Wright &Sim, 2003), as well as in meta-analysis (Moher et al., 1999). Weimputed missing data from patients who dropped out of a studyafter randomization with zero effects (Higgins & Green, 2005),assuming that dropouts were nonresponders.

Results

Characteristics of Included Studies

A descriptive overview of all integrated studies is presented inTable 1.

Extrinsic study characteristics. All studies were publishedbetween 1973 and 2001. Fifty percent of the integrated studieswere published until 1983, 75% until 1991. Studies were con-ducted in the United States (n � 30); Italy (n � 5); Canada (n �4); India (n � 3); Great Britain, Germany, and Denmark (each n �2); and France, Portugal, Belgium, Sweden, and the Netherlands(each n � 1).

Design and treatment characteristics. Twenty-one of the in-tegrated studies were pre-post trials, and 32 implemented CGs, 24of these being randomized controlled trials. Six studies success-fully implemented single or double blind designs. Studies inves-tigated 103 treatment conditions, of which 61 were active biofeed-back conditions including predominantly EMG-FB groups (k �57) as well as 2 temperature feedback (TEMP-FB) groups, 1galvanic skin response feedback (GSR-FB) group, and 1 electro-encephalography feedback (EEG-FB) group. Nine of the EMG-FBtreatments were conducted in combination with relaxation; in 3trials, only relaxation nonresponders were treated. The number ofbiofeedback sessions ranged from 6 to 20 (M � 10.8, SD � 4.1).Sessions lasted from 20 to 90 minutes with an average duration of41.4 minutes (SD � 15.3). In most of the active EMG-FB groups(k � 41), electrodes were placed bifrontal. In 12 groups, additionalelectrodes were placed on the neck (i.e., trapezius muscle) or jaw

3 Intercorrelations of headache variables were calculated at post-treatment and ranged from .35 for intensity and headache index to .83 forfrequency and headache index.

4 Significant two-way interactions with the methodological control vari-able resulted for EMG training site and diagnosis. Analyses for these twomoderators were restricted to directly calculated effects.

382 NESTORIUC, RIEF, AND MARTIN

Tab

le1

Des

ign,

Tre

atm

ent

Fea

ture

s,an

dE

ffec

tSi

zes

ofA

llIn

clud

edSt

udie

s

Stud

yau

thor

Des

ign

(na )

Follo

w-u

p(F

U)

Dro

pout

Tre

atm

ent

grou

ps(n

b)

Patie

nts

Out

com

em

easu

res

ES

calc

ulat

ion

Hea

dach

ere

duct

ion

Phys

iolo

gica

lou

tcom

ePs

ycho

logi

cal/

beha

vior

alou

tcom

e

Age

Hea

dach

edu

ratio

ndc

95%

CI

dc95

%C

Idc

95%

CI

And

rasi

k&

Hol

royd

,19

80R

CT (40)

0

3FU

EM

G-F

Bfr

onta

lis(1

0)19

.73.

5Pa

in,m

icV

,M

edfr

omt

stat

istic

0.37

�0.

16,0

.90

0.38

�0.

09,�

0.85

0.00

�0.

45,0

.45

And

rasi

k&

Hol

royd

,19

8336

mon

th1

EM

G-F

Bfr

onta

lisin

crea

se(9

)ac

tive

vs.

cont

rol

0.31

�0.

22,0

.84

——

0

4FU

EM

G-F

Bfr

onta

lisst

able

(10)

0.13

�0.

41,0

.66

——

0

4FU

Hea

dach

em

onito

ring

(10)

——

Are

naet

al.,

1995

RC

T (27)

—1

EM

G-F

Bfr

onta

lis(8

)34

.3(2

1–68

)16

.6(1

2.4)

Pain

,mic

Vfr

omM

SD

0.48

�0.

26,1

.22

0.42

�0.

30,1

.15

0E

MG

-FB

trap

eziu

s(1

0)1.

25d

0.40

,2.1

0�

0.22

�0.

84,0

.41

0PM

R(8

)0.

39�

0.33

,1.1

1—

—A

rena

etal

.,19

91Pr

e-po

st(8

)no

tgi

ven

EM

G-F

Bfr

onta

lis(8

)65

.0(6

2–71

)42

.4Pa

in,m

icV

,M

edfr

omM

SD

0.78

�0.

07,1

.63

0.66

�0.

11,1

.43

0.60

�0.

16,1

.36

Arn

dorf

er&

Alle

n,20

01Pr

e-po

st(5

)—

0T

EM

P-FB

(5)

11.0

(8–1

4)0.

5–3

Pain

from

tst

atis

tics

1.22

�0.

10,2

.54

——

Bill

ings

etal

.,19

84Pr

e-po

st(2

9)5

EM

G-F

Bfr

onta

lis

TE

MP-

FB(2

4)

35.3

(12.

8)8.

6Pa

in,m

icV

from

M

SD1.

46d

0.88

,2.0

40.

64d

0.20

,1.0

8—

Bis

chof

f&

Dah

linge

r,19

93R

CT (14)

not

give

nE

MG

-FB

trap

eziu

s,am

bula

nt(7

)

37.0

(10.

8)11

.0(1

1.6)

Pain

,mic

Vfr

omM

SD

1.05

d0.

10,2

.01

�0.

17�

0.92

,0.5

8—

—E

MG

-FB

trap

eziu

s,ho

me

base

d(7

)

39.0

(10.

5)20

.4(1

5.3)

0.97

d0.

03,1

.90

0.19

�0.

56,0

.94

Bla

ncha

rdet

al.,

1987

RC

T (28)

60m

onth

17FU

EM

G-F

Bfr

onta

lis

PMR

(4)

42.8

(36–

54)

14(1

–32)

Pain

from

M

SD0.

85�

0.34

,2.0

5—

—PM

R(5

)1.

17�

0.03

,2.3

6—

—B

lanc

hard

etal

.,19

82b

Pre-

post

(15)

—no

tgi

ven

EM

G-F

Bfr

onta

lisPM

Rnon

(15)

24–6

7no

tgi

ven

Pain

,Med

from

tst

atis

tics

0.62

d0.

05,1

.20

—0.

52d

0.10

,0.9

4

Bla

ncha

rdet

al.,

1982

aPr

e-po

st(1

5)—

1E

MG

-FB

fron

talis

PMR

non

(14)

40.2

(18–

68)

not

give

nPa

infr

omF

stat

istic

s0.

58�

0.05

,1.2

2—

Bor

geat

etal

.,19

91Pr

e-po

st(3

2)—

not

give

nE

MG

-FB

fron

talis

(32)

38.5

(20–

55)

12.3

(0.5

–38)

Pain

from

raw

data

0.53

d0.

16,0

.90

——

Bor

geat

etal

.,19

85Pr

e-po

st(3

3)—

not

give

nE

MG

-FB

fron

talis

,pai

nco

ntra

ctio

n(1

4)

38.6

(20–

55)

12.3

(0.5

–38)

Pain

,mic

V,

Med

from

pva

lue

0.45

�0.

11,1

.00e

0.67

d0.

08,1

.25

0.00

�0.

52,0

.52

——

EM

G-F

Bfr

onta

lis,

nopa

inco

ntra

ctio

n(1

9)

0.28

�0.

25,0

.81e

0.00

�0.

52,0

.52

0.56

d0.

08,1

.05

Bor

geat

etal

.,19

84Pr

e-po

st(1

6)—

not

give

nE

MG

-FB

fron

talis

(16)

35.2

(22—

56)

7.3

(0.1

—20

)Pa

in,m

icV

,M

edfr

omt

stat

istic

0.54

d0.

13,0

.95

0.66

d0.

10,1

.23

0.02

Bru

hnet

al.,

1979

RC

T (28)

1E

MG

-FB

fron

talis

m

asse

ter

(13)

35.1

11.5

Pain

,Med

from

odds

ratio

1.41

d0.

49,2

.33

—1.

09d

0.21

,1.9

7

4T

reat

men

tas

usua

l(1

0)ac

tive

vs.

cont

rol

——

(tab

leco

ntin

ues)

383META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

Tab

le1

(con

tinu

ed)

Stud

yau

thor

Des

ign

(na )

Follo

w-u

p(F

U)

Dro

pout

Tre

atm

ent

grou

ps(n

b)

Patie

nts

Out

com

em

easu

res

ES

calc

ulat

ion

Hea

dach

ere

duct

ion

Phys

iolo

gica

lou

tcom

ePs

ycho

logi

cal/

beha

vior

alou

tcom

e

Age

Hea

dach

edu

ratio

ndc

95%

CI

dc95

%C

Idc

95%

CI

Bud

zyns

kiet

al.,

1973

RC

T (18)

3m

onth

0E

MG

-FB

fron

talis

(6)

not

give

nPa

in,m

icV

from

pva

lue

——

0Ps

eudo

feed

back

(6)

activ

evs

.co

ntro

l1.

08d

0.36

,1.8

00.

48�

0.60

,1.5

7—

0H

eada

che

mon

itori

ng(6

)1.

08d

0.36

,1.8

0—

Bus

sone

etal

.,19

98R

CT (35)

12m

onth

0E

MG

-FB

fron

talis

PM

R(2

0)11

.1(2

.6)

2.6

(2.0

)Pa

in,m

icV

,A

nxfr

omM

SD

0.84

d0.

32,1

.35

0.00

�0.

44,0

.44

0.47

d0.

02,0

.92

5R

elax

atio

npl

aceb

o(1

0)13

.0(1

.5)

2.7

(2.0

)0.

96d

0.20

,1.7

30.

51�

0.16

,1.1

70.

03�

0.53

,0.5

8

Che

sney

&Sh

elto

n,19

76R

CT (24)

—no

tgi

ven

EM

G-F

Bfr

onta

lis(6

)no

tgi

ven

from

tst

atis

tic0.

00�

1.02

,1.0

2—

—PM

R(6

)Pa

inac

tive

vs.

cont

rol

1.62

d0.

60,2

.46

——

—E

MG

-FB

fron

talis

PM

R(6

)1.

87d

0.81

,2.9

3—

—H

eada

che

mon

itori

ng(6

)—

——

Col

let

etal

.,19

86R

CT (31)

7.5

mon

th3

TE

MP-

FB(1

3)39

.7(2

3–62

)7.

7(1

–20)

Pain

,Med

,A

nxfr

omM

SD

0.76

d0.

10,1

.41

—0.

79d

0.16

,1.4

2

3

5FU

AT

(12)

0.43

�0.

30,0

.99

—0.

25�

0.33

,0.8

2

Cot

tet

al.,

1981

RC

T (8)

12m

onth

0E

MG

-FB

fron

talis

R

elax

atio

n(4

)

not

give

nPa

in,M

edfr

omp

valu

e0.

86�

0.34

,2.0

5—

1.65

d0.

01,3

.30

0R

elax

atio

n(4

)0.

86�

0.34

,2.0

5—

1.65

d0.

01,3

.30

Cox

etal

.,19

75R

CT (27)

4m

onth

0

1FU

EM

G-F

Bfr

onta

lis(9

)39

(16–

64)

11(1

–39)

Pain

,Med

,Se

lffr

omM

SD

1.14

d0.

29,1

.99

—0.

69�

0.05

,1.4

2

0

1FU

PMR

(9)

Eff

icac

y1.

87d

0.87

,2.8

7—

0.25

�0.

41,0

.92

0

1FU

Plac

ebo

pill

(9)

0.37

�0.

40,1

.13

—0.

39�

0.29

,1.0

7

Dal

yet

al.,

1983

RC

T (26)

3m

onth

0E

MG

-FB

fron

talis

(9)

32.0

(11.

0)12

.2(9

.2)

Pain

from

M

SD0.

42�

0.35

,1.1

9—

1PM

R(8

)0.

48�

0.35

,1.3

0—

—0

TE

MP-

FB(8

)0.

80�

0.05

,1.6

6—

—E

pste

in&

Abe

l,19

77Pr

e-po

st(8

)—

2E

MG

-FB

fron

talis

(6)

32.8

(9.2

)8.

2(7

.8)

Pain

,mic

Vfr

omra

wda

ta0.

50�

0.36

,1.3

50.

14�

0.67

,0.9

4—

Gad

a,19

84R

CT (65)

—3

EM

G-F

Bfr

onta

lis(3

0)35

.2no

tgi

ven

Pain

from

M

SD1.

33d

0.85

,1.8

1—

4PM

R(2

8)36

.4no

tgi

ven

1.16

d0.

68,1

.64

——

Gra

yet

al.,

1980

RC

T (20)

3m

onth

5E

MG

-FB

fron

talis

,pai

nsi

de(5

)

33.8

(20–

70)

not

give

nPa

in,m

icV

from

tst

atis

tics

0.25

�0.

77,1

.28

1.15

�0.

40,2

.70

—E

MG

-FB

fron

talis

,no

npai

nsi

de(5

)

and

pva

lues

0.25

�0.

77,1

.28

0.85

�0.

48,2

.17

—R

elax

atio

n(5

)0.

51�

0.54

,1.5

60.

96�

0.45

,2.3

6—

384 NESTORIUC, RIEF, AND MARTIN

Tab

le1

(con

tinu

ed)

Stud

yau

thor

Des

ign

(na )

Follo

w-u

p(F

U)

Dro

pout

Tre

atm

ent

grou

ps(n

b)

Patie

nts

Out

com

em

easu

res

ES

calc

ulat

ion

Hea

dach

ere

duct

ion

Phys

iolo

gica

lou

tcom

ePs

ycho

logi

cal/

beha

vior

alou

tcom

e

Age

Hea

dach

edu

ratio

ndc

95%

CI

dc95

%C

Idc

95%

CI

Gra

zzi

etal

.,20

01Pr

e-po

st(5

4)36

mon

th6

10

FUE

MG

-FB

fron

talis

PM

R(3

8)11

.8(8

–17)

2.5

Pain

,mic

Vfr

omM

SD

,t4.

49d

3.64

,5.3

40.

51d

0.17

,0.8

5—

Gra

zzi

&B

usso

ne,

1993

Pre-

post

(14)

12m

onth

not

give

nE

MG

-FB

fron

talis

(14)

23.1

(7.2

)6.

0(4

.8)

Pain

,mic

V,

Anx

from

M

SD0.

06�

0.46

,0.5

90.

30�

0.23

,0.8

40.

86d

0.28

,1.4

4

Gra

zzi

etal

.,19

92Pr

e-po

st(1

9)12

mon

thno

tgi

ven

EM

G-F

Bfr

onta

lis(1

9)18

.3(7

.1)

5.6

(5.1

)Pa

in,m

icV

,A

nxfr

omM

SD

0.76

d0.

25,1

.27

0.92

d0.

38,1

.46

0.42

�0.

04,0

.88

Gra

zzi

etal

.,19

88Pr

e-po

st(2

0)—

not

give

nE

MG

-FB

fron

talis

(20)

24(1

5–26

)no

tgi

ven

Pain

,mic

V,

Anx

,Dep

from

M

SD1.

22d

0.63

,1.8

00.

80d

0.29

,1.3

00.

31�

0.13

,0.7

6

Gra

zzi

etal

.,19

90Pr

e-po

st(1

0)12

mon

thno

tgi

ven

EM

G-F

Bfr

onta

lis(1

0)10

.3(7

–14)

2.5

Pain

,mic

Vfr

omM

SD

1.27

d0.

41,2

.12

1.33

d0.

37,2

.29

Har

t&

Cic

hans

ki,

1981

RC

T (24)

—3

EM

G-F

Bfr

onta

lis(1

0)32

.611

.6Pa

in,m

icV

,M

edfr

omp

valu

e0.

53�

0.14

,1.2

00.

53�

0.14

,1.2

00.

53�

0.14

,1.2

0

1E

MG

-FB

trap

eziu

s(1

0)33

.621

.10.

53�

0.14

,1.2

00.

53�

0.14

,1.2

00.

00�

0.62

,0.6

2

Hay

nes

etal

.,19

75R

CT (21)

—no

tgi

ven

EM

G-F

Bfr

onta

lis(8

)20

.95.

2Pa

in,m

icV

from

tst

atis

tic0.

35�

0.21

,0.9

10.

52d

0.05

,0.9

9—

—R

elax

atio

n(8

)ac

tive

vs.

cont

rol

0.35

�0.

21,0

.91

——

—H

eada

che

mon

itori

ng(5

)—

——

Hea

ton,

1979

Pre-

post

(16)

—no

tgi

ven

EM

G-F

Bfr

onta

lis

trap

eziu

s(1

6)

not

give

nPa

in,M

edfr

omM

SD

0.61

d0.

08,1

.15

—0.

13�

0.37

,0.6

2

Hof

fman

,197

9Pr

e-po

st(4

)—

not

give

nE

MG

-FB

fron

talis

(4)

(21–

50)

(1–1

1)m

icV

from

M

SD—

�1.

18�

0.18

,2.5

4—

J.C

.Hol

royd

etal

.,19

80R

CT (30)

1m

onth

2E

MG

-FB

fron

talis

(8)

18.6

5.5

Pain

,EM

G,

Anx

from

odds

ratio

s1.

46d

0.03

,2.8

60.

91�

0.02

,1.8

40.

940.

00,1

.89

1Ps

eudo

ther

apy

(9)

and

tst

atis

tics

1.67

d0.

56,2

.78

—0.

81�

0.02

,1.6

40

Hea

dach

em

onito

ring

——

0.00

�0.

61,0

.61

K.A

.Hol

royd

&A

ndra

sik,

1982

RC

T (24)

24m

onth

5E

MG

-FB

fron

talis

(8)

not

give

nPa

infr

omM

SD

0.34

�0.

47,1

.15

——

—St

ress

copi

ng(1

1)1.

68d

0.82

,2.5

4—

—K

.A.H

olro

ydet

al.,

1977

RC

T (34)

4m

onth

1E

MG

-FB

fron

talis

(11)

27.0

6.0

Pain

,mic

V,

Med

from

pva

lues

0.28

�0.

23,0

.79

0.52

�0.

14,1

.20

0.38

�0.

09,0

.85

1St

ress

copi

ng(1

0)Ps

ycho

som

atic

activ

evs

.co

ntro

l0.

41�

0.11

,0.9

4—

0.39

�0.

09,0

.87

1H

eada

che

mon

itor/

wai

ting

list

(10)

Sym

ptom

s—

——

K.A

.Hol

royd

etal

.,19

84R

CT (43)

2m

onth

5E

MG

-FB

fron

talis

,hig

hsu

cces

s(9

)

18.9

4.6

Pain

,mic

V,

Med

from

pva

lues

0.53

�0.

26,1

.31

0.58

d0.

08,1

.08

0.64

�0.

19,1

.46

—E

MG

-FB

fron

talis

,hig

hsu

cces

s,in

cont

inge

nt(9

)

18.9

4.8

Self

-eff

icac

yan

dt

stat

istic

s0.

56�

0.23

,1.3

50.

29�

0.16

,0.7

30.

65�

0.15

,1.2

7

(tab

leco

ntin

ues)

385META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

Tab

le1

(con

tinu

ed)

Stud

yau

thor

Des

ign

(na )

Follo

w-u

p(F

U)

Dro

pout

Tre

atm

ent

grou

ps(n

b)

Patie

nts

Out

com

em

easu

res

ES

calc

ulat

ion

Hea

dach

ere

duct

ion

Phys

iolo

gica

lou

tcom

ePs

ycho

logi

cal/

beha

vior

alou

tcom

e

Age

Hea

dach

edu

ratio

ndc

95%

CI

dc95

%C

Idc

95%

CI

—E

MG

-FB

fron

talis

,m

oder

ate

succ

ess

(10)

18.7

5.2

0.36

�0.

37,1

.10

0.58

d0.

08,1

.08

0.00

�0.

69,0

.69

—E

MG

-FB

fron

talis

,m

oder

ate

succ

ess,

inco

ntin

gent

(10)

18.5

5.1

0.00

�0.

73,0

.73

0.29

�0.

16,0

.73

0.00

�0.

69,0

.69

Hud

zins

ki,1

983

Pre-

post

(18)

—2

EM

G-F

Bfr

onta

lis

trap

eziu

s,

rela

xatio

n(1

6)

36(2

0–55

)no

tgi

ven

mic

Vfr

omF

stat

istic

s—

1.75

d0.

87,2

.63

Hut

chin

gs&

Rei

nkin

g,19

76R

CT (18)

—no

tgi

ven

EM

G-F

Bfr

onta

lis(6

)23

not

give

nPa

in,m

icV

from

pva

lues

0.48

�0.

12,1

.08

0.69

�0.

22,1

.60

—E

MG

-FB

fron

talis

A

T(6

)ac

tive

vs.

cont

rol

0.48

�0.

12,1

.08

0.69

�0.

22,1

.60

—A

T(6

)—

0.69

�0.

22,1

.60

—Ja

nsse

n,19

83R

CT (18)

3m

onth

not

give

nE

MG

-FB

fron

talis

(6)

34.6

(17–

59)

12.8

(1.5

–50)

Pain

,mic

Vfr

omp

valu

es0.

40�

0.19

,1.0

00.

69�

0.22

,1.6

0—

—E

MG

-FB

fron

talis

PM

R(6

)ac

tive

vs.

cont

rol

0.40

�0.

19,1

.00

0.69

�0.

22,1

.60

—H

eada

che

mon

itori

ng/

wai

ting

list

(6)

—0.

00�

0.80

,0.8

0—

Joha

nsso

n&

Ost

,19

85R

CT (12)

—no

tgi

ven

EM

G-F

Bfr

onta

lis(6

)33

.3(1

9–54

)3.

0(1

–9)

Pain

,mic

Vfr

omt

stat

istic

s0.

74�

0.13

,1.6

11.

16d

0.08

,2.2

8—

—E

MG

-FB

fron

talis

,ge

nera

lizat

ion

1.40

d0.

08,2

.72

0.00

�0.

80,0

.80

trai

ning

(6)

Kro

ener

-Her

wig

etal

.,19

98R

CT (52)

6m

onth

2PM

R(2

0)8–

14no

tgi

ven

Pain

,Med

from

M

SD0.

21�

0.30

,0.7

2—

0.33

�0.

13,0

.79

—E

MG

-FB

fron

talis

(20)

0.88

d0.

33,1

.43

—0.

57d

0.09

,1.0

6

—H

eada

che

mon

itor/

wai

ting

list

(10)

0.10

�0.

62,0

.82

—�

0.14

�0.

79,0

.51

Mat

hew

etal

.,19

87R

CT (20)

—2

EE

G-F

Gal

pha

(8)

18–4

0Pa

in,m

icV

,E

EG

from

M

SD3.

15d

1.36

,4.9

51.

10d

0.20

,2.0

01.

13d

0.22

,2.0

4

6H

eada

che

mon

itori

ng/

wai

ting

list

(4)

Anx

0.86

�0.

34,2

.05

�0.

10�

1.08

,0.8

90.

08�

0.90

,1.0

7

Nef

fet

al.,

1983

Pre-

post

(8)

—no

tgi

ven

EM

G-F

Bfr

onta

lisPM

Rnon

(8)

not

give

nPa

infr

omt

stat

istic

1.43

�0.

88,3

.74

——

Nic

hols

on&

Bla

ncha

rd,1

993

Pre-

post

(8)

—4

EM

G-F

Bfr

onta

lis

PMR

(4)

66.7

(61–

80)

37.6

(11–

65)

Pain

,Dep

,A

nxfr

omra

wda

ta1.

63�

0.20

,3.4

6—

0.08

�0.

90,1

.06

386 NESTORIUC, RIEF, AND MARTIN

Tab

le1

(con

tinu

ed)

Stud

yau

thor

Des

ign

(na )

Follo

w-u

p(F

U)

Dro

pout

Tre

atm

ent

grou

ps(n

b)

Patie

nts

Out

com

em

easu

res

ES

calc

ulat

ion

Hea

dach

ere

duct

ion

Phys

iolo

gica

lou

tcom

ePs

ycho

logi

cal/

beha

vior

alou

tcom

e

Age

Hea

dach

edu

ratio

ndc

95%

CI

dc95

%C

Idc

95%

CI

Paiv

aet

al.,

1982

RC

T (36)

1m

onth

1E

MG

-FB

fron

talis

(8)

37.6

(17–

59)

not

give

nPa

in,m

icV

from

M

SD1.

99d

0.86

,3.1

20.

51�

0.02

,1.1

3—

1E

MG

-FB

fron

talis

,fal

sefe

edba

ck(8

)

and

pva

lues

1.08

d0.

18,1

.98

1.16

d0.

52,1

.80

1D

iaze

pam

(8)

0.94

d0.

07,1

.81

1.16

d0.

52,1

.80

—1

Plac

ebo

pill

(8)

0.41

�0.

39,1

.20

——

Peck

&K

raft

,197

7Pr

e-po

st(1

8)3

mon

th2

11

FUE

MG

-FB

fron

talis

m

asse

ter

(16)

(17–

67)

(1–4

2)Pa

in,m

icV

from

pva

lues

0.42

�0.

10,0

.93

0.00

�0.

57,0

.57

Phili

ps&

Hun

ter,

1981

RC

T (16)

2m

onth

1E

MG

-FB

fron

talis

(7)

31.9

(10.

8)(1

–30)

Pain

,mic

V,

from

M

SD0.

30�

0.51

,1.1

10.

71�

0.15

,1.5

71.

14d

0.25

,2.0

3

—E

MG

-FB

fron

talis

,in

crea

se(8

)

40.5

(10.

3)(1

–30)

cont

rol,

Dep

0.11

�0.

70,0

.91

�0.

08�

0.81

,0.6

50.

17�

0.36

,0.7

0

Rei

ch,1

989

RC

T (314

)36

mon

th3

24

FUE

MG

-FB

fron

talis

tr

apez

ius

(78)

not

give

nPa

infr

omp

valu

es0.

19�

0.04

,0.4

1—

—T

EN

S(7

4)ac

tive

vs.

rela

x0.

19�

0.04

,0.4

1—

—C

ombi

natio

ntr

eatm

ent

(81)

0.00

�0.

22,0

.22

——

—R

elax

atio

n(7

8)—

——

Rok

icki

etal

.,19

97R

CT (44)

—1

EM

G-F

Bfr

onta

lis

trap

eziu

s(2

9)

19.0

4.1

Pain

,con

trol,

Med

from

M

SD0.

95d

0.50

,1.4

10.

73d

0.29

,1.1

60.

69d

0.29

,1.1

0

1H

eada

che

mon

itori

ng(1

3)18

.64.

0m

icV

,sel

f-ef

fica

cy0.

18�

0.44

,0.7

9—

0.39

�0.

18,0

.95

Scho

enen

etal

.,19

91Pr

e-po

st(3

2)no

tgi

ven

EM

G-F

Bfr

onta

lis

trap

eziu

s

rela

xatio

n(3

2)

35(1

7–48

)no

tgi

ven

Pain

,mic

V,

Anx

from

M

SD1.

18d

0.72

,1.6

40.

89d

0.48

,1.3

00.

08�

0.41

,0.5

6

Seth

iet

al.,

1981

RC

T (16)

—2

EM

G-F

Bfr

onta

lis(6

)(1

6–45

)no

tgi

ven

Pain

from

raw

data

1.58

d0.

32,2

.84

——

1Y

oga

(7)

1.82

d0.

55,3

.09

——

Steg

er&

Har

per,

1980

RC

T (20)

1E

MG

-FB

fron

talis

(9)

34(1

9–51

)4

(1–2

1)Pa

in,m

icV

,D

epfr

omt

stat

istic

s0.

69�

0.10

,1.4

81.

55d

0.50

,2.6

01.

91d

0.76

,3.0

6

2H

ome

rela

xatio

n(8

)A

nx0.

33�

0.48

,1.1

40.

59�

0.17

,1.3

6—

Not

e.A

gean

dhe

adac

hedu

ratio

nin

year

s.D

ashe

sin

dica

teda

tano

tobt

aine

dor

notr

epor

ted

inth

est

udy.

ES

�ef

fect

size

;CI

�co

nfid

ence

inte

rval

;RC

T�

rand

omiz

edco

ntro

lled

tria

l;E

MG

-FB

�el

ectr

omyo

grap

hic

feed

back

;m

icV

�m

uscl

ete

nsio

nin

mic

rovo

lt;M

ed�

reco

rdin

gof

med

icat

ion

inta

ke;

PMR

�pr

ogre

ssiv

em

uscl

ere

laxa

tion

trai

ning

;T

EM

P-FB

�pe

riph

eral

skin

tem

pera

ture

feed

back

;PM

Rnon

�PM

Rno

nres

pond

er;

Anx

�an

xiet

ym

easu

res;

Dep

�de

pres

sion

mea

sure

s;A

T�

auto

geni

cre

laxa

tion

trai

ning

;E

EG

-FB

alph

a�

elec

troe

ncep

halo

grap

hyfe

edba

ckfo

ral

pha

freq

uenc

yba

nds;

TE

NS

�tr

ansc

utan

eous

elec

tric

alne

rve

stim

ulat

ion.

aN

umbe

rof

allo

cate

dpa

tient

spe

rst

udy.

bN

umbe

rof

patie

nts

com

plet

ing

each

trea

tmen

tgr

oup.

cd

indi

cate

sth

est

anda

rdiz

edm

ean

diff

eren

ceac

cord

ing

toH

edge

s’g

(Hed

ges

&O

lkin

,19

85)

and

itspr

e-po

steq

uiva

lent

(McG

aw&

Gla

ss,

1980

).d

Indi

vidu

alef

fect

size

diff

ers

sign

ific

antly

from

zero

(i.e

.,C

Ido

esno

tin

clud

eze

ro).

eM

ultip

lepu

blic

atio

n,he

adac

heda

tano

tin

tegr

ated

.

387META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

(i.e., masseter muscle) leaving the choice of training position to theindividual patient. Therapies were conducted with treatment man-uals in 44 studies.

The 42 control conditions included 10 untreated groups, 11placebo conditions, and 21 alternative treatments, of which 15were relaxation therapies, 3 pharmacotherapies (i.e., diazepam, 2combination therapies with unspecified drugs), 2 cognitive thera-pies (i.e., stress coping), and 2 physical treatments (i.e., transcu-taneous electrical nerve stimulation). Relaxation therapies weremainly progressive muscle relaxation (PMR; k � 8), autogenictraining (k � 4), meditation (k � 1), and relaxation not furtherspecified (k � 2). Placebo treatments consisted of 7 pseudofeed-back5 groups, 2 placebo pill conditions, 1 relaxation-without-instruction group, and 1 attention CG.

Patient characteristics and attrition. The total number ofpatients across all studies was 1,532 TTH patients. The averagenumber of patients per study was 23.4 (SD � 2.3, range � 8–54)excluding 1 outlier with 314 patients. In 27 studies, treatmentgroups contained less than 10 patients. In 46 studies, age and sexof the examined patient sample were reported. Age means rangedfrom 10.3 to 66.7 years; the average age of all patients was 35.9years (SD � 0.62); 71.7% (range � 42.9%–100%) of the patientswere female and 28.3% were male. In 34 studies, the history ofheadache problems was reported. The average number of yearswith TTH ranged from 1.2 to 42.4, averaging 13.9 (SE � 0.7).Two studies examined geriatric samples (grand age M � 66.1,range � 65.0–66.7, SE � 1.6; grand M of headache duration �39.2, range � 53–42.4, SE � 2.0), and 9 examined TTH inchildren and adolescents (grand age M � 14.2, range � 11–19,SE � 1.7; grand M of headache duration � 3.5, range � 2.5–5.6,SE � 1.7).

Diagnoses were made according to a standardized diagnosticsystem in 29 of the integrated studies (i.e., Ad Hoc Committee onthe Classification of Headache (1962), k � 17; Headache Classi-fication Committee of the International Headache Society (IHS;1988), k � 8, International Classification of Diseases (ICD)-9 k �1, other k � 2). In 16 studies, characteristic features of TTH wereapplied in screening interviews, while in 7 studies, diagnoses frompreceding medical examinations were accepted. CTTH patientswere investigated in 4 studies and ETTH in 2 studies, while mostof the studies (k � 47) made no explicit distinction between thetwo TTH forms. In 4 studies, the additional criterion of TTHassociated with pericranial tenderness was applied for patientinclusion. Neurological examinations to exclude organic headachebases were reported in 36 studies.

A total of 99 dropouts after treatment assignment were reportedacross 33 studies representing a completion rate of 91.3% atpost-treatment. Similar completion rates resulted for biofeedback(90.2%) and CGs (92.4%); differences were insignificant (riskratio: 1.29; 95% confidence interval [CI] � 0.89, 1.89). Thecompletion rate at follow-up was 84.6%. Three studies reported astatistical comparison of completers and dropouts; in 2 studies,reasons for discontinuation were given, while none of the inte-grated studies reported ITT analyses. In 20 of the integratedstudies, no information about treatment attrition was given.

Validity. According to our 12-point Validity Scale, the valid-ity of all integrated studies ranged from 1 to 11, averaging 5.6(SD � 2.1). Study validity was uncorrelated with headache reduc-tion effect sizes, r(58) � .15, p � .25; with effect sizes for

physiological outcome, r(35) � –.27, p � .12; and psychologicaloutcome, r(28) � .28, p � .15.

General Efficacy of Biofeedback

The primary outcome variable headache pain included fre-quency, duration, and intensity of TTH, consistently measuredwith a structured headache diary (k � 47).6 The headache diarieswere employed for an average of 20.1 days at baseline (SD � 8.5),18.2 at post-treatment (SD � 8.5), and 12.2 at follow-up (SD �7.6). Effect size calculation yielded 58 independent effect mea-sures for headache relief from pre- to post-treatment. The effectsizes ranged from d � 0.06 to d � 1.99 with one outlier d � 4.49(standardized residual 3 and largest impact on Q statistic)resulting from a study investigating biofeedback for children andadolescents with TTH (Grazzi et al., 2001). Effect size integrationresulted in a significant large average effect size (N � 826; d �0.82; 95% CI � 0.67, 0.97). The assumption of homogeneity heldwithin the REM (Q � 62.0, p � .30, 1 � � � .97), although aconsiderable amount of unexplained variance (�2 � .20) waspresent. The exclusion of the described outlier substantially dimin-ished this unexplained variance component (k � 57, N � 788, Q �47.8, pQ � .78, �2 � .06). The overall effect size, though numer-ically slightly reduced, persisted to be of medium-to-large magni-tude (d� � 0.73; 95% CI � 0.61, 0.84). The narrow CI confirms therobustness of this result. All further analyses (except those exam-ining efficacy for children and adolescents) were conducted underexclusion of this outlier.

For the controlled treatment comparisons, weighted averageeffect sizes, CIs, and homogeneity statistics are shown in Table 2.The significant medium-to-large effect size from pre- to post-treatment was replicated in comparison to untreated CGs. A sig-nificant medium effect size favoring biofeedback was found incomparison with that of placebo CGs and a significant small effectsize for biofeedback in comparison with that of relaxation CGs.For all reported comparisons, effect sizes were homogeneousaccording to the REM, although the statistical power to detectheterogeneity was rather low partly given the unexplained variancecomponents ranging from 6% to 14%. The comparisons ofbiofeedback with pharmacotherapy and physical and cognitivetherapies were insignificant and consisted of too few studies toprovide reliable conclusions.7

Sensitivity analyses. In a sensitivity analysis, we separatelyintegrated and compared effect sizes based on varying formulas.The results were robust for the comparison of Hedges’ d fromrandomized controlled trials (RCTs; untreated CGs) with Hedges’dpre-post from pre-post trials (Hedges’ d: k � 9; N � 180; d� � 0.81;95% CI � 0.46, 1.16; Q � 8.9; pQ � .35; �2 � .14; Hedges’dpre-post: k � 18; N � 297; d� � 1.00; 95% CI � 0.63, 1.36; Q �

5 Patients in pseudofeedback conditions were equally trained to influ-ence physiological parameters, but in the opposite direction (e.g., increaseof muscle tension) or under false feedback.

6 Of the remaining studies, five used headache questionnaires and clin-ical ratings, and one assessed only physiological outcome.

7 Note, however, that two of the three integrated pharmacotherapystudies revealed large effect sizes in favor of biofeedback, while the thirdfound no difference between biofeedback and a combination treatmentincluding pharmacotherapy.

388 NESTORIUC, RIEF, AND MARTIN

24.5; pQ � .11; �2 � .44), as well as for the comparison of effectsizes computed from means and standard deviations with thosecomputed from other test statistics (means and standard deviations:k � 29; N � 400; d� � 0.88; 95% CI � 0.73, 1.03; Q � 26.7; pQ �.54; �2 � .04; other test statistics: t, F statistics, odds: k � 13; N �164; d� � 0.74; 95% CI � 0.54, 0.95; Q � 10.9; pQ � .54; �2 �0). A significantly differing average effect size resulted for effectsizes calculated from significance bounds (significance bounds:k � 15; N � 224; d� � 0.33; 95% CI � 0.19, 0.48; Q � 4.8; pQ �.99; �2 � 0), indicating that this method of effects size reconstruc-tion leads to particularly conservative estimates. A subsequentexclusion of all effects sizes reconstructed from significancebounds resulted in a numerically but not statistically higher globaleffect size for biofeedback in TTH and a diminished amount ofunexplained variance in the pre-post effects (k � 42; N � 564; d�

� 0.85; 95% CI � 0.73, 0.97; Q � 39.3; pQ � .55; �2 � .03). Insum, these sensitivity analyses showed that our main effect size forthe study sample (d� � 0.73) might be a conservative but reliableestimate.

Stability of Treatment Effects

In 18 studies, the stability of treatment effects was examinedwith follow-up periods ranging from 3 to 60 months, averaging14.6 months (SD � 15.8). We calculated 20 independent follow-upeffect sizes (N � 309). They ranged from d � 0.15 to d � 1.27with one outlier d � 5.86, forming a unimodal and right-skeweddistribution. The outlier resulted again from the exceptionally largeeffects found by Grazzi and colleagues (2001). Aggregation in theREM with exclusion of the outlier resulted in a significant effectsize of medium-to-large magnitude (k � 19; N � 281; d� � 0.64;95% CI � 0.43, 0.86). Heterogeneity within the REM was given(Q � 14.5, pQ � .70, �2 � .08).

ITT analysis. To further investigate the stability of treatmenteffects, we calculated mean weighted effect sizes based on ITTanalyses. The ITT analysis for post-treatment data resulted in asignificant medium effect size (k � 57; NITT � 851; d� ITT � 0.62;95% CI � 0.53, 0.72), based on a homogeneous distribution ofsingle effects (Q � 49.0, pQ � .73, �2 � .03). Thus, the meantreatment effect of biofeedback decreased from d� � 0.73 to d� ITT �

0.62, when dropouts were considered as nonresponders. Evalua-tion of the follow-up effect sizes in ITT analysis resulted in asignificant small-to-medium average effect (k � 19; NITT � 355;d� ITT � 0.48; 95% CI � 0.29, 0.67), based on a homogeneousdistribution of single effect (Q � 16.8, pQ � .54, �2 � .07).Accordingly, ITT analysis led to a reduction of the averagebiofeedback effect at follow-up, while significant medium-to-small treatment improvements persisted.

Effects of Biofeedback on Different Types of OutcomeVariables

Headache frequency, intensity, and the headache index werereduced with large average effect sizes (see Table 3). Duration ofheadache episodes was reduced with a small-to-medium effectsize. Physiological outcome (i.e., muscle tension in microvolt) wasassessed as changes in muscle tension from baseline to post-treatment (k � 31). In 16 studies, within-session changes werereported. Muscle tension was reduced with a significant medium(CI including large effects) effect size within treatment sessionsand with a significant small (CI including medium effects) effectsize across sessions. Over the course of all biofeedback sessions,reductions in headache index and frequency were significantlylarger than reduction in muscle tension, indicated by the nonover-lapping CIs for the corresponding average effect sizes. Psycholog-ical/behavioral outcome was assessed with psychological symp-tom ratings (i.e., anxiety, depression, psychosomatic symptoms),cognitive variables (i.e., self-efficacy, locus of control), and be-havioral measures (i.e., use of pain medication). Self-efficacy,anxiety, and depression all yielded significant medium effect sizes,with corresponding CIs ranging from small to large effect sizes.Medication intake was reduced with a significant small-to-mediumeffect size. The average effect size for headache index was signif-icantly higher than the effect size for medication intake. The othersymptom categories did not show any significant differences. Forall outcome variables, effect sizes were homogeneous according tothe REM; unexplained variance components ranged from 3% to32%, while the power to detect heterogeneity was rather low insome of the sub-analyses.

Table 2Mean Weighted Effect Sizes for Headache Relief as a Function of Treatment Comparison

Comparison k N

Random effect model

d� 95% CI Q( p) �2 1 � �

BFB pre- vs. post-treatment 57 788 0.73a 0.61, 0.84 47.8 (.78) 0.06 .39BFB vs. no-treatment control 9 180 0.81a 0.46, 1.16 8.9 (.35) 0.14 .32BFB vs. placebo control 9 165 0.50a 0.27, 0.73 5.7 (.68) 0.00 .34BFB vs. relaxation 15 446 0.20a 0.09, 0.32 10.2 (.75) 0.00 .48BFB vs. pharmacotherapy 3 197 0.91 �0.12, 1.94 1.4 (.49) 0.70 .59BFB vs. physical therapy 2 165 0.00 �0.15, 0.16 0.2 (.88) 0.00 .05BFB vs. cognitive therapy 2 40 �0.45 �1.00, 0.10 1.0 (.32) 0.05 .22

Note. k � number of effect sizes; N � number of patients of tension-type headache; d� � weighted mean effect size; 95% CI � confidence interval ford� ; Q � homogeneity statistic for d� calculated via random effect model; �2 � random effects variance; 1 � � � power of the homogeneity test (post hoc�2 approximation; Hedges & Pigott, 2001); BFB � biofeedback.a Effect size differs significantly from zero (i.e., CI does not include zero).

389META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

Impact of Treatment Format and Patient Characteristics

Effect sizes (95% CI), homogeneity statistics, contrast weights,p values, and explained systematic variance within the moderatorlevels are shown in Table 4. Three of the a priori specifiedcontrasts were significant. EMG-FB yielded a medium averageeffect size and EMG-FB in combination with relaxation a largeaverage effect size. CIs for both modalities included medium-to-large effects. Other feedback modalities (i.e., TEMP-FB, EEG-FB,and GSR-FB) yielded a large average effect with a broad CI. Theobserved differences between modalities were significant (QB �37.9, p � .01; overall model fit was given: QW � 48.2, p � .70,�2

res � .04) and accounted for 26% of the systematic variancefrom the REM. The planned contrast was significant with a meaneffect size difference of 0.32, indicating that EMG-FB in combi-nation with relaxation was the most effective biofeedback modal-ity. Significantly differing average effect sizes resulted for thethree age groups (QB � 102.5, p � .01; overall model fit wasgiven: QW � 60.7, p � .28, �2

res � .001). Biofeedback was moreeffective in children and adolescents than in adults. The influenceof duration of TTH was further explored within the sub-samples ofadult and juvenile patients separately (the geriatric sample beingtoo small for sub-analyses). Contrary to our hypothesis, highereffect sizes resulted for adult patients with a longer history of TTH(QB � 41.5, p � .01; overall model fit was given: QW � 34.1, p �.80, �2

res � .004). Insignificant contrasts resulted for differenttraining sites within the EMG-FB treatments; study setting (i.e.,purely outpatient vs. including home training); treatment duration;diagnostic distinctions between CTTH, ETTH, and TTH withpericranial tenderness; and levels of chronicity in the sample ofchildren and adolescents.

Publication Bias

The distribution of effect sizes shown in Figure 2a assumes thetypical shape of a funnel plot, with one outlier effect size from anespecially large study (Reich, 1989). This study reports a small

effect (d � 0.19) for EMG-FB in comparison with that of relax-ation. Deviations from the normal distribution are minimal asdemonstrated with the normal quantile plot in Figure 2b (Wang &Bushmann, 1998).

In addition to this graphical method, we examined publicationbias by calculating fail-safe Ns for the critical effect sizes d�crit �0.01 and d�crit � 0.31. Four-thousand eighty-seven unpublishedstudies with zero effects would be necessary to reduce the ob-served average effect of d� � 0.73 to zero, while 76 would berequired to reduce it to a small average effect size. In sum,publication bias seems rather unlikely.

Discussion

General Efficacy and Specificity

The present meta-analysis provides an up-to-date evaluation ofthe efficacy of biofeedback as a behavioral treatment option forTTH. A robust treatment effect of medium-to-large magnitude (d�

� 0.73) was established through the integration of pre-post effectsizes from 53 studies. This finding was replicated in between-group comparisons with untreated CGs, resulting in a large aver-age effect size favoring biofeedback. These effects are clinicallymeaningful as they demonstrate symptom improvements of almostone standard deviation for patients suffering from TTH for 14years on average. With an average of 11 sessions, biofeedbacktreatments were altogether short and economical. Furthermore, thetreatment was generally very well accepted, as shown in the lowdropout rates.

Treatment specificity was investigated in between-groupcomparisons. Superior clinical results emerged for biofeedbackgroups compared with placebo CGs and relaxation therapies.Efficacy comparisons of biofeedback to pharmacotherapy,physical therapy, and cognitive therapy included only very fewstudies. For the interpretation of these comparisons, further

Table 3Mean Weighted Pre-Post Effect Sizes of Biofeedback as a Function of Symptom Category

Symptom category k N

Random effect model

d� 95% CI Q( p) �2 1 � �

Headache reductionFrequency 28 469 0.82a 0.57, 1.07 36.6 (.10) 0.32 0.89Intensity 27 406 0.69a 0.51, 0.86 26.9 (.42) 0.09 0.48Duration 13 123 0.46a 0.27, 0.65 9.5 (.66) 0.00 0.48Headache index 30 434 0.82a 0.58, 1.06 37.7 (.13) 0.30 0.97

Physiological outcome (EMG amplitudes)micV reduction (within session) 16 184 0.62a 0.46, 0.78 5.5 (.98) 0.00 0.24micV reduction (between sessions) 31 417 0.40a 0.26, 0.55 31.2 (.41) 0.07 0.53

Psychological/behavioral outcomeMedication index 18 249 0.42a 0.26, 0.57 16.4 (.50) 0.03 0.18Self-efficacy 5 65 0.63a 0.27, 0.99 4.2 (.37) 0.05 0.12Anxiety 9 147 0.53a 0.27, 0.79 9.4 (.31) 0.06 0.21Depression 5 61 0.54a 0.04, 1.03 4.6 (.33) 0.26 0.27

Note. k � number of effect sizes; N � number of migraine patients; d� � weighted mean effect size; 95% CI � confidence interval for d� ; Q � homogeneitystatistic for d� ; �2 � random effects variance, 1 � � � power of the homogeneity test (post hoc �2 approximation; Hedges & Pigott, 2001).a Effect size differs significantly from zero, (i.e., CI does not include zero).

390 NESTORIUC, RIEF, AND MARTIN

studies investigating relative or additional effects of biofeed-back are needed.

Results of previous reviews have consistently shown biofeed-back to be more effective than headache monitoring (Blanchard etal., 1980; Bogaards & ter Kuile, 1994; J. C. Holroyd & Penzien,1986; McCrory et al., 2001) but were inconclusive about thespecificity of biofeedback. Blanchard and colleagues (Blanchard etal., 1980) pointed out nearly 30 years ago that there were too fewstudies to draw conclusions about the equivalence of alternativeheadache treatments. J. C. Holroyd and Penzien (1986) reportedsignificant differences between behavioral treatments and placeboconditions but no differences within the active treatments, in spiteof a tendency for higher improvement rates in combined relaxationand biofeedback than in relaxation alone. However, none of thestudies published in the past two decades provided consistentsupport for this trend, with some reporting superior results for

relaxation, some favoring biofeedback, and others finding no dif-ference between the two. Likewise, the two most recent meta-analyses (Bogaards & ter Kuile, 1994; McCrory et al., 2001) couldnot shed light on the controversy of whether or not “the machinesare really necessary” (Silver & Blanchard, 1978, p. 217). Thesediscrepancies are, in fact, most easily explained by a lack ofstatistical power. Especially for the comparisons with alternativebona fide treatments, statistical power is crucial (Houle, Penzien,& Houle, 2005). Most of the initial comparisons of biofeedbackand relaxation in the primary studies were underpowered. A dif-ference between the current and previous meta-analyses was theextensive inclusion of studies to achieve sufficient power. Theintegration of data from over 400 patients demonstrated a consis-tent and significant advantage of biofeedback over relaxation.Consequently, the additional technical effort inherent in biofeed-back seems justified for the efficacious treatment of TTH.

Table 4Analysis of Variance in Headache Reduction Effect Sizes as a Function of Moderator Variable Levels

Moderator levelsContrastweights

Random effect model

k N d� 95% CI QWi C ( p) �explained2

Feedback modalityEMG-FB �1 44 626 0.66* 0.54, 0.78 40.9EMG-FB RT 1 9 124 0.98* 0.69, 1.27 6.2Other 0 4 38 0.94* 0.42, 1.45 1.1 0.32* (.04) 26%

EMG training siteBifrontal �1 30 398 0.81* 0.60, 0.96 28.4Neck 1/2 3 24 1.10* 0.52, 1.88 0.2Multiple muscles 1/2 5 104 1.01* 0.67, 1.38 2.6 0.25 (.18) 17%

Home trainingIncluded 1 35 509 0.74* 0.60, 0.89 21.5Not included �1 22 279 0.70* 0.52, 0.88 25.8 0.04 (.71) 0%

Treatment duration2.5 to 5 hours �1/2 16 193 0.67* 0.47, 0.88 8.35 to 7.5 hours 1/2 19 258 0.68* 0.50, 0.87 18.07.5 to 10 hours 1/2 15 182 0.89* 0.68, 1.11 15.3More than 10 hours �1/2 5 45 0.66* 0.26, 1.06 1.2Not reported 0 2 110 0.55* 0.18, 0.92 6.5 0.12 (.37) 27%

DiagnosisTTH not specified 0 32 397 0.80* 0.65, 0.96 22.2Chronic TTH �1 4 79 0.99* 0.62, 1.35 2.0Episodic TTH 1/2 1 20 0.84* 0.17, 1.50TTH with pericranial tenderness 1/2 5 68 1.10* 0.68, 1.51 7.9 �0.04 (.87) 17%

AgeChildren and adolescents 1 10 178 1.19* 0.84, 1.53 35.2Adults �1 46 636 0.73* 0.56, 0.89 25.0Geriatric 0 2 12 1.00 �0.02, 2.02 0.5 0.46* (.02) 12%

Duration of TTH (in adults)3.5 to 9 years 1 11 127 0.44* 0.20, 0.69 4.210 to 15 years �1/2 11 136 0.65* 0.39, 0.91 5.2More than 15 years �1/2 4 49 1.07* 0.63, 1.52 3.0Not reported 0 20 324 0.80* 0.61, 0.99 21.8 �0.42* (.02) 23%

Duration of TTH (children)Up to and including 3 years 1 4 73 1.95* 0.97, 2.94 8.93 to 6 years �1 5 85 0.80* 0.56, 1.39 0.7Not reported 0 1 20 0.88* 0.33, 1.43 1.15 (.08) 0%

Note. k � number of independent effect sizes; N � number of tension-type headache (TTH) patients; d� � weighted mean effect size with confidenceinterval (95% CI); QWi � homogeneity within each group; C � contrast with p value and percent reduction of the variance component (�2) through themoderating variable; EMG-FB � electromyographic feedback; EMG-FB RT � EMG-FB in combination with relaxation; other � peripheral skintemperature feedback (TEMP-FB), electroencephalography feedback for alpha frequency bands (EEG-FB), and/or galvanic skin response feedback(GSR-FB).* p � .05.

391META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

Maintenance of Treatment Effects

The established effects were shown to persist up to several yearsafter treatment; specifically, they remained stable over an averagefollow-up interval of 15 months. Though Blanchard et al. (1980)were already able to integrate data from six follow-up studiesindicating stability, and though recent qualitative reviews (Pen-zien, Rains, & Andrasik, 2002) have suggested that biofeedbacktreatment gains endure over time, the present findings constitutethe most comprehensive meta-analytical confirmation of the long-term efficacy of biofeedback for TTH. The reliability of theseencouraging findings was corroborated by ITT analysis, showingthat the established direct and follow-up effects persisted evenwhen dropouts were considered as nonresponders.

Treatment Moderators

Efficacy on multidimensional outcome measures. It is gener-ally agreed that the outcome assessment in psychological headachetrials needs to be multidimensional and should integrate somatic,

psychosocial, and behavioral information (K. A. Holroyd, Mali-noski, Davis, & Lipchik, 1999; Turk & Okifuji, 1999). In ourmeta-analysis, treatment success was confirmed not only for theprimary outcome variable headache but also for associated symp-toms of anxiety and depression, self-efficacy, physiological pa-rameters, as well as medication consumption. Biofeedback yieldedthe largest treatment effects for the reduction of headache fre-quency, which is the recommended primary outcome variable forheadache research (Andrasik, Lipchik, McCrory, & Wittrock,2005; IHS Clinical Trials Subcommittee, 2000). It can be con-cluded from our data that the reduction of headache throughbiofeedback is associated with improvements in comorbid psycho-logical symptoms. With regard to the question whether physiolog-ical learning occurs during biofeedback treatments, the data pre-sented provide some further insight. First, it is documented thataverage levels of muscle tension were in fact reduced throughbiofeedback. Second, these changes tended to be more pronouncedand to occur more reliably during each session as compared withover the courses of whole treatments. These results are relevant forthe debate of physiological versus psychological learning (seeRokicki et al., 1997) because this meta-analysis is the first toevaluate and confirm physiological changes during EMG-FB. Ona descriptive level, the present findings suggest that the ability toreduce dysfunctional muscle tension as well as an enhanced self-efficacy and reduced levels of anxiety and depression occur to-gether with the reduction of headache pain. Hypothesized causalrelationships among different outcome levels cannot be analyzedwith these data. Thus, it seems promising for future studies toincorporate repeated measures of potential mediators to addressthe question whether physiological or cognitive changes (i.e.,self-efficacy, K. A. Holroyd et al., 1984) are crucial for thetreatment success of biofeedback.

Influence of treatment and patient characteristics. The find-ings that EMG-FB in combination with relaxation proved to bemore effective than EMG-FB alone, while biofeedback in generalwas superior to relaxation, leads to the conclusion that the specificelements inherent in biofeedback add to the general efficacy ofrelaxation. Consequently, the therapeutic use of biofeedback notonly to control physiological parameters assumed to underlie head-ache but also to enhance the positive effects of general overallrelaxation is confirmed.

Our finding that behavioral headache treatments are particularlyefficacious in children and adolescents is in line with other meta-analyses of this age group (Eccleston, Morley, Williams, Yorke, &Mastroyannoploulou, 2002; Hermann, Kim, & Blanchard, 1995;Trautmann et al., 2006), emphasizing the prophylactic potential ofbiofeedback for juvenile headache patients. Within the adult pa-tient sample, duration of headache was a significant moderator oftreatment effects, as patients with a longer history of headacheexperienced higher treatment success. Given the generally highlevel of chronicity in the present patient sample, we conclude thatbiofeedback is a promising treatment alternative even for patientswho have suffered from TTH for a substantial part of their lives(see also efficacy results for relaxation nonresponders, Blanchardet al., 1982a, 1982b).

Influence of methodological factors. Following our ratherliberal inclusion criteria, we integrated a total of 53 studies withvarying methodological quality. On the one hand, this strategyallowed us to draw powerful conclusions regarding the efficacy,

Figure 2. Graphical analysis of publication bias. a: Funnel plot. Effectsizes d (independent pre-post effect sizes estimating the symptom reduc-tion on all headache variables through biofeedback therapy) displayed as afunction of the sample size of each study (k � 57). b: Normal quantile plot.Effect sizes d displayed against the expected quantiles of the normaldistribution.

392 NESTORIUC, RIEF, AND MARTIN

specificity, and maintenance of biofeedback, based on data frommore than 1,500 TTH patients. On the other hand, it led to the needto integrate effect sizes based on different reconstruction methodsand may have increased heterogeneity of the effect sizes. Weaccounted for this heterogeneity with the application of the REMthroughout the whole analysis. While results from fixed effectsmodels are valid only for the sample of integrated studies, infer-ences from REM have further generalizability and apply to treat-ment efficacy in general, at the cost of less powerful tests (Egger,Smith, & Phillips, 1997; Hedges & Vevea, 1998). The REM wasthe most appropriate generalization model for the integrated stud-ies, which were conducted in the past 3 decades, with patients frommany different countries and cultures. Potential biasing of thepresent results through varying effect size calculation formulaswas ruled out in sensitivity analyses. However, our method ofeffect size reconstruction from significance levels led to smallereffect estimates than the calculation from other test statistics.Similar findings have already been described by McCrory andcolleagues (McCrory et al., 2001), who reported considerablysmaller treatment gains (38% vs. 70%) in studies with insufficientreporting of statistics. To achieve more powerful moderator anal-ysis, we decided to integrate the conservative effect estimates,thereby potentially underestimating the overall effect size ofbiofeedback for TTH. Potential biasing influences of systematicrelations between the methodological quality of the integratedstudies measured with our validity rating and study findings, aswell as selective publishing of significant results (i.e., publicationbias), were investigated and proved unlikely.

Limitations and Requirements for Future Studies

The validity levels of the integrated studies were generallymedium. A number of very important methodological suggestionsand recommendations concerning behavioral headache researchhave recently been put forward (Penzien et al., 2005; Rains,Penzien, McCrory, & Gray, 2005), and future studies would un-doubtedly benefit from adopting these standards. The averagevalidity scores from our structured rating were smaller comparedwith prior results in the field of migraine (Nestoriuc & Martin,2007). Specifically, low sample sizes, resulting in the power prob-lems discussed above, failure to describe basic treatment andpatient characteristics, as well as the use of unstructured diagnosticsystems in some of the integrated studies had a negative impact onvalidity levels, thereby imposing an upper boundary for the valid-ity of whole analysis. The fact that the integration of only RCTsled to equal average effects size estimates underlines the validityof our findings.

In reviewing the selected studies, we identified several strengthsand weaknesses. Diagnostic distinctions between ETTH and theclinically more meaningful CTTH were only seldom made. Henceit is not yet possible to draw reliable conclusions regarding theequal effectiveness of EMG-FB for both headache conditions.Further studies directly comparing the efficacy of EMG-FB forepisodic and chronic TTH are needed. Regarding the differentEMG-FB protocols, it seems remarkable that protocols using mus-cles of the neck or multiple training sites for feedback (see, forexample, Bischoff & Dahlinger, 1993) are still rarely investigated.Therefore comparisons with the standard bifrontal training cannotbe interpreted reliably. We conclude that individualized training

protocols (e.g., adapted to patients’ perceived muscular abnormal-ities) may appear promising but need further investigation. Withregard to the measurement of headache impact, outcome wasconsistently recorded with structured headache diaries by usingmultidimensional pain measures and valid baseline phases. Manystudies assessed medication intake as secondary outcome. Theassessment of psychological symptoms and potential comorbidi-ties still seems underrepresented. Specifically, the assessment offunctional and behavioral outcome (i.e., lost work days, healthservice use, general activity level, social and role functioning) ismissing in the primary studies. Thus, in future studies we recom-mend to incorporate direct measures of the functional, social, andsocioeconomic “burden of headache.”

Suggestion for Established Level of Efficacy

The present meta-analysis provides scientifically sound evi-dence for the efficacy of biofeedback. The results have specificvalidity for children and adolescents, for adults, as well as forgeriatric TTH patients. Superior effect sizes for biofeedback incomparison with psychological placebo and relaxation therapieswere demonstrated. These effects were established with treatmentmanuals (in 83% of the integrated studies), under specification ofpatient characteristics (87%), and by multiple different researchteams. Thus, according to the American Psychological Associationcriteria for empirically supported therapies (Chambless & Hollon,1998), biofeedback constitutes a well-established treatment forTTH.

References*References marked with an asterisk indicate studies included in the

meta-analysis.

Ad Hoc Committee on the Classification of Headache. (1962). Classifica-tion of headache. Journal of the American Medical Association, 179,117–118.

*Andrasik, F., & Holroyd, K. A. (1980). A test of specific and nonspecificeffects in the biofeedback treatment of tension headache. Journal ofConsulting and Clinical Psychology, 48(5), 575–586.

*Andrasik, F., & Holroyd, K. A. (1983). Specific and nonspecific effectsin the biofeedback treatment of tension headache: 3-year follow-up.Journal of Consulting and Clinical Psychology, 51(4), 634–636.

Andrasik, F., Lipchik, G. L., McCrory, D., & Wittrock, D. A. (2005).Outcome measures in behavioral headache research: Headache param-eters and psychosocial outcome. Headache, 45, 429–437.

*Arena, J. G., Bruno, G. M., Hannah, S. L., & Meador, K. J. (1995). Acomparison of frontal electromyographic biofeedback training, trapeziuselectromyographic biofeedback training, and progressive muscle relax-ation therapy in the treatment of tension headache. Headache, 35(7),411–419.

*Arena, J. G., Hannah, S. L., Bruno, G. M., & Meador, K. J. (1991).Electromyographic biofeedback training for tension headache in theelderly: A prospective study. Biofeedback and Self Regulation, 16(4),379–390.

*Arndorfer, R. E., & Allen, K. D. (2001). Extending the efficacy of athermal biofeedback treatment package to the management of tension-type headaches in children. Headache, 41, 183–192.

*Billings, R. F., Thomas, M. R., Rapp, M. S., Reyes, E., & Leith, M.(1984). Differential efficacy of biofeedback in headache. Headache,24(4), 211–215.

*Bischoff, C., & Dahlinger, E. (1993). Behandlung von spannungskopf-

393META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

schmerz mit EMG-biofeedback im feld: Ein effizienzvergleich mit tra-ditionellem EMG-biofeedback [Treatment of tension-type headachewith EMG-feedback in the field: Comparative efficacy to traditionalEMG-feedback.] Verhaltenstherapie, 3(4), 286–295.

Blanchard, E. B., Andrasik, F., Ahles, T. A., Teders, S. J., & O’Keefe, D.(1980). Migraine and tension headache: A meta-analytic review. Behav-ior Therapy, 11(5), 613–631.

*Blanchard, E. B., Andrasik, F., Neff, D. F., Teders, S. J., Pallmeyer, T. P.,Arena, J. G., et al. (1982a). Biofeedback and relaxation training withthree kinds of headache: Treatment effects and their prediction. Journalof Consulting and Clinical Psychology, 50(4), 562–575.

*Blanchard, E. B., Andrasik, F., Neff, D. F., Teders, S. J., Pallmeyer, T. P.,Arena, J. G., et al. (1982b). Sequential comparisons of relaxation train-ing and biofeedback in the treatment of three kinds of chronic headacheor, the machines may be necessary some of the time. Behaviour Re-search and Therapy, 20(5), 469–481.

*Blanchard, E. B., Appelbaum, K. A., Guarnieri, P., & Morrill, B. (1987).Five year prospective follow-up on the treatment of chronic headachewith biofeedback and/or relaxation. Headache, 27(10), 580–583.

Bogaards, M. C., & ter Kuile, M. M. (1994). Treatment of recurrent tensionheadache: A meta-analytic review. The Clinical Journal of Pain, 10,174–190.

*Borgeat, F., Elie, R., & Castonguay, L. G. (1991). Muscular response tothe therapist and symptomatic improvement during biofeedback fortension headache. Biofeedback and Self Regulation, 16(2), 147–155.

*Borgeat, F., Elie, R., & Larouche, L. M. (1985). Pain response to volun-tary muscle tension increases and biofeedback efficacy in tension head-ache. Headache, 25(7), 387–391.

*Borgeat, F., Hade, B., Larouche, L. M., & Gauthier, B. (1984). Psycho-physiological effects of therapist’s active presence during biofeedbackas a simple psychotherapeutic situation. Psychiatric Journal of theUniversity of Ottawa, 9(3), 132–137.

*Bruhn, P., Olesen, J., & Melgaard, B. (1979). Controlled trial of EMGfeedback in muscle contraction headache. Annals of Neurology, 6(1),34–36.

*Budzynski, T. H., Stoyva, J. M., Adler, C. S., & Mullaney, D. J. (1973).EMG biofeedback and tension headache: A controlled outcome study.Psychosomatic Medicine, 35(6), 484–496.

*Bussone, G., Grazzi, L., D’Amico, D., Leone, M., & Andrasik, F. (1998).Biofeedback-assisted relaxation training for young adolescents with tension-type headache: A controlled study. Cephalalgia, 18(7), 463–467.

Castillo, J., Munoz, P., Guitera, V., & Pascual, J. (1999). Epidemiology ofchronic daily headache in the general population. Headache, 39, 190–196.

Chambless, D. L., & Hollon, S. D. (1998). Defining empirically supportedtherapies. Journal of Consulting and Clinical Psychology, 66, 7–18.

*Chesney, M. A., & Shelton, J. L. (1976). A comparison of musclerelaxation and electromyogram biofeedback treatments for muscle con-traction headache. Journal of Behavior Therapy and Experimental Psy-chiatry, 7(3), 221–225.

*Collet, L., Cottraux, J., & Juenet, C. (1986). GSR feedback and Schultzrelaxation in tension headaches: A comparative study. Pain, 25(2),205–213.

*Cott, A., Goldmann, J. A., Pavloski, R. P., Kirschberg, G. J., & Fabich, M.(1981). The long-term therapeutic significance of the addition of elec-tromyographic biofeedback to relaxation training in the treatment oftension headaches. Behavior Therapy, 12(4), 556–559.

*Cox, D. J., Freundlich, A., & Meyer, R. G. (1975). Differential effective-ness of electromyographic feedback, verbal relaxation instructions, andplacebo medication with tension headaches. Journal of Consulting andClinical Psychology, 43(6), 892–898.

*Daly, E. J., Donn, P. A., Galliher, M. J., & Zimmerman, J. S. (1983).Biofeedback applications to migraine and tension headaches: A double-blinded outcome study. Biofeedback and Self Regulation, 8(1), 135–152.

Eccleston, C., Morley, S., Williams, A., Yorke, L., & Mastroyannoploulou,K. (2002). Systematic review of psychological therapy for chronic painin children and adolescents, with a subset meta-analysis of pain. Pain,99, 157–165.

Egger, M., Smith, G. D., & Phillips, A. N. (1997). Meta-analysis: Princi-ples and procedures. British Medical Journal, 315, 1533–1537.

*Epstein, L. H., & Abel, G. G. (1977). An analysis of biofeedback trainingeffects for tension headache patients. Behavior Therapy, 8(1), 37–47.

*Gada, M. T. (1984). A comparative study of efficacy of EMG bio-feedback and progressive muscular relaxation in tension headache. In-dian Journal of Psychiatry, 26(2), 121–127.

Gibbons, R. D., Hedecker, D. R., & Davis, J. M. (1993). Estimation ofeffect size from a series of experiments involving paired comparisons.Journal of Educational Statistics, 18, 271–279.

Gleser, L. J., & Olkin, I. (1994). Stochastically dependent effect sizes. InH. Cooper & L. V. Hedges (Eds.), The handbook of research synthesis(pp. 229–255). New York: Russel Sage.

Granella, F., Farina, S., Malferrari, G., & Manzoni, G. C. (1987). Drugabuse in chronic headache: A clinico-epidemiologic study. Cephalalgia,7, 15–19.

*Gray, C. L., Lyle, R. C., McGuire, R. J., & Peck, D. F. (1980). Electrodeplacement, EMG feedback, and relaxation for tension headaches. Be-haviour Research and Therapy, 18(1), 19–23.

*Grazzi, L., Andrasik, F., D’Amico, D., Leone, M., Moschiano, F., &Bussone, G. (2001). Electromyographic biofeedback-assisted relaxationtraining in juvenile episodic tension-type headache: Clinical outcome atthree-year follow-up. Cephalalgia, 21(8), 798–803.

*Grazzi, L., & Bussone, G. (1993). Effect of biofeedback treatment onsympathetic function in common migraine and tension-type headache.Cephalalgia, 13(3), 197–200.

*Grazzi, L., D’Amico, D., & Bussone, G. (1992). Italian experience ofelectromyographic-biofeedback for tension headache: Clinical resultsand one-year follow-up. Headache Quarterly, 3(4), 421–425.

*Grazzi, L., Frediani, F., Zappacosta, B., Boiardi, A., & Bussone, G.(1988). Psychological assessment in tension headache before and afterbiofeedback treatment. Headache, 28(5), 337–338.

*Grazzi, L., Leone, M., Frediani, F., & Bussone, G. (1990). A therapeuticalternative for tension headache in children: Treatment and 1-yearfollow-up results. Biofeedback and Self Regulation, 15(1), 1–6.

Haddock, C. K., Rowan, A. B., Andrasik, F., Wilson, P. G., Talcott, G. W.,& Stein, R. J. (1997). Home-based behavioural treatments for chronicbenign headache: A meta-analysis of controlled trials. Cephalalgia, 17,113–118.

*Hart, J. D., & Cichanski, K. A. (1981). A comparison of frontal EMGbiofeedback and neck EMG biofeedback in the treatment of musclecontraction headache. Biofeedback and Self Regulation, 6(1), 63–74.

Hartmann, A., & Herzog, T. (1995). Varianten der effektstarkenberech-nung in meta-analysen: Kommt es zu variablen ergebnissen? [Calculat-ing effect size by varying formulas: Are there varying results?].Zeitschrift fur Klinische Psychologie und Psychotherapie, 24, 337–343.

*Haynes, S. N., Griffin, P., Mooney, D., & Parise, M. (1975). Electromyo-graphic biofeedback and relaxation instructions in the treatment ofmuscle contraction headaches. Behavior Therapy, 6(5), 672–678.

Headache Classification Committee of the International Headache Society.(1988). Classification and diagnostic criteria for headache disorders,cranial neuralgias and facial pain. Cephalalgia, 8, 1–96.

*Heaton, G. (1979). Biofeedback in headache management: An evaluationof 56 patients. Journal of the Indiana State Medical Association, 72(3),198–199.

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis.Orlando, FL: Academic Press.

Hedges, L. V., & Pigott, T. D. (2001). The power of statistical tests inmeta-analysis. Psychological Methods, 6, 203–217.

394 NESTORIUC, RIEF, AND MARTIN

Hedges, L. V., & Vevea, J. L. (1998). Fixed and random-effects models inmeta-analysis. Psychological Methods, 3, 486–504.

Hermann, C., Kim, M., & Blanchard, E. B. (1995). Behavioral and pro-phylactic pharmacological intervention studies of pediatric migraine: Anexploratory meta-analysis. Pain, 60, 239–256.

Higgins, J. P. T., & Green, S. (Eds.). (2005). Cochrane handbook forsystematic reviews of interventions 4.2.5 [updated May 2005]. In TheCochrane Library, Issue 3, 2005. Chichester, United Kingdom: Wiley.

*Hoffman, E. (1979). Autonomic, EEG and clinical changes in neuroticpatients during EMG biofeedback training. Research Communicationsin Psychology, Psychiatry and Behavior, 4(3), 209–240.

*Holroyd, J. C., Andrasik, F., & Noble, J. (1980). A comparison of EMGbiofeedback and a credible pseudotherapy in treating tension headache.Journal of Behavioral Medicine, 3(1), 29–39.

Holroyd, J. C., & Penzien, D. B. (1986). Client variables and the behav-ioural treatment of recurrent tension headache: A meta-analytic review.Journal of Behavioral Medicine, 9, 515–535.

*Holroyd, K. A., & Andrasik, F. (1982). Do the effects of cognitive therapyendure? A two-year follow-up of tension headache sufferers treated withcognitive therapy or biofeedback. Cognitive Therapy and Research,6(3), 325–334.

*Holroyd, K. A., Andrasik, F., & Westbrook, T. (1977). Cognitive controlof tension headache. Cognitive Therapy and Research, 1(2), 121–133.

Holroyd, K. A., Malinoski, P., Davis, M. K., & Lipchik, G. L. (1999). Thethree dimensions of headache impact: Pain, disability and affectivedistress. Pain, 83, 571–578.

*Holroyd, K. A., Penzien, D. B., Hursey, K. G., Tobin, D. L., Rogers, L.,Holm, J. E., et al. (1984). Change mechanisms in EMG biofeedbacktraining: Cognitive changes underlying improvements in tension head-ache. Journal of Consulting and Clinical Psychology, 52(6), 1039–1053.

Houle, T. T., Penzien, D. B., & Houle, C. K. (2005). Statistical power andsample size estimation for headache research: An overview and powercalculation tools. Headache, 45, 414–418.

*Hudzinski, L. G. (1983). Neck musculature and EMG biofeedback intreatment of muscle contraction headache. Headache, 23(2), 86–90.

*Hutchings, D., & Reinking, R. H. (1976). Tension headaches: What formof therapy is most effective? Biofeedback and Self Regulation, 1(2),183–190.

International Headache Society Clinical Trials Subcommittee. (2000).Guidelines for controlled trials of drugs in migraine (2nd ed.). Cepha-lalgia, 20, 765–786.

Jadad, A. R., Moore, R. A., Carroll, D., Jenkinson, C., Reynolds, D. J.,Gavaghan, D. J., & McQuay, H. J. (1996). Assessing the quality ofreports of randomized clinical trials: Is blinding necessary? ControlledClinical Trials, 17, 1–12.

*Janssen, K. (1983). Differential effectiveness of EMG-feedback versuscombined EMG-feedback and relaxation instructions in the treatment oftension headache. Journal of Psychosomatic Research, 27(3), 243–253.

*Johansson, J., & Ost, L. G. (1985). Self-control procedures in biofeedbacktreatment of tension headache: The effect of “generalization training.”Scandinavian Journal of Behaviour Therapy, 14(3), 99–112.

Johnson, B. T. (1989). DSTAT: Software for meta-analytical review ofresearch literature. Hillsdale, NJ: LEA.

Kaniecki, R. (2003). Headache assessment and management. Journal of theAmerican Medical Association, 289(11), 1430–1433.

Kroener-Herwig, B., Heinrich, U., & Morris, R. (2007). Headache inGerman children and adolescents: A population-based epidemiologicalstudy. Cephalalgia, 27, 519–527.

*Kroener-Herwig, B., Mohn, U., & Pothmann, R. (1998). Comparison ofbiofeedback and relaxation in the treatment of pediatric headache andthe influence of parent involvement on outcome. Applied Psychophysi-ology and Biofeedback, 23(3), 143–157.

*Mathew, A., Mishra, H., & Kumaraiah, V. (1987). Alpha feedback in the

treatment of tension headache. Journal of Personality and ClinicalStudies, 3(1), 17–22.

McCrory, D., Penzien, D. B., Hasselblad, V., & Gray, R. (2001). Behaviouraland physical treatments for tension-type and cervocogenic headaches. DesMoines, IA: Foundation for Chiropractic Education and Research.

McGaw, B., & Glass, G. V. (1980). Choice of metric for effect size inmeta-analysis. American Educational Research Journal, 17, 325–337.

Moher, D., Cook, D. J., Eastwood, S., Olkin, I., Rennie, D., & Stroup, D. F.(1999). Improving the quality of reports of meta-analyses of randomizedcontrolled trials: The QUOROM statement. Quality of reporting ofmeta-analyses. Lancet, 354, 1896–1900.

Molarius, A., & Tegelberg, A. (2006). Recurrent headache and migraine asa public health problem: A population-based study in Sweden. Head-ache, 46, 73–81.

*Neff, D. F., Blanchard, E. B., & Andrasik, F. (1983). The relationshipbetween capacity for absorption and chronic headache patients’ responseto relaxation and biofeedback treatment. Biofeedback and Self Regula-tion, 8(1), 177–183.

Nestoriuc, Y., & Martin, A. (2007). Efficacy of biofeedback for migraine:A meta-analysis. Pain, 128, 111–127.

*Nicholson, N. L., & Blanchard, E. B. (1993). A controlled evaluation ofbehavioral treatment of chronic headache in the elderly. Behavior Ther-apy, 24(3), 395–408.

*Paiva, T., Nunes, J. S., Moreira, A., Santos, J., Teixeira, J., & Barbosa, A.(1982). Effects of frontalis EMG biofeedback and diazepam in thetreatment of tension headache. Headache, 22(5), 216–220.

*Peck, C. L., & Kraft, G. H. (1977). Electromyographic biofeedback forpain related to muscle tension. A study of tension headache, back andjaw pain. Archives of Surgery, 112(7), 889–895.

Penzien, D. B., Rains, J. C., & Andrasik, F. (2002). Behavioural manage-ment of recurrent headache: Three decades of experience and empiri-cism. Applied Psychophysiology and Biofeedback, 27(2), 163–181.

Penzien, D. B., Rains, J. C., Lipchik, G. L., Nicholson, R. A., Lake, A. E.,& Hursey, K. G. (2005). Future directions in behavioural headacheresearch: Applications for an evolving healthcare environment. Head-ache, 45, 526–534.

*Philips, H. C., & Hunter, M. (1981). The treatment of tension headache:I. Muscular abnormality and biofeedback. Behaviour Research andTherapy, 19(6), 485–498.

Phillips, B., Ball, C., Sackett, D., Badenocch, D., Straus, S., Haynes, B., &Dawes, M. (2001). Levels of evidence and grades of recommendations.Oxford, United Kingdom: The Oxford Centre for Evidence-Based Med-icine.

Rains, J. C., Penzien, D. B., McCrory, D. C., & Gray, R. N. (2005).Behavioral headache treatment: History, review of empirical literature,and methodological critique. Headache, 45, 92–109.

Rasmussen, B. K., Jensen, R., Schroll, M., & Olesen, J. (1991). Epidemi-ology of headache in a general population: A prevalence study. Journalof Clinical Epidemiology, 44(11), 1147–1157.

Ray, J. W., & Shadish, W. R. (1996). How interchangeable are differentestimators of effect size? Journal of Consulting and Clinical Psychol-ogy, 64, 1316–1325.

*Reich, B. A. (1989). Non-invasive treatment of vascular and musclecontraction headache: A comparative longitudinal clinical study. Head-ache, 29(1), 34–41.

*Rokicki, L. A., Holroyd, K. A., France, C. R., Lipchik, G. L., France,J. L., & Kvaal, S. A. (1997). Change mechanisms associated withcombined relaxation/EMG biofeedback training for chronic tensionheadache. Applied Psychophysiology and Biofeedback, 22(1), 21–41.

Rosenthal, R. (1979). The “file drawer problem” and tolerance for nullresults. Psychological Bulletin, 86, 638–641.

Rosenthal, R. (1994). Parametric measures of effect size. In H. Cooper &L. V. Hedges (Eds.), The handbook of research synthesis (pp. 231–244).New York: Russel Sage.

395META-ANALYSIS OF BIOFEEDBACK FOR TENSION HEADACHE

*Schoenen, J., Gerard, P., de Pasqua, V., & Juprelle, M. (1991). EMGactivity in pericranial muscles during postural variation and mentalactivity in healthy volunteers and patients with chronic tension typeheadache. Headache, 31(5), 321–324.

Schwartz, B. S., Steward, W. F., Simon, D., & Lipton, R. B. (1998).Epidemiology of tension-type headache. Journal of the American Med-ical Association, 279(5), 381–383.

*Sethi, B. B., Trivedi, J. K., & Anand, R. (1981). A comparative study ofrelative effectiveness of biofeedback and shavasana (yoga) in tensionheadache. Indian Journal of Psychiatry, 23(2), 109–114.

Shadish, W. R., & Haddock, C. K. (1994). Combining estimates of effectsize. In H. Cooper & L. V. Hedges (Eds.), The handbook of researchsynthesis (pp. 261–281). New York: Russel Sage.

Silver, B. V., & Blanchard, E. B. (1978). Biofeedback and relaxationtraining in the treatment of psychophysiological disorders: Or are themachines really necessary? Journal of Behavioral Medicine, 217–239.

Smith, M. L., Glass, G. V., & Miller, T. I. (1980). The benefits ofpsychotherapy. London: Johns Hopkins.

*Steger, J. C., & Harper, R. G. (1980). Comprehensive biofeedback versusself-monitored relaxation in the treatment of tension headache. Head-ache, 20(3), 137–142.

Stillmann, M. J. (2002). Pharmacotherapy of tension-type headaches. Cur-rent Pain and Headache Reports, 6, 408–413.

Stovner, L. J., Hagen, K., Jensen, R., Katsarava, Z., Lipton, R. B., Scher,A. I., et al. (2007). The global burden of headache: A documentation ofheadache prevalence and disability worldwide. Cephalalgia, 27, 193–210.

Trautmann, E., Lackschewitz, H., & Kroner-Herwig, B. (2006). Psycho-logical treatment of recurrent headache in children and adolescents: Ameta-analysis. Cephalalgia, 26, 1411–1426.

Turk, D. C., & Okifuji, A. (1999). Assessment of patients’ reporting ofpain: An integrated perspective. Lancet, 353, 1784–1788.

Wang, M. C., & Bushmann, B. J. (1998). Using the normal quantile plot toexplore meta-analytic data sets. Psychological Methods, 3, 46–54.

Wortmann, P. M. (1994). Judging research quality. In H. Cooper & L. V.Hedges (Eds.), The handbook of research synthesis (pp. 97–109). NewYork: Russel Sage.

Wright, C. C., & Sim, J. (2003). Intention-to-treat approach to data fromrandomized controlled trials: A sensitivity analysis. Journal of ClinicalEpidemiology, 56, 833–842.

Yates, S. L., Morley, S., Eccleston, C., & Williams, A. (2005). A scalefor rating the quality for psychological trials for pain. Pain, 117,314 –325.

Received July 30, 2007Revision received November 26, 2007

Accepted December 21, 2007 �

AppendixTable A1Validity Rating (12 Dichotomous Items)

Item Rating (�max � 12)

Internal validity

Design 1: controlled; 0: uncontrolledTreatment allocation 1: randomized or matched; 0: not randomizedDropout 1: �20% pre-post, �35% follow-up; 0: 20%, 35%Outcome variable 1: at least two outcome measures, i.e., pain and psychological

or physiological outcome; 0: only one

External validity

Time point 1: follow-up of min. 6 month; 0: none or shorterPatient characteristic 1: statistics on age, sex and headache history

0: missing data in one or all categories

Construct validity

Treatment documentation 1: treatment manual with number and duration of sessionsdocumented; 0: no manual or missing information onnumber or duration of sessions

Diagnosis 1: use of structured diagnostic systems, i.e., ICD or IHS; 0:no structured criteria

Medication status 1: controlled; 0: uncontrolledBlinding 1: double or single blind; 0: no blinding

Statistical conclusion validity

Number of participants 1: 10 patients per treatment group; 0: �10Statistics for effect size calculation 1: means and standard deviations reported; 0: not reported

Note. ICD � International Classification of Diseases; IHS � International Headache Society.

396 NESTORIUC, RIEF, AND MARTIN