Did you add the same study twicein the meta-analysis?

I read with interest the systematic review of pharmacotherapy forpost-traumatic stress disorder (PTSD) by Mathew Hoskins andcolleagues. Looking at Fig. 2 of the paper and at the online-onlysupplemental file, however, it seems to me that two unpublishedstudies have been counted twice in the meta-analysis. As far as Ican see from the information reported in the review, the study‘Eli Lilly’ is the same as ‘Martenyi 2007’ and ‘Pfizer 589’ is thesame as ‘Friedman 2007’ (same drugs, same comparisons, samesample size). I would be grateful if the authors could clarify thematter.

1 Hoskins M, Pearce J, Bethell A, Dankova L, Barbui C, Tol QA, et al.Pharmacotherapy for post-traumatic stress disorder: systematic reviewand meta-analysis. Br J Psychiatry 2015; 206: 93–100.

Andrea Cipriani, Associate Professor, University of Oxford. Email:andrea.cipriani@psych.ox.ac.uk

doi: 10.1192/bjp.207.2.177

Authors’ reply: Professor Cipriani helpfully questions duplicationof data from two unpublished studies in our review. We could notget access to the unpublished material for ‘Eli Lilly’ and ‘Pfizer589’, even after contacting the pharmaceutical companies, andinstead relied on the raw data sets obtained from previous reviews.

We have contacted Dr Friedman, who has confirmed thatPfizer 589 was subsequently published as ‘Friedman 2007’. Wehave also contacted the authors of ‘Martenyi 2007’, but are, asyet, unable to confirm if this is the published ‘Eli Lilly’ paper. Thisseems distinctly possible, as the sample sizes are the same but it isimportant to note that Eli Lilly only released Treatment OutcomePTSD Scale data and not Clinician Administered PTSD Scale datato the National Institute of Health and Care Excellence reviewers.

When the ‘Pfizer 589’ data are removed, it changes the outcomefor sertraline, which now demonstrates a small but statisticallysignificant advantage over placebo in reducing the severity ofclinician-rated PTSD symptoms (8 studies, n= 1271, SMD70.16 (95%CI 70.31 to 70.02), w2 = 33%). This means thatparoxetine, fluoxetine, venlafaxine and sertraline can be consideredas potential treatments for PTSD.

The outcome for the trauma-type sub-analysis for sertraline isstill statistically insignificant (3 studies, n= 278, SMD 70.42(95%CI 71.03 to 0.19), w2 = 81%). ‘Eli Lilly’ was not includedin the meta-analysis of individual agents v. placebo. The overallmeta-analysis of selective serotonin-reuptake inhibitors (SSRIs)v. placebo, when ‘Eli Lilly’ and ‘Pfizer 589’ are removed, is nowslightly more in favour of SSRIs (19 studies, n= 3350, SMD70.27 (95%CI 70.37 to 70.16), w2 = 45%); see revised forestplot, here Fig. 1).

Mathew Hoskins, Jennifer Pearce, Andrew Bethell, Cardiff University, Cardiff,UK; Corrado Barbui, University of Verona, Verona, Italy; Wietse A. Tol, JohnsHopkins Bloomberg School of Public Health, Baltimore, USA; Soraya Seedat,Stellenbosch University, Stellenbosch, South Africa; Hanhui Chen, Shanghai JiaotongUniversity, Shanghai, China; Jonathan I. Bisson, Institute of Psychological Medicineand Clinical Neurosciences, Cardiff University School of Medicine, Cardiff, UK. Email:BissonJI@cardiff.ac.uk

doi: 10.1192/bjp.207.2.177a

177

Contents& Did you add the same study twice in the

meta-analysis?

& Schizophrenia and mixed-handedness

The British Journal of Psychiatry (2015)207, 177–178

Correspondence

Edited by Kiriakos Xenitidis andColin Campbell

Study or subgroup

Brady (2000)38

Brady (2005)61

Connor (1999)40

DavidsonA

Davidson (2001)42

Friedman (2007)21

Hertzberg (2000)47

Marshall (2001)50

Marshall (2004)64

Martenyi (2002)51

Martenyi (2007)23

Panahi (2011)26

Pfizer 588C

Shaleve (2011)27

SKB627E

Tucker (2001)53

Tucker (2003)68

Van der Kolk (2007)70

Zohar (2002)54

Total (95% CI)

Heterogeneity: t2 = 0.02; w2 = 32.43, d.f. = 18 (P= 0.02); I2 = 45%

Test for overall effect: Z= 5.04 (P50.00001)

70.2 70.1 0 0.1 0.2

Favours experimental Favours control

Mean

733

32.56

10.1

739.4

733

713.1

47

738.75

55.6

734.6

742.85

722.7

727.4

731.12

736.5

735.5

741.82

733.23

718.7

s.d.

28.1

15.69

9.8

27.12

23.76

27.5

8

27.2

33.4

28.1

25.5

7.3

27.12

29.63

26.1

24.58

29.09

22.11

6.7

Total

94

49

25

173

100

86

6

375

25

226

323

35

94

23

109

151

23

30

23

1968

Mean

723.2

32.7

20.5

734.17

726.2

715.4

42

725.3

62.8

26.8

736.6

717.5

727.9

727.8

730.8

724.7

738.7

730.95

713.5

s.d.

28.7

28.75

12.6

28.42

23.9

28.07

11

25.8

40.8

26.1

25.7

7.5

28.42

20.13

25.37

24.98

29.07

22.6

6.6

Total

93

45

22

179

108

83

6

188

27

75

88

35

94

23

103

156

10

29

19

1382 100.0%

Weight

5.8%

4.1%

2.4%

7.3%

6.1%

5.6%

0.8%

8.0%

2.8%

6.3%

6.8%

3.3%

5.9%

2.5%

6.2%

7.0%

1.7%

3.0%

2.2%

SMD IV, random, 95% CI

70.34 (70.63, 70.05)

70.01 (70.41, 0.40)

70.91 (71.52, 70.31)

70.19 (70.40, 0.02)

70.28 (70.56, 70.01)

0.08 (70.22, 0.38)

0.48 (70.68, 1.64)

70.50 (70.68, 70.32)

70.19 (70.73, 0.36)

70.28 (70.54, 70.02)

70.24 (70.48, 70.01)

70.69 (71.18, 70.21)

0.02 (70.27, 0.30)

70.13 (70.71, 0.45)

70.22 (70.49, 0.05)

70.43 (70.66, 70.21)

70.10 (70.85, 0.64)

70.10 (70.61, 0.41)

70.77 (71.40, 70.14)

70.27 (70.37, 70.16)

SMD IV, random, 95% CI

Experimental Control

8

7

8

Fig. 1 Revised meta-analysis of selective serotonin reuptake inhibitors v. placebo (SMD, standardised mean difference).

Schizophrenia and mixed-handedness

The new meta-analysis by Hirnstein & Hugdahl1 shows thatschizophrenia is robustly associated with non-right-handednessand thus makes a strong case for genetic links between schizo-phrenia, brain lateralisation and handedness. The associationwas obtained meta-analytically for psychometrically assessedhandedness, and it was even stronger when handedness wasassessed behaviourally. Of interest, the available evidence alsoallowed the tentative conclusion that this association might fore-most be driven by mixed-handedness, rather than left-handedness,thus suggesting that the strength of handedness, rather than itsdirection, might actually be linked to schizophrenia.

With regard to this possibility, we emphasise that assessmentand classification reliability is a pervasive problem in handednessresearch. Studies in this field often treat handedness as a binaryvariable, comprising only right v. left preferences, and frequentlymerely use single-item measures, or arbitrary criteria forhandedness classification, when multi-item inventories are used.

However, recent evidence2 shows that psychometricallyassessed handedness is in fact taxonic and discrete – that is,a matter of qualitative, rather than quantitative, differencesconsisting of the taxa of right-, left- and mixed-handedness.Similarly, psychometrically assessed footedness, earedness, andeyedness all consist of these three taxa, which, together withhandedness, might be explained by underlying sidedness, whichalso consists of the above three taxa. Further, handedness has beenfound to be a biased indicator of mixed-sidedness, which might beexplained by ubiquitous external and societal pressures promotingright-hand preference among actual non-right-handers.3 Ittherefore appears that footedness, obviously being less influencedby such external pressures, is the most important predictor ofsidedness.2

Given these facts, it is interesting to note that the Hirnstein &Hugdahl meta-analysis was still able to hint at mixed-handednessas the probable factor driving the association between

schizophrenia and non-right-handedness, thereby making a con-vincing case for the power of the meta-analytical approach, which,through data aggregation, might overcome methodologicalproblems of individual studies. In addition, increasing theclassification reliability of handedness has recently aided inclarifying the hypothesised seasonal pattern of birth monthsamong left-handed men,4 a hypothesis for which the availableprevious evidence had been highly inconsistent.

We thus recommend utilising psychometrically validated scales,along with a trichotomous classification, in investigations onschizophrenia and handedness, expecting that mixed-handednessmight turn out to be more relevant than left-handedness forexplaining this link. Further, we recommend intensifying researchspecifically with regard to footedness, which might be moreimportant with regard to brain lateralisation than handedness,5

and might provide more direct, and thus more robust, behaviouralevidence for the links between schizophrenia and brainlateralisation.

1 Hirnstein M, Hugdahl K. Excess of non-right-handedness in schizophrenia:meta-analysis of gender effects and potential biases in handednessassessment. Br J Psychiatry 2014; 205: 260–7.

2 Tran US, Stieger S, Voracek M. Evidence for general right-, mixed-, and left-sidedness in self-reported handedness, footedness, eyedness, andearedness, and a primacy of footedness in a large-sample latent variableanalysis. Neuropsychologia 2014; 62: 220–32.

3 Porac C, Coren S. Lateral Preferences and Human Behavior. Springer, 1981.

4 Tran US, Stieger S, Voracek M. Latent variable analysis indicates thatseasonal anisotropy accounts for the higher prevalence of left-handedness inmen. Cortex 2014; 57: 188–97.

5 Elias LJ, Bryden MP. Footedness is a better predictor of languagelateralisation than handedness. Laterality 1998; 3: 41–51.

Ulrich S. Tran, Martin Voracek, Department of Basic Psychological Researchand Research Methods, School of Psychology, University of Vienna, Vienna, Austria.Email: ulrich.tran@univie.ac.at

doi: 10.1192/bjp.207.2.178

178

Correspondence