Subgroup analysis and statistical power

Subgroup analysis and statistical power

European Journal of Obstetrics & Gynecology and Reproductive Biology 159 (2011) 244–245 Contents lists available at ScienceDirect European Journal o...

99KB Sizes 0 Downloads 37 Views

European Journal of Obstetrics & Gynecology and Reproductive Biology 159 (2011) 244–245

Contents lists available at ScienceDirect

European Journal of Obstetrics & Gynecology and Reproductive Biology journal homepage: www.elsevier.com/locate/ejogrb

LETTERS TO THE EDITOR—CORRESPONDENCE Subgroup analysis and statistical power Dear Editor, We read with great interest the recent published review and meta-analysis of RCT focused on the efficacy and adverse effects of using misoprostol for cervical ripening prior to hysteroscopy [1]. Although this study has several strengths, I have some concerns, especially about the authors’ conclusion. The authors observed a significant reduction in the need for cervical dilation considering all women (RR = 0.75, 95%CI = 0.59– 0.96), but a substantial heterogeneity between studies was also observed (I2 = 65%). One strategy to reduce heterogeneity between studies and identify in which situation(s) the intervention is more likely to work would be subgroup analysis, which was performed in this study, by separating in pre- and post-menopausal women. However, the heterogeneity remained significant and the RR in needing cervical dilation was similar in these two groups (RR = 0.73 in both cases). Therefore, other factors rather than menopausal status (e.g., misoprostol dose or route) are more likely to be responsible for the observed heterogeneity. Additionally, authors should have considered that by performing subgroup analysis a reduction in the power (or increased type II error) was expected, because fewer subjects are evaluated and any apparent lack of differential effect should be regarded with caution [2]. In this study, subgroup analysis only reduced the power, rather than helping to understand why differences between the studies occurred. Tables 2 and 3 have several mistakes: some of the 95%CI were wrongly reported; several (%) were not presented; and the heterogeneity was assessed only by p-value of Chi-squared test. Chi-squared is known to be poor at detecting true heterogeneity among studies; and currently I2 is the preferred test to evaluate inconsistency across studies, as this describes the percentage of the variability in effect estimates that is due to heterogeneity rather than chance [3]. However, when reading these tables we could observe several significant and/or trend to clinically relevant changes associated with misoprostol group on cervical ripening and adverse effects. The main problem of this article was the authors’ conclusion which is inconsistent with the presented results. The authors concluded that ‘‘based on the results of this review we cannot support the routine use of misoprostol prior to every hysteroscopy. Given that the lack of serious benefit from misoprostol is unlikely to be due to type II error, its use should be reserved for selected cases’’. In our opinion, the conclusion should be completely different, something like: ‘‘Misoprostol was associated with a significant reduction in the need for cervical dilation, a reduction in the duration of the procedure, and a trend to reduction in cervical tear. However, we observed a substantial heterogeneity between studies and the result may be different depending on misoprostol dose, route or patient characteristics. Additionally, misoprostol

0301-2115/$ – see front matter ß 2011 Elsevier Ireland Ltd. All rights reserved.

was associated with an increased risk of mild abdominal pain, bleeding, diarrhea, fever and nausea, and therefore the benefits should be balanced with the adverse effects when informing subjects before obtaining consent’’. Type II error was not assessed by this meta-analysis and should not be included in the conclusions. In order to estimate type II error, the authors need to define what magnitude of difference between the intervention and control groups should be considered as clinically relevant. This task might be controversial, is obviously dependent on the evaluated outcome characteristics, and should be performed before starting the review. For example, one might consider a relative change of 20% as clinically relevant [4] and type II error would be considered low only if the 95%CI did not overlap 0.80 or 1.20; or by evaluating the power of the performed tests, which requires additional analysis. References [1] Gkrozou F, Koliopoulos G, Vrekoussis T, et al. A systematic review and metaanalysis of randomized studies comparing misoprostol versus placebo for cervical ripening prior to hysteroscopy. Eur J Obstet Gynecol Reprod Biol )2011;(May 27). doi: 10.1016/j.ejogrb.2011.04.022. [2] Petticrew M, Tugwell P, Kristjansson E, Oliver S, Ueffing E, Welch V. Damned if you do, damned if you don’t: subgroup analysis and equity. J Epidemiol Community Health )2011;(June 6). doi: 10.1136/jech.2010.121095. [3] Higgins JP, Thompson SG, Deeks JJ, Altman DG. Measuring inconsistency in meta-analyses. BMJ 2003;327(September (7414)):557–60. [4] Martins WP, Rocha IA, Ferriani RA, Nastri CO. Assisted hatching of human embryos: a systematic review and meta-analysis of randomized controlled trials. Hum Reprod Update 2011;17(July–August (4)):438–53.

Wellington P. Martinsa,b, c, * Departamento de Ginecologia e Obstetrı´cia, Faculdade de Medicina de Ribeira˜o Preto da Universidade de Sa˜o Paulo (FMRP-USP), Ribeira˜o Preto, SP, Brazil b Instituto Nacional de Cieˆncia e Tecnologia (INCT) de Hormoˆnios e Sau´de da Mulher, Ribeira˜o Preto, SP, Brazil c Escola de Ultra-sonografia e Reciclagem Me´dica de Ribeira˜o Preto (EURP), Ribeira˜o Preto, SP, Brazil

a

Jose´ Vitor C. Zanardi Departamento de Ginecologia e Obstetrı´cia, Faculdade de Medicina de Ribeira˜o Preto da Universidade de Sa˜o Paulo (FMRP-USP), Ribeira˜o Preto, SP, Brazil *Corresponding author at: Departamento de Ginecologia e Obstetrı´cia da Faculdade de Medicina de Ribeira˜o Preto, Universidade de Sa˜o Paulo. Av. Bandeirantes, 3900, 8 andar., CEP: 14049-900 Ribeira˜o Preto, Sa˜o Paulo, Brazil. Tel.: +55 16 36022583; fax: +55 16 36330946 E-mail address: [email protected] (W.P. Martins). 13 June 2011 doi:10.1016/j.ejogrb.2011.07.046