864

short report

Re-evaluation of the subgroup analysis from the Royal College of Obstetricians and Gynaecologists randomized controlled trial of cervical cerclage

Kristin M. Knight & David N. hackney

Department of Obstetrics and Gynecology, Division of Maternal-Fetal Medicine, University of Rochester School of Medicine, Rochester, NY, USA

The Journal of Maternal-Fetal and Neonatal Medicine

2012

25

6

864

865

© 2012 Informa UK, Ltd.

10.3109/14767058.2011.594120

1476-7058

1476-4954

The Journal of Maternal-Fetal and Neonatal Medicine, 2012; 25(6): 864–865© 2012 Informa UK, Ltd.ISSN 1476-7058 print/ISSN 1476-4954 onlineDOI: 10.3109/14767058.2011.594120

Correspondence: Kristin M. Knight, MD, 601 Elmwood Avenue, Box 668, Rochester, NY 14642, USA. Tel: 585-275-7480. Fax: 585-256-1416. E-mail: kristin_knight@urmc.rochester.edu

History-indicated cervical cerclage is offered to patients who are at risk of spontaneous preterm birth (SPTB), though the indica-tions are controversial. A common practice of offering cerclage after three prior SPTBs or midtrimester losses (MTLs) is based on findings of the subgroup analysis of the 1993 Royal College of Obstetricians and Gynaecologists (RCOG) randomized trial of cervical cerclage. The subgroup analysis was performed by repeating the primary analysis within individual subgroups, which can lead to erroneous conclusions. We repeated the subgroup analysis by evaluating the interaction between the characteristic of interest and treatment allocation in a regression model. The interaction between cerclage and any prior PTB as a binary variable was non-significant. Among subjects deliv-ering at <37 weeks, there was a significant interaction between cerclage and prior PTBs as a continuous variable or ≥ 3 (p-values 0.04 and 0.03, respectively). There were no significant interac-tions between cerclage and the aforementioned outcomes among women who delivered at <33 weeks, though this may have been secondary to a smaller number of SPTB in this range. Our findings lend credence to the current recommendations regarding the use of history-indicated cerclage, though they remain subject to the inherent limitations of subgroup analyses.

Keywords: cervical cerclage, cervical insufficiency, prior preterm birth, randomized controlled trial, subgroup analysis

History-indicated cervical cerclage is offered to patients who are at risk of spontaneous preterm birth (SPTB), though the indica-tions remain controversial, leading to significant practitioner variation [1]. A common practice, endorsed by the American Congress of Obstetricians and Gynecologists among others [2,3], is to offer cerclage after 3 prior SPTBs or midtrimester losses (MTLs) without further risk stratification by cervical length. This stems from the subgroup analysis of the 1993 Royal College of Obstetricians and Gynaecologists (RCOG) randomized trial of cervical cerclage [4], which remains the largest such study. Though an overall reduction in SPTB was demonstrated among subjects with cerclage, secondary to the broad inclusion criteria this was a very heterogeneous group. Thus a subgroup analysis was performed, in which the benefit of cerclage appeared to be limited to subjects with three or more prior SPTB/MTLs.

The proper role of subgroup analyses in evidence based medicine, and whether or not clinical management should be

based on a study’s overall results versus individual subgroups, is controversial [5]. Though the RCOG analysis was pre-planned and used pre-randomization characteristics, it was performed by repeating the primary analysis within individual subgroups, a technique that can lead to erroneous conclusions and is not considered appropriate by many biostatisticians. A more proper technique, as reviewed by Klebanoff [5] and others, is to evaluate the interaction between the characteristic of interest and treat-ment allocation in a regression model. Because the data in Table 4 of the RCOG study is described as being complete and mutually exclusive, and because this particular subgroup analysis serves as the basis for many treatment recommendations, we sought to re-analyze it through an interaction analysis.

The data from Table 4 of the RCOG study was abstracted in parallel by the two authors, and an artificial dataset was gener-ated with outcome and clinical characteristics (twins, prior cone biopsy and “other”) coded as binary variables. Logistic regres-sion was performed using Stata 11 (College Station, TX, USA) with delivery <33 or <37 weeks gestational age (GA) as the outcome and clinical characteristics as predictors. The analysis was repeated with prior PTB coded as just present or absent, ≥ or <3 or as a continuous variable (coded 0-3), along with PTB history x cerclage interaction variables. A p-value of <0.05 was significant.

Table I presents the p-values of the coefficients of the interaction variables for both GA cut offs. The interaction between cerclage and any prior PTB as a binary variable was non-significant for both GA cutoffs. Among subjects delivering at <37 weeks, there was a significant interaction between cerclage and prior PTBs as a continuous variable or ≥3. There were no significant interac-tions between cerclage and the aforementioned outcomes among women who delivered at <33 weeks.

Table I. Significance (p-value) of the interaction between cerclage and preterm birth history in multivariate logistic regression. Delivery before

33 completed weeksa 37 completed weeksa

Any prior PTB (yes/no) 0.83 (0.93, 0.47–1.82) 0.22 (0.72, 0.42–1.22)Number of prior PTBs (0–3)b 0.26 (0.82, 0.59–1.15) 0.04 (0.75, 0.57–0.99)≥3 prior PTBs (yes/no) 0.08 (0.41, 0.15–1.13) 0.03 (0.37, 0.15–0.89)PTB(s) = preterm birth(s) or midtrimester losses.aAll data reported as p-value (odds ratio, 95% confidence intervals).bContinuous variable.

(Received 08 April 2011; revised 17 May 2011; accepted 31 May 2011)

J M

ater

n Fe

tal N

eona

tal M

ed D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y B

iblio

teka

Uni

wer

syte

tu W

arsz

awsk

iego

on

11/1

8/14

For

pers

onal

use

onl

y.

Cerclage subgroups 865

Copyright © 2012 Informa UK, Ltd.

Our findings support the notion that the benefit of history-in-dicated cerclage in reducing PTB prior to 37 weeks is primarily among women with ≥3 prior PTBs. Unlike the published analysis, we did not demonstrate a significant interaction for deliveries at <33 weeks, though a trend (p = 0.08) did exist, and this group had a smaller sample size. These data lend credence to the current recommendations regarding the use of history-indicated cerclage. Of note, however, even a properly performed subgroup analysis should not be viewed as equivalent to a separate prospective, randomized trial of subjects in that group. Clinicians, therefore, remain limited to the available data and the inherent limitations of subgroup analyses.

Declaration of interest: The authors report no conflicts of interest.

References1. Fox NS, Gelber SE, Kalish RB, Chasen ST. History-indicated cerclage:

practice patterns of maternal-fetal medicine specialists in the USA. J Perinat Med 2008;36:513–517.

2. ACOG Practice Bulletin No. 48. Cervical Insufficiency. American Congress of Obstetricians and Gynecologists. Obstet Gynecol 2003; 102:1091–1099.

3. Fox NS, Chervenak FA. Cervical cerclage: a review of the evidence. Obstet Gynecol Surv 2008;63:58–65.

4. MacNaughton MC, Chalmers IG, Dubowitz V, Dunn PM, Grant AM, McPherson K, Pearson JF, Peto R, Turnbull AC. Final report of the Medical Research Council/Royal College of Obstetricians and Gynaecologists Multicentre Randomised Trial of Cervical Cerclage. Br J Obstet Gynaecol 1993;100:516–523.

5. Klebanoff MA. Subgroup analysis in obstetrics clinical trials. Am J Obstet Gynecol 2007;197:119–122.

J M

ater

n Fe

tal N

eona

tal M

ed D

ownl

oade

d fr

om in

form

ahea

lthca

re.c

om b

y B

iblio

teka

Uni

wer

syte

tu W

arsz

awsk

iego

on

11/1

8/14

For

pers

onal

use

onl

y.