Systematic review and simulation study of ignoring clustered data in surgical trials

Dell-Kuster, S. and Droeser, R. A. and Schäfer, J. and Gloy, V. and Ewald, H. and Schandelmaier, S. and Hemkens, L. G. and Bucher, H. C. and Young, J. and Rosenthal, R.. (2018) Systematic review and simulation study of ignoring clustered data in surgical trials. British Journal of Surgery, 105 (3). pp. 182-191.

[img] PDF
Restricted to Repository staff only


Official URL: https://edoc.unibas.ch/71510/

Downloads: Statistics Overview


Multiple surgical procedures in a single patient are relatively common and lead to dependent (clustered) data. This dependency needs to be accounted for in study design and data analysis. A systematic review was performed to assess how clustered data were handled in inguinal hernia trials. The impact of ignoring clustered data was estimated using simulations. PubMed, Embase and the Cochrane Library were reviewed systematically for RCTs published between 2004 and 2013, including patients undergoing unilateral or bilateral inguinal hernia repair. Study characteristics determining the appropriateness of handling clustered data were extracted. Using simulations, various statistical methods accounting for clustered data were compared with an analysis ignoring clustering by assuming 100 hernias, with a varying percentage of patients having bilateral hernias. Of the 50 eligible trials including patients with bilateral hernias, 20 (40 per cent) did not provide information on how they dealt with clustered data and 18 (36 per cent) avoided clustering by assessing the outcome by patient and not by hernia. None of the remaining 12 trials (24 per cent) considered clustering in the design or analysis. In the simulations, ignoring clustering led to an increased type I error rate of up to 12 per cent and to a loss in power of up to 15 per cent, depending on whether the patient or the hernia was the randomization unit. Clustering was rarely considered in inguinal hernia trials. The simulations underline the importance of considering clustering as part of the statistical analysis to avoid false-positive and false-negative results, and hence inappropriate study conclusions.
Faculties and Departments:03 Faculty of Medicine > Departement Klinische Forschung > Clinical Epidemiology and Biostatistics CEB
UniBasel Contributors:Ewald, Hannah
Item Type:Article, refereed
Article Subtype:Research Article
Publisher:John Wiley & Sons Ltd
Note:Publication type according to Uni Basel Research Database: Journal article
Identification Number:
edoc DOI:
Last Modified:26 May 2021 08:23
Deposited On:06 Jan 2020 09:18

Repository Staff Only: item control page