Non-compliance with randomised allocation and missing outcome data in randomised controlled trials evaluating surgical interventions: a systematic review

Randomised controlled trials are widely acknowledged as the gold standard in medical research although their validity can be undermined by non-compliance with the randomly allocated treatment and missing data. Due to the nature of the intervention, surgical trials face particular threat to compliance and data collection. For example, ineligibility for the intervention may only become apparent once the operation has commenced. It is unclear how such cases are reported and handled. The objective was to assess non-compliance and missing data in reports of trials of surgical interventions. Searches for reports of trials involving at least one surgical procedure and published in 2010 were carried out in the Medical Literature Analysis and Retrieval System Online (MEDLINE(®)). Data on missing data, non-compliance and methods of handling missing data were extracted from full texts. Descriptive data analyses were carried out on the data. Forty-five (55%) studies reported non-compliance with treatment allocation and 52 (63%) reported primary outcome missing data. The median levels of non-compliance and missing data were 2% [IQR (0, 5), range (0-29)] and 6% [IQR (0, 15), range (0-57)], respectively. Fifty-two (63%) studies analysed as randomised, 17 (21%) analysed per protocol and 3 (4%) analysed as treated. Complete case analysis was the most common method used to deal with missing data, 35/52 (67%). The reporting of non-compliance to allocation and the handling of missing data were typically suboptimal. There is still room for improvement on the use of the CONSORT statement particularly in accounting for study participants. Transparency in reporting would facilitate evidence synthesis.