Abstract
The continuous development and refinement of techniques is a key attribute of the discipline of surgery. Robust evidence derived from randomised clinical trials and subsequent synthesis using meta-analysis methods is therefore an indispensable tool in modern evidence-based surgical practice. This thesis aimed to address two challenges to the development of this evidence, namely the synthesis of trial data where multiple treatment comparisons exist, and the influence of sources of bias on the results of surgical randomised trials.
Research synthesis typically employs pairwise meta-analysis methods to summarise trial data. However, there are more than two treatment options available for most conditions, meaning that data pertaining to all treatments cannot be incorporated using pairwise methodology. Network meta-analysis allows for the simultaneous comparison of multiple treatments and enables their ranking in terms of benefit and harm. This thesis used two different examples to explore this method’s utility in surgical research synthesis.
Firstly, network meta-analysis methodology was used to investigate the efficacy of preoperative carbohydrate loading for patients undergoing elective surgery. None of the previously published pairwise meta-analyses had been able to account for the different doses and control treatments used in trials. This network meta-analysis represents the most comprehensive synthesis of the available evidence, and showed that carbohydrate loading confers a small reduction in length of stay when compared to fasting, but no significant difference when compared to water or placebo, and no other clinically important effect on postoperative outcomes.
Secondly, all described anti-reflux operations were assessed and ranked using network metaanalysis methods to summarise the entirety of available randomised trial data, to determine the optimal procedure for the treatment of gastro-oesophageal reflux disease. This had remained an unresolved question despite numerous pairwise meta-analyses. The results showed that a posterior partial fundoplication provides the best balance of long-term, durable reflux control with less dysphagia, compared to other treatments.
The second half of this thesis explored the effect of non-blinding and other methodological deficiencies on surgical trials. Recent pooled studies have provided empirical evidence that these potential sources of bias significantly influence reported trial results by exaggerating the effect estimate of the studied treatment. Surgical randomised trials have important differences to drug trials, and it is important to determine whether this effect also applies to them. A study using meta-epidemiological methods at an individual trial level was conducted to determine how lack of bias-minimisation measures including blinding affects the reported outcomes of surgical randomised trials, using data from over 300 trials. The results showed that trials that did not use blinding or adequate random sequence generation reported a significantly greater difference between treatment groups compared to trials that used such measures.
This thesis has demonstrated the utility of network meta-analysis in surgical research synthesis, and produced definitive evidence-based answers to two questions. Many more clinically relevant questions can be answered in future using the same methods. This thesis has also empirically proven the importance of implementing blinding in surgical trials, which is relevant for appraising published trials and designing future randomised trials.