Data synthesis and meta-bias

The complications extracted from studies will initially be graded according to the CD classification. Complications will then be grouped for analysis into either primary and secondary outcome; primary if they are CD3 and above (major complications), and secondary if they are CD1 and above (overall complications). Secondary outcomes will be further categorised into any complications, pneumonia, cardiac complication, surgical site infection, urinary tract infection, venous thromboembolism, renal failure, readmission, return to theatre, death and sepsis. The number of postoperative complications for both the lower HbA1c and higher HbA1c group will be used to derive pool estimates for each postoperative complication. HbA1c will be treated as a binary variable as this is the format in which most primary studies report their complications.

Statistical analyses will be performed using Stata Statistical Software (StataCorp. 2019. Stata Statistical Software: Release V.16, StataCorp). Funnel plots, together with Begg’s rank correlation test and Egger’s regression asymmetry test will be used to assess publication bias.²⁸ In addition, the Duval and Tweedie non-parametric ‘trim and fill’ method of accounting for publication bias will be performed to formalise the use of funnel plots and adjust the meta-analysis by incorporating the theoretical missing trials.²⁸ Q-statistic will be used to investigate the degree of heterogeneity between studies. As a limitation of Cochran’s Q-test is the fact that it might be underpowered when few studies have been included or when event rates are low, standard practice is to adopt a higher p value (rather than 0.05) as threshold for statistical significance. Thus, a p>0.1 will be interpreted as evidence by chance alone. I² statistical test²⁹ will be carried out to describe the proportion of total variation caused by heterogeneity because the Q-statistic has low power in common situations of few studies and excessive power to detect clinically unimportant heterogeneity when there are many studies.³⁰ I² of less than 30% of the variability in point estimate will be considered as mild heterogeneity, more than 50% as notable heterogeneity, whereas anything in between considered as moderate heterogeneity. The random effects model (DerSimonian-Laird method) will be used to derive pool estimates and generate forest plots to account for interstudy heterogeneity, to see the effect of low and elevated HbA1c on the various postoperative complications. A meta-regression will be performed to evaluate the effect of different HbA1c cut-off values on the outcome. If a meta-analysis is not possible, a qualitative synthesis will be performed instead.