Roel Bogie

Chapter 3

calculated using random effect models. We used the R statistical program version 3.2.2 to process all the collected data. 22 Using raw numbers from the included articles and a binomial distribution, the prevalences and their 95%CIswere calculatedper study. A random-effectsmodel withDerSimonian- Laird estimator, which takes into consideration both within and between study heterogeneity, was applied to the raw proportions and odds ratios using the Metafor package version 1.9. 23 For the pooled prevalence of LSTs, a double arc-sinus correction was applied (Freeman-Tukey). 24 This method adjusts for variance instability in situations of low prevalence. After correction, the results were converted to the original proportion scale for interpretation. Heterogeneitywas testedwith the Q test for significance and with the inconsistency index (I 2 ), where a value of >50% was considered as substantial heterogeneity between studies. Funnel plots with Egger’s test for asymmetry were constructed to test the possible effect of publication bias. 25 Additionally, sensitivity analyses were performed after exclusion of studies with a high risk of bias. Potential effect modification of

Electronic search:

• Pubmed: 2500 • Embase: +200 • Scopus: +131 • Cochrane library: +118

2949 unique records

197 articles selected on the basis of the title and abstract

Exclusion of papers: • No information on LSTs: 80 • Selected population: 34 • Abstract / reviews: 32 • Duplicates: 3

48 articles eligible after reading full text

7 eligible for LST prevalence estimates

12 eligible for both prevalence estimates and risk of submucosal invasion

29 eligible for risk of submucosal invasion

Figure 3.2: Flow-chart of the search, and inclusion and exclusion of articles on prevalence and submucosal invasion in LSTs.

42