HSC Section 3 - Trauma, Critical Care and Sleep Medicine

M OUBAYED ET AL .

prespecified statistical plan to account for effects of hetero- geneity included subgroup and meta-regression analyses. Assessment of study quality A validated tool to assess bias in observational studies without a control or comparison group does not yet exist and when created has to be population specific. For this reason and to assess key differences between studies, data were collected on a number of additional study features, which may be considered as surrogates of study quality. Data on the study design and observation period for VTE complications were recorded. Furthermore, the studies were classified according to the Oxford Center for Evidence-Based Medicine 2011 level of evidence (LOE; www.cebm.net). Statistical handling For all studies, we report the proportions in percentage of VTE and bleeding complications. For studies that compared chemoprophylaxis to no chemoprophylaxis, specific propor- tions for VTE and bleeding were reported for each group. For studies in which data on the number of cancer resections with or without free flap reconstruction cases could be extracted or were reported separately, proportions of VTE and bleeding were reported for each group separately as this had the potential to contribute to subgroup analyses. Because of anticipated heterogeneity, a more conserva- tive approach applying the random effects model based on the DerSimonian–Laird method was chosen for all analyses. This method takes into account within-study variation, between-study variation, and the heterogeneity inherent in comparing results from different centers, care settings, and patient subgroups. Forest plots were con- structed for all outcomes displaying the random-effects model summary effect measure, 95% confidence interval (CI), and the I 2 statistic. We were able to compare incidence of VTE and bleeding between groups (chemoprophylaxis vs no chemoprophylaxis, cancer/free flap subgroup vs no can- cer/free flap), and for these odds ratios (ORs) with 95% CIs were reported. To assess publication bias, we used a standard funnel plot, a random-effects Duval and Tweetie’s trim and fill, and calculated a Fail Safe N measure. All statistical anal- yses were performed using Comprehensive Meta-Analysis version 2 (Biostat, Englewood, NJ). RESULTS Study selection The search strategy yielded 729 nonduplicate articles (see Figure 1). Of these, 6 articles were identified using our previously described additional methodology to con- firm a thorough search. After title and abstract review, 46 articles remained. After a full article review, 23 studies were included in our systematic review. The most com- mon reason for exclusion was a review or commentary article. No contact with authors was necessary. Study characteristics and key moderators The characteristics of the included studies are shown in Table 1. 12–32 Studies were published between 1987 and 2015. The search did not identify any randomized trials

reviews were assessed, the “related articles” to key publications in PubMed were reviewed, forward citations were used, and one expert in the field (S.P. Most) was consulted to make sure we were not missing any key references. This systematic review and meta-analysis was performed following the guide- lines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 10 and, given that the studies reviewed were all observational studies, we also utilized the Meta-analysis of Observational Studies in Epidemiology (MOOSE) guidelines. 11 The protocol for this review has not been previously published or registered. Study selection After database search and duplicate removal, studies were initially screened for relevance based on their title and abstract, and either excluded or included in this study. Full texts were consulted on an as-needed basis for studies for which the title and abstract could not determine eligibility. Two investigators (S.P. Moubayed and M.W.M.) indepen- dently performed the search and assessed each article for eli- gibility and disagreement between reviewers was resolved using consensus, whereas a third reviewer (S.P. Most) was used rarely when agreement could not be achieved and when articles in doubt were included for a full article review. Inclusion and exclusion criteria We included English-language studies pertaining to any OHNS patient, whether inpatient or outpatient. Studies reporting DVT and/or PE complications were included. Cohort studies, clinical trials, and outcome studies were retained for inclusion, whereas case series of 100 patients or less were excluded. Studies in languages other than English, studies pertaining to unrelated diseases (ie, sigmoid sinus thrombosis), reviews, commentaries, and editorials were excluded. Data collection process, moderators, bias across studies, and heterogeneity Incidence of DVT, PE, VTE, and bleeding complica- tions were recorded, including sample size, so that event and incidence rates could be calculated. Risk of bias across studies may be present, particularly with regard to publication bias. Because we were reviewing surgical literature, it was very likely that smaller studies or those with unfavorable outcomes may not be published in the literature. A funnel plot was created to assess publica- tion bias, and the Duval and Tweedie’s trim-and-fill method was used for the analysis, particularly as it related to each OHNS population subgroup to assess when the missing studies would be plotted. For the aforementioned reasons, the a priori hypothesis was that a large degree of heterogeneity existed among studies. Important areas of variation include the type of thromboprophylaxis used, population subgroup within OHNS (general, endocrine, skull base, or cancer with or without free flap reconstruction), and year of publication. We also collected information on the outcome study period (ie, how long the wait period was to assess the outcome of interest), the year the study was performed, and the type of study (administrative database, cohort study with either prospective or retrospective review). The

HEAD & NECK—DOI 10.1002/HED JUNE 2017

164