2018 Section 5 - Rhinology and Allergic Disorders

Otolaryngology–Head and Neck Surgery 156(1)

Results Search Results

the primary repair, this was not considered a failure of repair; thus, effort was made to exclude these cases from failed repair rates when possible. Level of evidence was assigned to each study according to the Oxford Centre for Evidence-Based Medicine. 15 Quality Assessment Quality assessment of case series was performed with guide- lines from the National Institute for Health and Care Excellence (NICE). There is currently no widely accepted quality assessment tool for evaluation of case series; how- ever, many meta-analyses have demonstrated the utility of the NICE grading system. 16-18 Each study is scored on a scale from 0 to 8, with higher scores indicative of higher quality. Assessment of risk of bias in randomized prospective studies was performed with the Cochrane handbook (version 5.1.0). 19 The following potential biases were examined: selection bias, performance bias, detection bias, reporting bias, and other uncategorized biases. Judgment of the risk for each type of bias was designated as ‘‘low risk,’’‘‘high risk,’’ or ‘‘unclear risk.’’ Data Analysis The primary outcome measure examined was success of pri- mary endoscopic CSF leak repair based on the use of perio- perative LD use. Success of the intervention was defined as no reported postoperative incidence of CSF rhinorrhea or positive tests confirming presence of extracranial CSF (ie, beta-trace protein, b 2-transferrin) within the study follow- up period. Statistical heterogeneity among the included stud- ies was measured with Cochran’s Q , t 2 , and I 2 statistics, as well as visual inspection of forest plots. I 2 is the preferred statistic for assessment of heterogeneity, and values \ 25% to 30% usually suggest low heterogeneity. 20,21 Mantel- Haenszel odds ratios (ORs) with a random effects model (DerSimonian-Laird) were calculated for each study and overall to estimate the intervention effect. For studies that had a success rate of 100% in both arms, continuity correction was performed by adding 0.5 to the rate of successful repair for each arm (and 1 to the number of total subjects in each arm) to prevent calculation of an unde- fined OR (ie, 0/1). For example, if a treatment group (n = 10) and a control group (n = 10) both have a 100% success rate, the calculation of the OR would be (0/0) / (10/10)—not a true calculation of the OR. Rather, adding 0.5 to the rate and 1 to the total number of subjects instead gives (0.5/0.5) / (11/11), yielding an OR of 1.0. As is the case in most meta- analyses, the relatively small number of pooled studies was power limiting. As such, to increase the sensitivity of detect- ing between-study heterogeneity, an interpretation of the Q statistic was made with a P value significance threshold of 0.10 rather than the conventional 0.05 (a low P value pro- vides evidence of heterogeneity). 19 Likelihood of publication bias was evaluated with funnel plots and Egger’s test. All analyses were performed with SPSS 18.0 for Windows (PASW Statistics, Chicago, Illinois).

The literature search yielded a total of 1314 nonduplicate articles, of which 51 were identified from the references of relevant articles (see Figure 1 for PRISMA flowchart). Of the 116 full-text articles that were critically reviewed, 104 were excluded, with the most common reason (54 articles) being inability to determine postoperative recurrence rates of CSF leak following endoscopic repair by LD placement. Twelve studies were suitable for inclusion in the meta- analysis: 1 randomized controlled trial and 11 case series. 3,13,22-31 The randomized controlled trial was evalu- ated for risk of bias with the Cochrane handbook and found to be ‘‘low risk’’ across all potential biases (selection, per- formance, detection, attrition, reporting, and other). When the remaining 11 studies were assessed for quality with the NICE criteria ( Table 1 ), 1 study received an 8; 2 received a 7; 1 received a 6; 3 received a 5; and 4 received a 4. Indications for perioperative lumbar drainage as an adjunct to endoscopic repair varied extensively among stud- ies. Given the large proportion of retrospective studies, sig- nificant selection bias between the study arms is possible. In multiple studies, perioperative lumbar drainage had a ten- dency to be utilized in patients who were at greater risk of CSF leak recurrence suggested by difficult reconstruction, 28 uncertainty of leak control after repair 13,24 or persistent CSF leak after reconstruction, 28 CSF leaks of spontaneous etiol- ogy, 30 large dural defects, 13,24 secondary CSF leak repair, 13 defects involving the frontal or sphenoid sinus, 26 reconstruc- tion performed with an underlay closure, 26 or leaks more active than those considered slow or small. 29 In 5 retrospec- tive studies, criteria for LD placement were absent or poorly defined. 3,23,25,27,31 No studies delineated LD placement by CSF leak flow rate (high vs low); however, 2 series con- tained only patients with high-flow leaks (Zanation et al 31 and Garcia-Navarro et al 25 ). In regard to potential differences in patient risk factors between study arms, only Albu et al 22 conducted a formal comparison of patient characteristics and comorbidities between 1 LD and –LD groups, in which they found no sig- nificant differences. In all other studies, this comparison was not performed, nor was this information available for extraction. Assessment for Heterogeneity and Publication Bias There was no significant between-study heterogeneity ( Q = 13.1, P . .10, I 2 = 16.0%). Figure 2 shows the funnel plot for each included study, with the dashed line representing 95% confidence intervals (95% CIs). Linear regression of the funnel plot performed with Egger’s test demonstrated P . .05, suggestive of insignificant publication bias. Study Characteristics and Results The data of eligible studies are presented in Table 2 . Of the 12 studies, 1 had level of evidence 1b, and the remaining had level of evidence 4. All studies with the exception of 1

266