Non-echo Planar Diffusion-Weighted Imaging in the Detection of Recurrent or Residual Cholesteatoma: A Systematic Review and Meta-Analysis of Diagnostic Studies

Abstract

We performed a systematic review and meta-analysis of patients with suspected recurrent cholesteatoma who underwent non-echo planar imaging (non-EPI) using diffusion-weighted magnetic resonance imaging (MRI), with surgery as the reference standard. We searched Medline, Google Scholar, and the Cochrane database for diagnostic test accuracy studies. The following prespecified subgroup analyses were performed: patient age, number of radiologists interpreting MRI, study design, and risk of bias. We used a bivariate model using a generalized linear mixed model to pool accuracies. Of the 460 records identified, 32 studies were included, of which 50% (16/32) were low risk of bias. The overall pooled sensitivity was 92.2% (95% CI 87.3-95.3%), and specificity was 91.7% (85.2-95.5%). The positive likelihood ratio was 11.1 (4.5-17.8), and the negative likelihood ratio was 0.09 (0.04-0.13). The pooled diagnostic odds ratio was 130.3 (20.5-240). Heterogeneity was moderate on visual inspection of the hierarchical summary receiver operating characteristic curve. Subgroup analyses showed prospective studies reporting higher accuracies (p=0.027), which were driven by higher specificity (prospective 93.1% (88.4-96.0%) versus retrospective 81.2% (81.0-81.4%)). There was no difference in subgroups comparing patient age (p=0.693), number of radiologists interpreting MRI (p=0.503), or risk of bias (p=0.074). No publication bias was detected (p=0.98). In conclusion, non-EPI is a highly sensitive and specific diagnostic test able to identify recurrent cholesteatomas of moderate to large sizes. This test can be considered a non-invasive alternative to second-look surgery.

Keywords: cholesteatoma; diagnostic accuracy; meta-analysis; non-epi; recurrent; second-look surgery.

Figure 1. PRISMA flow chart showing the study selection process.

PRISMA: preferred reporting items for systematic reviews and meta-analyses.

Figure 2. Characteristics, accuracies, and forest plots of the 32 included studies.

Source: [13-44]. TP: true positive, FP: false positive, FN: false negative, TN: true negative. Note: This image is the author's own creation.

Figure 3. Hierarchical summary receiver operating characteristic curve showing a pooled sensitivity of 92.2% and a pooled specificity of 91.7%. The 95% confidence intervals and 95% prediction intervals show a moderate distance between one another, indicating moderate heterogeneity.

HSROC: hierarchical summary receiver operating characteristic curve. Note: This image is the author's own creation.

Figure 4. Funnel plot constructed using Deeks model, showing no asymmetry on visual inspection or regression test (p=0.98).

ESS: effective sample size. Note: This image is the author's own creation.