The authors of abstracts and studies not reporting with sufficient data were contacted to request additional information. Data extraction and quality assessment.  The same two investigators who performed the database searches also performed the relevant data extraction

independently. In order to resolve disagreement between reviewers, a third reviewer assessed all discrepant items, LBH589 order and the majority opinion was used for analysis. Relevant studies were further examined with QUADAS criteria again. To perform accuracy analyses, we extracted data on characteristics of studies and patients, measurements performed, and results. For each report, we extracted the following items: author; journal; year of publication; sample size; description of study population (age); study design (prospective, retrospective PD0325901 in vivo or unknown); patient enrollment (consecutive or not); inclusion and exclusion criteria, reasons for exclusions from the analysis. For each study, we recorded the number of true-positive, false-positive, true-negative, and false-negative findings for DWI or PET/CT in diagnosing pancreas lesions. Data synthesis and analysis.  The sensitivity and specificity of the techniques assessed

in a given study were extracted or calculated MCE using 2 × 2 contingency tables. We combined sensitivities and specificities across studies using a hierarchical regression

model.14 A fully Bayesian approach to model fitting was taken. This model allows more between- and within-study variability than do fixed-effect approaches, by allowing both test stringency and test accuracy to vary across studies.14 Uniform distributions were used as prior information for the specification of the unknown parameters of the hierarchical model. The inverse gamma prior was chosen for the between-study variance parameters. Different prior ranges that cover all plausible values were chosen for sensitivity analyses. Goodness-of-fit measures were computed for each diagnostic method to evaluate model fitting. The hierarchical regression model allows the calculation at the same time of the summary sensitivity (true positives) and specificity (1-false positives), taking into account the interdependence of these metrics. Moreover, the summary receiver operating characteristic (SROC) curve can be derived from the estimation of the parameters of the model. The SROC curve shows the summary trade-off between sensitivity and specificity across the included studies and the summary likelihood ratios. Likelihood ratios are also metrics that combine both sensitivity and specificity in their calculation.