Bloodiness also was graded into 4 levels: none, absent blood cells; low, a few blood cells without affecting cytopathology diagnosis; moderate, partially obscured by blood cells but possible cytopathology diagnosis; high, obscured by blood cells leading to inadequate interpretation. Diagnostic yield was evaluated by accuracy,

sensitivity, and specificity, which were calculated by using a final diagnosis that was defined as a diagnosis obtained from the integration of all the results of cytopathology, biopsy, PS-341 or surgical pathology, or clinical observation after at least 12 months with necessary follow-up studies. The degree of cytopathologic atypia was graded into 5 levels: definite LBH589 chemical structure for malignancy, suspicious for malignancy, atypical, negative for malignancy, inadequate for diagnosis. An experienced cytopathologist (K.-T. J.), who was blinded to the use of suction during puncturing and the expression techniques, interpreted all of the slides. A 2 × 2 factorial design with a 2-way

analysis was used to evaluate effectiveness of the two separate techniques simultaneously; that is, the samples for which suction was applied during puncturing (expressed either by reinserting the stylet or air flushing) were compared with the samples for which suction was not applied, and the samples expressed by reinserting the stylet (with or without suction) were compared with the samples expressed by air flushing. Thiamet G The 2-way analysis was chosen because

the 2 techniques should be performed separately in chronologic sequence and would not interact with each other, although all of these 4 methods were to be performed in every patient. Also, it should be mentioned that there were no interactions involving other technical factors such as the order of sampling and the needle size, excluding patient factors that were uncontrollable in nature. We calculated descriptive statistics for sensitivity, specificity, accuracy, and the distribution of cellularity and bloodiness. The mean values and standard deviation (SD) of continuous variables were presented as mean (SD) and categorical variables as frequencies. To calculate P values for 2-way comparisons between each method, we used generalized estimating equations with logit-link and unstructured correlation matrix, taking into account the correlated nature of the 4 samples obtained from the same patient. P values were further adjusted for multiple comparisons, and odds ratios (OR) with confidence intervals (CI) were calculated where relevant.