Likelihood of neoplasia for diagnoses modified by probability terms in canine and feline lymph node cytology: How probable is probable?

Background: Descriptive probability modifiers are used often to convey the uncertainty of a pathology diagnosis, but they also contribute to ambiguity in communication between pathologists and clinicians. Objectives: Our goal was to determine the frequency and use of probability modifiers in canine and feline lymph node cytology diagnoses, and to determine the actual likelihood of neoplasia for diagnoses with and without modifiers, based on the histologic outcome. Methods: Canine and feline lymph node cytology and histology reports over an 11-year period (2001-2011; n = 367) were evaluated retrospectively. Diagnoses were categorized as neoplastic/malignant (lymphoma, metastatic) or non-neoplastic/benign. The frequency and type of modifier, and the sensitivity, specificity, and predictive values for neoplasia were determined for modified and unmodified diagnoses using histology as the gold standard. Results: Ninety-one of 367 (24.8%) cytology diagnoses were modified by probability terms, including 25/204 (12.2%) diagnoses of non-neoplastic lesions and 66/163 (40.5%) diagnoses of neoplasia. In addition, 26 unmodified diagnoses of neoplasia were followed by a probability phrase indicating specific tumor type. Based on the histologic outcome, modified diagnoses had higher sensitivity (87.3%, confidence interval [CI] 75.5, 94.7%) but lower specificity (50.0%, CI 32.9, 67.1%) for neoplasia than did unmodified diagnoses (60.6 and 100%, respectively; P < 0.0001, Chi square). Modified phrases indicating the probability of a specific tumor type were accurate in 22/26 (84.6%) cases. Positive predictive values for neoplasia were 100% (CI 96.2, 100%) for unmodified and 72.7% (CI 60.4, 83.0%) for modified diagnoses. Negative predictive values were 65.9% (CI 58.5, 72.8%) for unmodified and 72.0% (CI 60.4, 83.0%) for modified diagnoses. No significant difference was found in the likelihood of neoplasia for individual terms used to modify a cytologic diagnosis except for "cannot rule out" (P = 0.0368). Conclusions: Most modified diagnoses of cancer in canine and feline lymph node cytology have a 60-83% likelihood of neoplasia based on histologic outcome, compared with 96-100% for unmodified diagnoses. Non-neoplastic lesions, regardless of modifiers, have a 12-49% likelihood of neoplasia. A limited number of risk categories based on these likelihoods may be a more effective and accurate way to communicate the risk of malignancy in lymph node cytology.