Negative predictive value of r-ebus underestimated the probability of cancer: a meta-analysis of dichotomous vs multichotomous test approaches

SESSION TITLE: Lung Cancer Posters 5 SESSION TYPE: Original Investigation Posters PRESENTED ON: 10/11/2023 12:00 pm - 12:45 pm PURPOSE: Diagnostic tests can be dichotomous or multichotomous (if there are more than 2 possible results). Diagnostic bronchoscopy has been described using sensitivity, specificity, positive predictive value, and negative predictive value (NPV). This is a dichotomous formulation, with a procedure being either diagnostic of a particular disease (e.g., cancer) or negative. However, bronchoscopy can make multiple diagnoses (e.g., TB) so there are more than 2 possible results. This study aimed to assess whether a dichotomous or multichotomous formulation of bronchoscopy more accurately describes the probability of cancer (pCA) following a non-diagnostic bronchoscopy in patients with peripheral pulmonary lesions (PPL). METHODS: A previously published meta-analysis was used to identify studies of r-EBUS for diagnosis of PPL. Only studies that provided enough detail to calculate dichotomous parameters and multichotomous LRs were included. For the dichotomous formulation, 2 x 2 contingency tables were constructed to calculate prevalence of cancer, sensitivity, specificity, NPV, and the pCA following a negative test result. For the multichotomous formulation, bronchoscopy results were classified as positive for cancer, positive for another specific benign disease, or non-diagnostic. LRs for each category were calculated. We used Stata SE 17 metandi command using the method of DerSimon and Laird; and metan using a random effects model for estimates of LRs. RESULTS: A total of 7506 patients from 46 studies were included. Average cancer prevalence was 0.76. Using a dichotomous formulation, pooled sensitivity was 0.74 (95% CI (Confidence Interval), 0.71 – 0.77) and specificity was 0.97 (95% CI, 0.96 – 0.98). Using a multichotomous formulation, the LR for a non-diagnostic test was 0.49 (95% CI 0.42 – 0.58, I2 = 84.4%). The pCA given a non-diagnostic result using the dichotomous formulation was always less than the multichotomous formulation in all studies. The median difference between multichotomous and dichotomous pCA given a non-diagnostic result was 0.13 (IQR (interquartile range) 25-75, 0.06 – 0.24). Using a prevalence of 0.76, the pCA was 0.62 and 0.47 for the multichotomous and dichotomous methods respectively. CONCLUSIONS: LR methods avoid the information loss that occurs from dichotomizing the possible test results into just two bins. This study shows the using traditional dichotomous formulations to calculate NPV and pCA following a non-diagnostic test results in significant underestimation of the pCA, while the multichotomous LR method is accurate. Future studies of bronchoscopy should use the multichotomous method with LRs to calculate the pCA following a non-diagnostic result rather than NPV. CLINICAL IMPLICATIONS: The NPV underestimates the probability of cancer following a non-diagnostic bronchoscopy result. DISCLOSURES: No relevant relationships by Horiana Grosu grant relationship with Intuitive Surgical Please note: $20001 - $100000 by David Ost, value=Grant/Research Support Consultant relationship with Beckton Dickinson Please note: 2021 to present by David Ost, value=Consulting fee Consultant relationship with Johnson and Johnson Please note: 2021 to present by David Ost, value=Consulting fee No relevant relationships by Paula Sainz