RESUMO
BACKGROUND AND OBJECTIVE: Pleuroscopy with pleural biopsy has a high sensitivity for malignant pleural effusion (MPE). Because MPEs tend to recur, concurrent diagnosis and treatment of MPE during pleuroscopy is desired. However, proceeding directly to treatment at the time of pleuroscopy requires confidence in the on-site diagnosis. The study's primary objective was to create a predictive model to estimate the probability of MPE during pleuroscopy. METHODS: A prospective observational multicentre cohort study of consecutive patients undergoing pleuroscopy was conducted. We used a logistic regression model to evaluate the probability of MPE with relation to visual assessment, rapid on-site evaluation (ROSE) of touch preparation and presence of pleural nodules/masses on computed tomography (CT). To assess the model's prediction accuracy, a bootstrapped training/testing approach was utilized to estimate the cross-validated area under the receiver operating characteristic curve. RESULTS: Of the 201 patients included in the study, 103 had MPE. Logistic regression showed that higher level of malignancy on visual assessment is associated with higher odds of MPE (OR = 34.68, 95% CI = 9.17-131.14, p < 0.001). The logistic regression also showed that higher level of malignancy on ROSE of touch preparation is associated with higher odds of MPE (OR = 11.63, 95% CI = 3.85-35.16, p < 0.001). Presence of pleural nodules/masses on CT is associated with higher odds of MPE (OR = 6.61, 95% CI = 1.97-22.1, p = 0.002). A multivariable logistic regression model of final pathologic status with relation to visual assessment, ROSE of touch preparation and presence of pleural nodules/masses on CT had a cross-validated AUC of 0.94 (95% CI = 0.91-0.97). CONCLUSION: A prediction model using visual assessment, ROSE of touch preparation and CT scan findings demonstrated excellent predictive accuracy for MPE. Further validation studies are needed to confirm our findings.