Study objectives: To determine the predictive accuracy of pH for identifying patients with malignant pleural effusions who will fail pleurodesis.
Design: Analysis of published and unpublished individual patient-level data retrieved from a MEDLINE search and correspondence with primary investigators.
Study selection: Studies that reported pleural fluid pH values and outcomes of pleurodesis for patients with malignant pleural effusions.
Data collection and analysis: Primary investigators supplied data for 433 patients. Receiver operating characteristic analysis and logistic regression estimated the predictive accuracy, decision thresholds, and value of pleural fluid pH compared with other clinical factors. The primary investigations were graded for study design.
Results: Pleural fluid pH was the only independent predictor of pleurodesis failure (odds ratio, 4.46; 95% confidence interval [CI], 2.69 to 7.45; p < 0.0001) and had an area under the receiver operating characteristic curve (decision threshold, < or = 7.28) of 0.671 (95% CI, 0.624 to 0.715). The pH model fit the data well (p = 0.48) with the probability of pleurodesis failure increasing as pH decreased; specificity and negative predictive values for pleurodesis failure exceeded 90% and 80%, respectively, with a positive predictive value of 45.7% at pH values < or = 7.15. The primary studies had several important design limitations.
Conclusions: Using patient-level data, this study showed that pleural fluid pH has only modest predictive value for predicting symptomatic failure and should be used with caution, if at all, in selecting patients for pleurodesis. The limitations of the primary studies and low predictive accuracy should be considered when using pleural fluid pH for patient care.