Pleural Fluid Analytes Collected at an Academic Medical Center, February 2011 – January 2026
Description
This dataset accompanies a retrospective cohort study examining pleural fluid pH in adult thoracentesis specimens. The study had three objectives: (1) to estimate pleural fluid pH in specimens meeting prespecified non-pathological criteria, (2) to evaluate the diagnostic performance of pH alone for classifying exudative versus transudative effusions, and (3) to identify biochemical predictors of pleural fluid pH using linear mixed-effects models. Data were collected from (mostly) adult patients who underwent thoracentesis at UC Davis Medical Center in Sacramento, California, between February 2011 and January 2026. Pleural fluid pH was measured using a blood gas analyzer, and pleural fluid chemistry (cholesterol, lactate dehydrogenase [LDH], total protein, glucose) was analyzed using standard automated clinical laboratory analyzers. Effusions were classified as exudative if pleural fluid LDH exceeded 141 IU/L or cholesterol exceeded 55 mg/dL; specimens meeting neither criterion were classified as transudative. A subset of specimens was classified as non-pathological based on pH >7.45, thoracentesis volume <500 mL, and absence of documented pathological aetiology. The dataset comprises 2,315 pleural fluid specimens from 1,953 unique patients. Each record includes the pleural fluid pH, concentrations of cholesterol (mg/dL), LDH (U/L), total protein (g/dL), and glucose (mg/dL), the upper limit of normal for serum LDH, patient demographics (age at collection, sex, height, weight, BMI), and collection-to-result times (minutes) for pH and the fluid chemistry panel. A binary Condition variable classifies each specimen as non-pathological (0) or pathological (1). Identifying information has been removed, and patients have been assigned sequential de-identified subject numbers. Some patients contributed multiple specimens; these are ordered chronologically by collection date within each subject via the SampleNumber variable. Key findings include that non-pathological pleural fluid pH (median 7.49) was lower than historically cited values. Pleural fluid pH provided moderate discrimination between exudative and transudative effusions (AUC 0.742; 95% CI: 0.719–0.766). In mixed-effects modelling, LDH and glucose were the primary biochemical predictors of pH (marginal R² = 0.28; ICC = 0.75), while protein was not a significant predictor. Researchers using this dataset should account for within-subject correlation, as 15% of patients contributed more than one specimen (range: 1–5). Mixed-effects models or generalized estimating equations are recommended. Missing data are present for some biochemical and anthropometric variables and were not missing completely at random (Little's MCAR test p < 0.001); multiple imputation under the missing-at-random assumption is appropriate. The Variable Dictionary sheet in the accompanying Excel file provides variable names, labels, data types, and coded value descriptions.
Files
Steps to reproduce
This retrospective cohort study used data extracted from the electronic health record (EHR) at UC Davis Medical Center, Sacramento, California. The study population included adult patients who underwent thoracentesis between February 2011 and January 2026. Demographic variables (age, sex, height, weight, race, ethnicity) were obtained from the EHR. Body mass index was calculated from recorded height and weight. Living status and dates of death were obtained from institutional records for survival analyses. Pleural fluid specimens were collected during clinically indicated thoracentesis procedures. Pleural fluid pH was measured using a blood gas analyzer. Pleural fluid chemistry — including cholesterol, lactate dehydrogenase (LDH), total protein, glucose, albumin, amylase, triglycerides, and hematocrit — was analyzed using standard automated clinical chemistry analyzers. The upper limit of normal for serum LDH was recorded to facilitate LDH ratio calculations. Collection-to-result times were captured in minutes for each assay. Effusion classification followed a two-step approach. The primary classification used cholesterol and LDH criteria: effusions were exudative if pleural fluid LDH exceeded 141 IU/L or cholesterol exceeded 55 mg/dL, and transudative if neither criterion was met. Light's criteria classification was also recorded. A subset was classified as non-pathological if all of the following were satisfied: pH >7.45, thoracentesis volume <500 mL, and no identifiable pathological aetiology. Two reviewers independently evaluated candidate non-pathological specimens; inclusion required concordance. Statistical analyses were performed in R. Linear mixed-effects models (lme4 and lmerTest packages) with random intercepts for patient identifier accounted for multiple specimens per patient. Missing biochemical data (25–71% across analytes) were addressed using multiple imputation with predictive mean matching (mice package), generating 20 imputed datasets pooled using Rubin's rules. Little's MCAR test confirmed data were not MCAR, supporting multiple imputation under the missing-at-random assumption. Diagnostic performance was evaluated using ROC analysis (pROC package) with cluster bootstrap confidence intervals (2,000 iterations) resampled at the patient level. Model fit was quantified using marginal and conditional R² via the Nakagawa-Schielzeth method (MuMIn package). Specimens were excluded if assays were compromised by technical error, clinical documentation was insufficient, or key demographic variables were missing. To reproduce these analyses, researchers should use the de-identified dataset, employ mixed-effects models to account for within-subject correlation, and apply multiple imputation for the remaining missing data biochemical variables. The Variable Dictionary sheet defines all variable names, labels, and coded values.
Institutions
- University of California, DavisCalifornia, Davis