PBW vs FVC - Predicted Body Weight versus Forced Vital Capacity in Mechanical Ventilation

This study compared two methods of estimating lung volumes for mechanical ventilation settings: predicted body weight (PBW), currently used in clinical practice, versus predicted forced vital capacity (PFVC) calculated using spirometric reference equations. We found that PBW systematically overestimates lung volumes in older, shorter, female, and non-White patients compared to younger, taller, male, and White patients, which leads to these groups receiving relatively higher tidal volumes during mechanical ventilation. This algorithmic bias was associated with increased mortality risk, suggesting that more personalized approaches to adjusting ventilators could lead to better outcomes in the ICU.

Initial Preprint: SSRN

Published Paper: The Lancet Respiratory Medicine

Repository Structure

Preprint Analysis Files

• 1 - VTFVC - Simulated data.Rmd - Generates simulated patient data and calculates PBW-based and FVC-based tidal volumes across different demographics
• 2 - VTFVC - RICU data import.Rmd - Imports and processes data from MIMIC-IV and eICU databases
• 3 - VTFVC - Incusion criteria.Rmd - Applies inclusion/exclusion criteria to create the final analytical cohort
• 4 - VTFVC - Table 1.Rmd - Creates baseline characteristics table (Table 1) for the study population
• 5 - VTFVC - Regression and sensitivity analyses.R - Performs primary regression analyses and sensitivity analyses
• 6 - VTFVC - Regression Tables.Rmd - Formats regression results into publication-ready tables
• 7 - VTFVC - Survival curves.R - Generates survival curves and related analyses

Final Manuscript Analysis

• PBW vs FVC analyses and figures.qmd - This file includes generates figures and analysis used in the final manuscript. Notable changes from the original preprint include a direct comparison of PBW and predicted FVC (rather than VT/PBW and VT/FVC) and additional investigation of driving pressure and normalized elastance.

Requirements

R Version

• R 4.0.0 or higher recommended

Key R Packages

Data Manipulation & Core

• tidyverse - Core data science packages
• data.table - Fast data manipulation
• lubridate - Date/time handling

Statistical Analysis

• lme4, lmerTest - Mixed-effects models
• survival, survminer - Survival analysis
• marginaleffects - Marginal effects calculations
• broom, broom.mixed - Tidy model outputs

Visualization

• ggplot2 (via tidyverse) - Core plotting
• patchwork - Combine plots
• ggpubr, ggbeeswarm - Publication-ready plots
• ggdag, dagitty - DAG visualization

Specialized

• rspiro - Spirometry calculations
• furrr - Parallel processing
• officer, flextable - Document generation

Running the Analysis

For the Preprint Analysis:

1. Run files 1-7 in numerical order
2. Each R Markdown file can be knitted to generate outputs
3. Results and figures will be saved to results/ and figures/ directories

For the Published Paper Analysis:

1. Ensure the preprint analysis has been completed first
2. Run PBW vs FVC analyses and figures.qmd to generate additional analyses

Output

• Tables: Generated as Word documents and CSV files in the results/ directory
• Figures: Saved as high-resolution images in the figures/ directory
• Statistical Results: Regression outputs and model summaries saved as RDS files

Data Sources

This analysis uses data from:

• MIMIC-IV: Medical Information Mart for Intensive Care IV database
• eICU: eICU Collaborative Research Database

Both databases are available from PhysioNet and are subject to PhysioNet's data use agreements.

Reference

Sarma, et al. Evaluating the generalisability of formulas used to set tidal volumes in mechanically ventilated patients: an observational, multicohort, retrospective study. Lancet Respiratory Medicine 2025