Machine Learning-Based Identification of Patients with Elevated Central Venous Pressure Using Features Extracted from...

Area of Science:

• Biomedical Engineering
• Critical Care Medicine
• Machine Learning Applications

Background:

• Central venous pressure (CVP) is crucial for hemodynamic monitoring and fluid resuscitation in critically ill patients.
• Current CVP measurement via catheterization is invasive, time-consuming, and carries risks.
• There is a need for non-invasive methods to assess CVP.

Purpose of the Study:

• To investigate the feasibility of using machine learning (ML) to analyze non-invasive photoplethysmography (PPG) signals for elevated CVP detection.
• To develop and validate an ML model capable of distinguishing between normal and elevated CVP using PPG features.

Main Methods:

• A Light Gradient-Boosting Machine (LightGBM) model was trained on a large perioperative dataset (MLORD) of 1665 patients with simultaneous PPG and CVP waveforms.
• 843 PPG features per cardiac cycle were extracted, along with average and standard deviation features per patient.
• Recursive Feature Elimination with Cross-Validation (RFECV) selected 246 features; hyperparameters were tuned, and the model was validated using bootstrapping.

Main Results:

• The LightGBM classifier achieved a mean area under the receiver operating characteristic curve (AUC) of 0.79 (95% CI: 0.71-0.84).
• Mean accuracy was 0.71 (95% CI: 0.65-0.77), indicating good discriminatory power.
• The model successfully distinguished between normal (5–15 mmHg) and elevated (CVP > 15 mmHg) CVP.

Conclusions:

• PPG-derived features, when analyzed by ML, can effectively discriminate between normal and elevated CVP.
• This study demonstrates the potential of non-invasive PPG analysis as a surrogate for invasive CVP monitoring.
• These findings pave the way for developing novel, non-invasive CVP assessment tools in clinical practice.