PulseTrendingJournal ClubResearchersJournalsExplore
Instagram
HomeTrendingJournal ClubExplore
Synapse
⌘+K
Synapse
May 31, 2026Journal of Echocardiography

Explainable machine learning for estimation of elevated left ventricular filling pressure: a multicenter validation

View Full Paper
Ask AI
Bookmark
Share

Why the study?

Guideline algorithms often result in indeterminate LVFP, and machine learning methods lack interpretability, necessitating explainable models for clinical use.

Does an explainable machine learning model improve the estimation of elevated left ventricular filling pressure compared to guideline-recommended algorithms in patients undergoing echocardiography and right heart catheterization?

Population

956 patients undergoing echocardiography and RHC at three hospitals

Comparison

Two explainable machine learning models vs guideline-recommended algorithm

Design

Retrospective multicenter validation study

Key result

Explainable machine learning models significantly outperformed guideline-recommended algorithms in estimating elevated left ventricular filling pressure (AUROC 0.83 vs 0.72; p=0.016).

Authors

YNYutaka NakamuraNKNobuyuki KagiyamaSSSirish Shrestha

Discussion

Loading...

Member takes

Overview

Explainable ML may aid LVFP estimation when guidelines are indeterminate; leaves open prospective validation before clinical adoption.

Key Points

  • To develop an explainable machine learning model for estimating left ventricular filling pressure with patient-level interpretation.
  • Retrospective enrollment of 956 patients undergoing echocardiography and right heart catheterization across three hospitals.
  • Two extreme gradient boosting models were trained using data from 621 patients, estimating elevated pulmonary artery wedge pressure.
  • Models' performance was compared using external test data from another hospital with 335 patients.
  • 31.0% of patients had elevated pulmonary artery wedge pressure, and 42.7% were classified as indeterminate by the GL-algorithm.
  • Model 1 AUROC: 0.82 (95% CI 0.73–0.92); Model 2 AUROC: 0.83 (95% CI 0.75–0.91), both significantly outperforming the GL-algorithm AUROC of 0.72 (95% CI 0.60–0.83, p=0.020).
  • Model 2 performed equally well for indeterminate cases with visualized contributions of variables using SHAP.

Study Design

Type

Observational (n=956)

Multicenter

Yes

Structured PICO

Does an explainable machine learning model improve the estimation of elevated left ventricular filling pressure compared to guideline-recommended algorithms in patients undergoing echocardiography and right heart catheterization?

P
Population
956 patients who underwent echocardiography and right heart catheterization (RHC) at three hospitals within a median of 3 days.
I
Intervention
Explainable machine learning models (extreme gradient boosting) using variables from guideline algorithms (Model 1) or SHAP-selected variables (Model 2) to estimate elevated pulmonary artery wedge pressure (PAWP ≥ 18 mmHg).
C
Comparator
Guideline-recommended algorithms (GL-algorithm).
O
Outcome
Area under the receiver-operating characteristic curve (AUROC) for elevated LVFP (PAWP ≥ 18 mmHg).surrogate

Main Result

Effect estimate: AUROC 0.83 (95% CI 0.75-0.91)

Absolute Event Rate: 0.83% vs 0.72%

p-value: p=0.016

Explainable machine learning models using echocardiographic parameters significantly improve the estimation of elevated left ventricular filling pressure compared to conventional guideline algorithms, while providing patient-level interpretability.

Cite This Study

Nakamura et al. (2026) conducted an observational in Elevated left ventricular filling pressure (n=956). Explainable machine learning models (XGBoost) vs. Guideline-recommended algorithms was evaluated on Area under the receiver-operating characteristic curve (AUROC) for elevated LVFP (PAWP ≥ 18 mmHg) (AUROC 0.83, 95% CI 0.75-0.91, p=0.016). Explainable machine learning models significantly outperformed guideline-recommended algorithms in estimating elevated left ventricular filling pressure (AUROC 0.83 vs 0.72; p=0.016).

synapsesocial.com/papers/6a1bd5915783ba022b6fe3a2https://doi.org/10.1007/s12574-026-00738-x
View Full Paper
Ask AI
Bookmark
Share