Demographic subgroup-aware analysis of blood pressure predictive models

Q: What is the clinical evidence from this study?

Population: Patients in intensive care units requiring blood pressure monitoring (n=2315). Intervention: Machine learning models with signal quality assessment (WiSQI) vs. Baseline models without signal quality assessment (WoSQI). Primary outcome: Mean Absolute Error (MAE) for systolic blood pressure (SBP) estimation.

Q: What does this research mean for the field?

Incorporating signal quality assessment into photoplethysmography-based machine learning pipelines achieves blood pressure prediction accuracy meeting AAMI standards, with a mean absolute error of 3.36 mmHg for systolic and 1.76 mmHg for diastolic blood pressure. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Blessing Kolawole; Sinuo Fan; Yingjie Lao; Valencia Joyner Koomson

doi:10.1371/journal.pdig.0001687

October 15, 2026PLOS Digital HealthOpen Access

Demographic subgroup-aware analysis of blood pressure predictive models

View Full Paper

Ask AI

Bookmark

Share

Key result

Signal quality assessment cuts ML systolic blood pressure prediction error by ~6 mmHg.

n=2,315

Figures

Why the study?

Blood pressure estimation using PPG waveform analysis requires a demographic-aware approach due to individual physiological and age-related vascular changes affecting prediction accuracy.

Does integrating signal quality assessment and demographic-aware training improve the accuracy of blood pressure prediction models using PPG waveforms in ICU patients?

Key Points

To evaluate the predictive accuracy and demographic fairness of conventional machine learning models for cuff-less blood pressure estimation using photoplethysmography waveform analysis.
Extracted features from photoplethysmography waveforms in the MIMIC III waveform database to train three conventional machine learning models: Support Vector Regression (SVR), Random Forest (RF), and Gradient Boosting Regression (GBR).
Compared a baseline pipeline including all records (WoSQI) against an enhanced approach (WiSQI) that filtered records at a 70% Signal Quality Index threshold and applied demographic-aware partitioning across race, age, and gender.
On the independent held-out test set, the enhanced SVR model achieved an MAE of 3.36 mmHg (RMSE 4.98, R2 = 0.95) for systolic blood pressure and an MAE of 1.76 mmHg (RMSE 2.58, R2 = 0.94) for diastolic blood pressure.
In cross-validation, incorporating signal quality filtering reduced SVR MAE from 9.10 ± 0.14 mmHg to 3.51 ± 0.15 mmHg for SBP, and from 4.72 ± 0.14 mmHg to 1.84 ± 0.10 mmHg for DBP.
Subgroup evaluations showed lower MAE for younger cohorts (ages 18–25 and 26–35) and Asian and Hispanic subgroups, whereas older cohorts (ages 56–65 and 65–89) and Black and 'Other' subgroups exhibited higher error rates.

Structured PICO

Does integrating signal quality assessment and demographic-aware training improve the accuracy of blood pressure prediction models using PPG waveforms in ICU patients?

P

Population

2,315 adult ICU patients from the MIMIC-III database with concurrent PPG and ABP waveforms, utilized to train demographic-aware machine learning models for cuffless blood pressure prediction.

I

Intervention

Machine learning models (Support Vector Regression, Random Forest, Gradient Boosting) for cuffless blood pressure prediction, trained on features extracted from photoplethysmography (PPG) waveforms with a rigorous Signal Quality Index (SQI) threshold > 70%.

C

Comparator

Baseline machine learning models trained using all available PPG records without signal quality assessment (WoSQI approach).

O

Outcome

Mean Absolute Error (MAE) for systolic blood pressure (SBP) and diastolic blood pressure (DBP) predictions.surrogate

Main Result

Absolute Event Rate: 3.36% vs 9.1%

Integrating signal quality assessment and demographic-aware approaches significantly improves the accuracy of machine learning models for cuffless blood pressure estimation using PPG waveforms.

Limitations

Generalizability of findings from ICU cohorts to healthy populations is limited due to hemodynamic instability and medical interventions in critically ill patients.
Demographic imbalance with a heavily White and older population.
Generalizability of findings from ICU cohorts to healthy populations may be limited due to hemodynamic instability and medical interventions
Dataset is inherently biased with an older (70% aged 56-89) and predominantly White (78%) distribution

Implication

Signal quality assessment refines PPG-based cuffless BP models; leaves open prospective validation before clinical adoption.

Abstract

Blood pressure (BP) is an important cardiovascular health indicator, usually monitored non-invasively with a sphygmomanometer, which can be bulky and inconvenient for real-time monitoring. Blood pressure estimation using photoplethysmography (PPG) waveform analysis offers a promising alternative for intermittent or continuous cuff-less BP monitoring. However, changes in vascularization due to individual physiology and age can directly affect the morphology of the PPG waveform. Thus, prediction accuracy requires a demographic-aware approach. To address these challenges, we trained three conventional ML models for BP prediction and conducted a subgroup-aware analysis to evaluate model performance across individual and intersectional demographic groups. Additionally, we show that signal quality assessment is a pivotal step in the signal-processing pipeline before feature extraction to mitigate external perturbations (e.g., motion artifacts) in the PPG waveform. We achieve a mean absolute error (MAE) of 3.36 mmHg for systolic blood pressure (SBP) and 1.76 mmHg for diastolic BP (DBP) with models trained when signal quality assessment is incorporated.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Kolawole et al. (2026) studied Patients in intensive care units requiring blood pressure monitoring (n=2,315). Machine learning models with signal quality assessment (WiSQI) vs. Baseline models without signal quality assessment (WoSQI) was evaluated on Mean Absolute Error (MAE) for systolic blood pressure (SBP) estimation. Machine learning models incorporating signal quality assessment achieved a mean absolute error of 3.36 mmHg for systolic and 1.76 mmHg for diastolic blood pressure prediction.

synapsesocial.com/papers/6ad2187bfc66cec880534e33 https://doi.org/10.1371/journal.pdig.0001687

View Full Paper

Ask AI

Bookmark

Share

Demographic subgroup-aware analysis of blood pressure predictive models

Figures

Key Points

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Demographic subgroup-aware analysis of blood pressure predictive models

Figures

Key Points

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

More related papers

More from PLOS Digital Health

Keep reading

More related papers

More from PLOS Digital Health