PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
Synapse
⌘+K
Synapse
September 17, 2026Journal of Medical Internet ResearchOpen Access

Evaluation of a National Health Service Machine-Learning Model for Hypertension Case-Finding: Retrospective Cohort Study

View Full Paper
Ask AI
Bookmark
Share

Why the study?

A substantial proportion of adults with hypertension remain undiagnosed, limiting opportunities for early intervention, motivating evaluation of a predictive model for case-finding.

Does a machine-learning model accurately identify undiagnosed hypertension in adults without a prior diagnosis compared to standard logistic regression?

Population

1,802,920 individuals aged 16 years or older without prior hypertension diagnosis in North West London

Comparison

Machine-learning predictive model vs more interpretable logistic regression model

Design

Retrospective cohort study

Key result

A machine-learning model for hypertension case-finding yielded an overall sensitivity of 62.7% (95% CI 62.5-62.8) and specificity of 60.7%, with low positive predictive value.

Authors

GIGloria IhenetuPublic Health EnglandAAAhmad AlkhatibGeorgetown UniversityVNVesselin NovovImperial College London

Discussion

Loading...

Member takes

Overview

Modest sensitivity, low PPV, and demographic biases caution against clinical use; leaves open whether complex ML outperforms simpler regression for hypertension case-finding.

Key Points

  • To independently evaluate the predictive performance, demographic variation, and practical clinical utility of a National Health Service machine-learning model designed to identify undiagnosed hypertension.
  • Retrospective cohort study conducted among 1,802,920 individuals aged 16 years or older registered with general practices in North West London without prior hypertension diagnoses from May 2023 to May 2024.
  • Assessed model classifications against clinical records and blood pressure measurements using logistic regression to evaluate performance across demographic groups and against a simpler regression approach.
  • The model achieved an overall sensitivity of 62.7% (95% CI 62.5–62.8) and specificity of 60.7% (95% CI 60.5–60.8), with positive predictive values spanning 31.5% (95% CI 31.2–31.8) to 42.9% (95% CI 42.5–43.2) and negative predictive values spanning 77.6% (95% CI 77.2–77.9) to 84.9% (95% CI 84.7–85.2).
  • Predictions exhibited marked age and racial divergence: 96.2% (58,951/61,281) of patients aged 70–79 were predicted to have hypertension versus 0.08% aged 20–39, with higher sensitivity observed in older, Black, and more socioeconomically deprived populations.
  • The predictive accuracy of the complex machine-learning algorithm did not demonstrate superiority over a standard, interpretable logistic regression model.

Study Design

Type

Cohort (n=1,802,920)

Multicenter

Yes

Structured PICO

Does a machine-learning model accurately identify undiagnosed hypertension in adults without a prior diagnosis compared to standard logistic regression?

P
Population
1,802,920 individuals aged 16 years or older with no prior diagnosis of hypertension, evaluated retrospectively between May 2023 and May 2024.
E
Exposure
Machine-learning predictive model for hypertension case-finding
C
Comparator
More interpretable logistic regression model
O
Outcome
Predictive performance (sensitivity, specificity, positive predictive value, negative predictive value) against recorded hypertension status using medical diagnoses and blood pressure records

A machine-learning model for hypertension case-finding demonstrated modest sensitivity and specificity with low positive predictive value and significant demographic variation, suggesting simpler, more transparent models may be preferable.

Limitations

  • Low positive predictive value
  • Significant proportion of true cases remained undetected
  • Considerable variation in performance associated with demographic characteristics

Cite This Study

Ihenetu et al. (2026) conducted a cohort in Undiagnosed hypertension (n=1,802,920). Machine-Learning Model for Hypertension Case-Finding vs. Interpretable logistic regression model was evaluated on Sensitivity for predicting hypertension (95% CI 62.5-62.8). A machine-learning model for hypertension case-finding yielded an overall sensitivity of 62.7% (95% CI 62.5-62.8) and specificity of 60.7%, with low positive predictive value.

synapsesocial.com/papers/6aabb7975f706d05830e6c0fhttps://doi.org/10.2196/87084
View Full Paper
Ask AI
Bookmark
Share