Cross-sectional predictive models demonstrated acceptable discriminatory capacity for detecting hypertension with AUCs ranging from 0.70 to 0.89, with minimal differences between logistic regression and machine learning.
Systematic Review (n=4,384,100)
Do cross-sectional predictive models accurately detect undiagnosed hypertension in adults?
Simplified clinical predictive models perform nearly as well as complex machine learning approaches for detecting undiagnosed hypertension, offering valuable tools for population screening.
Effect estimate: AUC 0.70-0.89
Introduction: Hypertension (HTN) affects approximately 1.3 billion people worldwide, nearly half of whom remain undiagnosed, underscoring the urgent need for efficient diagnostic models to enable timely detection. This study aimed to identify and critically evaluate cross-sectional predictive models developed for diagnosing hypertension in adults globally. Methods: A systematic review was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) and TRIPOD guidelines, searching MEDLINE/PubMed, Scopus, Web of Science, and EMBASE from January 2000 to March 2024. We included observational studies that developed or validated cross-sectional models for adult HTN detection. The risk of bias was assessed using the PROBAST (Prediction Model Risk of Bias Assessment Tool). Results: Eight studies met the inclusion criteria, predominantly from Asia (four from China, one from Korea, one from Qatar, one from the United Arab Emirates, and one from Bangladesh). The models demonstrated acceptable discriminatory capacity area under the receiver operating characteristic curve (AUC) 0.70-0.89, with minimal differences between traditional statistical approaches (logistic regression) and machine learning methods. The most consistent predictors were age (present in >90% of models), body mass index (85%-90%), sex/gender (75%-80%), and diabetes (70%-75%). Laboratory biomarkers provided only marginal improvements in predictive performance (AUC increase of +0.02-0.04) compared with models based exclusively on clinical variables, raising concerns about their cost-effectiveness. Notably, no predictive models were identified for Latin American populations, despite the region's high prevalence of HTN. This absence highlights a critical research gap. Conclusion: Cross-sectional predictive models represent valuable tools for the detection of HTN, with simplified clinical models performing nearly as well as more complex approaches. Future research should prioritize the development of models tailored to underrepresented populations, integrating social determinants of health and adopting accessible formats to facilitate implementation in resource-limited settings.
Vera-Ponce et al. (Fri,) conducted a systematic review in Hypertension (n=4,384,100). Cross-sectional predictive models was evaluated on Discriminatory capacity (AUC) for hypertension detection (AUC 0.70-0.89). Cross-sectional predictive models demonstrated acceptable discriminatory capacity for detecting hypertension with AUCs ranging from 0.70 to 0.89, with minimal differences between logistic regression and machine learning.