Key result
All current risk prediction models for incident hypertension or increased blood pressure have a high risk of bias and most lack external validation, with the Framingham model showing a pooled AUC of 0.758 across validation studies.
Why the study?
Models predicting the risk of developing essential hypertension or increased blood pressure can identify high-risk individuals, but a summary and assessment of models developed in the general adult population was needed.
Systematic Review (n=29)
Effect estimate: Pooled AUC 0.758 (95% CI 0.692-0.825)
Existing risk prediction models for incident hypertension have a high risk of bias and most lack external validation, indicating a need for better adherence to reporting guidelines in future studies.
Current hypertension risk models warrant clinical caution; challenges their use and leaves open rigorously developed, validated alternatives.
Background and aims Models that can predict the risk of developing essential hypertension or increase in blood pressure (BP) can be used to identify high risk individuals. We aimed to summarize and assess prediction models developed in the general adult population using longitudinal data, as well as any external validation of such models. Methods For this systematic review, we searched the literature on Medline and Embase for studies published between database inception and February 5, 2021. We conducted a narrative synthesis of all models and assessed the risk of bias (ROB) in included studies using PROBAST. We also performed a meta-analysis of all external validation studies validating the Framingham hypertension risk model. We excluded models based on cross-sectional data, or those based on specific patient populations. Results Our review includes 29 articles which contain 42 prediction models and 11 external validation studies of existing prediction models. Among model development studies, only five models performed both internal and external validation. Among the validation studies, only two existing models were externally validated by researchers other than the ones who developed the model. Most models had low ROB in the predictors and outcomes domains, and half had low ROB in the participants domain. However, all had high ROB in the analysis domain due to inappropriate handling of missing data and/or lack of adequate performance measures, which resulted in high overall ROB for all models. Conclusions All current risk prediction models predicting hypertension or increased blood pressure have high ROB and most have not been externally validated. New studies should aim to reduce their ROB using standard reporting guidelines and externally validated existing models.
No takes yet. Share an insight, caveat, or question.
Bajwa et al. (2023) conducted a systematic review in Incident hypertension or increase in blood pressure (n=29). Risk prediction models for hypertension was evaluated on Predictive performance (AUC) of the Framingham Hypertension Risk model in external validation studies (Pooled AUC 0.758, 95% CI 0.692-0.825). All current risk prediction models for incident hypertension or increased blood pressure have a high risk of bias and most lack external validation, with the Framingham model showing a pooled AUC of 0.758 across validation studies.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: