Controversy in Hypertension: Con-side of the argument using AI for HTN diagnosis and management

Abstract

Artificial intelligence (AI) applications in hypertension diagnosis and management are described as promising despite limited evidence of safety and efficacy. This perspective examines the substantial risks of premature AI implementation in hypertension care. We reviewed the current evidence base for AI approaches in hypertension, analyzed data quality issues in electronic health records, examined regulatory frameworks, and considered the readiness of health care systems and stakeholders. AI approaches for hypertension lack demonstrated efficacy and safety from randomized clinical trials. Training data from electronic health records suffer from poor quality control, with blood pressure measurements differing from true resting values. Clinicians routinely disregard recorded measurements, obtaining supplemental data inaccessible to algorithms. The Microsoft Aurora-Blood Pressure study, the most comprehensive evaluation of cuffless blood pressure technology, failed to meet accuracy standards except in highly constrained circumstances. Current regulatory frameworks struggle with AI's evolving nature, and informed consent becomes meaningless when clinicians cannot explain algorithmic decision-making. Accountability for AI-generated errors remains unclear, creating new patient safety risks. Infrastructure barriers prevent implementation in low-resource settings where AI could theoretically address personnel shortages. While AI may eventually transform hypertension care, current evidence does not support widespread adoption. The substantial risks-including poor data quality, regulatory gaps, accountability concerns, and potential skill atrophy among clinicians-outweigh unproven benefits. We recommend moving slowly and requiring rigorous clinical trial evidence before implementing AI systems in hypertension diagnosis and management.

Keywords: blood pressure; calibration; goals; humans; hypertension.