CONTRIBUTION OF ARTIFICIAL INTELLIGENCE TO THE DECISION-MAKING PROCESS WHEN PERFORMING A HOME BLOOD PRESSURE MONITORING

Abstract

Objective: Analyze home blood pressure monitoring data with the help of algorithms used in artificial intelligence and compare this method to traditional statistical analysis to categorize a subject with respect to its blood pressure status. Design and method: From an anonymized database of self-measurements performed in hypertensive patients followed in a hypertensive specialist center, artificial intelligence algorithms were used (logistic regression, gradient boosting tree, neural network) and compared with usual statistical methods. The self-measurement protocol consisted of 6 sequences over 3 days with 3 measurements per sequence, and 2 sequences per day. The mean of the 18 measures was used as a baseline to categorize the decision-making process with the thresholds SBP > = 135 or DBP>= 85 to define hypertension status. Results: The data base included 3910 HBPM which generated over 110,000 values to enter the analysis model. The figure shows that the AI allows to confirm or deny ‘hypertension’ in 46% of the subjects, as soon as sequence 2 is performed and that 82% of the subjects have a definite diagnosis after 4 sequences. In addition, AI is systematically more efficient than the statistical method whose performance, at 70% of patients with a definite diagnosis, remains at a plateau after sequence 4. Conclusions: Artificial intelligence algorithms are more efficient than conventional statistical methods to categorize the blood pressure status of a subject after only 2 sequences of a Home Blood Pressure Monitoring. From this hospital database it appears that 4 sequences performed over 2 days are sufficient to confirm or deny the diagnosis of hypertension in more than 80% of subjects.