Predictive capacity of events in hypertensive patients using artificial neural network analysis from ambulatory blood pressure monitoring data compared with clinical risk stratification

Background There is no evidence available on the comparison of the predictive value of serious events in hypertensive patients using artificial neural network (ANN) analysis of ambulatory blood pressure monitoring (ABPM) data versus clinical risk stratification (CRS). Methods ABPM studies were analyzed that included a total of 27 measurements per each study: mean systolic, diastolic, mean, pulse and 24-h mean blood pressure and heart rate; daytime and nighttime mean pressures and heart rate; standard deviations of pressures and heart rate; circadian rhythm. The dependent variable was the combined end point of occurrence of death, stroke, acute myocardial infarction, heart failure and kidney disease during the follow-up..For clinical risk stratification, the argentine consensus on hypertension was used as a model. The discriminative capacity of the clinical endpoint with ANN-ABPM and the CRS were evaluated through the analysis of the area under the ROC curve. Both areas under the ROC curve were compared by De Long test. SPSS 23.0 Statistics was used for statistical analyses and NR modeling. Results Data from 491 studies were analyzed in total; mean age of the population: 69 ± 14 years, 53 % female, 11.6 % diabetic, 51 % dyslipidaemic, body mass index (mean 26 ± 4), 14.3 % smokers. Median follow-up was 6.6 years (95% CI 4.5-8). The ANN model with the best predictive ability was the Multilayer Perceptron model with a hidden layer; neuronal architecture (27/7/2) See figure 1. Nocturnal sistolic blood pressure had an independent standardized significance of 100% for algorithm performance. The area under the ROC curve for the endpoint by neural network analysis with ABPM variables was 0.810 (95% CI 0.77-0.90), for risk stratification it was 0.67 (95% CI 0.56-0.77) De Long's test p < 0.001. SEe figure 2 Conclusions We observed a higher discriminative capacity for follow-up events using ANN analysis with the ABPM variables compared to conventional clinical risk stratification. This observation raises a research hypothesis to be validated prospectively with a larger multicenter study. The addition of stratification risk from RNA analysis should improve decision making in hypertensive patients.