Abstract
OBJECTIVE:The goal of the study was to assess the performance of artificial intelligence in predicting upper limits of normal (ULN = 95th percentile for a given age and height) office blood pressure (BP) in children whose ULN varies with age and height.
DESIGN AND METHOD:The most recent pediatric normative office BP data (Flynn J, 2017:140:e2017) were used as a training data set for the Fuzzy Rules Based System (FRBS), to generate rules for office BP prediction. FRBS was then applied to office BP measured in 756 patients (405 boys, 351 girls) aged 4.3 to 16.9 years (median = 14.1), who were seen in the clinic from 2012 to 2018; median height was 160 cm (range = 117 to 187). Systolic and diastolic BP ULN predicted by FRBS (PredSBP, PredDBP) were compared with calculated ULN (CalcSBP, CalcDBP) using descriptive statistics, correlation analysis and Bland-Altman statistics.
RESULTS:Systolic and diastolic BP ULN ranged from 109 to 138 mmHg and from 68 to 86 mmHg respectively. The mean ± SD difference between PredSBP and CalcSBP was 0.45 ± 0.89 mmHg. Similarly, the PredDBP differed from CalcDBP by 0.18 ± 0.67 mmHg. The correlation coefficient between predicted and calculated SBP and DBP was 0.99 and 0.98 respectively. The mean bias (on Bland-Altman analysis) between PredSBP and CalcSBP was −0.45 mmHg, with lower and upper limits of agreement (LOA) ranging from −2.19 to +1.29 mmHg. The bias for PredDBP and CalcDBP was even lower (mean bias = −0.18 mmHg, LOA = −1.49 to +1.13 mmHg).
CONCLUSIONS:The Fuzzy Rules Based System accurately predicted the upper limits of systolic and diastolic office blood pressure readings for children of different ages, heights and blood pressures.