Simulation and validation for count-based binary decision of target blood pressure achievement in home blood pressure monitoring data analysis for clinical practice

Abstract

Introduction Home blood pressure monitoring (HBPM) is a useful tool to identify hypertension and to decide whether a patient's blood pressure (BP) is controlled. The use of automatized oscillometric BP measurement devices has become increasingly popular with help of information technology and internet of things to the devices. However, applying HBPM to daily clinical practices is still challenging, because most patients with hypertension are in age groups not familiar to digital devices and internet and high BP criteria using average home BP values are often useless in outpatient clinics without easily accessible average BP calculation tools. Therefore, we developed a simple and straightforward method to interpret HBPM through counts of BP ≥135/85 mmHg. Methods We simulated 400 cases of HBPM using a random number generator function in statistical software. The simulated average home systolic BP (SBP) and its standard deviation (SD) were 125±15 mmHg and 12±5 mmHg and the number of HBP readings was 24 times. The simulated diastolic BP (DBP) was randomly selected to 50–75% of the SBP. The validation of the binary interpretation method was conducted using actual HBPM data from 386 subjects in a rural area of South Korea. Receiver operating characteristics curve analysis was conducted, and linear regression and logarithmic models were fitted between the numbers of home BP ≥135/85 mmHg and mean BP. Hypertension was defined with average home BP ≥135/85 mmHg. Results In the simulated cohort, hypertension was presented in 197 cases (49.3%). The C-index of the numbers of BP readings ≥135/85 mmHg was 0.994 (95% confidence interval [CI] 0.990–0.998), and ≥12 of 24 BP readings ≥135/85 mmHg showed a sensitivity of 95.4%, a specificity of 95.1% and an accuracy of 95.3% for the diagnosis of hypertension. In validation cohort, the numbers of home BP measurements varied from 8 to 81 times. The validation cohort similarly showed that the C-index of the ratio between the number of high BP readings (≥135/85 mmHg) to the number of BP measurements (R-NHBP/NBP) was 0.985 (95% CI, 0.976–0.994) and the best accuracy was shown at R-NHBP/NBP of ≥0.45. R-NHBP/NBP of ≥0.5 showed a sensitivity of 0.957, a specificity of 0.907 and an accuracy of 0.927. The accuracy of the R-NHBP/NBP of ≥0.5 decreased as SD and the range of SBP increased, whereas it did not change with the number of measurements (Figure 1). R-NHBP/NBP <0.2 predicted normotension and R-NHBP/NBP >0.8 predicted hypertension in 95% confidence. Mean widths of the 95 prediction intervals for the average SBP and DBP were 18.2 mmHg and 12.6 mmHg, respectively (Figure 2). Conclusion Counting the number of BP ≥135/85 mmHg can provide accurate assessments for the BP levels. R-NHBP/NBP of ≥0.5 is a simple and accurate marker of high BP in HBPM, and R-NHBP/NBP could be a useful tool to assess BP levels in patients practicing HBPM. FUNDunding Acknowledgement Type of funding sources: None. Figure 1 Figure 2