Screening for Atrial Fibrillation During Automatic Blood Pressure Measurements

Abstract

Atrial fibrillation (Afib) contributes significantly to overall cardiovascular risk. Widespread screening for Afib in primary care is sometimes performed by palpation, but suffers from low accuracy and is dependent on clinician experience. Algorithms implemented on oscillometric blood pressure devices can detect Afib with high sensitivity and specificity, but information on factors affecting accuracy is scant. Concurrent diagnostic electrocardiogram (ECG) and oscillometry were measured in participants in ECG clinics at two sites. Root mean squared successive difference (RMSSD) and irregularity index (Irrx) were calculated from oscillometric data and used to train logistic regression classifiers. Monte Carlo cross validation with 20 splits was performed to estimate confidence intervals for mean sensitivity and specificity, with various weightings, in the absence or presence of ectopics, and with or without repeated measurements. 707 measurements, including 168 Afib, were collected from 569 participants with mean (standard deviation) age of 63 (16) years. Sensitivity/specificity of RMSSD and Irrx were 0.982/0.908 and 0.986/0.960 respectively when ectopics were included. Excluding ectopics from the data improved specificity by up to 5%. Nevertheless, based on this performance and after accounting for prevalence of Afib in the population aged over 60 years, and estimated costs of healthcare, oscillometric screening for Afib in this age group could return a positive net health-economic benefit.

Atrial fibrillation detection in primary care during blood pressure measurements and using a smartphone cardiac monitor

“…The accuracy of the PRV parameters (percentage of classifications that were correct) across cut-points reached over 90% (Supplementary Tables 1 – 3 ), which is in the range of values observed for this metric in previous AF diagnostic studies (nearly all on outpatient samples) of other BP monitors 4 . Further, the 95% confidence intervals for our AUC values encompassed the AUC values reported in a hospital-based study of the BP+ device 15 . Our work extends these prior findings as we evaluated this device in a different setting (primary care) using different analytical approaches: in combination with the KMCM and using both decision and precision-recall curve analyses.…”

“…Our work extends these prior findings as we evaluated this device in a different setting (primary care) using different analytical approaches: in combination with the KMCM and using both decision and precision-recall curve analyses. Additional novelty is derived from our use of parameters not assessed (sARV, SD and relative range) in these past studies 4 or examined each only in one prior study (RMSSD 15 and IPP 24 ).…”

“…The waveforms spanned approximately 10 s; thus analysis was performed on approximately 10–12 pulse intervals. Six measures were used: (1) standardised average real variability (standardised ARV; sARV), the mean of the absolute pulse-interval differences across consecutive beats (analogous to ARV in BP variability measurement 17 ), as a percentage of the mean of these intervals, (2) root mean square of successive differences (RMSSD, square root of the mean of the squared differences of the duration of successive pulse intervals) 15 , 18 , (3) standard deviation (SD) 21 – 23 , (4) coefficient of variation (CV; SD as a percentage of the mean pulse interval; reported as irregularity index in previous studies 15 , 18 , 22 , 23 ), (5) relative range, the range (maximum minus minimum) of all pulse intervals across beats divided by their mean (comparable to relative range in BP variability assessment 17 ) and, (6) irregular pulse percentage (IPP), the number of irregular pulses as a percentage of the total number of pulses 24 . For IPP, pulses were defined as irregular if they were > ± 15% 24 the duration of the mean beat.…”

“…We performed multiple subgroup analyses. First, we excluded people with a non-AF arrhythmia as this could reduce specificity 15 . Second, we omitted individuals with a rhythm that was paced as this could be regular in those with AF (reducing sensitivity).…”

“…Given the knowledge gap, we evaluated how well PRV measured using a BP monitor ( BP + device ; Uscom, Sydney, Australia) screens for AF in primary care. We hypothesised that this could detect AF accurately 15 , thereby making it an attractive screening method as PRV can be assessed during measurement of both BP and other parameters that predict cardiovascular events 16 – 18 . To build on past research, we assessed screening performance across a range of PRV thresholds for AF detection by this instrument.…”