To what extent can the chosen blood pressure measurement technique affect the outcomes of an observational survey?

The study represents a secondary analysis of EUROASPIRE survey data from the Czech Republic, a cross-sectional analysis of three pooled independent cohorts (EUROASPIRE III, IV and V examined in 2006–2007, 2012–2013 and 2016–2017) of patients with stable manifest coronary heart disease (CHD), in other words, baseline examination was done at least 6 months after its first manifestation. Patients aged less than 80 years and hospitalized for any of the following discharge diagnosis were retrospectively identified from hospital records. The diagnoses included: first coronary artery bypass grafting, first percutaneous transluminal coronary angioplasty and acute myocardial infarction or ischemia. Recruitment of patients started with the most recent hospital record and proceeded backward until the required sample of 600 subjects in each campaign (EUROASPIRE III, IV and V) was achieved. These patients were invited for an interview/clinical examination and responders (77.3% of the initially identified pool of patients) were included in the survey. All three campaigns of the EUROASPIRE survey were conducted in the same two centers in the Czech Republic: University Hospital in Pilsen and Department of Cardiology, Institute for Clinical and Experimental Medicine in Prague under almost identical protocol. Each interview/clinical examination took place 6–24 months after the qualifying index event. For the purpose of present analysis, we used a subsample of 913 subjects (65.6% of total sample). This subsample covers all subjects from EUROASPIRE III plus those who were interviewed in Pilsen center in the course of EUROASPIRE IV and V (serial BP measurements, necessary for this subanalysis was done only in one of both Czech centers of this project).

The standard protocol of EUROASPIRE (EA) survey was followed as described elsewhere [9,15]. Information on personal and demographic characteristics, personal and family history of CHD, lifestyle and pharmacotherapy were obtained. Height and weight were measured in light indoor clothes without shoes using SECA 701 scales and measuring stick (SECA, Hamburg, Germany). Waist circumference was measured using a tape measure. Breath carbon monoxide was measured by a SMOKERLYSER device (Bedfont Scientific, Upchurch, UK) to verify smoking status (with 10 ppm of breath carbon monoxide as the cut-off point). Venous blood samples were drawn after at least 12 h of overnight fast. Laboratory examinations included estimation of total and HDL cholesterol, triglycerides (TG) and glucose, and were performed in the central study laboratory of the respective EUROASPIRE survey [9,15]. LDL cholesterol was calculated using the Friedewald equation, in other words, LDL = total cholesterol – HDL – (TG/2.22) (all in mmol/l). Symptoms of depression and/or anxiety was ascertained using standardized self-reported Hospital Anxiety and Depression Scale (HADS) questionnaire (validated Czech version) [16].

In all BP measurement modalities, BP was measured in a sitting position and after at least 10 min rest, the patient should sat with the right arm on the table (the cuff should be in the same horizontal level, as patient's heart), palm facing upward and uncross his/her legs. Standardized adult cuff (16 × 30 cm) was used for patients with arm-circumference 22–32 cm, while an oversized cuff (16 × 36 cm) for those with arm-circumference larger than 32 cm. The cuff was applied to the upper arm so that the bottom edge of the cuff was positioned 1–2 cm above the elbow joint and the center of the bladder was over the brachial artery.

We use five different modalities of BP measurement (Figure 1): First, ‘per-protocol’ measurement was performed according to the defined EUROASPIRE survey methodology, in other words, at the very beginning of survey procedures (namely before taking blood sample) by study nurse. An automated oscillometric device was used (OMRON M5-I in EUROASPIRE III, while OMRON M6 in EUROASPIRE IV and V, OMRON Corp., Kyoto, Japan). BP readings were obtained at least twice with 5 min apart, but if the difference between both values exceeded 10 mmHg, repeated readings were realized (for the purpose of present analysis average of last two readings was used). Second measurement was realized by physician during interview at least 30 min after ‘per-protocol’ measurement. In the second BP measurement, we used auscultatory device (standard mercury sphygmomanometer with 2 mm of mercury precision in EUROASPIRE III and IV or NISSEI DM-3000 electronic sphygmomanometer [NISSEI, Tokyo, Japan] in EUROASPIRE V). The same approach as in ‘per-protocol’ BP measurement regarding number and BP values used for analysis was applied. Additional three BP measurement modalities were realized only in subsets patients. First, in a random subsample (named ‘cohort 2’) we repeated automated oscillometric measurement, similarly as in ‘per-protocol’ measurement and by the same study nurse, but at the end of the survey procedures. Second, in EUROASPIRE V survey we performed another oscillometric measurement using hybrid NISSEI DM-3000 device (which also enable to switch for oscillometric mode) obtained by (the same) physician during an interview (this subsample is marked ‘cohort 3’). Last BP measurement modality (done in ‘cohort 3’ again) represents unattended automated oscillometric (uAO) measurement using BPTru device (model BPM 200, BpTRU Medical Devices, Coquitlam [BC], Canada). It was set to measure BP six-times in 1-min intervals, while the patient had been seated alone in a separate room. The BPM 200 model of BPTru skips the first measurement and calculates the mean of the second to the sixth measurement.

BP values were compared intra-individually across all five BP measurement modalities (Figure 1). Descriptive statistics (means, medians and frequencies) and conventional statistical methods (Wilcoxon's paired test and Spearman's correlation) were used. We also tested whether any covariate might influence the intra-individual difference between per-protocol and auscultatory technique, by means of multiple step-wise logistic regression. For this purpose, conventional risk factors were dichotomized using cut-off points proposed by the 5th Joint European Guidelines for Cardiovascular Prevention [5]. Raised BP (failure to achieve BP target values) was defined as systolic BP ≥140 mmHg and/or diastolic BP ≥90 mmHg (irrespective of diabetes). HADS score for depression or anxiety ≥8 points was categorized as respective mood disorder present. Statistical analysis was performed using STATISTICA 8 (StatSoft, Inc., OK, USA) and STATA 8 (STATA Corp LP, TX, USA). The p-values < 0.05 were considered significant.

We compared automatic oscillometric measurement assessed at the very beginning of survey procedures with auscultatory measurement realized by physician during interview with patients; full sample (cohort 1) was used for this analysis. Per-protocol oscillometric BP was significantly higher than standard auscultatory technique, with median intra-individual difference between the both methods 9/6 mmHg (Table 2, top panel & Figure 2, panels A). Despite the fact that BPs were significantly correlated, the linearity and tightness of this association were relatively poor (Figure 2, panels B). Moreover, the differences between both methods were not symmetrically distributed either for systolic or for diastolic BP (see Bland–Altman plots, panels C of Figure 2). Use of oscillometric (per-protocol) technique over-rated prevalence of inappropriate BP control (defined as systolic BP ≥140 mmHg and/or diastolic BP ≥140 mmHg) by more than 24% (≈ 59 vs 35%), compared with auscultatory technique.

In a next step, we analyzed if any covariate (age, gender, current smoking, obesity, increased waist circumference, hypercholesterolemia, diabetes mellitus, anxiety, depression, treatment with specific drug classes, survey and center) may influence the difference between per-protocol and auscultatory BP value. In unifactorial analysis, we found that significant differences in either systolic or diastolic BP were found among age subgroups (11/8 vs 8/5 mmHg [p-values < 0.0001/<0.0001], in those aged ≥65 years vs <65 years, respectively) waist circumference categories (10/7 vs 8/5 [p- values 0.008/0.002], increased vs normal waist circumference, respectively) and survey (8/4 vs 11/6 vs 11/11 [p-values for trend < 0.0001/<0.0001], in EUROASPIRE III, IV and V, respectively). However, in multivariate analysis only age and current smoking group remained significant covariates of both, systolic and diastolic BP (see Supplementary Material for details). We also tested the potential effect of the observer (i.e., a physician who performed the auscultatory BP measurement). We did not find any significant influence of the observer on the established difference between per-protocol and auscultatory BP value, if adjusted for above-mentioned covariates (in multiple step-wise regression model; not in table). Moreover, since one observer was involved in all three surveys (EUROASPIRE III, IV and V), we used it as reference and compared with each of other observers, with no apparent effect again (see Supplementary Table).

Additional oscillometric BP measurement was done in a random subsample of 295 subjects (cohort 2), and we evaluated to what extent BP differ when registered at the very beginning (per-protocol) and at the end of survey procedures. Using the same device for both measurements (OMRON M6), BP somewhat decreased over time, giving median intra-individual difference 3/4 mmHg (Table 2). Spearman correlation coefficients were 0.70 (p < 0.0001) and 0.63 (p < 0.0001) for systolic and diastolic BP's, respectively. Also nonachievement of BP target was by repeated measurement oscillometric measurement (at the end) very similar to per-protocol rates (at the beginning), in other words, approximately 57 versus 52% (Table 2). In contrast, nonachievement of BP target was in this subsample only 28.5% (not in table), when auscultatory technique was used for definition.

In a subset of 202 patients (cohort 3, EUROASPIRE V campaign), we performed additional BP readings using another two different devices. First, we compared per-protocol oscillometric readings (OMRON) versus another type of oscillometric device (NISSEI), the latter measurement was done during interview with the physician. NISSEI device measurements were 8/10 mmHg lower than per-protocol BP values (Table 2). In contrast, comparing NISSEI oscillometric readings with auscultatory readings (both performed at the interview with physician), the difference between both measurements was substantially lower (-3/0 mmHg, auscultatory minus NISSEI, Table 2). Also concordance of both techniques was better (Figure 3) than in per-protocol versus auscultatory comparison (Figure 2).

Finally, we compared unattended automated oscillometric (uAO) device (BP Tru) with ‘per-protocol’ oscillometric and auscultatory measurement. Oscillometric per-protocol measurement was by 13/12 mmHg higher than the average of 5 reading from uAO device and over-rated prevalence of inappropriate BP more than twofold (58 vs 22%). Only auscultatory systolic measurements were higher than uAO-based values (with median difference 4/0 mmHg). Moreover, corresponding prevalence of BP ≥140/90 was the same using these two methods (Table 2). General concordance of both methods was relatively good (Figure 4).