Comparisons of home and daytime ambulatory blood pressure measurements

The characteristics recorded for the 366 consecutively recruited participants are summarised in Table 1. Of these, 96 were excluded because they failed to follow instructions to make 3 triplicate HBPMs per day and/or contributed fewer than 27 measurements in total (Figure 1). The remaining 270 participants provided 5997 triplicates, − an average of 22.2 triplicates (out of a maximum of 30) per participant. Table 2 indicates the distributions of values of systolic and diastolic HBPM and heart rate (HR) for the 1st, 2nd and 3rd values in the total of 5997 triplicates. Table 3 indicates the corresponding distributions for the 1st triplicate contributed by each of the 270 participants. Friedman testing on the data from all triplicates (Table 2 indicated differences, significant at the p < 0.001 level, between the 1st and second and the 1st and 3rd readings. The general rule was that 1st systolic and diastolic HBPM and HRs were greater than the 2nd and 3rd values, but there was sufficient variability for the general rule to be inapplicable in about one third of the triplicates.

When all participants were considered, maleness was associated with a significantly higher median HBPM than femaleness, even when age, height and weight were allowed for, but this was not so for the 175 patients who were selected for ABPMs. In the sample of 347 participants patients, the ratio of males to females was 1.012:1. In the sample of 175 selected by the blood pressure criterion, the ratio was 1.1875 to 1. Details are given on the webpage.

Of the participants whose average HBPM exceeded 125/80, 175 had reasonably complete BPM sets with a mean of 45 systolic HBPM daytime readings (15 triplicates) per participant and 63 systolic ABPM daytime readings (Figure 1).The distributions of the HBPMs over up to 10 days and of ABPMs over up to 18 hours were compared in terms of their minimum, 1st, 2nd and 3rd quartile, and maximum values. The five indices of ABPM were plotted against the corresponding indices of HBPM in Figure 2. The minima are scattered mostly below the line of equality, but without much adherence to its slope. The first quartile, median and third quartile adhere quite well to the slope of the line of equality, the first quartile being closest to it and the median and third quartile being displaced progressively above it. The maxima are scattered, mostly above the line of equality, but without much adherence to it. The Pearson correlation coefficients show a modest correlation between the respective first quartiles, medians and third quartiles, but weak correlation between the minima and maxima. The dotted rectangles enclose the ranges ABPM (vertical) and HBPM (horizintal) observations in each plot. They are not strikingly different between plots and between ABPM and HBPM.

Further data on the distributions of ABPM and HBPM in terms of the median values of their five indices of distribution (Minimum, 1st Quartile, Median, 3rd Quartile and Maximum) are placed on the above mentioned website.

In each sub-plot of Figure 2, the vertical distance between each plotting symbol and the line of equality is the difference between ABPM and HBPM (ABPM – HBPM) for the corresponding index of distribution for that participant. Within each dotted rectangle in each sub-plot there will be a set of indices of distribution (Minimum, 1st Quartile, Median and 3rd Quartile, Maximum) for ABPM minus HBPM. These are shown in Figure 3. The vertical distance between the uppermost and lowermost lines shows the span between the maxima and minima of the respective indices of distribution. It is of the order of 70 mm Hg for the 1st Quartiles, Medians and 3rd Quartiles, but increased to about 130 mm Hg. For the 1st Minima and about 170 mm Hg for the Maxima. The Minima tended to be distributed towards negative values of (ABPM-HBPM), but towards progressively more positive values for the Medians, 3rd Quartiles and Maxima.

There is disagreement between guidelines for self-measured BPMs at home: (i) the European Society of Hypertension [1] and the 2012 Canadian Hypertension Education Program [2] recommend duplicate HBPMs in the morning and evening; (ii) the American Society of Hypertension [3] recommends triplicate HBPMs [4]; (iii) the Japanese Society of Hypertension recommends at least one HBPM [5]. As reported here and elsewhere [6, 7], the 1st HPPM in a triplicate tends to be higher than the 2nd and 3rd. The differences are quite small. The statistical significances attached to the values in Table 2 and Table 3 arise more from their consistency than from their size. They amount, on average, to 3 – 4 mmHg for systolic BP, 1 mmHg for diastolic and 1 – 2 mmHg for the heart rate. These amount respectively to about 18–25%, 9% and 9–10% of the associated standard deviations. Inverting the simple power calculations to obtain numbers of triplicates needed to have 80% power to attach significance at the 5% level gives 268 – 475 triplicates for systolic BP, 2077 – 2453 for diastolic, and 393–2077 for heart rate. The repeated measures element of the Friedman test that was applied would have provided somewhat more discriminating power because it separated differences in measurement within triplicate from differences between triplicates. However the extra discriminating power was not sufficient for the significance of the differences in diastolic pressure and heart rate that were seen for the 5997 triplicates represented in Table 2 to be apparent also in the 270 triplicates in Table 3.

Nonetheless, the differences between triplicates have been sufficient in at least one study for the diagnosis of “hypertension” to be made more frequently with the first than with the mean of the 2nd and 3rd [8]. Consequently, they may have some effect on selecting patients for inclusion in trials of antihypertensive medication, or conceivably commencement or modification of anti-hypertensive therapy. However it is hardly conceivable that patients would be started on such therapy solely on a BPM if that measurement was so close to the chosen criterion that the differences in measurement in a triplicate would be crucial. The decision would surely have to be based on a more comprehensive assessment of risk.

Some of the guidelines recommend discarding the higher HBPMs that are often measured initially, but this refinement had no impact in explanatory modelling of cardiovascular risk [9] or organ damage [10]. In keeping with the greater precision evident in Table 2 than in Table 3, one would expect the precision in estimating CVR to increase with the number of measurements contributing to the estimate of BP, though possibly only up to a point. For example, Niiranen et al. [10] found that the most precise predictions were achieved by including all values of duplicates measured twice per day over 7 days, though the measurements made over the first 3 days were more influential than those made later.

Timing of the HBPMs and choice of subject may also be important. For example, in a mixture of normotensives and hypertensive patients seen in the clinic, evening HBPMs predicted stroke more reliably than casual HBPM regardless of the number of measurements [11]. On the other hand, morning HBPMs predicted stroke more reliably than evening HBPMs in patients on hypertensive medication but not in normotensive patients [12]. Likewise, the variability in HBPM had an additional predictive power beyond mean systolic HBPM only in hypertensive patients [13].

Kawabe et al. [14] reported that the 1st systolic HBPM in morning triplicates had a larger coefficient of variation (CV) than the 2nd and was probably affected by the subject’s gender and smoking habits. In our Tables 2 and 3, the standard deviations did not seem very different between measurements in the triplicates.

The gender differences obtained in our small sample were consistent with those in the review by Reckelhoff [15].

ABPM was developed before HBPM and remains more copiously documented. It can also identify a larger number of features than either clinic or HBPM (e.g. “white coat hypertension”, night-time ABPM, the night-to-day BP ratio and “dipping” status - which are significant predictors of outcome [16]). It has therefore acquired ”gold standard” status though this does not necessarily mean that average or median ABPM is, by itself, the most accurate indicator of true normo-/hypertensive status. Unlike ABPMs, self-measured HBPMs can only be obtained while the patient is awake. The best agreement between ABPM and HBPM appeared to be in the morning between the first HBPM of the day and the mean of the 4 observations taken in the second hour after the patient had wakened [17]. We have restricted our comparison to daytime ABPMs and HBPMs.

A systematic review including six comparisons of HBPM and ABPM found that blood pressures over the criterion of 135/85 mmHg were obtained more frequently overall with HBPMs [5]. However, in the three studies with the largest numbers of HBPMs (29 to 56), the average ABPMs were higher than the HBPMs, to a lesser or greater degree. In one [18], the difference was similar to that found in our previous study [19] with a similar number of HBPM measurements.Our present results were from a data set that included an average of 45 HBPM and 63 ABPM measurements per participant. The comparison are in terms of the indices of distribution of the sets of observations of HBPM and ABPM in each subject across a set of subjects, as illustrated in our Figure 2 and Figure 3. For the 1st, 2nd and 3rd quartiles: (i) increases in individual ABPMs between patients tended to correlate reasonably well with increases of the corresponding HBPMs (r close to 0.7 in plots 2, 3 and 4 in Figure 2); (ii) ABPM tended to be greater than the corresponding HBPM by amounts that increased with increase in BP (by 10 mm Hg for the median of the 1st quartiles, 14 mm Hg for the 2nd and 19.5 mm Hg for the 3rd, Figure 3). At the extremes of the distributions: (i) there was less correspondence between the increases in ABPM and HBPM (r less than 0.3 for plots 1 and 5 in Figure 2); (ii) the spans of the distributions of the minima and maxima were wider, and the minima of the distributions of the 1st, 2nd 3rd quartiles were at negative values of (ABPM – HBPM, Figure 3).

It is a plausible conjecture that the behaviour of the measurements at the extremes of their range may be partly due to movement-related measurement artefacts in the ABPMs: the subject may be less inclined to interrupt his or her activity every 15 minutes for the duration of each ABPM whereas he or she might be more likely to stop and be settled for the HBPMs, at least by the time the 2nd and 3rd measurement in each triplicate has been made. The relative symmetry of both distributions in our Figure 2 about their medians is consistent with the observation of Lijarcio [20] who demonstrated, that for HBPM, the medians and means are more or less interchangeable – which was corroborated by detailed comparisons within our data set. The observations in Figure 2 are consistent with our previous observations [20] that, with duplicate HBPMs 3 times per day for a week, the coefficient of variation of the HBPMs was smaller than that of daytime ABPM obtained between 7 am and 10 pm on a single day, (4.2% compared with 5.5%).

Irrespective of whether it is the ABPM or HBPM that gives a “truer” picture of whether or not a subject should be labelled “hypertensive” at a particular time, we have argued previously on pragmatic grounds [19] that the smaller CVs associated with HBPM should make them more useful for detecting sooner and more reliably if, when, and how quickly a change in the subject’s underlying blood pressure has taken place, whether in response to institution or modification of treatment, or to other intercurrent influences. A further pragmatic advantage of HBPMs for longer-term monitoring is that because of resource limitations, facilities for ABPMs are usually only available over a shorter time and are more intrusive on activities of daily living.