Use of the median for the evaluation of blood pressure self-measurement (BPSM)

1. Introduction 

In recent years, there has been increasing use of semi-automatic electronic devices that allow patients to measure their blood pressure (BP) in their own homes, carrying out what is known as BP self-measurement (BPSM). This method of measuring BP prevents observer bias and the effect that the presence of a health professional may cause during the measurement [1], [2], [3]. In different studies, the mean values of a minimum number of home BPSM have been similar to those obtained through ambulatory monitoring of the BP, and it is therefore considered an exact and repeatable measurement method when carried out in standardized conditions [4] and with duly validated devices [5], [6], [7].

In clinical practice, different protocols have been used to carry out BPSM for a variable number of measurements, though generally equal to or greater than 12 [8]. Both the measurements obtained for SBP and those obtained for DBP must be added up in order to subsequently mathematically calculate the mean. Although some models of semi-automatic sphygmomanometers allow the values measured to be stored in the memory and even directly transmitted to the computer, this system is not always useful in clinical practice for reasons related to both the patient and the health professional. Under these conditions, making repeated, multiple calculations implies a significant investment in time for the doctor or nurse.

The median is a much simpler calculation parameter than the mean. In this study, we evaluated the use of the arithmetic median for the diagnosis of high blood pressure in both treated and untreated patients who carry out BPSM. We believe this is the first time that this parameter has been used, as we found no bibliographical references.

2. Patients and methods 

The study included 253 consecutive patients who were followed in primary care centers. The average age of the patients was 51.1±12.2 years. Forty-eight percent were male and 52% female. Some 61.7% of the patients were being treated with antihypertensive medication at the time of the study; the rest were receiving no treatment.

In all cases, BP measurements were taken at the center (two consecutive measurements with a 2-minute interval). Then, the patients were each handed an Omron model 705 semi-automatic self-measurement device, together with a sheet specifying the measurement protocol in the format of a table to be filled in and instructions for correct BPSM at home. All patients were instructed orally on these aspects. Patients were considered to be suffering from high blood pressure (when the patient was receiving no medication) or to not be adequately controlled (when the patient was already receiving pharmaceutical treatment) if they presented an average BP in BPSM of 135/85 mmHg or higher.

All patients carried out 12 measurements over 3 days. Each day two groups of two measurements were carried out with a 2-minute interval, in accordance with a published protocol [8]. For all patients, the mean SBP and DBP, as well as the median of both values (middle value of the interval of the measurements obtained), were calculated.

Continuous variables were expressed as mean±standard deviation. Frequency variables were compared by means of the chi-square test and the kappa concordance coefficient. Statistical analysis of the data was carried out using the computer program SPSS 10.0. For the calculation of sensitivity and specificity, the results of the arithmetic mean of the BP were taken as the benchmark test. Validation of the test was carried out by means of Bland-Altman plots, with their corresponding confidence intervals.

3. Results 

The calculated SBP was practically the same when the average of the means of all twelve measurements (143.04±18.53 mmHg) was used as when the average of the medians of the measurements was used (143.01±18.85 mmHg). There were also no statistically significant differences in the mean DBP when comparing the mean of the twelve measurements and the median of these measurements (mean of the medians 82.25±9.56; mean of the means 82.26±9.39 mmHg).

With regard to the proportion of patients with normal BP based on each of the two parameters, 47.3% of the measurements had a mean SBP below 135 mmHg and 79.8% had a mean DBP below 85 mmHg. Using the median, the SBP was below 135 mmHg in 48.4% of the cases and the DBP was below 85 mmHg in 82.01% of the cases. The differences were not significant (Fig. 1). The kappa concordance coefficient was 0.994 for SBP and 0.882 for DBP. The Bland-Altman plots are shown in Fig. 2. The mean difference in the measurements was 0.02 mmHg for SBP (95% CI − 0.54– 0.50) and − 0.18 mmHg for DBP (95% CI − 0.47– 0.10).

View Large Image Download to PowerPoint

Fig. 1. Percentage of blood pressure control when using the mean versus the median. Values are very similar with both methods.

View Large Image Download to PowerPoint

Fig. 2. Bland-Altman graphics of concordance between mean and median values. Differences are small and distribution seems to be regular, even for extreme values.

The sensitivity of the median for SBP was 97.0% (95% CI 95–100) and the specificity 97.0% (95% CI 95–100). False-positives amounted to 1.60% and false-negatives were 1.18%. The positive likelihood ratio was 33 and the negative likelihood ratio was 0.03. The sensitivity of the median for DBP was 94% (95% CI 90–98) and the specificity 97.0%. (95% CI 93–100). False-positives were 4.25% and there were no false-negatives (0.0%). The positive likelihood ratio was 27 and the negative likelihood ratio 0.06.

4. Discussion 

The use of BPSM makes it possible to detect a significant percentage of ill patients who, although they appear to have high BP according to the measurements taken in the doctor's office, are normotensive or whose BP is adequately controlled with the treatment they are receiving, as has been shown in previous studies. This diagnostic technique has also been shown to be a good predictor of morbidity and mortality associated with high BP, as it achieves better correlation rates than those obtained with casual measurements taken in the doctor's office [9].

The main disadvantage of BPSM is that it requires the doctor to do cumbersome calculations with numerous figures in order to obtain the mean BP measurements taken by the patient. Using the median – i.e., the mean of the highest and lowest values obtained during the BPSM – is much simpler than calculating the mean of all measurements (usually 12 or more). The most important point is that, according to the final results of this study, the median seems to be as efficient as the mean for diagnostic purposes.

An important aspect to point out with regard to BSM is the use of accepted and validated devices in accordance with adequate procedures. The list of this type of semi-automatic device is not very long, but it includes the model used in this study [10].

There are some device models for BPSM that can automatically calculate the mean, but with these devices one cannot correct the extreme values and/or erroneous values that may occur during the measurement procedure at the patient's home. Other devices are able to transmit data directly to the computer for processing and subsequent filing by the doctor. This represents an important advance, from a technical point of view, but it is not feasible to assume such a device will be used by the majority of patients. The reason for this is that, in general, all of these self-measurement devices (validated, capable of calculating the mean or of transmitting data) are significantly more expensive than those models, not always recommended or validated, that patients who resort to this technique tend to use.

The solution to this problem is not easy, but data suggest that the measurements obtained with non-validated devices can be accepted and assessed from a clinical point of view, even though it means having to apply a higher degree of exigency from an experimental point of view [11]. This is important because, in most cases, the doctor will receive numerous measurements that cannot be processed by the device or computer to calculate the mean value of the BP measurements taken, which he uses to base his diagnosis and treatment decision on. The calculation of the median, as used in this study, could offer a simple alternative to avoid this problem.

An important point when it comes to programming the BPSM is the choice of when and how many measurements are to be taken. The longer the time period and the greater the number of measurements, the higher the value of the prognosis and the reliability, as a greater number of measurements will be obtained. However, this will lengthen the period of time necessary to carry out these measurements, reducing the availability of the measurement devices and increasing the discomfort of the patient and the calculations the doctor must make in order to conclude the process. The recommendation from a recent consensus conference is to take 12 measurements over a period of 72 h [8], and that is the protocol that was used in this study.

Using the median to assess the results of measurements obtained by BPSM with semi-automatic devices seems to be a reliable method. It offers very similar results to those obtained when, as is customary, the mean of all the measurements obtained is the diagnostic value used.

5. Learning points