# Confirmation or Exclusion of Stage I Hypertension by Ambulatory Blood Pressure Monitoring

## Abstract

*Abstract *Criteria for the diagnosis or exclusion of hypertension using ambulatory blood pressure monitoring have not been agreed upon. We designed this study to provide a statistically based guide for using results of ambulatory blood pressure monitoring to resolve this issue. To generate this information, we used a database of 228 subjects (135 men, 93 women; average age, 45 years) referred by their primary physicians over the past 7 years for evaluation of borderline or stage I hypertension (average blood pressure, 148/92 mm Hg; SD, ±17.5/12.2 mm Hg). In this population, the pooled SDs of systolic and diastolic ambulatory blood pressures were 13.8 and 11.6 mm Hg, respectively. Using the pooled SD, we calculated the probability that a patient’s blood pressure falls within the hypertensive range (>140/90 mm Hg). The 95% confidence interval for each subject’s blood pressure was also determined. For example, if 40 ambulatory blood pressure measurements are performed on a subject and the average systolic ambulatory blood pressure is 137 mm Hg, then there is a 10% probability that the patient’s “true” average blood pressure is actually in the hypertensive range. By contrast, if the systolic pressure is 143 mm Hg, there is a 90% probability that the patient is hypertensive. This approach may be useful for clinical decision making and also for the design of clinical trials.

Accurate determination of usual or average blood pressure (BP) is necessary for the assessment of risk for stroke and coronary heart disease. The variability inherent in casual office BP measurements,^{1} ^{2} ^{3} ^{4} the variability of BP itself throughout the day,^{5} and the possibility of bias in the observer^{6} may lead to inaccurate assessment of true BP in subjects with borderline or stage I hypertension (for example, diastolic BPs between 85 and 95 mm Hg). Reliance on a small number of BP measurements may lead to inaccurate assessment of a subject’s true BP due to regression dilution.^{7} By providing many measurements of BP within 1 day, ambulatory BP (ABP) monitoring can reduce error due to regression dilution and thus improve the estimate of the true BP.^{8} ^{9} If a value of 140/90 mm Hg is used as the basis for classifying patients as hypertensive,^{10} it is important to know how to interpret average pressures derived from ambulatory monitoring near this level (ie, the likelihood that the true average pressure is definitely above or below 140/90 mm Hg).

ABP monitoring, now studied for more than 30 years, improves one’s ability to determine which hypertensive individuals will develop cardiovascular disease^{11} ^{12} ^{13} ^{14} ; however, the statistical criteria for the diagnosis or exclusion of hypertension using ABP monitoring have not been defined. Without the application of proper statistical criteria, this technique, like any other, can lead to misclassification of subjects. Therefore, the goal of this study was to develop valid criteria for the diagnosis or exclusion of hypertension with reasonable statistical certainty using ABP monitoring.

## Methods

### Subjects

A total of 228 subjects (135 men, 93 women), aged 16 to 89 years (average age, 45), were referred by their primary physicians for ABP monitoring over the past 7 years for evaluation of borderline or stage I hypertension. All subjects had at least two casual office BP measurement of greater than or equal to 135/85 mm Hg. Antihypertensive therapy had been withdrawn for at least 2 months, and no subject had isolated systolic hypertension of the elderly. Subjects with clinical or laboratory signs of secondary hypertension were excluded.

### BP Monitoring

In 51 subjects, BP was recorded with a SpaceLabs 5200 automatic BP monitor. In 147 subjects, BP was recorded with a SpaceLabs 90202 (or 90207) automatic BP monitor. Finally, for the remaining 30 subjects, the Suntech Accutracker BP monitor (Suntech Medical Instruments) was used. These BP monitors have been validated in previous studies.^{15} Readings were taken every 15 to 20 minutes during the daytime, which was arbitrarily defined as being from 6 am to midnight. Readings were deleted automatically by the system software for poor quality or if the pulse pressure was less than 15 mm Hg.

### Statistical Procedures

#### Calculation of the Probability That the Average of ‘n’ ABP Measurements Represents ‘True’ Hypertension

The central limit theorem states that for sample sizes greater than or equal to 30 (ie, n≥30), the sample distribution can be assumed to be normally distributed even if the population distribution is not. The average of the sample distribution is equal to the average of the population distribution, with the SD of the sample distribution equal to ς/, where ς is the SD of the population distribution.^{16} ^{17} Fig 1⇓ shows how this applies to ABP measurements. The “true” BP (population distribution) of a subject over the course of a day has an unknown probability distribution with an average (μ) and SD (ς) (see Fig 1⇓, solid curve). The shape of the probability distribution of BPs over the course of a day is influenced by variables such as physical activity, sleep, and morning surge. In contrast, the sample distribution, which is the probability distribution of the average of “n” ABP measurements, will be normally distributed (for n≥30) about an average (μ) and SD (ς/) (see Fig 1⇓, dotted curve).

As the null hypothesis, we assume that the test subject is hypertensive, with an average BP equal to 140/90 mm Hg (ie, μ=140/90 mm Hg). Based on the fifth report of the Joint National Committee on the detection, evaluation, and treatment of high blood pressure (JNC-V), a BP of 140/90 mm Hg was chosen as the lower limit of hypertension. Then for any given average of “n” ABP measurements performed on a test subject, the probability that this average ABP came from a hypertensive subject (ie, μ=140/90 mm Hg) is given by the area under the sample distribution curve to the left of the measurement. If this probability is small, then we have little confidence that the null hypothesis is true and therefore we reject it. However, as this probability increases, we have greater confidence that the null hypothesis is true (ie, the subject is “truly” hypertensive).

Therefore, to calculate the probability that the average of “n” ABP measurements came from a “truly” hypertensive subject, one must first calculate the respective test statistics with the null hypothesis being that the average ABP is 140/90 mm Hg: where ABP_{sys} and ABP_{dias} are average systolic and diastolic ABPs, respectively; and ς_{sys} and ς_{dias} are systolic and diastolic SDs, respectively.

Then, using standard tables for a one-tailed test for a normal distribution, one can calculate the probability that the average of the “n” ABP measurements came from a “truly” hypertensive subject. We used this technique to calculate the probability that the average ABPs of the 228 subjects enrolled in this study represent “true” hypertension.

#### Individual and Pooled SDs

As discussed above, the probability that the average of “n” ABP measurements came from a “truly” hypertensive subject is a function of the SD (ς_{i}, where i is the ith subject), the average ABP, and the number of ABP measurements taken (n). To develop a generalized methodology applicable to all subjects with borderline or stage I hypertension, we assessed whether use of the pooled SD (ς_{p}), as opposed to each individual’s SD, significantly changed the calculated probability. If ς_{p} (which is a constant) can be substituted for ς_{i} (which varies from subject to subject), then the probability that the average of “n” ABP measurements came from a “truly” hypertensive subject becomes only a function of the average ABP and the number of ABP measurements taken (n). For each subject, therefore, we performed the calculations using both the pooled SD and each subject’s individual SD and compared the results. The pooled SD was calculated as follows: where ς_{p}= and ς_{p}^{2} is the pooled estimated variance of all 228 subjects in the study; n_{i} is the number of ABP measurements taken on the ith subject in the study; and ς_{i} is the SD of the ABP measurements taken on the ith subject in the study.

## Results

### Ambulatory BP Characteristics

An average (±SD) of 47±11 readings was obtained on the 228 subjects in the study. The average systolic ABP was 137±8.8 mm Hg (range, 109 to 172 mm Hg); the average diastolic ABP was 85±7.2 mm Hg (range, 58 to 112 mm Hg).

### Individual and Pooled SDs

The pooled SDs for all three ABP monitoring devices (SpaceLabs 5200, SpaceLabs 90202 or 90207, and Accutracker) were calculated (see Table 1⇓) and found to be similar. When the SDs for all 228 subjects studied with the three devices were combined to calculate the total pooled SD (ς_{p}), the systolic and diastolic values were 13.8 and 11.6 mm Hg, respectively.

### Probability of Hypertension in the 228 Subjects Using Individual SDs

Using each subject’s individual SD, we calculated the probability that the average of “n” ABP measurements taken on each of the 228 subjects represents “true” hypertension (see Table 2⇓). If one arbitrarily defines as normotension an average ABP that has less than a 10% probability of coming from a “truly” hypertensive subject, then 126 (55.3%) and 142 (62.3%) of the subjects are normotensive based on systolic and diastolic BPs, respectively. Similarly, if one defines as hypertension an average ABP that has a greater than 90% probability of coming from a “truly” hypertensive subject, then 69 (30.3%) and 43 (18.9%) of the subjects are hypertensive based on systolic and diastolic BPs, respectively.

### Probability of Hypertension in the 228 Subjects Using Pooled SDs (ς_{p})

Using the pooled SDs, we calculated the probability that the average of “n” ABP measurements taken on each of the 228 subjects represents “true” hypertension (see Table 2⇑). Once again, if one defines as normotension an average ABP that has less than a 10% probability of coming from a “truly” hypertensive subject, then 126 (55.3%) and 136 (59.6%) of the subjects are normotensive based on systolic and diastolic BPs, respectively. Similarly, 69 (30.3%) and 42 (18.4%) of the subjects are hypertensive based on systolic and diastolic BPs, respectively.

### Correlation of Results Using Individual and Pooled SDs

Table 2⇑ shows all 228 subjects categorized with respect to the probability that their average ABP represents “true” hypertension (140/90 mm Hg). The calculations were done for both individual and pooled SDs. When the categorization of all subjects using the individual SDs was compared with the categorization of all subjects using the pooled SDs, the correlation coefficients were .9997 and .998 for systolic and diastolic BPs, respectively. The average difference in the calculated probabilities using the pooled SD rather than the individual SD was 0.2% for both systolic and diastolic BPs.

The correlation coefficients were similar (ie, >.99) for men, women, subjects younger than 50 years, and subjects 50 years and older when the categorization of subjects using each individual’s SD was compared with the categorization of subjects using the pooled SD.

Therefore, it appears that for this population of borderline or stage I hypertensive subjects, the probability that the average of “n” ABP measurements represents “true” hypertension is not significantly changed when the pooled SD is used rather than each individual’s SD. Substituting the pooled SD (ς_{p}, which is a constant) for the individual’s SD in the equations for the test statistic (Z) allows us to develop a generalized methodology defining the probability that the average of “n” ABP measurements represents “true” hypertension as a function of the average ABP and the number of ABP measurements taken.

### Generation of Probability Curves

With the use of the pooled SDs of 13.8 and 11.6 mm Hg for systolic and diastolic BPs, respectively; the test statistics become:

Therefore, if the number of ABP measurements (n) is held constant, the test statistics become only a function of the average measured ABP; similarly, the probability that the average of “n” ABP measurements represents “true” hypertension becomes only a function of the average measured ABP. Figs 2⇓ and 3⇓ are a set of curves that show the probability that the average of “n” ABP measurements represents “true” hypertension as a function of the average measured ABP for n=30, n=40, and n=50. The 95% confidence intervals as a function of the average measured ABP are shown in Table 3⇓.

We realized that 140/90 mm Hg, chosen as the lower limit of hypertension for this study, is somewhat arbitrary. A meta-analysis performed by Staessen et al^{18} on 3746 normotensive subjects showed that the lower limit of hypertension (ie, 2 SDs above the mean) for daytime ABP measurements was 146/91 mm Hg. If this value is used for the lower limit of hypertension, then the curve for systolic BP (Fig 2⇑) would be shifted 6 mm Hg to the right and the curve for diastolic BP (Fig 3⇑) would be shifted 1 mm Hg to the right. The 95% confidence intervals, shown in Table 3⇑, would remain unchanged.

## Discussion

The goal of the present study was to develop valid criteria to aid the clinician in the diagnosis or exclusion of hypertension with reasonable statistical certainty using ABP monitoring. If a value of 140/90 mm Hg is used as the basis for classifying individuals as hypertensive, it is important to know how to interpret average ABPs near this level. The potential for the misclassification of borderline or stage I hypertensive individuals using ABP monitoring is primarily due to the inherent uncertainty associated with taking a finite sample of a large population. The average obtained by sampling a large population is only an approximation of the “true” average of that population. Similarly, the average obtained from the finite number of ABP measurements taken over the course of a day is only an approximation of the “true” average BP and therefore also has an inherent uncertainty associated with it. However, if the number of ABP measurements (n) is at least 30, we may assume that the sampling distribution (ie, the probability distribution of the average of “n” ABP measurements) approximates a normal curve. Therefore, the uncertainty inherent in taking a finite number of ABP measurements can be quantified in terms of the probability that the average of “n” ABP measurements came from a “truly” hypertensive subject (ie, average BP=140/90 mm Hg). This method is used by many quality-assurance monitoring systems in which the error terms are known to hold across populations.^{19}

Not only can the uncertainty inherent in ABP monitoring be quantified, it can be minimized to a large extent. Previous studies have shown that the average ABP of a given subject varies when that subject is tested and then retested several days later (ie, test-retest variability) and that this variability decreases as the number of ABP measurements (n) recorded over 24 hours increases.^{20} ^{21} The decrease in test-retest variability (ie, the increase in test-retest reproducibility) is due to the fact that as n increases, the probability distribution curve of the average of “n” ABP measurements becomes narrower (ie, ς^{2} ∝ 1/n). This means that as “n” increases, one becomes more confident that the measured average ABP is an accurate representation of the subject’s “true” average BP.

Using this approach, we generated curves that will aid the clinician in the confirmation or exclusion of hypertension with reasonable statistical certainty. These curves are shown in Figs 2⇑ and 3⇑. One immediately notices that the probability curves vary little when the number of ABP measurements (n) is increased from 30 to 50. When the number of ABP measurements is increased from 30 to 50, there is, at most, a 6% increase in confidence that an individual is not hypertensive when the average measured ABP is less than 140/90 mm Hg. Similarly, there is, at most, a 6% increase in confidence that an individual is hypertensive when the average measured ABP is greater than 140/90 mm Hg. Also, the 95% confidence intervals narrow only slightly when the number of ABP measurements is increased from 30 to 50 (see Table 3⇑).

Given the fact that very little is gained in statistical certainty when the number of ABP measurements is increased from 30 to 50, we believe that 40 ABP measurements are adequate when analyzing ABPs with this statistical method. Forty ABP measurements satisfy the criteria of the central limit theorem (ie, n≥30), while at the same time not inconveniencing the subject with an excessive number of ABP measurements.

An example using this method is as follows: A clinician might ask, if 40 ABP measurements are performed on a patient over the course of a day, what must the average systolic ABP be to have reasonable statistical certainty that the patient’s “true” average systolic BP is not in the hypertensive range? The answer depends on the amount of uncertainty the clinician is willing to accept.

If the clinician is willing to accept no more than a 10% probability that a patient’s “true” average systolic BP is in the hypertensive range, then an average systolic ABP of less than 137 mm Hg could reasonably be defined as normotension (see Fig 2⇑ for n=40). Average systolic ABPs above 137 mm Hg would exceed the level of uncertainty the clinician is willing to accept, and at this point, the location of the patient’s average systolic ABP on the probability curve shown in Fig 2⇑ must be assessed. For example, if the average systolic ABP were 143 mm Hg, there is a 90% probability that the “true” average systolic BP is in the hypertensive range (see Fig 2⇑ for n=40), with a 95% confidence interval of 138.7 to 147.3 mm Hg (see Table 3⇑ for n=40). Clearly, the patient would be considered hypertensive with reasonable statistical certainty and the appropriate nonpharmacological or antihypertensive therapy instituted. On the other hand, if the average systolic ABP were 141 mm Hg, there is a 70% probability that the patient’s average systolic BP is in the hypertensive range, with a 95% confidence interval of 136.7 to 145.3 mm Hg. At this point, the clinician can assess, on the basis of other risk factors (such as age, diabetes, high cholesterol, family history of coronary artery disease, history of stroke, or tobacco use), whether it is appropriate to treat the patient for hypertension or to hold off treatment and remonitor the patient at a later date.

Finally, one potential limitation of our model is that it assumes that the variability of BP for a given individual (ς_{i}^{2}) does not change significantly from day to day. This allows us to make the assumption that the variance (ς^{2}) of the probability distribution of the average of “n” ABP measurements is equal to ς_{i}^{2}/n. However, although several studies have shown that the test-retest reproducibility of average ABP improves with increasing “n,” the improvement is not as great as would be predicted by the equation ς^{2}=ς_{i}^{2}/n. There appears to be a point reached (probably around n=30-60) where increasing “n” does not result in corresponding improvement in test-retest ABP reproducibility.^{20} ^{21} It has been theorized that the limitation on test-retest reproducibility is related to the nonstandardization of subject activity and position during ABP monitoring, both of which are likely to vary from day to day (ie, ς_{i}^{2} in a given individual significantly varies from day to day). Evidence that this theory is true is supported by the fact that Gerin et al^{22} showed an increase in ABP test-retest reproducibility when they meticulously controlled for subject activity and position. Subject reaction to the medical environment and changes in body weight have also been implicated as factors that may influence ABP test-retest reproducibility.^{23}

Because of the observed unresponsiveness of test-retest reproducibility to increasing “n” above a particular threshold, Reeves and Myers^{24} proposed an expanded model for the variance (ς^{2}) of the average of “n” ABP measurements as ς^{2}=ς_{i}^{2}/n+ς_{b}^{2}, where ς_{b}^{2} represents a test-retest variance that may be significant between any two measurements. Clearly, further studies need to be done for determination of the etiology of the test-retest variance (ς_{b}^{2}) that has been observed in ABP monitoring to better quantify and minimize its effect on test-retest variability. Already, Gerin et al^{22} have shown that controlling for subject activity and position can significantly decrease ς_{b}^{2}. Once ς_{b}^{2} is accurately quantified, it can be incorporated into our model, thus resulting in an even more precise estimation of the probability of hypertension. Until then, we believe that the model presented in this article will be a useful tool for the clinician to confirm or exclude hypertension with reasonable statistical certainty in those individuals diagnosed with borderline or stage I hypertension by office BP measurement.

- Received October 25, 1996.
- Revision received October 29, 1996.
- Accepted October 29, 1996.

## References

- ↵
- ↵
- ↵
Gordon T, Sorlie P, Kannel WB. Problems in the assessment of blood pressure: the Framingham study. Int J Epidemiol
*.*1976;4:327-334. - ↵
- ↵
Pickering G.
*The Nature of Essential Hypertension*. New York, NY: Grune & Stratton; 1961:9-21. - ↵
- ↵
MacMahon S, Peto R, Cutler J, Collins R, Sorlie P, Neaton J, Abbott R, Godwin J, Dyer A, Stamler J. Blood pressure, stroke, and coronary heart disease, part 1: prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias. Lancet
*.*1990;335:765-774. - ↵
Pearce KA, Grimm RH, Rao S, Svendsen K, Liebson PR, Neaton JD, Ensrud K. Population-derived comparisons of ambulatory and office blood pressures: implications for the determination of usual blood pressure and the concept of white coat hypertension. Arch Intern Med
*.*1992;152:750-756. - ↵
- ↵
- ↵
Perloff D, Sokolow M, Cowan R. The prognostic value of ambulatory blood pressures. JAMA
*.*1983;249:2793-2798. - ↵
Devereux RB, Pickering TG, Harshfield GA, Kleinert HD, Denby L. Left ventricular hypertrophy in patients with hypertension: importance of blood pressure response to regularly recurring stress. Circulation
*.*1983;68:470-476. - ↵
- ↵
Verdecchia P, Porcellati C, Schillaci G. Ambulatory blood pressure: an independent predictor of prognosis in essential hypertension. Hypertension
*.*1994;24:793-801. - ↵
- ↵
Frank H, Althoen SC.
*Statistics: Concepts and Applications*. New York, NY: Cambridge University Press; 1994:380-417. - ↵
Zar JH.
*Biostatistical Analysis*. 2nd ed. Englewood Cliffs, NJ: Prentice-Hall; 1984:97-121. - ↵
- ↵
Davies OL.
*Design and Analysis of Industrial Experiments*. New York, NY: Cambridge University Press; 1954:31-37. - ↵
- ↵
- ↵
- ↵
Palatini P, Mormino P, Canali C, Santonastaso M, De Venuto G, Zanata G, Pessina AC. Factors affecting ambulatory blood pressure reproducibility. Hypertension
*.*1994;23:211-216. - ↵

## This Issue

## Article Tools

- Confirmation or Exclusion of Stage I Hypertension by Ambulatory Blood Pressure MonitoringCarlton R. Moore, Lawrence R. Krakoff and Robert A. PhillipsHypertension. 1997;29:1109-1113, originally published May 1, 1997https://doi.org/10.1161/01.HYP.29.5.1109
## Citation Manager Formats