how many self-measured blood pressure readings are needed to

8/18/2019 How Many Self-Measured Blood Pressure Readings Are Needed To

1/22

Journal of Behavioral Medicine, Vol. 22 , No. 1, 1999

How Many Self-Measured Blood Pressure ReadingsAre Needed to Estimate Hypertensive Patients’ªTrueº Blood Pressure?

Mar Âõa Paz Garc Âõa-Vera1,2 and Jes Âus Sanz1

Accepted for publication: April 12, 1998

The present research was aimed at determining the number of self-measured

blood pressure (BP) readings needed to attain reliable estimates of true BPin hypertensive patients. Correlation coef®cients and standard deviation of

differences between pairs of measurements as well as generalizability theory

were applied to data from a controlled study on stress management training for essential hypertension (Garc Âõa-Vera et al., 1997). Forty-three hypertensive patients self-recorded 48 readings of BP (at home and at work) at both the

pretreatment and the posttreatment (separated by a period of 2 months) and 24 readings of BP at follow-up (6 months after the pretreatment). The results

showed that it is enough to take two readings, one at work and the other at

home, from each of 3 consecutive days to get reliable estimates of SBP andDBP across settings, over 1 week and over 2 months. This same criterion would

be valid to get reliable estimates of DBP over 6 months, but two readings, oneat work and the other at home, from 8 or more consecutive days may need tobe taken to achieve similarly reliable results for SBP.

KEY WORDS: blood pressure measurement; self-monitoring; measurement reliability.

INTRODUCTION

Since the early 1970s, studies applying behavioral procedures to the treat-


2/22

Garc Âõa-Vera and Sanz94

of the best-studied areas in behaviora l medicine. However, the ef®cacy of behav-

ioral treatments is still subject to controversy. This state of affairs is related to

the differences in results between studies with only one baseline session and

those with four or more baseline sessions: the shorter the baseline, the better the

results for behavioral treatments (Jacob et al., 1991). These differences are due,

in part, to the variability of blood pressure (BP) measurements. As Jacob et al. pointed out: ªblood pressure is variable at all times; we have become increas-

ingly concerned that what we see as being the result of relaxation therapy is an

epiphenomenon of this variability, and that this variability cannot be harnessed

as persistent blood pressure reductionsº (Jacob et al., 1991, p. 14).

Therefore, the inherent variability of BP implies that a single BP measure-

ment is not likely to provide a stable estimate of patients’ ªtrueº or usual BP. The

true BP is conceived as the average level of pressure over time and, also, the BP that

is ultimately responsible for the adverse effects of hypertension (Pickering, 1991).

Any individual measurement is an attempt to estimate the patient’s true level of

BP, from which the observed value will differ to a greater or lesser extent. Thus,

the goal of BP measurement is attaining a stable or reliable estimate of a patient’s

true BP. Given that reliability can, in theory, be improved by increasing the num-

ber of readings, the practical question linked to this issue is how many readings are

needed to estimate hypertensive patients’ true BP adequately.

A few studies have examined the reliability of BP measurement, expressedas the correlation between pairs of measurements or as the standard deviation

of the differences between such pairs of measurements, in order to estimate the

number of readings required for obtaining adequate reliability when assessing

hypertensive patients. However, these studies have been restricted to clinic or

ambulatory BPs (e.g., Coats, 1990; Mancia et al., 1994; Shepard, 1981). No pre-

vious studies are known to have examined this question in relation to self-mea-

surement of BP, in spite of several studies suggesting that the predictive value of

self-measurement for hypertensive cardiovascular morbidity may be superior to that of clinic readings because self-measured BPs show a closer correlation with

24-hr ambulatory monitoring and hypertensive target organ damage (Kleinert et

al., 1984; Marolf et al., 1987; Verdecchia et al., 1985). The aim of the present

research was to determine the number of self-measured readings needed to attain

reliable estimates of true BP in hypertensive patients. To this end, we analyzed

data from a controlled study on stress management training for essential hyper-

tension (Garc Âõa-Vera et al., 1997). Following Coats’s suggestion (1990), we think that an investigation of the

reliability of BP measurements should be carried out with subjects similar to


3/22

Self-Measured Blood Pressure Reliability 95

despite previous pharmacological treatments or despite current medication. Con-

sequently, this study examined reliability of self-measured BP in such a sample.

On the other hand, most behavioral treatments are delivered for 2±6 months,

and therefore, we examined the reliability of self-measured BP when the interval

between measurements was both 2 and 6 months. In addition, we also examined

the reliability over a shorter period (1 week) to compare our results thoroughlywith those reported by previous studies with clinic and ambulatory BPs.

Although most of the previous literature has examined the reliability of

BP measurement in terms of correlations and standard deviations of the differ-

ences between pairs of measurements, Llabre et al. (1988a,b) have convincingly

argued that generalizability (G ) theory (Cronbach et al., 1972) provides a powerful

methodology for studying the reliability of BP measurement. Whereas test±retest

correlations and standard deviations of the differences between pairs of measure-

ments only consider one source of variability at a time (e.g., number of visits or

days), G theory offers variance component estimates and coef®cients of depend-

ability (G *), analogous to reliability coef®cients, that enable one to study both sys-

tematic and random sources of error variance simultaneously and to determine the

number of readings required to attain a certain level of reliability while considering

those sources of variance. Llabre et al. (1988b) examined, in mild hypertensives,

the stability of BPs taken in the laboratory and of ambulatory BPs taken at home

and at work. In the present study, we extended the work of Llabre et al. into anadditional type of BP measurement: self-measured BPs.

METHOD

Subjects

Over a 3-year period 65 male patients, diagnosed as having essential hypertension by their family physicians, volunteered to participate in a research study

on behavioral treatment of essential hypertension after being referred by their

primary care physicians for an exhaustive evaluation of their BP. Subjects were

referred if their BP was not well controlled [i.e., diastolic BP (DBP) greater than

or equal to 90 mm Hg or systolic BP (SBP) greater than or equal to 140 mm

Hg] despite previous pharmacological treatments or despite taking currently one

or two medications. All subjects were recruited from two health centers, one

in Pozuelo de Alarc Âon and the other in Monterrozas, two cities near Madrid.Potential patients were considered for this research if (a) the diagnosis was con-


4/22


the investigation, and (c) their recent medical history elaborated by their pri-

mary care physician did not include a diagnosis of mental disorder according to

the ICD-9 system. All patients who met criter ia and who were referred by their

family physicians were asked to participate in this study and all of them con-

sented to participate. Nevertheless, 22 patients were excluded from the study:

17 patients were excluded because their family physician changed their medi-cation at some point during the pretreatment assessment or during the treatment

(although, obviously, they continued receiving the treatment), 4 patients discon-

tinued the study because they moved out of town, and 1 patient abandoned the

study because of unknown reasons.

In sum, the ®nal sample of this study was composed of 43 patients. In

this sample, 11 patients did not receive any medication at the beginning of

the study and 29 patients received just one antihyperten sive drug: either an

angiotensin-converting enzyme inhibitor (14 patients), a calcium-channel antag-

onist (8 patients), a b -blocker (6 patients), or a diuretic (1 patient). Two patients

were on a combination of two drugs, the combination including in all cases a

diuretic, and one patient was on a combination of three drugs. Demographic and

clinical characteristics of the sample appear in Table I.

Table I. Demographic and Clinical Data

Mean SD Minimum Maximum n

Age (years) 45.37 8.87 26 69 41

SBP at clinic 150.32 8.51 137 170 43

DBP at clinic 99.46 7.03 86 122 43

Self-measured SBP 130.26 10.78 111 153 43

Self-measured DBP 84.12 7.62 69 100 43

Duration of hypertension (months) 67.42 67.26 2 240 43

Compliance with the medication (%) 95.69 13.36 25 100 32

No. of previous treatments 0.98 1.10 0 4 43No. of cigarettes per day 3.30 7.99 0 30 43

Regularity of meals (0 ± 10 scale) 7.70 2.61 0 10 43

Discomfort by medication (0 ± 10 scale) 0.13 0.71 0 4 32

Compliance with BP self-monitoring (%)

Pretreatment (®rst 8 days) 99.42 2.15 87.5 100 43

Pretreatment (second 8 days) 99.61 1.53 91.67 100 43

Posttreatment 99.31 2.33 87.5 100 43

Follow-up 98.72 3.47 83.33 100 39

Note. SBP and DBP at clinic: average of three clinic measurement s of BP taken, upon entry to the study, by a nurse at the health center patients attended regularly following the guidelines


5/22


Procedure

The investigation consisted of four phases: (a) a pretreatment assessment

of BP, and of physiological and psychosocial hypertension-related variables; (b)

a stress management training phase for the treated group and a waiting list for

the control group; (c) a posttreatment assessment identical to the pretreatmentassessment; and (d) a follow-up assessment of BP. Written informed consent was

obtained for each phase. Both assessment and treatment were conducted by the

same therapist (the ®rst author).

After being referred by their physicians, patients were invited to an individ-

ual assessment of their hypertension composed of two 60-min sessions. The ®rst

session was held in the health center the patient attended regularly. Patients com-

pleted an interview that assessed several variables related to their hypertension

problem (stressful events, duration of hypertension, adherence to medication, etc.).

Then patients were carefully instructed on how to self-m easure and self-record BP

readings correctly, and carried out several practice trials. Patients were asked to

self-measure their BP on three occasions per day for 8 days: two times at home

(when getting up in the morning and before bedtime) and one at work. In order

to avoid an arti®cial BP increase due to squeezing the in¯ation bulb, each patient

was instructed to place the cuff on the nondom inant arm and to use the hand of the

opposite arm to in¯ate the cuff while keeping the cuff arm relaxed and placed on a plain surface at heart level. Patients were instructed to urinate and to refrain from

smoking, eating, taking caffeine or doing physical exercise, during at least 30 min

before self-measuring their BP. Each patient was also directed to measure BP after

a 5-min rest period in a sitting position. All these directions were given to patients,

not only orally, but also on a printed sheet to take home. In addition, patients were

also given self-recording sheets that included reminders on how to carry out self-

measurement of BP correctly [see Garc Âõa-Vera et al. (1993) for more details con-

cerning self-measurement technique training]. The second session of the pretreatment assessment was aimed to get addi-

tional physiological information and to monitor self-measurements of BP. In

order to complete the physiological assessment, this second session was held

8 days after the ®rst one in the laboratories of the Faculty of Psychology of the

Universidad Com plutense de Madrid. In this session, instructions for self-mea-

surement of BP were reviewed and patients were asked to monitor their BP for

other 8 days. After pretreatment assessment, all patients were randomized into

two groups: 22 to the stress-management group and 21 to the control group.

Two weeks after the end of stress management training which lasted 2


6/22


Four months after the end of stress management training (follow-up),

patients were asked to self-monitor their BP on three occasions per day for 8

days. At this time, four patients were dropped from the data analyses because

of medication change during follow-up period and failure to complete the self-

measurements at follow-up due to unknown whereabouts. Of these patients, one

was in the stress management group and three in the control group.Compliance to the request for BP self-monitoring was high. This compli-

ance was expressed, after counting BP readings, as a percentage of the BP read-

ings required for each period of self-monitoring. For example, during the ®rst 8

days of pretreatment assessment, 24 BP readings were required and, therefore,

if patients returned 24 BP readings their compliances would be 100%. Mean

percentages of compliance for the sample are shown in Table I.

Materials

Each patient was given an OMRON HEM-403C digital BP monitor to self-

record BP. This semiautomatic monitor uses an oscillometric method to measure

BP with an accuracy of ±3 mm Hg and generates a digital display giving SBP,

DBP, and pulse rate. Before each assessment phase, monitors were rechecked for

accuracy and, if necessary, recalibrated by the technical service of manufacturers

of the digital blood pressure monitor in Spain.

Statistical Analyses

Anomalous BP readings were deleted according to the rules set by Chatel-

lier et al. (1996): (1) SBP greater or equal to 250 mm Hg or lower or equal to

70 mm Hg and (2) DBP greater or equal to 150 mm Hg or lower or equal to

40 mm Hg. The number of anomalous BP readings was negligible: 1 of 1032readings during the ®rt 8 days of pretreatment assessment, 1 of 2064 readings

during the posttreatment assessment, and 3 of 936 readings during the follow-up

assessment. Valid data were analyzed by either Pearson correlation coef®cients

or standard deviation of the differences between pairs of measurements (®rst and

second measurement).

We considered the readings taken during the ®rst 8-day period of the pre-

treatment assessment as the ®rst measurement and analyzed their reliability by

number of days, number of readings, and type of setting (at home in the morn-

ing at home in the evening and at work) Thus we considered successively as


7/22


average of two readings (at home in the morning and at work) of the ®rst day,

then the average of these two readings and two readings of the second day, then

the average of these four readings and two readings of the third day, and so on.

Finally, we considered successively as ®rst measurement the average of three

readings (at home in the morning, at home in the evening, and at work) of the

®rst day, then the average of these three readings and three readings of the sec-ond day, then the average of these six readings and other three readings of the

third day, and so on.

As second measurement, we considered successively (a) the average of the

24 readings (3 readings per dayÐat home in the morning, at home in the evening,

and at workÐduring 8 days) taken during the second 8-day period of the pre-

treatment assessment; (b) the average of the 48 readings (3 readings per dayÐat

home in the morning, at home in the evening, and at workÐduring 16 days)

taken during the posttreatment assessment, and (c) the average of the 24 read-

ings (3 readings per dayÐat home in the morning, at home in the evening, and at

workÐduring 8 days) taken during the follow-up assessment. These three de®ni-

tions of individual BP level allow us to examine the reliability of self-measured

BP over 1 week, over 2 months, and over 6 months, respectively. Given that

patients in the stress management group signi®cantly lowered their self-mea-

sured BPs at posttreatment and follow-up, whereas patients in the control group

did not (see Garc Âõa-Vera et al., 1997), the reliability over 2 and 6 months wasexamined only in the patients in the control group.Differences in reliability over time between types of BP (SBP versus DBP)

and among types of settings (homeÐmorning, home±evening, and work) were

tested by comparing correlation coef®cients using a z -statistic for comparison

of two dependent sample correlations based on Fisher’s r-to- z transformation,

and by comparing standard deviations of differences using a t -statistic for com-

parison of two samples variances from two dependent samples. To control for

experimentwise Type I errors, those tests were conducted using levels of con®-dence corrected by Bonferroni’s procedure for each kind of reliability over time

(1 week, 2 months, and 6 months, that is, .05 / 40 = .00013).Valid data were also analyzed using the G -theory methodology followed by

Llabre et al. (1988a,b). To ensure an adequate sample size, we restricted G -the-

ory analyses to data from the 16-day pretreatment phase (N = 43) and, therefore,

we studied the following (P ´ S ´ D) design: Person ´ Setting (homeÐmorning,

homeÐevening, work) ´ Day (16 days). Estimates of the variance components

for a random effect model were obtained via the BMDP-8V program (BMDP,

1990) This program generates an ANOVA summary table and from it calcu-


8/22


RESULTS

Reliability of Self-Measured BP Over 1 Week,

Over 2 Months, and Over 6 Months

Table II shows, for all patients, correlation coef®cients and standard deviations of the differences between BP values taken during the ®rst 8-day period of

the pretreatment assessment and the average of the BP readings taken during the

second 8-day period of the pretreatment assessment. Taking into account one BP

reading per day, a visual inspection of the correlation coef®cients and standard

deviations of the differences revealed that (a) DBP self-measurements showed

slightly lower standard deviations of the differences and higher correlation coef-

®cients over time than SBP self-measurements, and (b) self-measured BPs taken

at work led to slightly more reliable estimates of BP over time than self-mea-

sured BP taken at home as re¯ected by higher correlation coef®cients and lower

standard deviations of the differences. This pattern of results was similar to those

found over 2 months and over 6 months for the patients in the control group (see

Tables III and IV). However, in spite of this consistency of ®ndings, in all cases

results yielded by the z - and t -statistics did not reveal any statistically signi®cant

difference between DBP and SBP or between work BP and home BP.

A visual inspection of the data displayed in Tables II±IV also revealed a pattern of results consistent across 1-week, 2-month, and 6-month reliability data:

independently of the type of setting in which self-measured BPs were taken (at

home in the morning or in the evening or at work) and of the type of BP mea-

surement (DBP or SBP), increasing the number of days led to more reliable

measurements as re¯ected by higher correlation coef®cients and lower standard

deviations of the differences. Nevertheless, the gains in reliability were modest

when more than four or ®ve measurements were taken into account. For example,

taking into account ®ve SBP readings registered at work, we obtained test±retestcoef®cients of .88, .87 and .78 and standard deviations of differences of 5.7, 6.1,

and 8.2, for 1 week, 2 months, and 6 months, respectively. These values did not

increase when six or more SBP readings were taken into account. Likewise, the

values obtained with ®ve DBP readings registered at work (.87 / 4.5, .88 / 4.2,.89 / 4.6) scarcely increased when six or more readings were taken into account.

Several investigators have chosen a test±retest correlation coef®cient of .80

or higher as indicative of reliable measurements of BP (Shepard, 1981; Llabre

et al., 1988). Using this standard, our results pointed out that, using only one

measurement per day measurements may need to be taken on as many as 2 to


9/22


Table II. Correlation Coef®cients (r) and Standard Deviations of the Differences (SDD)

Between BP Values Taken During the First 8-Day Period of the Pretreatment Assessment and the

Average of the BP Readings Taken During the Second 8-Day Period of the Pretreatment

Assessment for All Patients (N = 43)

SBP DBP

No. of days r SDD r SDD

One reading per dayAt home in the morning

1 .67 13.3 .61 10.42 .72 10.4 .71 7.93 .74 9.9 .76 7.14 .76 8.7 .79 6.45 .80 8.2 .83 6.06 .81 7.6 .84 5.57 .83 6.9 .85 5.1

8 .85 6.6 .86 4.9At home in the evening

1 .67 11.6 .70 8.72 .72 9.0 .78 6.73 .76 8.4 .80 6.14 .78 8.3 .81 5.85 .78 8.0 .78 6.36 .83 6.8 .82 5.37 .85 6.2 .85 4.88 .85 6.2 .87 4.3

At work

1 .73 9.1 .64 8.42 .83 6.8 .74 6.23 .85 6.4 .83 5.04 .88 5.7 .85 4.85 .88 5.7 .87 4.56 .88 5.7 .87 4.47 .89 5.4 .89 4.08 .90 5.2 .89 4.2

Two readings per daya

1 .73 9.9 .66 8.52 .81 7.4 .76 6.43 .83 7.0 .82 5.44 .85 6.3 .85 5.05 .87 6.0 .87 4.76 .88 5.7 .88 4.47 .89 5.3 .89 4.18 .90 4.9 .90 3.9

Three readings per dayb

1 .76 9.0 .72 7.42 .82 6.9 .81 5.83 .84 6.6 .85 5.04 .86 6.0 .87 4.75 .88 5.6 .88 4.5

6 .90 5.2 .89 4.17 .92 4.6 .91 3.78 93 4 2 92 3 4


10/22


Table III. Correlation Coef®cients (r) and Standard Deviations of the Differences (SDD)

Between BP Values Taken During the First 8-Day Period of the Pretreatment Assessment and the

Average of the BP Readings Taken 2 Months Later During the Posttreatment Assessment for

the Patients in the Control Group (n = 21)

SBP DBP



1 .61 14.9 .69 9.02 .74 9.7 .73 7.03 .76 8.7 .79 5.94 .78 7.9 .82 5.35 .81 7.3 .86 4.76 .82 7.1 .87 4.57 .83 7.0 .88 4.3


1 .76 9.6 .86 8.12 .78 8.3 .87 6.03 .80 8.3 .88 5.54 .79 8.5 .88 5.05 .79 8.2 .85 6.06 .83 7.3 .87 5.37 .84 7.0 .89 4.88 .80 7.9 .90 4.5

At work

1 .69 9.3 .59 8.52 .79 7.7 .73 6.43 .79 8.1 .81 5.54 .85 6.7 .87 4.55 .87 6.1 .88 4.26 .87 6.1 .88 4.47 .89 5.7 .90 4.08 .88 5.8 .88 4.6


1 .67 10.7 .68 7.82 .77 8.3 .76 6.13 .80 8.0 .81 5.44 .83 7.0 .86 4.55 .86 6.4 .88 4.26 .86 6.3 .89 4.17 .88 5.9 .91 3.78 .89 5.6 .90 3.4


1 .74 9.2 .79 6.92 .82 7.3 .84 5.13 .82 7.3 .87 4.84 .84 6.9 .89 4.25 .86 6.2 .91 3.9

6 .88 5.9 .91 3.77 .89 5.7 .93 3.48 89 5 5 93 3 4


11/22


Table IV. Correlation Coef®cients (r) and Standard Deviations of the Differences (SDD)

Between BP Values Taken in the First 8-Day Period of the Pretreatment Assessment and the

Average of the BP Readings Taken 6 Months Later During the Follow-up Assessment for the

Patients in the Control Group (n = 18)

SBP DBP



1 .66 11.8 .81 7.82 .72 9.7 .85 5.53 .67 10.2 .87 5.24 .68 9.7 .88 5.05 .72 9.0 .88 4.96 .71 9.1 .87 4.97 .72 9.0 .88 4.8


1 .69 10.0 .82 9.42 .69 9.7 .85 6.53 .70 9.6 .86 5.94 .68 10.1 .85 5.65 .67 10.0 .82 6.56 .71 9.2 .84 5.67 .73 8.8 .86 5.28 .62 10.4 .87 4.8

At work

1 .65 9.6 .70 7.42 .71 9.1 .80 5.73 .70 9.6 .83 5.54 .75 8.6 .88 4.85 .78 8.2 .89 4.66 .78 8.3 .89 4.67 .79 7.9 .90 4.28 .79 7.9 .90 4.5


1 .70 9.4 .81 6.62 .72 9.1 .85 5.13 .70 9.6 .86 5.14 .73 8.9 .89 4.65 .77 8.2 .90 4.46 .76 8.4 .89 4.57 .77 8.1 .90 4.38 .78 8.0 .90 4.3


1 .74 8.8 .82 7.32 .74 8.7 .86 5.43 .72 9.1 .86 5.34 .73 8.9 .88 4.85 .75 8.4 .89 4.6

6 .76 8.3 .89 4.57 .77 8.0 .90 4.28 78 8 0 90 4 1


12/22


may need to be taken on as many as 1 (at home) or 2 (at work) days for a reliable

estimate of DBP, but even averaging eight SBP readings, we did not achieve a

correlation coef®cient of .80 over 6 months either for self-measured BP taken

at work or for self-measured BP taken at home.

On the other hand, a visual inspection of Tables II±IV revealed that, in

comparison to the use of one measurement per day (e.g., one measurement takenat work), the use of two measurements per day (one reading taken at home in

the morning and the other one taken at work) did not decrease the minimum

number of days needed to get reliability coef®cients of around .80 over 1 week,

over 2 months, or over 6 months, whereas the use of three measurements per

day (two readings taken at home and the third one taken at work) decreased it

slightly.

To our knowledge, most medications tested in literature have failed to alter

BP variability (cf. Pickering, 1991). However, using noninvasive BP monitoring,

Novo et al. (1989) found that verapamil (a calcium antagonist) decreased signif-

icantly BP variability. Therefore, it may be that any of medications or combina-

tion of medications administered to patients in this study could have affected BP

reliability and, therefore, have become a potential confound. To rule out this pos-

sibility, data from the unmedicated patients were analyzed separately, although

we restricted this analysis to 1-week reliability (pretreatment phase) because only

six subjects remained unmedicated at posttreatment or at follow-up. These anal-yses revealed that the pattern of test±retest coef®cients and standard deviations

of differences was almost-identical to that found in the whole sample.

Individual Differences and BP Reliability

Several factors, such as level of BP, age, alcohol, exercise, and smoking,

may affect BP variability (cf. Pickering, 1991) and, therefore, the reliability ofBP self-measurements. Consequently, we examined the relationship between the

demographic / clinical variables mentioned in Table I and BP variability. To do this, we de®ned variability as the differences between pairs BP measurements.

We considered as ®rst BP measurement the BP values (averages of three readings

per day) taken during the ®rst 8-day period of the pretreatment assessment and

took them into account by num ber of days (as we did in the lower parts of Tables

II±IV). As second BP measurement, we considered successively (a) the average

of the BP readings taken during the second 8-day period of the pretreatment

assessment (1-week variability) (b) the average of the BP readings taken during


13/22


ined its relationship only with 2- and 6-month variabilities (long-term variabil-

ity). Correlations were calculated between demographic / clinical variables and the different variability estimations.

Age and initial level of self-measured BP were the only demographic / clin-ical variables that showed a consistent pattern of relationship with BP variability,

relationships that were maintained even after partialling out the effects due to theother signi®cant variable (i.e., the relationship between age and variability, for

example, was maintained after partialling out the effects due to the initial level of

self-measured BP). The 1-week variability increased progressively with aging,

as indicated by signi®cant correlations ranging between .38 (8 days) and .52

(1 day) for SBP and between .32 (8 days) and .45 (1 day) for DBP. Interestingly,

the long-term variability, also increased progressively with aging, although only

for SBP; we did not ®nd any relationship between long-term variability and age

for DBP. Thus, for SBP, correlations between age and 2-month variability ranged

between .39 (8 days) and .50 (3 days), whereas correlations between age and 6-

month variability ranged between .40 (8 days) and .56 (1 day); nevertheless, in

these two cases some correlations were not signi®cant (r’s lower than .42 and

.46, respectively, for two-tailed p > .05) due to decrease in power.

On the other hand, increased long-term BP variability was also related to

higher initial self-measured BPs, although only for DBP. We found signi®cant,

moderate correlations between 6-month variability and the initial level of self-measured DBP, ranging between .57 (2 days) and .63 (8 days); correlations

between long-term variability and initial levels of DBP also were moderate in

size, ranging between .24 (2 days) and .40 (8 days), but they were not signi®cant

due to the decrease in power.

Generalizability of Self-Measured BPs Across Days and Settings

Estimates of the variance components for both SBP and DBP for Person

´ Setting ´ Day (P ´ S ´ D) design are displayed in Table V. These results

showed that systematic variation due to either settings (S), days (D), or their

interaction (S ´ D) was minimal or zero. Therefore, our data did not indicate

a consistent pattern of decrease (or increase) across days, as may be the result

of habituation, or systematic setting differences. The results in Table V also

indicated that, for SBP, random variation across settings (P ´ S) had more than

twice the magnitude than does random variation across days (P ´ D), whereas

for DBP both sources of random variation had similar magnitudes That is for


14/22


Table V. Variance Component Estimates for Person ´ Setting ´ Day (P ´ S ´ D)

Design (Data from 16-Day Pretreatment Assessment; N = 43)

Variance component estimation

Source SBP DBP

Person (P) 107.57 54.47

Setting (S) 0.00a 0.53

Day (D) 0.00a 0.32

P ´ S 20.91 7.17

P ´ D 9.22 7.74

S ´ D 0.30 0.00a

P ´ S ´ D 59.66 34.06

aActual estimates were negative and close to zero and, therefore, were replaced with

zeros as recommended by Cronbach et al. (1972).

substituted into the G *-coef®cient formula. The resulting coef®cients are pre-

sented in Table VI. Taking the .80 standard, those coef®cients indicated

that, when only one setting is sampled, 7 or more DBP readings (days;

one reading per day) would permit generalizations across days and across

settings; however, 12 or more SBP readings (days; one reading per day)

may need to be taken to achieve similarly reliable results. Sampling two or

three settings reduced the number of readings needed to be taken in order

to obtain measures that were generalizable across days and across settings,

particularly for SBP. For both SBP and DBP, three BP readings taken in

each of two settings, or two BP readings taken in each of three settings

Table VI. G * Coef®cients for Person ´ Setting ´ Day (P ´ S ´ D) Design (Data from 16-Day

Pretreatment Assessment; N = 43)

G * coef®cient

SBP DBP

No.

of days 1 setting 2 settings 3 settings 1 setting 2 settings 3 settings

1 .45 .68 .75 .52 .65 .71

2 .66 .78 .83 .65 .77 .81

3 .71 .82 .86 .71 .82 .86

4 .74 .84 .88 .75 .84 .885 .75 .85 .89 .77 .86 .89


15/22


would allow one to attain BP measures generalizable across settings and across

days.

DISCUSSION

The principal aim of this study was to determine how many self-measured

BP readings are enough to attain reliable estimates of true BP in hypertensive

patients. Taking a test±retest correlation coef®cient of .80 or higher as indica-

tive of reliable measurement, our results concerning test±retest correlations and

standard deviations of the differences between pairs of measurements indicate

that, in general, a single reading taken either at home in the morning, at home in

the evening, or at work on 1±6 consecutive days would yield reliable BP mea-

surements over 1 week and 2 months. To achieve reliable estimates of BP over

6 months, a single reading on 1±2 consecutive days seem to be required in the

case of DBP, whereas the readings from more than 8 days seem to be required

in the case of SBP.

Test±retest correlations and standard deviations of the differences between

pairs of measurements enable one to study only reliability over time, that is, to

determine the number of readings that are needed to ensure acceptable levels of

generalizability across days. When generalizations across days and settings areintended, G -theory provides a more adequate methodology for answering reli-

ability questions in BP measurement. Thus, for example, given that test±retest

correlations and standard deviations of the differences do not take into account

the variability across settings, their results indicate that the sampling of two or

three settings, in comparison to the use of a single setting, did not decrease the

minimum number of days needed to get reliability coef®cients of around .80

over 1 week, over 2 months, or over 6 months or decreased it slightly. Never-

theless, the results based on G -methodology show that sampling two or threesettings reduces the number of readings needed to be taken in order to obtain

measures that are generalizable across days and across settings, particularly for

SBP because of its greater random variation across settings.

This discrepancy illustrates the usefulness of G -theory for studying the reli-

ability of BP measurement, but also it underscores the importance of specifying

the universe to which practitioners or researchers wish to generalize the BP self-

measurements. When generalizations across days and settings are intended, the

results of the present study indicate that a single reading from each of 3 days

taken in each of two settings (six readings in total) or a single reading from each


16/22


indexes of reliability has its limitations, their results may become very informa-

tive when they are integrated with those obtained with G -methodology. Thus, for

example, a visual inspection of our results that the reliability indexes (test±retest

correlation coef®cient and standard deviation of differences) obtained from self-

measured BP taken at work were consistently, although not signi®cantly, greater

than those obtained from self-measured BP taken at home. Thus, if we wereasked to set simple criteria in order to determine the baseline for studies on

behavioral treatment of hypertension and we decided to sample two settings

instead of three, we would select work as one of these settings and conclude

that it is enough to take two readings, one at work and the other at home, from

each of 3 consecutive days to get reliable estimates of SDP and DBP across

settings, over 1 week and over 2 months. This same criterion would be valid

to get reliable estimates of DBP over 6 months, but 16 or more SBP read-

ings (2 readings per 8 days) may need to be taken to achieve similarly reliable

results.

These recommendations are consistent with those suggested by Llabre et al.

(1988b) after applying G -theory to ambulatory BPs in normotensives and mild

hypertensives. These investigators concluded that three readings in each of two

settings permitted generalizations of SBP, but ®ve or six DBP readings in each

of two settings were necessary for proper generalizability across settings. In fact,

reliability indexes found in the present study for self-measured BP are similar to those found in previous studies for ambulatory BPs. A summary of research

on reliability over time of ambulatory, clinic, and self-measured BP readings

in hypertensive patients is shown in Table VII. Although is very dif®cult to

reach ®rm conclusions from this table because studies differ in various aspects

(different intervals, different BP monitors, etc.), a visual inspection of Table

VII suggests that correlation coef®cients and standard deviations of differences

found in the present study for self-measured BP are at least comparable to those

found in the literature for clinic and ambulatory BPs.Although this study was concerned with subjects similar to those likely to

be recruited in studies on behavioral treatment of hypertension, that is, subjects

whose BPs remain in the hypertensive range despite previous pharmacological

treatments or despite the current use of medication, the results of the whole

sample (medicated and unmedicated) and the unmedicated subgroup were quite

similar and led to comparable recommendations in terms of the minimal adequate

number of BP readings to obtain reliable estimates of BP over 1 week.

Several studies (e.g., Gellman et al., 1986, 1990) have demonstrated that

factors such as position or activity can in¯uence BP variability Therefore the


17/22


Table VII. Studies of the Reliability of Ambulatory, Clinic, and Self-Measured BP Readings in

Hypertensive Patients

Reference(s) Subject Interval Corr elation SDD

Ambulatory BP

Coats (1990) 100 1 month Ð / 0.83 Ð / 6.5Des Combes et al. (1984) 84 3 months 0.82 / 0.78 Ð Fotherby & Potter (1993) 22 10 weeksa 0.93 / 0.51 6.3 / 4.8Mansoor et al. (1994) 24 15 monthsa 0.87 / 0.90 9.8 / 4.7Marolf et al. (1987) 31 1 week Ð 8 / 9Weber et al. (1982) 6 2 weeks 0.72 / 0.76 Ð

Clinic BP

Coats (1990) 100 1 month Ð / 0.59 Ð / 12.6Des Combes et al. (1984) 101 3 months 0.82 / 0.79 Ð

Fotherby & Potter (1993) 22 10 weeksa 0.51 / 0.89 17.4 / 7.0Mancia et al. (1994) 13 4 weeks 0.90 / 0.91

b 10 / 6.3b

0.93 / 0.95 c 6.7 / 4.2c

Mansoor et al. (1994) 24 15 monthsa 0.48 / 0.31 17 / 10Marolf et al. (1987) 31 1 week Ð 11 / 8Watson et al. (1987) 32 1 week Ð 10 / 7

Self-measured BP

Kleinert et al. (1984) 27 2 weeks 0.77 / 0.80 Ð Present studyd 43 1 week 0.88 / 0.85

e 5.7 / 4.8e

0.90 / 0.89 f 5.2 / 4.2

f

21 2 months 0.85 / 0.87e 6.7 / 4.5

e

0.88 / 0.88 f 5.8 / 4.6

f

19 6 months 0.75 / 0.88 e 8.6 / 4.8e0.79 / 0.90

f 7.9 / 4.5 f

Note Interval: interval between measurements or visits. Correlation: test±retest correlation

coef®cient for SBP / DBP. SDD: standard deviation of the differences between measurements forSBP / DBP (expressed as mm Hg).aMedian time interval between measurements.bValues estimated from the average of three readings per day.cValues estimated from the average of 25 readings per day.d Data from medicated and unmedicated patients.eValues estimated considering the average of four readings taken at work during 4 days (one reading

per day) as the ®rst measurements. f Values estimated considering the average of eight readings taken at work during 8 days (one reading

per day) as the ®rst measurement.

lines, studies on ambulatory and clinical blood pressure have found system-

atic setting differences among home, work, and clinic readings (e.g., Picker-

ing et al., 1985). However, in the present study, systematic variability due to

setting (home±morning, home±evening, work) was minimal and self-measured

BPs seemed to be generalizable across settings This ®nding may be also partly


18/22


to standardize measurement conditions and, hence, to increase the reliability of

self-measured BPs.

In all three self-measurement settings considered in the present study, we

found that DBP self-measurements tended to be more reliable over 1 week, over

2 months, and over 6 months than SBP self-measurements, although the dif-

ferences were not statistically signi®cant. As Table VII shows, several studies have also reported higher test-retest correlations and / or lower standard deviations of the differences between measures of DBP than between measures of

SBP [Fotherby and Potter (1993) for clinic BP; Kleinert et al. (1984), Mansoor

et al. (1994), and Weber et al. (1982) for ambulatory BP]. However, other studies

have reported the opposite pattern, especially concerning the test±retest correla-

tions [Fotherby and Potter (1993) for ambulatory BP; Des Combes et al. (1984);

Mansoor et al., (1994) for clinic BP]. Anyway, the trend in our data indicating

that SBP measurements seems to be less reliable than DBP measurements, if

con®rmed, would have signi®cant implications, because, although DBP has been

commonly regarded as being more important clinically than SBP, this consid-

eration has shifted in the last 10±15 years. A large number of epidemiological

studies have shown, almost without exception, that SBP is a more important

determinant of the risk of cardiovascular morbidity than DBP (see the review

by Rutan et al., 1988). Therefore, if the focus of therapeutic decisions shifts

from DBP to SBP, clinicians and researchers should be alerted about the need to obtain more readings of SBP than of DBP to achieve reliable estimates of

BP and, therefore, alerted about the possibility that standard measurement pro-

cedures for BP determination may not be useful enough for obtaining reliable

estimates of SBP, although they can obtain reliable estimates of DBP.

On the other hand, our results indicate that some demographic and clini-

cal factors in¯uence self-measured BP variability and, therefore, reliability of

BP self-measurements. Particularly, we found that self-measured BP variabil-

ity, especially self-measured SBP variability, increased progressively with aging. This ®ndings is consistent with the results reported by previous studies assessing

BP variability by ambulatory monitoring (e.g., Floras et al., 1988; Mancia et al.,

1980; Mora and Oc Âon, 1991); in fact, some of these previous studies have alsofound that the relation between BP variability and age is more evident for SBP

than for DBP (Mora and Oc Âon, 1991). This relation may be partly accounted forby the diminished barore¯ex sensitivity associated with aging, although other

factors may also be involved (Floras et al., 1988; Mancia et al., 1980). We also

found that increased self-measured DBP variability was related to higher initial

levels of self-measured DBP partly replicating the results of previous studies


19/22


cially of SBP, with older patients, more self-measured BP readings seem to be

required in comparison to the baseline criteria proposed above. Likewise, more

self-measured BP readings seem to be required to achieve reliable estimates of

DBP with patients who initially show high levels of self-measured DBP.

Following Shepard’s (1981) suggestion, we arbitrarily chose a reliability

coef®cient of .80 as indicative of reliable measurement over time. For each spe-ci®c work, practitioners and researchers have to decide whether this standard is

appropriate or not. Results reported here (see Tables II±IV and VI) allow prac-

titioners and researchers to determine different baseline lengths and to sample

two or three settings to get higher standards of reliability, even around .90. In

any case, this study presents reliability data of self-measured BP across settings

and over short and long time intervals that suggest that self-measurement of

BP allows one to attain stable readings of BP which are at least representative

of the natural environments in which patients spend a large part of their day.

Therefore, self-measurement of BP may be useful to overcome the limitations

of clinic BP both for the initial evaluation of hypertension and for checking the

response to behavioral treatment over extended periods, ranging, at least, from

2 to 6 months.

ACKNOWLEDGMENTS

We are grateful to Dr. Mar Âõa Rosario Mart Âõnez-Arias for her valuable sug-gestions on generalizability analyses. We would also like to thank two anony-

mous reviewers for their helpful comments on an early version of this article.

REFERENCES

American Society of Hypertension (1992). Recommendations for routine blood pressure measure-

ment by indirect cuff sphygmomanom etry. Am. J. Hypertens. 5(4): 207 ± 209.

BMDP (1990). BM DP Statistical Software , University of California Press, Berkeley.

Chatellier, G., Dutrey-Dupagne, C., Vaur, L., Zannad, F., Gen Áes, N., Elkik, F., and M Âenard, J. (1996).Home self blood pressure measurement in general practice. The SMART study. Am. J. Hyper-

tens. 9(7): 644 ± 652 .

Coats, A. J. S. (1990). Reproducibility or variability of casual and ambulatory blood pressure data:

Implications for clinical trials. J. Hipertens. 8 (Suppl. 6): S17 ±S20.

Cronbach, L. J., Glaser, G. C., Nanda, H., and Rajaratnam, N. ( 1972). The Dependability of Behav-

ioral Measurements, Wiley, New York.Des Combes, B. J., Porchet, M., Waeber, B., and Brunner, H. R. (1984). Ambulatory blood pressure

di R d ibili d di bili H i 6 110 114


20/22


Garc Âõa-Vera, M. P., Labrador, F., and Arribas, J. M. ( 1993). El autorregistro de la presi Âon arte-rial para el diagn Âostico y manejo de la hipertensi Âon arterial [Self-recorded blood pressure for

the diagnosis and management of essential hypertension]. In Arribas, J. M., and Caballero, F.

(eds.), Manual de cirug Âõa menor y otros procedimientos en la consulta del m Âedico de familia[Handbook of minor surgery and other procedures in family physician’s of®ce], Merck Sharp

& Dohme, Madrid, pp. 517 ± 521 .

Garc Âõa-Vera, M. P., Labrador, F., and Sanz, J. (1997). Stress management training for essential hypertension: A controlled study. Appl. Psychophysiol. Biofeedback 22(4): 261 ± 283 .Gellman, M. D., Ironson, G. H., Spitzer, S. B., Keenan, M., Schneiderman, N., and Weidler, D. J.

(1986). Ambulatory blood pressure as a function of race, gender, place, and mood. Circulation

74 (Suppl. II): 319.

Gellman, M. D., Spitzer, S. B., Ironson, G. H., Llabre, M., Saab, P., Pasin, R. D., Weidler, D. J.,

and Schneiderman, N. (1990). Posture, place, and mood effects on ambulatory blood pressure.

Psychophysiology 27(5): 544 ± 551.

Jacob, R. G., Chesney, M. A., Williams, D. M., Yijun Ding, B. S., and Shapiro, A. P. (1991). Relax-

ation therapy for hypertension: Design effects and treatment effects. Ann. Behav. Med. 13(1):

5 ± 17 .

Kleinert, H. D., Harsh®eld, G. A., Pickering, T. G., Devereux, R. B., Sullivan, P. A., Marion, R. M.,

Mallory, W. K., and Laragh, J. H. (1984). What is the value of home blood pressure measure-

ment in patients with mild hypertension? Hypertension 6(4): 574 ± 578 .

Llabre, M. M., Ironson, G. H., Spitzer, S. B., Gellman, M. D., Weidler, D. J., and Schneiderman,

N. (1988a). How many blood pressure measurements are enough?: An application of general-

izability theory to the study of blood pressure reliability. Psychophysiology 25(1): 97 ± 106 .

Llabre, M. M., Ironson, G. H., Spitzer, S. B., Gellman, M. D., Weidler, D. J., and Schneiderman, N.

(1988b). Blood pressure stability of normotensives and mild hypertensives in different settings.

Health Psychol. 7 (Suppl.): 127 ± 137.

Mancia, G., Ferrari, A., Gregorini, L., Parati, G., Pomidossi, G., Bertinieri, G., Grassi, G., and

Zanchetti, A. (1980). Blood pressure variability in man: its relation to high blood pressure,

age, and barore¯ex sensitivity. Clin. Sci. 59 (Suppl. 6): 401S± 404S.

Mancia, G., Ulian, L., Parati, G., and Trazzi, S. (1994). Increase in blood pressure reproducibility by

repeated semi-automatic blood pressure measurements in the clinic environment. J. Hypertens.

12(4): 469 ± 474.

Mansoor, G. A., McCabe, E. J., and White, W. B. (1994). Long-term reproducibility of ambulatory

blood pressure. J. Hypertens. 12(6): 703 ± 708.

Marolf, A. P., Hany, S., B Èattig, B., and Vetter, W. (1987). Comparison of casual, ambulatory andself-determined blood pressure measurement. Nephron 47 (Suppl. 1): 142 ± 145.

Ministerio de Sanidad y Consumo (1990). Consenso para el control de la hipertensi Âon arterial en

Espa Äna [Consensus for the control of hypertension in Spain], Ministerio de Sanidad y Consumo,Madrid.

Mora, J., and Oc Âon, J. (1991). Variabilidad de la presi Âon arterial y de la tasa card Âõaca in hipertensi Âonarterial esencial [Variability of arterial pressure and of heart rate in essential arterial hyperten-

sion]. Med. Clin. (Barc.). 97(8): 292 ± 296.

Novo, S., Alaimo, G., Abrignami, M. G., Longo, B., Muratore, G., and Strano, A. (1989). Noninva-

sive blood pressure monitoring evaluation of verapamil slow-release 240-mg antihypertensive

effectiveness. J. Cardiovasc. Pharmacol. 13 (Suppl): 38 ± 41 .

Pickering, T. G. (1991). Ambulatory Monitoring and Blood Pressure Variability , Science Press, Lon-

don.

Pickering, T. G., Harsh®eld, G. A., Devereux, R. B., and Laragh, J. H. (1985). What is the role ofambulatory blood pressure monitoring in the management of hypertension patients? Hyperten-

i 7 171 177


21/22


Verdecchia, P., Bentivoglio, M., Provvidenza, M., Savino, K., and Corea, L. (1985). Reliability of

self-recorded arterial pressure in essential hypertension in relation to the stage of the disease.

In Rome, G. G. (ed.), Blood Pressure Recording in the Clinical Management of Hypertension ,

Edizioni L. Pozzi, Roma, pp. 40 ± 42 .

Watson, R. D. S., Lumb, R., Young, M. A., Stallard, T. J., Davies, P., and Littler, W. A. (1987).

Variation in cuff blood pressure in untreated outpatients with mild hypertension: Implications

for initiating antihypertensive treatment. J. Hypertens. 5: 207 ± 211.

Weber, M. A., Drayer, J. I. M., Wyle, G. A., and Young, J. L. ( 1982). Reproducibility of the whole-

day blood pressure pattern in essential hypertension. Clin. Exp. Hypertens. A4: 1377 ± 1390.


22/22

Reproducedwithpermissionof thecopyrightowner. Further reproductionprohibitedwithoutpermission.

how many self-measured blood pressure readings are needed to

Documents