www.thelancet.com Vol 375 March 13, 2010 895
Lancet 2010; 375: 895–905
See Comment page 867
See Articles page 906
See Review page 938
See Lancet Neurol
Stroke Prevention Research
Unit, University Department of
Clinical Neurology, John
Radcliff e Hospital, Oxford, UK
(Prof P M Rothwell FMedSci,
S C Howard DPhil); Stroke and
Hypertension Unit, Connolly
Hospital, Dublin, Ireland
(E Dolan MRCP); Conway
Institute of Biomolecular and
University College Dublin,
Belfi eld, Dublin, Ireland
(Prof E O’Brien FRCP);
International Centre for
Circulatory Health, Imperial
College London, London, UK
(J E Dobson MSc,
Prof P S Sever FRCP,
N R Poulter FMedSci); and
Department of Medicine,
University of Goteborg,
(Prof B Dahlöf MD)
Prof Peter M Rothwell, Stroke
Prevention Research Unit,
University Department of Clinical
Neurology, Level 6, West Wing,
John Radcliff e Hospital,
Headington, Oxford OX3 9DU, UK
Prognostic signifi cance of visit-to-visit variability,
maximum systolic blood pressure, and episodic
Peter M Rothwell, Sally C Howard, Eamon Dolan, Eoin O’Brien, Joanna E Dobson, Bjorn Dahlöf, Peter S Sever, Neil R Poulter
Background The mechanisms by which hypertension causes vascular events are unclear. Guidelines for diagnosis and
treatment focus only on underlying mean blood pressure. We aimed to reliably establish the prognostic signifi cance
of visit-to-visit variability in blood pressure, maximum blood pressure reached, untreated episodic hypertension, and
residual variability in treated patients.
Methods We determined the risk of stroke in relation to visit-to-visit variability in blood pressure (expressed as
standard deviation [SD] and parameters independent of mean blood pressure) and maximum blood pressure in
patients with previous transient ischaemic attack (TIA; UK-TIA trial and three validation cohorts) and in patients with
treated hypertension (Anglo-Scandinavian Cardiac Outcomes Trial Blood Pressure Lowering Arm [ASCOT-BPLA]). In
ASCOT-BPLA, 24-h ambulatory blood-pressure monitoring (ABPM) was also studied.
Findings In each TIA cohort, visit-to-visit variability in systolic blood pressure (SBP) was a strong predictor of
subsequent stroke (eg, top-decile hazard ratio [HR] for SD SBP over seven visits in UK-TIA trial: 6·22, 95% CI
4·16–9·29, p<0·0001), independent of mean SBP, but dependent on precision of measurement (top-decile HR over
ten visits: 12·08, 7·40–19·72, p<0·0001). Maximum SBP reached was also a strong predictor of stroke (HR for top-
decile over seven visits: 15·01, 6·56–34·38, p<0·0001, after adjustment for mean SBP). In ASCOT-BPLA, residual
visit-to-visit variability in SBP on treatment was also a strong predictor of stroke and coronary events (eg, top-decile
HR for stroke: 3·25, 2·32–4·54, p<0·0001), independent of mean SBP in clinic or on ABPM. Variability on ABPM
was a weaker predictor, but all measures of variability were most predictive in younger patients and at lower (<median)
values of mean SBP in every cohort.
Interpretation Visit-to-visit variability in SBP and maximum SBP are strong predictors of stroke, independent of
mean SBP. Increased residual variability in SBP in patients with treated hypertension is associated with a high risk of
Hypertension is the most prevalent treatable risk factor
for stroke and other vascular events.1,2 Underlying usual
blood pressure (conceived as the true underlying average
blood pressure over a period of time) is widely considered
to be of primary importance in the cause of vascular
disease,3,4 and hence in diagnosis and treatment of
hypertension,5–7 and this notion underpins all major
clinical guidelines.8–11 Yet, the mechanisms by which
raised blood pressure causes stroke and other vascular
events are poorly understood. Mean blood pressure is
clearly important, but other factors, such as variability or
maximum blood pressure reached, might also play a
part,12 particularly at older ages when most vascular
events occur.13 However, visit-to-visit variability in blood
pressure is usually dismissed as random, noteworthy
only as an obstacle to the reliable estimation of usual
blood pressure.14–18 Consequently, although substantial
visit-to-visit variability in clinic blood pressure is
common,19–24 episodic hypertension tends not to be
treated.12 In patients with occasional high blood pressure,
guidelines recommend continued monitoring or 24-h
ambulatory blood-pressure monitoring (ABPM),8–11 with
treatment decisions based on mean blood pressure. Yet,
although situational variability in blood pressure has
been studied,25,26 the prognostic value of visit-to-visit
variability and episodic hypertension in the same setting
has not been reliably established.
We showed previously that visit-to-visit variability in
blood pressure is increased in cohorts at high risk of
stroke,19,20 that it is consistent within individuals over
time (ie, not random),27 and that it seems to predict stroke
independently of mean systolic blood pressure (SBP).28
Prompted by these observations and by shortcomings in
the usual blood-pressure hypothesis,12 we aimed to
reliably establish the prognostic signifi cance of visit-to-
visit variability in blood pressure, maximum blood
pressure reached, episodic hypertension, and residual
variability in blood pressure in patients already receiving
antihypertensive drugs. We studied a large cohort of
patients with previous transient ischaemic attack (TIA;
UK-TIA aspirin trial),29 with validation in three similar
www.thelancet.com Vol 375 March 13, 2010
cohorts,30–32 and a broad population of patients with
hypertension in the Anglo-Scandinavian Cardiac
Outcomes Trial Blood Pressure Lowering Arm
(ASCOT-BPLA).32,33 In ASCOT-BPLA, we also measured
the prognostic value of short-term variability during
individual visits and on 24-h ABPM.
The UK-TIA aspirin trial was a double-blind randomised
trial of aspirin (1200 mg vs 300 mg vs placebo) in
2435 patients with a recent TIA or ischaemic stroke,
which was undertaken from 1979 to 1985.29 Visit-to-visit
variability in blood pressure was not aff ected by the
randomised treatment.27 To avoid confounding due to
any eff ect of recent stroke on variability in blood
pressure,34,35 analysis was confi ned to 2006 patients
presenting with TIA only. Sitting blood pressure was
measured once at every 4-month follow-up visit with a
mercury sphygmomanometer and with the patient
rested. Details of all vascular events and deaths were
recorded during follow-up, and reviewers were masked
to treatment allocation and data for blood pressure.
The main results from the UK-TIA analysis were tested
in three other TIA and stroke cohorts. The fi rst cohort
was from the European Stroke Prevention Study
(ESPS-1),30 in which 2500 patients were randomly
assigned to dipyridamole 75 mg plus aspirin 325 mg
versus placebo three times daily. Blood pressure (mean
of left and right arm; sitting after rest; mercury
sphygmomanometer) was measured at follow-up visits
every 3 months for 2 years. Because dipyridamole is itself
vasoactive, we studied only the placebo group. The second
cohort was from the Dutch TIA trial,31 in which
3150 patients were randomly assigned to aspirin 30 mg
versus aspirin 283 mg. A subgroup of 1473 patients was
randomly assigned to atenolol 50 mg versus placebo.
Blood pressure (sitting
sphygmomanometer) was measured at follow-up visits
every 4 months for a mean of 2·6 years. The third cohort
was from the ASCOT-BPLA trial,32,33 which included
2011 patients with previous TIA or stroke.
The generalisability of the fi ndings in the TIA cohorts
was tested in the main ASCOT-BPLA trial.32,33 Patients
with hypertension, aged 40–79 years, with three or more
other vascular risk factors, but no coronary heart disease,
were randomly assigned, by the PROBE design, to one of
two antihypertensive regimens instead of any existing
antihypertensive drugs: amlodipine adding perindopril
as needed (amlodipine-based) versus atenolol adding
bendrofl umethiazide and potassium as needed (atenolol-
based). Treatment was titrated to achieve a clinic blood
pressure of less than 140/90 mm Hg, or less than
130/80 mm Hg in patients with diabetes. Patients with
total cholesterol 6·5 mmol/L or higher could also be
randomly assigned to atorvastatin 10 mg daily or to
placebo. At every visit (baseline, 6 weeks, 3 months,
after rest; mercury
Decile of SD SBP
Hazard ratio (95% CI)
Hazard ratio (95% CI)
Hazard ratio (95% CI)
Figure 1: Hazard ratios for risk of any subsequent stroke by deciles of SD SBP
based on the fi rst seven measurements (baseline to 2 years) in the UK-TIA
trial, with the fi rst decile as the reference category
Analyses for all patients (A), excluding those with a past history of stroke (B),
and excluding those with either a past history of stroke or infarction on baseline
CT brain imaging (C). SBP=systolic blood pressure.
www.thelancet.com Vol 375 March 13, 2010 897
6 months, and every 6 months thereafter) clinic blood
pressure was measured three times in the sitting position
after 5 min rest with a validated, semiautomated
oscillometric device (Omron HEM-705CP, OMRON
Healthcare, Kyoto, Japan).36 Participants at four centres
had yearly 24-h ABPM (SpaceLabs 90207, SpaceLabs,
Hertford, UK) with blood-pressure readings every
30 min.37–39 No editing criteria were applied to individual
readings. Mean time-weighted daytime (0900–2100 h),
night-time (0100–0600 h), and 24-h SBP and diastolic
blood pressure (DBP) were calculated.39
Visit-to-visit variability of blood pressure was defi ned as
the standard deviation (SD) or coeffi cient of variation
(SD/mean). Since the coeffi cient of variation and mean
blood pressure can still be correlated, we created a further
proportional to SD/meanx, with x derived from curve
fi tting. In the UK-TIA cohort, variables were calculated
from visits 1 to 7 (0–24 months) and from 1 to 10
(0–36 months)—the latter providing more reliable
estimates of blood-pressure values (webappendix p 1),
but less subsequent follow-up to predict outcomes. In
the other cohorts, variability was calculated over all
follow-ups and separately over an initial period up to
We measured the contributions of within-individual
visit-to-visit variability versus
diff erences in mean blood pressure to the distribution of
group blood pressure at every follow-up visit. Expected
variance of SBP values at one follow-up was estimated as
the sum of: between-individual variance in mean SBP,
within-individual visit-to-visit variance in SBP, and
between-individual variance in within-individual visit-to-
visit SBP. The proportion of variance attributable to
within-individual variability was estimated as the sum of
the latter two components divided by the total variance
(webappendix p 2).
In the UK-TIA cohort, mean and visit-to-visit variability
in blood pressure during initial follow-up were related to
risk of subsequent stroke. Patients who had events in the
measurement period were excluded. To allow for non-
linearity, variables were split into deciles and hazard
ratios (HRs) calculated in relation to the fi rst decile.
Analyses were adjusted for mean blood pressure during
the measurement period, and for age, sex, and baseline
vascular risk factors. We also measured: (1) overall
predictive power (area under receiver operating
characteristic curve) of each blood-pressure parameter;
(2) eff ect of increasing reliability of estimation of
variability by analyses limited to patients with increasing
numbers of blood-pressure readings before stroke;
(3) predictive value of maximum and minimum blood
pressure; and (4) risk in four exclusive categories based
on maximum and minimum SBP: stable normotension
(maximum ≤140 mm Hg), episodic moderate hyper-
independent of mean,
tension (minimum ≤140 mm Hg and maximum
140–179 mm Hg), episodic severe hypertension
(minimum ≤140 mm Hg and maximum ≥180 mm Hg),
and stable hypertension (minimum >140 mm Hg).
In ASCOT-BPLA, the mean of the second and third
readings at every visit was used to calculate blood
pressure and heart rate. Estimation of visit-to-visit
variability took account of the initial reduction in blood
pressure attributable to treatment initiation and dose
adjustment: (1) mean blood pressure and variability were
based only on readings from the 6-month follow-up
onwards; (2) we calculated average successive variability
(average absolute diff erence between successive values);
and (3) for patients with three or more visits after
6 months we calculated variability over and above any
linear time trend in blood pressure—residual SD, the
square root of the total squared deviation of data points
from a linear regression of blood pressure against time,
divided by (n–2), where n is the number of readings.
Blood-pressure parameters, split into deciles with both
treatment groups combined, were related to the risks of
stroke and of the predefi ned ASCOT total coronary events
outcome,32,33 taking the fi rst decile of the amlodipine
group as the reference. Analyses included only events
occurring after the 6-month follow-up visit. Several
sensitivity analyses were done: (1) adjusted for mean
blood pressure during the measurement period, age, and
sex; (2) further adjusted for all baseline variables in the
HR for mean SBP HR for variability in SBP
HR (95% CI)p value HR (95% CI)p value
Every row shows the estimates from a Cox model applied to data from patients who survived for at least n follow-up
visits, where n ranges from 2 (3 months) to 10 (3 years). Quintiles were used rather than deciles to provide suffi cient
group sizes to extend the analysis to ten blood-pressure readings. SBP=systolic blood pressure. HR=hazard ratio.
CV=coeffi cient of variation. VIM=variation independent of mean.
Table 1: Hazard ratios (top vs bottom quintile) for risk of subsequent stroke (ie, after the measurement
period) in the UK-TIA trial from a model combining mean SBP and visit-to-visit variability in SBP (SD or
CV or VIM), repeated with increasingly precise estimates of both variables
See Online for webappendix
www.thelancet.com Vol 375 March 13, 2010
previous ASCOT-BPLA analysis;33 (3) stratifi ed by mean
SBP during follow-up (above vs below the median);
(4) blood-pressure parameters calculated from only
6–30-month visits (n=5) and related to risk of events after
30 months; (5) on-treatment analysis (ie, patients who
were consistently complaint with medication, as defi ned
in previous reports);32,33 (6) adjusted for randomisation to
statin treatment; (7) excluding patients with past history
of TIA or stroke; and (8) adjusted for visit-to-visit
coeffi cient of variation of heart rate from all visits from
6 months onwards (SD and mean heart rate were
correlated; r²=0·34, p<0·0001). The number of
antihypertensive drugs being taken at last follow-up was
also related to visit-to-visit variability and maximum SBP.
Within-visit variability in blood pressure was expressed
as the SD and range of the three readings at each visit,
averaged across all visits from 6 months onwards.
In the ABPM substudy, we correlated daytime and night-
time variability in ABPM blood pressure with visit-to-visit
variability. To measure any eff ect of diff erences in time of
day of blood-pressure measurement on observed visit-to-
visit variability, we used data from repeated ABPMs to
compare variability between ABPM in SBP measured at
the same time of day during offi ce hours (0900–1700 h),
with readings selected at random during offi ce hours. We
also correlated the morning surge (highest SBP from
0900–1100 h minus lowest from 0600–0800 h) with visit-
to-visit variability in clinic blood pressure. We compared
predictive value of ABPM variability versus visit-to-visit
variability in blood pressure for stroke and coronary events
with Cox models both for the fi rst ABPM during follow-up
and for parameters averaged across all ABPMs. In view of
the smaller numbers of outcomes in the ABPM substudy,
HRs were calculated for continuous variables with both
randomised treatment groups combined.
Role of the funding source
There was no funding for this study. ASCOT was an
investigator designed and led study. None of the sponsors
of ASCOT-BPLA had any input into the design,
performance, analysis, or reporting of the analyses
reported in this Article. The report was sent to the major
sponsor for information before fi nal acceptance for
publication. The contractual agreement between Imperial
College, London, UK, and the sponsor allows the sponsor
the opportunity to see and comment on any report, but
not to exercise any right of veto. The corresponding
author had full access to all the data in the study and had
fi nal responsibility for the decision to submit for
In the UK-TIA cohort, 2006 patients (1438 men; mean
age 60·3 years, SD 9·1; median time since TIA 23 days,
IQR 8–46) had a median of 10 (range 1–20) follow-up
visits before stroke or death. Results of analyses based on
pulse pressure and SBP were similar (data not shown).
Mean SBP was 150·3 mm Hg (SD 25·3) at baseline and
fell to 146·4 mm Hg (23·3) at 1 year, but was stable
thereafter (webappendix p 13). However, systolic blood
pressure in individuals was highly variable from one visit
to the next (r²=0·25–0·35, webappendix p 14). Within-
individual visit-to-visit variability accounted for 41·5%
of the variance in group SBP at each follow-up
(webappendix p 2).
Visit-to-visit coeffi cient of variation of SBP correlated
with mean SBP (r=0·22), but variation independent of
mean of SBP (SD/mean¹·⁶⁷) did not (r=0·01). Coeffi cient of
variation of DBP was not correlated with mean DBP (data
not shown). Reproducibility of variability was moderate
(eg, intraclass correlation coeffi cient [ICC] for SD
SBP=0·34, 95% CI 0·26–0·41, for visits 1–7 vs 8–14). Visit-
to-visit variability of SBP over visits 1–7 was unrelated to
subsequent mean SBP or DBP. Simulations showed that
mean blood pressure based on seven to ten readings
provided a reasonable estimate of usual blood pressure
(webappendix p 1), and reproducibility of mean blood
pressure (visits 1–7 vs 8–14) was good (ICC=0·76, 95% CI
Of 1324 (66%) patients who reached visit 7, 270 had a
subsequent stroke (n=104) or coronary event (166). Mean
SBP over visits 1–7 predicted stroke (HR 1·43, 95% CI
1·18–1·74 per 20 mm Hg, p<0·0001; top-decile HR 2·44,
1·56–3·82, after adjustment for age, sex, and baseline risk
factors). However, visit-to-visit variability in SBP was a
stronger predictor (top-decile HR for SD SBP: 6·22,
Hazard ratio (95% CI)
Category of maximum SBP
Figure 2: Hazard ratios for risk of any subsequent stroke by categories of
maximum SBP of the fi rst seven measurements of blood pressure during the
fi rst 2 years of follow-up in the UK-TIA trial, adjusted for mean SBP during the
The fi rst category is the reference category. Digit preference in the recording of
high values limited maximum SBP to eight roughly equally sized categories
based on visits 1–7 (≤140, 141–150, 151–160, 161–170, 171–180, 181–190,
191–219, and ≥220 mm Hg). SBP=systolic blood pressure.
www.thelancet.com Vol 375 March 13, 2010 899
4·16–9·29, fi gure 1; HR 4·37, 2·73–6·99, after adjustment
for mean SBP, age, sex, and other risk factors; HR 12·08,
7·40–19·72, when based on visits 1–10; p<0·0001). Visit-to-
visit variation independent of mean of SBP also predicted
stroke (visit 1–7 top-decile HR: 3·27, 2·06–5·21, p<0·0001),
both in patients receiving (3·67, 2·34–5·75, n=808) and
not receiving (2·27, 1·41–3·67, n=516) antihypertensive
drugs at baseline; however, coeffi cient of variation of DBP
was a weak predictor (1·37, 0·81–2·31).
Predictive power of visit-to-visit variability of SBP
increased with precise estimation (table 1). Results were
unaff ected by exclusion of 43 patients with previous
stroke (visit 1–7 top decile HR for SD SBP: 8·23,
5·51–12·30, p=0·001) and a further 90 patients with
asymptomatic infarction on baseline CT (HR 10·44,
6·65–16·38, p=0·002; fi gure 1); adjustment for any
temporal trend in blood pressure in individuals (residual
SD, data not shown); and adjustment for mean arterial
pressure or mean pulse pressure (data not shown).
Predictive value of visit-to-visit variability was similar in
men and women, but decreased with age (visit 1–7 top-
quartile HR of SD SBP by tertile of age: 9·43, 1·96–45·5
at <56 years; 3·01, 0·97–9·36 at 56–64 years; and 1·71,
0·74–3·98 at ≥65 years).
Maximum SBP predicted stroke independently of mean
SBP (fi gure 2; adjusted HR for top-decile over seven visits:
15·01, 6·56–34·38, p<0·0001, after adjustment for mean
SBP), with the strength of the association increasing with
the number of visits used (top-quintile adjusted HR: 3·03,
1·18–7·76, p=0·021, based on four visits and increasing to
11·74, 3·23–42·57, p<0·0001, based on ten visits). Maximum
SBP was more predictive of stroke than was mean SBP
(webappendix p 3) and maximum minus mean (peak) was
more predictive than was mean minus minimum (trough)
(webappendix p 4). Maximum SBP was most predictive at
lower values of mean SBP (based on visits 1–7, excluding
the maximum): HR for maximum SBP adjusted for mean
SBP was 4·95 (1·28–22·4, p=0·007) at mean SBP less than
130 mm Hg; 3·19 (1·65–6·23, p=0·0001) at 130–159 mm
Hg; and 1·13 (0·50–2·53, p=0·75) at 160 mm Hg or higher.
Patients with episodic severe hypertension had a higher
risk of stroke than did those with stable hypertension (36
[13·7%] vs seven [4·5%], p=0·003, webappendix p 5; age
and sex adjusted HR 3·58, 1·58–8·10) despite a lower mean
SBP (157·9 mm Hg [SD 8·7] vs 167·3 mm Hg [7·2],
Visit-to-visit variability in SBP was similar in the four
TIA cohorts (table 2), as was its contribution to the
variance in group SBP at follow-up visits (table 2,
webappendix p 2). Visit-to-visit variability of SBP was
consistently more predictive of stroke than was mean
SBP (table 2), and tended to be most predictive in patients
with lower baseline SBP (<median vs ≥median).
In the main ASCOT-BPLA study, 18 530 (96%) patients
had two or more scheduled follow-up visits from
6 months onwards (median 10, IQR 9–11). Visit-to-visit
variability in SBP was similar to that in the four TIA
cohorts and accounted for more than 50% of the
UK-TIA Aspirin trialASCOT–BPLA trial* ESPS-1†Dutch TIA trial‡
Atenolol groupAmlodipine group
Number of cases
Frequency of follow-up (months)
Mean (SD) baseline SBP
Mean (SD) 1 year SBP
Mean (SD) within-individual visit-to-visit variability in SBP
Range of SBP ≥50 mm Hg (%)
Group variance in SBP attributable to
intra-individual variation (%)
HR (95% CI) for stroke (unadjusted)
HR (95% CI) for stroke adjusted for mean SBP
1324 10129991247 3150
SBP=systolic blood pressure. CV=coeffi cient of variation. HR=hazard ratio. VIM=variation independent of mean. *Subgroup of patients with previous stroke or transient ischaemic
attack (TIA). †Placebo group only. ‡Results were similar after exclusion of the atenolol substudy (data not shown).
Table 2: Blood-pressure parameters and their predictive values (HRs and 95% CI for risk of stroke in the top vs bottom decile of each measure) in the four
independent cohorts of patients with TIA and minor stroke
www.thelancet.com Vol 375 March 13, 2010
variance in group SBP at each follow-up (webappendix
p 2). Reproducibility (visits at 6–36 vs 42–72 months) of
visit-to-visit (ICC 0·30, 95% CI 0·27–0·33) and within-
visit (0·43, 0·40–0·45) SD SBP was independent of
treatment group or seasonal trends in blood pressure.
Visit-to-visit SD SBP correlated with mean SBP (r=0·37)
as did coeffi cient of variation of SBP (r=0·17). Variation
independent of mean of SBP was SD/mean¹·⁷⁸.
Mean SBP was a weak predictor of stroke and coronary
events, whereas visit-to-visit variability was a strong
predictor of both (fi gure 3), independent of any time
trend in SBP during follow-up (eg, residual SD top-
decile HR for stroke in atenolol group: 3·96, 2·54–6·18).
Visit-to-visit variability in SBP was greater in the
atenolol group than in the amlodipine group,40 but the
risk relations were similar in both groups (fi gure 3),
and in the on-treatment cohort, and remained similar
after adjustment for age, sex, and mean SBP
(webappendix p 6) and all other baseline risk factors
(eg, adjusted HRs for variation independent of mean in
amlodipine group: stroke 2·97, 1·32–6·71; coronary
events 3·41, 1·98–5·88). Variability during the fi ve visits
at 6–30 months predicted risk of events after 30 months
(webappendix p 7). Variability in DBP was less predictive
(webappendix p 15). With the same categories of
behaviour of SBP defi ned in webappendix p 5 (based on
the visits at 6–30 months), episodic severe hypertension
was associated with a higher subsequent risk of stroke
than was stable hypertension (33/815 [4·0%] vs 75/2828
[2·7%], p=0·03), despite a lower mean SBP during the
risk period (142·1 mm Hg [SD 14·8] vs 147·3 mm Hg
Visit-to-visit variability in SBP was more predictive of
ischaemic than haemorrhagic stroke (webappendix p 8),
remained predictive after exclusion of patients with
previous TIA or stroke (top-decile HRs: average successive
variability=4·04, 2·39–6·83; variation independent of
mean=2·52, 1·58–4·03), was a stronger predictor of
stroke in patients with less than median (142·8 mm Hg)
mean SBP during follow-up (interaction p=0·006 for
variation independent of mean, fi gure 4), and predicted
risks of myocardial infarction, angina, and heart failure
(all p<0·0001, webappendix p 16). Prediction of stroke or
coronary events varied only in relation to age (interaction
p=0·01), the strongest association being in the youngest
(≤57 years) quartile (eg, top-decile HR for stroke: 5·06,
2·09–12·26, for visit-to-visit variation independent of
mean of SBP).
Since a high follow-up blood pressure would often
trigger patients to receive an add-on drug, use of three or
more agents increased with variation independent of
mean of SBP and with maximum SBP (webappendix p 9).
However, risks of stroke and coronary events still
increased in relation to maximum SBP (eg, top-decile
HR: 2·51, 1·69–3·73, p=0·0008, for risk of stroke in the
atenolol group). Minimum SBP did not predict stroke or
coronary events (data not shown).
Heart rate was not correlated with SBP at baseline or
on follow-up in either treatment group, or with visit-to-
visit SD SBP (r=0·01 in both groups). Visit-to-visit
coeffi cient of variation of heart rate was weakly correlated
with visit-to-visit variation independent of mean of SBP
(r²=0·02 in both groups), but was of little prognostic
value (webappendix p 17).
Within-visit SD SBP was a weak predictor of vascular
events (webappendix p 18; top-decile HR for stroke
adjusted for mean SBP: 1·52, 1·09–2·13). The
association with stroke risk was stronger in patients
with lower mean SBP and varied with age (interaction
p=0·04), with the strongest relations in the youngest
(≤57 years) quartile (eg, top-decile HR for stroke: 3·22,
1·16–8·89). However, the diff erence between the fi rst
SBP during a clinic visit and the mean of the second
two (white-coat eff ect) was not predictive of stroke or
coronary events (data not shown), and was not correlated
with visit-to-visit variability (r=0·01 for visit-to-visit SD,
coeffi cient of variation, and variation independent of
mean). Overall within-visit SD SBP was weakly
correlated with visit-to-visit variability (SD r=0·18,
p<0·0001; variation independent of mean r=0·23,
In the ASCOT-BPLA ABPM study, 1905 patients had a
mean of 3·25 (range 1–10) ABPMs from 6 months
onwards. In 843 patients with four or more ABPMs, the
diff erence in mean daytime SBP between adjacent
ABPMs after adjustment for any linear trend in that
individual (mean 7·9 mm Hg [SD 4·8]) correlated with
visit-to-visit variability in clinic SBP over the same
period (both as residual SD: r=0·34 based on ≥four
ABPMs; r=0·43 based on ≥fi ve ABPMs; both p<0·0001).
Inter-ABPM variability for single SBP readings at the
same time of day was consistent during 0900–1700 h;
SD SBP was 13·3 mm Hg at 1000 h versus 14·3 mm Hg
for times taken at random. The morning surge in SBP
was not correlated with visit-to-visit variability in clinic
SBP (r²=0·02) and did not predict stroke (HR per SD
increase: 0·88, 0·65–1·10, p=0·44).
Intra-ABPM SD of daytime SBP correlated with mean
daytime SBP (r=0·26, p<0·0001), but coeffi cient of
variation of daytime SBP did not (data not shown). Intra-
ABPM coeffi cient of variation SBP correlated with visit-to-
visit coeffi cient of variation in clinic SBP (atenolol group
r=0·38, amlodipine group r=0·29; both p<0·0001), but
visit-to-visit variability in clinic SBP was more predictive of
vascular events (webappendix p 10). However, both
parameters predicted risk of vascular events independently
of average daytime mean SBP across all ABPMs (HRs per
SD increase: daytime mean SBP=1·09, 95% CI 0·94–1·27,
p=0·26; daytime coeffi cient of variation SBP=1·17,
1·01–1·36, p=0·04; visit-to-visit coeffi cient of variation of
clinic SBP=1·48, 1·28–1·71, p<0·0001). Visit-to-visit
coeffi cient of variation of SBP during 6–30 months of
follow-up also predicted events thereafter independently of
mean and SD daytime SBP on ABPM (HR per SD
www.thelancet.com Vol 375 March 13, 2010 901
Number of patients Risk of strokeRisk of coronary events
Decile of measureDecile of measureDecile of measure
Hazard ratio (95% CI)
Hazard ratio (95% CI)
Hazard ratio (95% CI)
Hazard ratio (95% CI)
Figure 3: Distribution of patients in the two treatment groups in ASCOT-BPLA according to deciles of mean, SD, VIM, and ASV of SBP (left) and the associations of each of these variables with
risk of stroke (middle) and risk of coronary events (right)
The middle and right columns show the hazard ratios (95% CI) for risks of stroke and acute coronary events, respectively, by deciles of the same parameters. The fi rst decile in the amlodipine-based
group is the reference category. Numbers of outcome events by decile and treatment group are given below each graph. SBP=systolic blood pressure. VIM=variation independent of mean. ASV=average
www.thelancet.com Vol 375 March 13, 2010
increase=1·29, 1·08–1·55, p=0·007). Daytime coeffi cient
of variation of SBP on ABPM was most predictive in
patients with lower (<median) mean daytime SBP (HR per
SD increase: 1·42, 1·18–1·71).
Maximum daytime SBP on ABPM predicted risk of
stroke and coronary events (HR per SD for both: 1·23,
1·08–1·41, p=0·005; webappendix p 10), particularly after
adjustment for mean daytime SBP on ABPM (1·44,
1·12–1·84, p=0·002). Minimum daytime SBP on ABPM
was less predictive than was maximum daytime SBP
(webappendix p 10).
Variability in SBP is related to baseline characteristics
in webappendix p 11 for ASCOT-BPLA and in
webappendix p 12 for UK-TIA. Randomisation to
atorvastatin in the lipid-lowering group of ASCOT
resulted in a small reduction in variation independent
of mean of SBP (diff erence=0·33, 95% CI 0·54–0·13,
p=0·001), but all risk relations described above were
independent of this eff ect (data not shown).
We have shown that visit-to-visit variability in SBP is a
powerful predictor of stroke and coronary events
independent of mean SBP, that maximum SBP is more
predictive than is mean SBP (on clinic readings or on
ABPM), that residual variability in SBP on treatment
has a poor prognosis, and that stable hypertension has
a better prognosis than does episodic hypertension.
Along with accompanying reports,12,40,41 these fi ndings
challenge the usual blood-pressure hypothesis and have
implications for diagnosis, treatment, and monitoring
of patients with hypertension.
Patients at risk
Patients at risk
Hazard ratio (95% CI)
Hazard ratio (95% CI)
VIM SBPVIM SBP
SD SBPSD SBP
Risk of strokeRisk of coronary events
Decile of measure
Decile of measure
Figure 4: Hazard ratios for risks of stroke and acute coronary events in ASCOT-BPLA patients with mean SBP during follow-up less than the median value for
the trial population (<142·8 mm Hg) by deciles of SD and VIM SBP
Treatment groups are combined, but allocation is adjusted for in the model. The fi rst decile is the reference category. Deciles are based on the cut-points used for the
full trial population. SBP=systolic blood pressure. VIM=variation independent of mean.
www.thelancet.com Vol 375 March 13, 2010 903
There has been uncertainty about the prognostic value
of variability in blood pressure on ABPM.12 There is some
evidence that day-to-day variability in home blood
pressure predicts fatal stroke,42 but no previous studies of
the prognostic value of maximum blood pressure, and
only a few small studies of the prognosis of visit-to-visit
variability in clinic blood pressure.43–47 Hata and
colleagues44 showed that coeffi cient of variation of
previous clinic SBP from case notes was slightly greater
in 138 patients with stroke than in controls,43 but not in
patients with myocardial infarction.44 Variability in offi ce
blood pressure in 144 patients on renal dialysis predicted
mortality, but was based on only six vascular deaths.45
Havlik and co-workers46 reported a weak association
between SBP variability (three clinic readings over
6 years) and white matter lesions on brain imaging
25 years later. However, a substudy of the Syst-Eur trial47
recorded no relation between the SD of six blood-pressure
readings (from three visits over 8 weeks) and vascular
risk.47 None of these studies was adequately powered or
fully adjusted for mean blood pressure, or had enough
readings to estimate usual blood pressure reliably
(webappendix p 1). Our study of four large cohorts
provides reliable evidence that visit-to-visit variability in
SBP is a strong independent predictor of stroke and other
We studied trial cohorts with previous TIA from the
1980s because the high risk of stroke and the scrutiny of
stroke outcomes meant that we could defi ne risk relations
reliably, and because high blood pressure was often left
untreated after TIA or stroke at that time,48,49 allowing us
to compare the prognoses of episodic and stable
hypertension. Moreover, by contrast with many trials
investigating lowering of blood pressure, patients with
variable blood pressure were not excluded by screening
before the trials, and patients with a full range of baseline
blood-pressure values were recruited, so that the
prognostic value of mean SBP would not be
underestimated. We studied the UK-TIA trial in most
detail because usual SBP was a strong predictor of stroke
in that cohort,50 such that blood-pressure readings must
have been reliable, and because the frequency (every
4 months) and length (up to 6 years) of follow-up allowed
reasonably reliable estimation of visit-to-visit variability
and mean blood pressure with suffi cient subsequent
follow-up to measure predictive values.
Analysis of ASCOT-BPLA (based on 1·12 million blood-
pressure readings) provided six important insights. First,
residual visit-to-visit variability in blood pressure was as
high as in the TIA cohorts, despite standardised
measurement and aggressive blood-pressure lowering.
Second, variability in clinic SBP was a strong predictor of
stroke, heart failure, angina, and myocardial infarction.
Third, the extent and prognostic value of visit-to-visit
variability in SBP were independent of heart rate and
variability in heart rate. Fourth, within-visit variability in
SBP correlated with visit-to-visit variability, but was a weak
predictor of vascular events. Fifth, visit-to-visit variability
in SBP was unrelated to the white-coat eff ect. Finally, the
lower event rate in the amlodipine group than in the
atenolol group, which could not be explained by changes
in mean blood pressure or other risk factors,33 can be
explained by reduced visit-to-visit variability in SBP.40
The ASCOT-BPLA ABPM study provided four further
insights. First, visit-to-visit variability in clinic blood
pressure was not due to variation in the time of
measurement within offi ce hours (0900–1700 h). Second,
variability in mean daytime SBP on repeated ABPMs
correlated with visit-to-visit variability in clinic SBP,
indicating a contribution from fl uctuations in underlying
blood pressure. Third, variability in blood pressure on
ABPM was a weaker predictor of vascular events than
was visit-to-visit variability, and it accounted less well for
the reduced event rate in the amlodipine group than the
atenolol group,40 suggesting that average variability from
minute to minute (best exemplifi ed as average successive
variability on ABPM) does not capture elements of
variability that are associated with risk of stroke. Finally,
the predictive value of visit-to-visit variability in SBP was
independent of mean value averaged across all clinic
visits and averaged across several ABPMs.
Visit-to-visit variability in SBP was related to factors
that correlate with arterial stiff ness, including age, female
sex, smoking, diabetes, and peripheral vascular disease,
but only age and mean blood pressure aff ected the
prognostic value of variability. Variability increased with
age, but its eff ect on stroke risk was greatest at young
ages, perhaps because of fewer competing causes of
stroke or death or because of greater susceptibility to
target organ damage.
Our study had several potential shortcomings. First, in
the TIA cohorts, some variability in clinic blood pressure
could have been due to non-adherence to guidelines for
measurement or inadequate calibration of measuring
devices. However, such errors would not account for a
visit-to-visit range of SBP of 50 mm Hg or greater, which
was noted in about a third of patients, including the
atenolol group of ASCOT-BPLA, in which blood-pressure
measurement was standardised. Second, blood pressure
was measured only once at every visit in the UK-TIA and
Dutch TIA trials, but the alerting response was of no
prognostic value in ASCOT-BPLA. Third, in some
analyses, we related variability in blood pressure to
outcomes during the measurement period. However,
visit-to-visit, within-visit, and ABPM variability during
6–30 months in ASCOT-BPLA all predicted vascular
events thereafter, and measurement and outcome periods
were separate in all analyses of the UK-TIA cohort.
Fourth, we had no data for use of, or compliance with,
antihypertensive drugs during follow-up in the older TIA
cohorts. However, visit-to-visit variability was a strong
predictor of stroke in the ASCOT-BPLA on-treatment
cohort that was fully compliant. Fifth, mean SBP over
visits 7–10 will not have fully accounted for usual blood
www.thelancet.com Vol 375 March 13, 2010
pressure (webappendix p 1), but it was highly reproducible
in the UK-TIA cohort, much more so than was visit-to-
visit SD SBP. Sixth, our fi ndings cannot be generalised to
Our fi ndings do not prove a causal link between
variability in blood pressure (or maximum SBP) and
stroke. However, the risk relations were strong and
consistent in several cohorts, despite imprecision in
estimation of variability. Pre-existing cerebral ischaemia
could lead to both altered central autonomic control of
blood pressure34,35 and an increased risk of stroke, but the
risk relation in the UK-TIA trial strengthened after
exclusion of patients with previous stroke or cerebral
infarction and it was present in patients without previous
TIA or stroke in ASCOT-BPLA. Experimental data from
animal models also lend support to a causal link;12 there
are plausible mechanisms,12 and eff ects on visit-to-visit
variability in SBP explain diff erences between classes of
antihypertensive drugs in their eff ect on stroke risk.12,40,41
However, more research is needed to fully understand
the association between visit-to-visit variability in blood
pressure and risk of vascular events, and large-scale
pooled analyses of multiple cohorts will be required. The
risk association for coronary events needs further study,
particularly since antihypertensive drug class eff ects are
less obvious for coronary events in most trials than for
More work is also needed to identify measures that
would combine the prognostic information associated
with visit-to-visit variability in blood pressure with ease
of use in routine clinical practice. For example,
associations with postural instability in blood pressure
and the pressor response to other stimuli should be
determined. However, our fi ndings do have immediate
implications for the diagnosis and management of
hypertension, choice of drug, design and reporting of
trials, and drug development.12 Briefl y, patients with
episodic hypertension should no longer be excluded from
trials of antihypertensive drugs; increased residual
variability in SBP in treated patients has a poor prognosis,
despite greater use of add-on drugs; and stabilisation of
blood pressure should be regarded as a potentially
important target in the development of new agents and
new combinations of drugs. Furthermore, in secondary
prevention after TIA or stroke, for which rates of
treatment with antihypertensive drugs are low in routine
clinical practice51 despite good evidence of benefi t,52 the
high risk of stroke in patients with episodic hypertension
draws attention to the false reassurance of a few normal
PMR derived the hypothesis, collated the data from the TIA trials,
planned and supervised all analyses, and wrote the paper. SCH did the
analyses, with help from JED. NRP and PSS advised on analyses of the
main ASCOT-BPLA cohort, designed the ASCOT-BPLA trial, along with
BD, and commented on drafts of the report. ED and EO’B were
investigators in the ASCOT ABPM study, advised on analyses of the
ABPM data, and commented on drafts of the report.
Confl icts of interest
JED has institutional research funds from Pfi zer. PSS has received
payment from Pfi zer for lectures, travel, and accommodation and has
received research grants from Pfi zer and Servier. NRP has received
research grants from Pfi zer and Servier as well as payment for
consultancy, travel, lectures, advisory boards, and preparation of reports.
BD has received payment for consultancy or board membership from
MSD, Novartis, Boehringer Ingelheim, Daiichi Sankyo, and Pfi zer, and has
received speaker fees from all these companies. He also has stock options
in Mintage Scientifi c. PMR has no confl icts of interest in relation to the
topic of this paper, but has received payment from Servier for lectures and
steering committee membership in relation to the PERFORM trial.
SCH, ED, and EO’B declare that they have no confl icts of interest.
We thank the investigators of the UK-TIA aspirin trial, the European
Stroke Prevention Study (ESPS-1), the Dutch TIA aspirin trial, and the
ASCOT-BPLA study. We acknowledge the collaboration of the other
ASCOT-BPLA ABPM substudy researchers: Alice V Stanton,
Simon Thom, Mark Caulfi eld, Gordon McInnes, and David Collier.
SCH was funded by the UK Medical Research Council and by the National
Institute for Health Research (NIHR) Biomedical Research Centre, Oxford,
UK. NRP and PSS were supported by a Biomedical Research Centre Award
to Imperial College NHS Health Care Trust, are in receipt of a British
Heart Foundation Centre for Excellence Award, and along with PMR are
recipients of NIHR Senior Investigator Awards. Pfi zer was the major
funding source of the main ASCOT trial with additional support provided
by Servier Research Group, Paris, France.
1 Warlow C, Sudlow C, Dennis M, Wardlaw J, Sandercock P. Stroke.
Lancet 2003; 362: 1211–24.
2 Lawes CM, Vander Hoorn S, Rodgers A, for the International Society
of Hypertension. Global burden of blood-pressure-related disease,
2001. Lancet 2008; 371: 1513–18.
3 MacMahon S, Peto R, Cutler J, et al. Blood pressure, stroke, and
coronary heart disease. Part 1, Prolonged diff erences in blood
pressure: prospective observational studies corrected for the
regression dilution bias. Lancet 1990; 335: 765–74.
4 Prospective Studies Collaboration. Age-specifi c relevance of usual
blood pressure to vascular mortality: a meta-analysis of individual
data for one million adults in 61 prospective studies. Lancet 2002;
5 Williams B, Lindholm LH, Sever P. Systolic pressure is all that
matters. Lancet 2008; 371: 2219–21.
6 Leenen FH. Blood pressure lowering, not vascular mechanism of
action, is the primary determinant of clinical outcome.
Can J Cardiol 2004; 20 (suppl B): 77B–82B.
7 Blood Pressure Lowering Treatment Trialists’ Collaboration,
Turnbull F, Neal B, Ninomiya T, et al. Eff ects of diff erent regimens
to lower blood pressure on major cardiovascular events in older and
younger adults: meta-analysis of randomised trials. BMJ
2008; 336: 1121–23.
8 OBrien E, Asmar R, Beilin L, et al, on behalf of the European
Society of Hypertension Working Group on BP Monitoring.
Practice guidelines of the European Society of Hypertension for
clinic, ambulatory and self BP measurement. J Hypertens 2005;
9 Chobanian AV, Bakris GL, Black HR, et al, and the National High BP
Education Program Coordinating Committee. Seventh report of the
Joint National Committee on Prevention, Detection and Evaluation,
and Treatment of High BP. Hypertension 2005; 42: 1206–52.
10 Mancia G, De Backer G, Dominiczak A, et al. Management of
Arterial Hypertension of the European Society of Hypertension;
European Society of Cardiology. 2007 guidelines for the
management of arterial hypertension: the Task Force for the
Management of Arterial Hypertension of the European Society of
Hypertension (ESH) and of the European Society of Cardiology
(ESC). J Hypertens 2007; 25: 1105–87.
Parati G, Stergiou GS, Asmar R, et al. ESH Working Group on BP
Monitoring. European Society of Hypertension guidelines for BP
monitoring at home: a summary report of the Second International
Consensus Conference on Home BP Monitoring. J Hypertens 2008;
www.thelancet.com Vol 375 March 13, 2010 905
12 Rothwell PM. Limitations of the usual blood-pressure hypothesis and
importance of variability, instability, and episodic hypertension.
Lancet 2010; 375: 938–48.
13 Rothwell PM, Coull AJ, Silver LE, et al, for the Oxford Vascular Study.
Population-based study of event-rate, incidence, case fatality, and
mortality for all acute vascular events in all arterial territories (Oxford
Vascular Study). Lancet 2005; 366: 1773–83.
14 Armitage P, Fox W, Rose GA, Tinker CM. The variability of
measurements of causal blood pressure. II Survey experience.
Clin Sci 1966; 30: 337–44.
15 Klungel OH, de Boer A, Paes AH, Nagelkerke NJ, Seidell JC,
Bakker A. Estimating the prevalence of hypertension corrected for the
eff ect of within-person variability in blood pressure. J Clin Epidemiol
2000; 53: 1158–63.
16 Turner MJ, van Schalkwyk JM. Blood pressure variability causes
spurious identifi cation of hypertension in clinical studies: a computer
simulation study. Am J Hypertens 2008; 21: 85–91.
17 Marshall T. When measurements are misleading: modelling the
eff ects of blood pressure misclassifi cation in the English population.
BMJ 2004; 328: 933.
18 Keenan K, Hayen A, Neal BC, Irwig L. Long term monitoring in
patients receiving treatment to lower blood pressure: analysis of data
from placebo controlled randomised controlled trial. BMJ 2009;
19 Howard SC, Rothwell PM. Regression dilution of systolic and
diastolic BP in patients with established cerebrovascular disease.
J Clin Epidemiol 2003; 56: 1084–91.
20 Cuff e RL, Howard SC, Algra A, Warlow CP, Rothwell PM.
Medium-term variability of BP and potential underdiagnosis of
hypertension in patients with previous transient ischemic attack or
minor stroke. Stroke 2006; 37: 2776–83.
21 Colandrea MA, Griedman GD, Nichaman MZ, et al. Systolic
hypertension in the elderly: an epidemiological assessment.
Circulation 1970; 41: 239.
22 Marshall T. When measurements are misleading: modelling the
eff ects of BP misclassifi cation in the English population. BMJ 2004;
23 Hypertension Detection and Follow-up Program Cooperative Group.
Variability of BP and the results of screening in the hypertension
detection and follow-up program. J Chronic Dis 1978; 31: 651.
24 Whelton PK, Klag MJ. Epidemiology of high BP. Clin Geriatr Med
1989; 5: 639–55.
25 Verdecchia P, O’Brien E, Pickering T, et al, for the European Society
of Hypertension Working Group on Blood Pressure Monitoring.
When can the practicing physician suspect white coat hypertension?
Statement from the Working Group on Blood Pressure Monitoring
of the European Society of Hypertension. Am J Hypertens 2003;
26 Pickering TG, Eguchi K, Kario K. Masked hypertension: a review.
Hypertens Res 2007; 30: 479–88.
27 Howard SC, Rothwell PM. Reproducibility of measures of visit-to-visit
variability in blood pressure after transient ischaemic attack or minor
stroke. Cerebrovasc Dis 2009; 28: 331–40.
28 Cuff e RL, Rothwell PM. Medium-term variability in systolic blood
pressure is an independent predictor of stroke. Cerebrovasc Dis 2005;
19 (suppl 2): 51 (abstr).
29 Farrell B, Godwin J, Richards S, Warlow C. The United Kingdom
transient ischaemic attack (UK-TIA) aspirin trial: fi nal results.
J Neurol Neurosurg Psychiatry 1991; 54: 1044–54.
30 The ESPS Group. The European Stroke Prevention Study (ESPS).
Principal end-points. Lancet 1987; 330: 1351–54.
31 The Dutch TIA Study Group. A comparison of two doses of
aspirin (30mg vs 283mg a day) in patients after a transient
ischaemic attack or minor ischaemic stroke. N Engl J Med 1991;
32 Dahlöf B, Sever PS, Poulter NR, et al, for the ASCOT Investigators.
Prevention of cardiovascular events with an antihypertensive
regimen of amlodipine adding perindopril as required versus
atenolol adding bendrofl umethiazide as required, in the Anglo-
Scandinavian Cardiac Outcomes Trial-Blood Pressure Lowering Arm
(ASCOT-BPLA): a multicentre randomised controlled trial. Lancet
2005; 366: 895–906.
33 Poulter NR, Wedel H, Dahlöf B, et al, for the ASCOT investigators.
Role of blood pressure and other variables in the diff erential
cardiovascular event rates noted in the Anglo-Scandinavian Cardiac
Outcomes Trial-Blood Pressure Lowering Arm (ASCOT-BPLA).
Lancet 2005; 366: 907–13.
34 Zhang Z, Oppenheimer SM. Electrophysiological evidence for
reciprocal insulo-insular connectivity of baroreceptor-related neurons.
Brain Res 2000; 863: 25–41.
35 Robinson T, Ward-Close S, Potter J. A comparison of beat-to-beat BP
variability in acute and subacute stroke patients with cerebral
infarction. Cerebrovas Dis 1997; 7: 214–19.
36 O’Brien E, Mee F, Atkins N. An accurate automated device for home
BP measurement at last! The Omron HEM-705C. J Hypertens 1994;
37 O’Brien E, McInnes GT, Stanton A, et al; Anglo-Scandinavian Cardiac
Outcomes Trial. Ambulatory blood pressure monitoring and 24-h
blood pressure control as predictors of outcome in treated
hypertensive patients. J Hum Hypertens 2001; 15 (suppl 1): S47–51.
38 Dolan E, Stanton AV, Thom S, et al, for the ASCOT Investigators.
Ambulatory blood pressure monitoring predicts cardiovascular events
in treated hypertensive patients—an Anglo-Scandinavian cardiac
outcomes trial substudy. J Hypertens 2009; 27: 876–85.
39 O’Brien E, Mee F, Atkins N, O’Malley K. Accuracy of the SpaceLabs
90207 determined by the British Hypertension Protocol. J Hypertens
1991; 9: 573–74.
40 Rothwell PM, Howard SC, Dolan E, et al, on behalf of the ASCOT-
BPLA and MRC Trial Investigators. Eff ects of β blockers and calcium-
channel blockers on within-individual variability in blood pressure
and risk of stroke. Lancet Neurol 2010; published online March 12.
41 Webb AJS, Fischer U, Mehta Z, Rothwell PM. Eff ects of
antihypertensive-drug class on interindividual variation in blood
pressure and risk of stroke: a systematic review and meta-analysis.
Lancet 2010; 375: 906–15.
42 Kikuya M, Ohkubo T, Metoki H, et al. Day-by-day variability of blood
pressure and heart rate at home as a novel predictor of prognosis: the
Ohasama study. Hypertension 2008; 52: 1045–50.
43 Hata Y, Kimura Y, Muratani H, et al. Offi ce BP variability as a
predictor of brain infarction in elderly hypertensive patients.
Hypertens Res 2000; 23: 553–60.
44 Hata Y, Muratani H, Kimura Y, et al. Offi ce BP variability as a
predictor of acute myocardial infarction in elderly patients receiving
antihypertensive therapy. J Hum Hypertens 2002; 16: 141–46.
45 Tozawa M, Iseki K, Yoshi S, Fukiyama K. BP variability as an adverse
prognostic risk factor in end-stage renal disease.
Nephrol Dial Transplant 1999; 14: 1976–81.
46 Havlik RJ, Foley DJ, Sayer B, Masaki K, White L, Launer LJ. Variability
in midlife systolic BP is related to late-life brain white matter lesions:
the Honolulu-Asia Aging study. Stroke 2002; 33: 26–30.
47 Pringle E, Phillips C, Thijs L, et al, for the Syst-Eur investigators.
Systolic BP variability as a risk factor for stroke and cardiovascular
mortality in the elderly hypertensive population. J Hypertens 2003;
48 Jansen PA, Schulte BP, Meyboom RH, Gribnau FW. Antihypertensive
treatment as a possible cause of stroke in the elderly. Age Ageing 1986;
49 PROGRESS Management Committee. Blood pressure lowering for
the secondary prevention of stroke: rationale and design for
PROGRESS. Perindopril Protection Against Recurrent Stroke Study.
J Hypertens 1996; 14 (suppl): S41–45.
50 Rodgers A, MacMahon S, Gamble G, Slattery J, Sandercock P,
Warlow C. BP and risk of stroke in patients with cerebrovascular
disease. The United Kingdom Transient Ischaemic Attack
Collaborative Group. BMJ 1996; 313: 147–48.
51 Toschke AM, Wolfe CD, Heuschmann PU, Rudd AG, Gulliford M.
Antihypertensive treatment after stroke and all-cause mortality—an
analysis of the General Practitioner Research Database (GPRD).
Cerebrovasc Dis 2009; 28: 105–11.
52 PROGRESS Collaborative Group. Randomised trial of a perindopril-
based blood-pressure-lowering regimen among 6105 individuals with
previous stroke or transient ischaemic attack. Lancet 2001;