Relationship between clinic and ambulatory blood pressure and mortality: an observational cohort study in 59 124 patients

Summary Background Ambulatory blood pressure provides a more comprehensive assessment than clinic blood pressure, and has been reported to better predict health outcomes than clinic or home pressure. We aimed to examine associations of clinic and 24-h ambulatory blood pressure with all-cause and cardiovascular mortality in a large cohort of primary care patients referred for assessment of hypertension. Methods We did an observational cohort study using clinic and ambulatory blood pressure data obtained from March 1, 2004, to Dec 31, 2014, from the Spanish Ambulatory Blood Pressure Registry. This registry included patients from 223 primary care centres from the Spanish National Health System in all 17 regions of Spain. Mortality data (date and cause) were ascertained by a computerised search of the vital registry of the Spanish National Institute of Statistics. Complete data were available for age, sex, all blood pressure measures, and BMI. For each study participant, follow-up was from the date of their recruitment to the date of death or Dec 31, 2019, whichever occurred first. Cox models were used to estimate associations between usual clinic or ambulatory blood pressure and mortality, adjusted for confounders and additionally for alternative measures of blood pressure. For each measure of blood pressure, we created five groups (ie, fifths) defined by quintiles of that measure among those who subsequently died. Findings During a median follow-up of 9·7 years, 7174 (12·1%) of 59 124 patients died, including 2361 (4·0%) from cardiovascular causes. J-shaped associations were observed for several blood pressure measures. Among the top four baseline-defined fifths, 24-h systolic blood pressure was more strongly associated with all-cause death (hazard ratio [HR] 1·41 per 1 – SD increment [95% CI 1·36–1·47]) than clinic systolic blood pressure (1·18 [1·13–1·23]). After adjustment for clinic blood pressure, 24-h blood pressure remained strongly associated with all-cause deaths (HR 1·43 [95% CI 1·37–1·49]), but the association between clinic blood pressure and all-cause death was attenuated when adjusted for 24-h blood pressure (1·04 [1·00–1·09]). Compared with the informativeness of clinic systolic blood pressure (100%), night-time systolic blood pressure was most informative about risk of all-cause death (591%) and cardiovascular death (604%). Relative to blood pressure within the normal range, elevated all-cause mortality risks were observed for masked hypertension (HR 1·24 [95% CI 1·12–1·37]) and sustained hypertension (1·24 [1·15–1·32]), but not white-coat hypertension, and elevated cardiovascular mortality risks were observed for masked hypertension (1·37 [1·15–1·63]) and sustained hypertension (1·38 [1·22–1·55]), but not white-coat hypertension. Interpretation Ambulatory blood pressure, particularly night-time blood pressure, was more informative about the risk of all-cause death and cardiovascular death than clinic blood pressure.


Introduction
Ambulatory blood pressure provides a more compre hensive assessment of blood pressure over the course of a 24h period than clinic blood pressure, and has been reported to better predict health outcomes than clinic or home pressure. [1][2][3][4] Evidence for the influence of ambulatory blood pressure on prognosis comes mainly from populationbased studies, 4-8 and a few relatively small clinical investigations. [9][10][11][12] However, in most of these previous studies, the number of participants studied or clinical outcomes were often limited (even in studies that included information about nonfatal outcomes), 5,6,[8][9][10][11][12] reducing the ability to discriminate the predictive value of the clinic versus ambulatory blood pressure. Moreover, studies including a larger number of participants were mainly obtained by pooling databases from previous small studies. 4,8 Furthermore, uncertainty exists about whether the average ambulatory blood pressure over the nighttime, daytime, or the full 24 h is the strongest predictor of death. [3][4][5][6][7][8][9]11,13,14 Some 4,6,9,11 but not all studies 5,10 have suggested that nighttime blood pressure is the strongest predictor of death. Moreover, the implications of hypertension phenotypes such as whitecoat hypertension and masked hypertension with regard to mortality have remained illdefined-namely, because of the small number of deaths in previous studies. 15 -24 In this largest study to date, we aimed to report the associations between longterm usual levels of clinic and ambulatory blood pressure indices, and of hypertension phenotypes, with total and cardiovascular death in patients in primary care practice recruited into the Spanish Ambulatory Blood Pressure Monitoring Mortality Study.

Study design and patient population
We did an observational cohort study using clinic and ambulatory blood pressure data obtained from March 1, 2004, to Dec 31, 2014, from the Spanish Ambulatory Blood Pressure Registry, a national study of patients selected by their physicians at 223 primary care centres from the Spanish National Health System in all 17 regions of Spain. 25,26 Eligible participants were 18 years or older and had to meet guidelinerecommended indications for ambulatory blood pressure monitoring, [27][28][29] which included suspected whitecoat hypertension, refractory or resistant hypertension, assessment of drug treatment efficacy, highrisk hypertension, labile or borderline hypertension, and the study of circadian blood pressure pattern. All patients included into the registry provided written informed consent. The protocols for the registry analyses were approved by the Institutional Review Boards of the participating centres.

Blood pressure indices and study variables
Clinic blood pressure was measured with validated oscillometric devices or calibrated mercury sphygmo manometers after a 5min rest while seated, using standardised procedures. 27, 30 We used the mean of two clinic pressure readings. Thereafter, ambulatory blood

Research in context
Evidence before this study The link between elevated clinic blood pressure and reduced life expectancy has been recognised for decades. Several studies have reported that ambulatory blood pressure measured over 24 h is more strongly associated with health outcomes than conventional clinic blood pressure measurement. These studies have varied in size and statistical power, particularly with respect to all-cause mortality, or have only achieved scale by pooling databases from previous smaller studies. Some studies have also assessed clinic, 24-h, daytime, or night-time average blood pressures to evaluate which are most strongly associated with death. The majority, but not all, suggest that night-time blood pressure is most predictive of cardiovascular morbidity or death. Specific blood pressure phenotypes that are distinct from sustained hypertension exist-notably, masked hypertension (clinic blood pressure normal, ambulatory blood pressure elevated) or white-coat hypertension (clinic blood pressure elevated, ambulatory blood pressure normal). The relationship between these phenotypes and death has remained ill-defined because of the small number of deaths in previous studies in which these phenotypes have been characterised.

Added value of this study
This study is by far the largest, single population-based study of the relationship of both clinic and ambulatory blood pressures with death, undertaken in primary care, involving almost 60 000 patients, over a long follow-up duration (median 9·7 years), during which 7174 deaths were reported, including 2361 from cardiovascular disease. The study shows that ambulatory blood pressure was more informative about the risk of all-cause death and cardiovascular death than conventional clinic blood pressure. Indeed, after adjustment for clinic blood pressure, 24-h blood pressure remained strongly associated with death (hazard ratio 1·43 per 1 -SD increment [95% CI 1·37-1·49]), whereas most of the association between clinic blood pressure and death was lost after adjustment for 24-h blood pressure (1·04 [1·00-1·09]). Night-time ambulatory systolic blood pressure was six times more informative for death than clinic systolic blood pressure and nearly twice as informative as daytime ambulatory systolic blood pressure. These findings were similar whether patients were treated for hypertension at baseline (35 128 [59%] of 59 124) or not, (23 996 [41%]) and were consistent for all ages and both sexes. Finally, relative to those with normal blood pressure, masked hypertension was associated with an increased risk of death whereas white-coat hypertension was not.

Implications of all the available evidence
The findings of this study reinforce and substantially extend the information from previous studies linking methods of blood pressure measurement to patient outcomes. This very large study unequivocally demonstrates the dominance of ambulatory blood pressure over clinic blood pressure in the association between blood pressure and all-cause death and cardiovascular death, whether treated for hypertension or not. The dominance of the association of night-time blood pressure with death confirms some previous reports and is striking, highlighting a need to both evaluate and control night-time blood pressure, particularly in higher risk patients. The risk of death associated with masked hypertension is concerning as these patients usually remain undetected with screening using clinic blood pressure alone. The absence of an association between whitecoat hypertension and increased risk of death appears reassuring; however, many of these patients will progress to sustained hypertension.
pressure monitoring was done using validated, oscillometric devices (Spacelabs 90207; Snoqualmie, WA, USA), programmed to record blood pressure at 20min intervals for the day and at 30min intervals for the night. Appropriate cuff sizes were used. We used the mean of all valid readings (based on predefined quality criteria for a valid 24 h blood pressure average, including successful recording of ≥70% of systolic and diastolic pressure readings during 24 h, and at least one valid measurement per h). 66% of patients achieved valid readings of more than 75%, 42% achieved valid readings of 80%, 25% achieved valid readings of 85%, 16% achieved valid readings of 90%, and 10% achieved valid readings of 95%.
Day and night periods were defined according to the patient's selfreported data for sleeping and waking times. Other variables were obtained at entry from personal interview and physical examination at study visits or from clinical records. The case report form included a question about previous cardiovascular disease with a box for investigators to tick to indicate presence of previous disease. However, if the box was not checked, distinguishing whether this omission was because the participant had no history of previous disease or the question was not answered was not possible; as such, some cases of previous cardiovascular disease might have been missed. For a subset of patients, clinic and ambulatory blood pressure was measured at two different timepoints (an average of 18 months apart).

Mortality data
The date and cause of death were ascertained by a computerised search of the vital registry of the Spanish National Institute of Statistics (contract 20534 between the University of Barcelona and the National Institute of Statistics), which has been shown to be accurate and reliable with complete coverage. 31 Individuals were designated as having died if they were recorded in the vital registry. Cause of death was determined from the death certificate, and was coded according to the International Classification of Diseases (10th Revision). Deaths with codes in the range I00-I99 were classified as of cardiovascular origin. Cardiovascular deaths were also further subdivided into deaths from coronary heart disease (I21-I25), stroke (I60-I69), or heart failure (I50). For each study participant, followup was from the date of their recruitment to the date of death or Dec 31, 2019, whichever occurred first.

Statistical analysis
Complete data were available for age, sex, all blood pressure measures, and BMI. Missing data for current smoking, diabetes, and dyslipidaemia status was less than 1%, so patients with missing data were assumed not to have the condition. Univariable group comparisons of continuous variables were done using ANOVA tests, and categorical variables were done using χ² tests. Pearson correlation coefficients were used to estimate correlation between blood pressure indices. For each measure of blood pressure, we created five groups (ie, fifths) defined by quintiles of that measure among those who subsequently died, to ensure similar number of deaths in each group. We used Cox regression to estimate mortality hazard ratios (HRs) for each of the top fourfifths of each blood pressure measure relative to the lowest onefifth. Assessment of the proportional hazards assumption found some evidence against proportionality for some of the blood pressure measures. However, even in the presence of nonproportionality, the Cox HR still provides a useful summary statistic to describe the average association of the blood pressure index to risk over the followup period. These HR estimates were adjusted for age, sex, smoking status (current vs not), BMI, diabetes status (previous record of diabetes vs no record), dyslipidaemia status (previous record of dyslipidaemia vs no record), previous cardiovascular disease (previous record of ischaemic heart disease, stroke, or heart failure vs no record), and number of antihypertensive drugs (zero or untreated vs one vs two or more). The association of each blood pressure measure independently of other measures was then assessed by inclusion of those other measures into the model. Data from 2928 patients with repeated measurements of clinic and ambulatory blood pressure were used to calculate regression dilution ratios using Rosner's method. 32 Log HRs and their associated standard errors were then divided by these regression dilution ratios to correct for regression dilution bias (ie, the underestimation of the association of longterm usual blood pressure with risk caused by measurement error and withinperson variability in blood pressure).
In the figures, each HR (including the reference group with an HR of 1•0) is shown with a groupspecific confidence interval that reflects the amount of information only in that single category. 33 Means of the repeated blood pressure measurements were calculated for each baselinederived fifth and the HRs were plotted at these values to show associations between usual blood pressure values and mortality. On the basis of nonlinear associations, a posthoc decision was made to estimate the mortality HR per 1 -SD increment in usual blood pressure across the top fourfifths of each distribution (analyses using all fifths were also done). The informativeness of the different blood pressure measures (compared with clinic systolic blood pressure) for prediction of allcause and cardiovascular death was calculated with the following formula: 34 Wald tests were used to calculate the χ² statistic for each blood pressure measure, which provides an assessment of the goodnessoffit of the confounderadjusted Cox Informativeness = 100 × χ² for given measure χ² for clinic systolic pressure models. Separate associations in subgroups defined by age (<60, ≥60 years), sex, obesity (BMI <30, ≥30 kg/m²), diabetes, and antihypertensive drug use were estimated by including an appropriate interaction term into a regression model. Additionally, to guard against potential estimation problems arising from collinearity in the mutually adjusted regression models, additional sensitivity analyses were done using the residuals of each blood pressure measure adjusted for other blood pressure measures (rather than models in which the correlated blood pressure measures were entered into the same regression model). Subsequently, hypertension phenotypes were defined as blood pressure in the normal range (clinic systolic blood pressure <140 mm Hg and clinic diastolic blood pressure <90 mm Hg; and 24h systolic <130 mm Hg and 24h diastolic <80 mm Hg); whitecoat hypertension (clinic systolic ≥140 mm Hg or clinic diastolic ≥90 mm Hg, but 24h systolic <130 mm Hg and 24h diastolic <80 mm Hg); masked hypertension (clinic systolic <140 mm Hg and clinic diastolic <90 mm Hg, but 24h systolic ≥130 mm Hg or 24h diastolic ≥80 mm Hg); and sustained hypertension (clinic systolic ≥140 mm Hg or clinic diastolic ≥90 mm Hg, and 24h systolic ≥130 mm Hg or 24h diastolic ≥80 mm Hg). [27][28][29] HRs for whitecoat, masked, and sustained hypertension (all compared with blood pressure in the normal range) were estimated using Cox regression In the model with additional adjustment for clinic or 24-h blood pressure, clinic blood pressure is adjusted for 24-h blood pressure and ambulatory blood pressure measures are adjusted for clinic blood pressure. (C) In the model additionally adjusted for components of 24-h blood pressure, daytime blood pressure is adjusted for night-time blood pressure and night-time blood pressure is adjusted for daytime blood pressure. HR=hazard ratio.   Analyses excluding patients with clinic diastolic blood pressure of less than 70 mm Hg from the normal range category and estimating the associations separately in treated and untreated patients were also done. Twotailed p<0•05 were considered to indicate statistical significance; no correction for multiple testing was done. We did the analyses using SAS (version 9.4) and R (version 4.2.1).

Role of the funding source
The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report.

Results
In total, 59 746 individuals who were 18 years or older in the Spanish Ambulatory Blood Pressure Registry could be linked to the vital registry of the Spanish National Institute of Statistics. Of these individuals, 622 (1•0%) were excluded because of incomplete information about demographic or clinical characteristics; as a result, 59 124 patients were included in this analysis. The mean age of the study participants at recruitment into the registry was 58•7 years (SD 14•1), the mean clinic systolic blood pressure was 148•0 mm Hg and diastolic blood pressure was 86•5 mm Hg, and the mean 24h ambulatory systolic blood pressure was 128•8 mm Hg and diastolic blood pressure was 76•2 mm Hg. 31 337 (53·0%) of 59 124 were men and 27 787 (47·0%) were women (table 1; appendix pp 4, 11-12). 35 128 (59·4%) of 59 124 were treated for hypertension. During a median followup of 9•7 years (IQR 7•7-11•3), 7174 (12•1%) participants died, including 2361 (4•0%) from a cardiovascular cause, which included 685 coronary heart disease deaths, 503 stroke deaths, and 302 heart failure deaths.
Clinic and ambulatory blood pressure measurements were modestly correlated (correlation coefficients of 0•43 for systolic blood pressure and 0•52 for diastolic blood pressure; appendix pp 5, 13). Daytime and night time blood pressure measurements were more strongly correlated (correlation coefficients of 0•73 for systolic blood pressure and 0•74 for diastolic blood pressure; appendix p 5). Substantial variability was observed in clinic and ambulatory blood pressure monitoring measurements taken at two different visits (regression dilution ratios for measurements an average of 18 months apart were 0•46 for clinic systolic blood pressure and 0•48 for 24h systolic blood pressure; appendix pp [14][15]. Associations between systolic blood pressure measures and allcause death were Jshaped, particularly for clinic blood pressure, but the associations were loglinear when excluding the fifth with the lowest blood pressure values (figure 1; appendix p 6). In the confounderadjusted model, 24h systolic blood pressure was more strongly associated with allcause death (  In the model with additional adjustment for clinic or 24-h systolic blood pressure, clinic blood pressure is adjusted for 24-h blood pressure and ambulatory blood pressure indices are adjusted for clinic blood pressure. In the model additionally adjusted for components of 24-h blood pressure, daytime blood pressure is adjusted for night-time blood pressure and night-time blood pressure is adjusted for daytime blood pressure. HR=hazard ratio. . Results were similar in sensitivity analyses using the residuals of each blood pressure measure adjusted for other blood pressure measures (appendix p 7). Associations were consistent across subgroups (appendix pp [16][17] and, except for nighttime blood pressure, were attenuated when refitted across all fifths of each distribution (table 2). For cause specific cardiovascular death, nighttime systolic blood pressure was most strongly associated with risk of death, even after adjustment for daytime systolic blood pressure (appendix p 7).

Discussion
In this large study of clinic and ambulatory blood pressure, blood pressure measures obtained through ambulatory blood pressure monitoring were more informative about the risk of allcause death or cardiovascular death than conventional clinic blood pressures. Importantly, once 24h blood pressure was known, most of the informativeness of clinic systolic blood pressure was lost, whereas associations for ambulatory blood pressure measures were largely unaffected by adjustment for clinic systolic blood pressure. The relative informativeness of 24h ambulatory systolic blood pressure for risk of death was almost fivetimes greater than clinic systolic blood pressure. Furthermore, with respect to ambulatory blood pressure measures, night time systolic blood pressure was about sixtimes more informative for death than clinic systolic blood pressure and nearly twice as informative as daytime systolic blood pressure. Although masked and sustained hypertension were associated with an increased risk of death compared with patients with 24h blood pressure within normal range, whitecoat hypertension was not.
Previous population and clinical studies have shown that ambulatory systolic blood pressure predicts death and cardiovascular outcomes better than clinic systolic blood pressure. [4][5][6][8][9][10][11] The present study extends these findings in a single population with more than 50 000 patients and also demonstrates the effect of commonly encountered blood pressure phenotypes on death. A recent study by Yang and colleagues also demonstrated the dominance of 24h ambulatory systolic pressure and, in particular, nighttime systolic pressure compared with clinic systolic pressure in predicting death. 4 However, unlike the present study, Yang and colleagues' study did not include data for hypertension phenotypes.
The superiority of nighttime blood pressure over daytime blood pressure has been observed in other previous studies, 4,9,11,13,14,35,36 including a small group of highrisk patients from the Spanish Registry. 37 Although daytime blood pressure lost its predictive ability after adjustment for nighttime blood pressure, nighttime blood pressure maintained its predictive ability after adjusting for daytime blood pressure. Several mechanisms might be responsible for this nocturnal blood pressure superiority, the most likely being the more standardised conditions under which blood pressure is recorded during sleep, relative to much more variable activity during daytime, which is supported by the reproducibility data in the appendix (p 14), in which nighttime systolic pressure had a higher regression dilution ratio than daytime or clinic systolic pressure. Studies have also reported that elevated nighttime blood pressure is associated with target organ damage (eg, chronic kidney disease 38 and heart failure 36 ), which might contribute to a higher risk of death.
Interesting relationships were also observed between clinic or ambulatory diastolic blood pressure and death, which were Jshaped or Ushaped. The strength of the association between increasing diastolic blood pressure and mortality appears weaker than for systolic blood pressure, as observed in previous studies, 4,5,7,8,11,13,35 although these studies did not explicitly assess the shape of the underlying associations. The increased risk of death at lower values of diastolic blood pressure is likely to reflect, at least in part, reverse causation due to the effects of arterial ageing, stiffening, and subclinical disease, all of which are associated with a widened pulse pressure and lower diastolic pressure. 39 Our findings that masked and sustained hypertension are associated with increased risks of death are consistent with other studies. 2,12,15,17,19 Poorer outcomes in patients with masked hypertension most likely relates to a delayed recognition of hypertension and undertreatment in such individuals as, at the time of this study, guidelines generally recommended that the diagnosis of hypertension and its drug treatment be guided predominantly by clinic blood pressure. More recent guidelines 30,40 promote the wider use of socalled outofoffice blood pressure that will better facilitate the diagnosis and treatment of masked hypertension. The prognosis of whitecoat hypertension has been a matter of debate, with some studies showing a cardiovascular risk similar to that of normotensives, 15,17,[19][20][21] whereas others found increased incidence of events. 41 Our study suggests that this phenotype is not associated with an increased risk of mortality compared with those with blood pressure in the normal range. Ambulatory blood pressure measurements were also taken for clinical reasons, and therefore, patients with normal blood pressures are not necessarily a healthy population, which might explain why a lower risk of death, relative to normotensive patients, was observed for patients with whitecoat hypertension in this study. Supporting this idea, when patients with a diastolic blood pressure of less than 70 mm Hg were excluded from the analysis, whitecoat hypertension became neither protective nor deleterious for allcause death or cardiovascular death.
This study has some limitations. First, clinic blood pressure represented the average of only two readings at each clinic visit; thus, the mean clinic blood pressure could be overestimated. Moreover, variability existed in both clinic and ambulatory blood pressure monitoring where repeated measures were taken, although adjustment for regression dilution bias was used to account for this variability where possible. Second, data for medication, although available at baseline, were not available during the followup period. Third, some selection bias might have occurred from the inclusion criteria for ambulatory blood pressure monitoring, obtained on the basis of indications for this procedure contained in guidelines contemporary to the study design and followup. Fourth, we did an observational study on the prognostic value of blood pressure monitoring and, thus, no direct inference can be made regarding the benefit of basing treatment on ambulatory blood pressure measurements. Fifth, the present study does not consider the association of ambulatory blood pressure monitoring with nonfatal events, as only data for deaths were available. The introduction of revascularisation therapies after a cardiovascular event is known to influence the probability of death. Finally, the patients we studied were predominantly a White population, and the results might not apply to people of other races.
In conclusion, in this large study, systolic blood pressure obtained through ambulatory blood pressure monitoring, and particularly nighttime systolic blood pressure, were more informative measures for the risk of allcause death and cardiovascular death, compared with clinic systolic blood pressure. Masked and sustained hypertension were also associated with an increased risk of death compared with patients with 24h blood pressure within normal range. Conversely, we found no evidence that whitecoat hypertension was associated with increased risk of death.

Contributors
AdlS, LMR, and BW designed the study and contributed to data gathering. NS, JRE, and CB analysed the data. All authors drafted and revised the paper, and approved the final version of the manuscript.

Declaration of interests
NS reports institutional grant support from Boehringer Ingelheim, Lilly, and Novo Nordisk for investigator led/designed research unrelated to this project. AdlS reports lecture fees from Viatris, and support for travel to attend meetings from Menarini and Lacer. LMR reports consulting fees from Novartis, Bayer, Pfizer, Medtronic, ReCor, Sanofi, and Sandoz and lecture fees from Novartis, Bayer, Pfizer, Medtronic, ReCor, and Sanofi. CB reports institutional grant support from Boehringer Ingelheim for investigator designed/led research unrelated to this work. BW reports lecture fees from Daichi Sankyo, Servier, and Pfizer, and institutional grant support from Omron for investigator designed/led research unrelated to this work. All other authors declare no competing interests. The Clinical Trial Service Unit and Epidemiological Studies Unit has a staff policy of not accepting honoraria or other payments from the pharmaceutical industry, except for the reimbursement of costs to participate in scientific meetings. BW on behalf of University College London has received an investigatorled grant award from Omron, Japan and honoraria for lectures on blood pressure measurement, and is supported by the National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre.

Data sharing
The data that support the findings of this study are available from the corresponding author upon reasonable request.