Prospective external validation of the Predicting Out-of-OFfice Blood Pressure (PROOF-BP) strategy for triaging ambulatory monitoring in the diagnosis and management of hypertension: observational cohort study

Abstract Objective To prospectively validate the Predicting Out-of-OFfice Blood Pressure (PROOF-BP) algorithm to triage patients with suspected high blood pressure for ambulatory blood pressure monitoring (ABPM) in routine clinical practice. Design Prospective observational cohort study. Setting 10 primary care practices and one hospital in the UK. Participants 887 consecutive patients aged 18 years or more referred for ABPM in routine clinical practice. All underwent ABPM and had the PROOF-BP applied. Main outcome measures The main outcome was the proportion of participants whose hypertensive status was correctly classified using the triaging strategy compared with the reference standard of daytime ABPM. Secondary outcomes were the sensitivity, specificity, and area under the receiver operator characteristic curve (AUROC) for detecting hypertension. Results The mean age of participants was 52.8 (16.2) years. The triaging strategy correctly classified hypertensive status in 801 of the 887 participants (90%, 95% confidence interval 88% to 92%) and had a sensitivity of 97% (95% confidence interval 96% to 98%) and specificity of 76% (95% confidence interval 71% to 81%) for hypertension. The AUROC was 0.86 (95% confidence interval 0.84 to 0.89). Use of triaging, rather than uniform referral for ABPM in routine practice, would have resulted in 435 patients (49%, 46% to 52%) being referred for ABPM and the remainder managed on the basis of their clinic measurements. Of these, 69 (8%, 6% to 10%) would have received treatment deemed unnecessary had they received ABPM. Conclusions In a population of patients referred for ABPM, this new triaging approach accurately classified hypertensive status for most, with half the utilisation of ABPM compared with usual care. This triaging strategy can therefore be recommended for diagnosis or management of hypertension in patients where ABPM is being considered, particularly in settings with limited resources.


Extended methods
The protocol for this study have been published previously. 1

Study design
This study used a prospective, multi-centre observational cohort design, recruiting patients from Primary Care and Secondary Care.

Study participants and setting
Consecutive patients attending participating centres in Primary or Secondary Care, for whom ambulatory blood pressure monitoring was considered appropriate, were enrolled between May 2015 and January 2017. Eligible patients were those undergoing ambulatory blood pressure monitoring (ABPM) as a result of routine blood pressure screening or monitoring in Primary Care or via referral to Secondary Care with suspected hypertension, newly diagnosed or treated hypertension, resistant hypertension, secondary hypertension or other conditions requiring specialist advice. Anonymised data were collected on all patients fulfilling the eligibility criteria: Inclusion Criteria -Age 18 years or above -Attending clinical practice for routine ABPM Exclusion Criteria -Aged under 18 years old -Lack of availability of basic clinical information -Multiple clinic blood pressure readings (obtained on at least three occasions within the same visit) not recorded -Ambulatory blood pressure monitor not worn as instructed

Procedures
All participants underwent ABPM, clinic blood pressure measurement and collection of patient characteristics. Data collected for each participant are detailed in the protocol 1 and included: blood pressure measurements (values and measurement technique), previous treatment prescriptions, body mass index (BMI), smoking status and history of diabetes, chronic kidney disease, atrial fibrillation and cardiovascular disease. Antihypertensive medication changes made after ABPM were not recorded. All data were collected from electronic health records and entered directly onto the study database by trained staff at each data collection site.
To capture as close to 'routine' blood pressures as possible, all participating sites were asked to measure clinic and ambulatory blood pressure according to their usual practice. No specific protocol for measurement was used. 'Routine blood pressure' was defined as readings taken by the consulting healthcare professional as part of routine clinical practice.
A minimum of three clinic readings taken at the time of referral for ambulatory blood pressure monitoring or at monitor fitting were required for inclusion in the study. It is acknowledged that this may not always reflect routine practice 2 (even though it is recommended in guidelines), 3 but this was necessary to permit validation of the triaging algorithm. Each site was offered a validated automated blood pressure monitoring device (Omron M10-IT, Omron Corporation, Kyoto, Japan) to assist with the collection of multiple clinic blood pressure readings, but were given the option to continue using their own monitor, so long as at least three readings were taken and recorded. To our knowledge, all readings were taken with the physician or nurse present in the room. ABPM was conducted using the practice/hospital's own ambulatory monitor and fitted by a trained nurse or allied health professional. Some practices in Primary Care only collected daytime ambulatory pressures. Details of clinic and ambulatory blood pressure monitors used at each site are given in the online appendix (eTable 1, online appendix).
For the primary analysis, clinic blood pressure was defined by the 1st clinic reading taken at each visit, replicating the definitions used in the original PROOF-BP analyses 4 and better reflecting how blood pressure is captured in routine practice. 2,5 Mean of daytime ambulatory blood pressure was defined as it was used for routine practice -i.e. by patient diary, and/or monitor settings preprogrammed by the treating physician or nurse.

The PROOF-BP triaging approach
The triaging strategy applies an algorithm to clinic blood pressure readings and patient characteristics to identify three groups: those with definitively normal blood pressure, those with definitively high blood pressure and those requiring further investigation using ABPM. The algorithm utilises three blood pressures taken at the clinic appointment combined with information from the patient's electronic health record: age, sex, BMI, hypertensive and treatment history and the presence of cardiovascular disease (see eFigures 2 and 3, online appendix). 4 The triaging strategy is applied in three stages: 4 1) Using the algorithm incorporating three clinic BP readings and patient characteristics, estimate the predicted difference between clinic and daytime ambulatory blood pressure. 2) Add this difference to the clinic blood pressure to generate an 'adjusted clinic blood pressure'. 3) Triage an individual for ambulatory monitoring (or not) depending on the level of adjusted clinic blood pressure (<130/80 mmHg = normal blood pressure not requiring treatment; 130/80-144/89mmHg = uncertain blood pressure requiring additional monitoring with ABPM; >145/90 mmHg = high blood pressure requiring treatment).

Primary outcome
The primary outcome of this study was the proportion correctly classified with hypertension using the triaging strategy compared to the reference standard of daytime ABPM (using a threshold for hypertension of ≥135/85 mmHg). 3,6,7 This was defined as the proportion of patients with sustained hypertension (true positives), normotension (true negatives), white coat hypertension (false positives) and masked hypertension (false negatives).

Secondary outcomes
The sensitivity (for detecting hypertension in patients with the condition), specificity (for ruling out hypertension in those without the condition), positive and negative predictor values were estimated and compared to guideline strategies for measuring blood pressure from the UK, 3 US, 7 Europe, 6 Canada 8 and Japan 9 (eTable 2, online appendix). Further secondary outcomes included accuracy of the triaging strategy in different sub-groups: setting (Primary vs. Secondary Care), age (<65 years vs. >65 years), sex, smoking status (never/ex-smoker vs. current smoker), BMI (<30kg/m 2 vs. >30kg/m 2 ), previous history of hypertension, diabetes, chronic kidney disease and cardiovascular disease.

Data analysis
Descriptive statistics were used to describe the number of patients classified with sustained hypertension, white coat hypertension, normotension and masked hypertension with the triaging approach (the primary outcome) using daytime ABPM as the reference standard. These were used to calculate the sensitivity, specificity, positive and negative predictor values of the triaging approach and the total proportion of participants with correctly classified hypertensive status and proportion that would have been referred for ABPM.
To examine model performance, a logistic regression model was constructed with true hypertension (defined by daytime ABPM) as the dependant outcome variable and classification using the triaging approach as the independent predictor variable. From this model the area under the receiver operating characteristic (AUROC) curve statistic was estimated. For secondary analyses, classification of patients' hypertensive status and utilisation of ABPM with the PROOF-BP strategy was compared to guideline recommended strategies for the diagnosis of hypertension 3,6-9 (eTable 2) with McNemar's chi-squared test.
Further analyses were conducted examining the primary outcomes using different definitions of clinic and ambulatory blood pressure: 1) Mean of the 2nd & 3rd clinic readings 2) Mean of all three clinic readings 3) Mean daytime ambulatory blood pressure defined as 'up-to-standard' (estimated from a minimum of 14 readings) 3 4) Mean daytime ambulatory blood pressure defined as 'up-to-standard', with outlying readings excluded 10 and standardised times for defining the daytime period (7am-11pm; derived from raw ambulatory blood pressure data) 5) Mean 24 hr ambulatory blood pressure (all available readings) 6) Mean 24 hr ambulatory blood pressure defined as 'up-to-standard' (estimated from a minimum of 14 daytime readings and 7 night-time readings [if readings are taken at 30 minute intervals] or 70% of attempted night-time readings) 3 7) Mean night-time ambulatory blood pressure (all available readings) 8) Mean night-time ambulatory blood pressure defined as up-to-standard readings (estimated from a minimum of 7 readings [if readings are taken at 30 minute intervals] or 70% of attempted readings). 9) Mean night-time ambulatory blood pressure defined as 'up-to-standard', with outlying readings excluded 10 and standardised times for defining the daytime period (11pm-7am; derived from raw ambulatory blood pressure data)

Post-hoc analyses
Post-hoc analyses were undertaken to examine performance of the PROOF-BP algorithm on its own (without additional ambulatory monitoring) and this was compared to other blood pressure measurement strategies (employed without ABPM). 10 11 30-32 Subgroup analyses of the sensitivity and specificity of the PROOF-BP triaging approach were undertaken and accompanied by one additional non pre-specified subgroup: patients in whom the treating clinician's own monitor was used to measure clinic blood pressure vs. those where monitors were provided by the research team.
All analyses were conducted using STATA version 13.1 (MP parallel edition, StataCorp, Texas, USA).
Results are presented as means or proportions, with standard deviations or 95% confidence intervals, unless otherwise stated.

Sample size
Based on the original validation of the PROOF-BP prediction model, 4 accrual of data from at least 800 patients was required for estimation of hypertensive status with an accuracy of ±1-3% (see eTable 3, online appendix). 1 A sample size of up to 1000 patients was specified to ensure that the pre-specified sub-group analysis could be adequately powered. Approximately 364 patients (182 in each group) were required in each sub-group to examine the secondary outcomes proposed in the proposed study. This was based on a likelihood ratio test of two proportions detecting a 10% difference in the classification of hypertensive status between two sub-group populations, with a significance level of 0.05 and 90% power.

Patient and public involvement
Patients with a history of hypertension were approached to discuss the study at the design phase of the project. In particular their opinions were sought on the methods of recruitment and patient facing study literature, prior to ethics and NHS R&D applications.

STUDY DESIGN
The aim of this study is to examine novel strategies for the diagnosis and management of hypertension using data from routine clinical practice. This will be achieved using a prospective, multi-centre observational cohort study design, setting up a Prospective Register Of patients undergoing repeated OFfice and Ambulatory Blood Pressure Monitoring (PROOF-ABPM) in Primary Care, Secondary Care and at pharmacies. Data contained within the register will include patient characteristics, repeated clinic and ambulatory blood pressure, clinical assessment data and subsequent admissions to hospital and mortality. The PROOF-ABPM will be unique in its consideration of multiple clinic blood pressure measurements in relation to ambulatory blood pressure readings taken in routine clinical practice.

Primary objective
Establish the accuracy of the PROOF-BP prediction tool at predicting out-ofoffice blood pressure in routine clinical practice The proportion of true positive, true negative, false positive and false negative classifications of hypertension according to out-of-office monitoring.

Secondary objectives
Establish the accuracy of the PROOF-BP prediction tool compared to current strategies in routine clinical practice Improvement in the classification of patients' hypertensive status of >10% or reduction in the utilisation of out-of-office monitoring of >20% compared to existing strategies.
Establish the accuracy of the PROOF-BP prediction tool in different clinical settings (Primary Care, Secondary Care and pharmacies) Difference in the proportion of patients correctly classified as hypertensive (according to out-of-office monitoring) by setting.
Establish the accuracy of the PROOF-BP prediction tool in specific populations (older vs. younger patients, males vs. females, those at high cardiovascular disease risk vs. low risk, those taking antihypertensive medications vs. those not) Difference in the proportion of patients correctly classified as hypertensive (according to out-of-office monitoring) by patient characteristic.
Exploratory objectives Examine whether the 'adjusted clinic blood pressure' generated by the prediction model predicts long term clinical outcomes (e.g. hospital admission with myocardial infarction or stroke, mortality) better than standard clinic blood pressure Hazard ratio describing the association between adjusted clinic blood pressure and total mortality, cardiovascular mortality, hospital admission with stroke, myocardial infarction or heart failure.

PRIMARY OUTCOME
The primary outcome of this study will be to define the accuracy of the PROOF-BP prediction tool in terms of the proportion of true/false positive/negative results in the general population attending routine clinical practice.

SECONDARY OUTCOMES
Secondary outcomes will include assessment of model accuracy in different sub-groups -age (young vs. old), cardiovascular disease risk (high vs. low risk according to previous history and risk scores using data where available), those with Chronic Kidney Disease, Diabetes and across healthcare settings: Primary Care, Secondary Care and pharmacy settings.
In the longer term, linked data from the registry will be used to examine whether the 'adjusted clinic blood pressure' generated by the prediction model can better predict long term clinical outcomes (e.g. hospital admission with myocardial infarction/stroke and mortality) than standard clinic blood pressure. After the initial period of data collection is complete, the resources and funding required to continue ongoing data collection and follow-up will be reviewed. Where possible, patient accrual and data collection will continue and the study will become a research database permitting further investigations into blood pressure monitoring by a variety of means and cardiovascular disease risk factor data linked to cardiovascular disease morbidity and mortality in routine clinical practice.

TARGET POPULATION
The register is web-based to permit access from a variety of healthcare settings. Data collection procedures will be piloted in the hypertension clinic at University Hospitals Birmingham. Following successful role out in this clinic, other centres from Primary care will be invited to contribute patient data to the registry. Eligible patients will meet the following inclusion/exclusion criteria:  Lack of availability of basic clinical information  Clinic blood pressure readings obtained on at least three occasions within the same visit not recorded  Ambulatory blood pressure monitor not worn as instructed and/or invalid readings

SAMPLE SIZE
The proposed study will collect data on consecutive patients referred for ambulatory blood pressure monitoring in routine clinical practice. Based on the initial validation phase of the PROOF-BP prediction model, conducted using data from previous studies, 4 accrual of data from approximately 1000 patients would allow for estimation of hypertensive status with an accuracy of ±1-3%. In this previous data, 71% of patients were classed as true positives, 24% were classed as true negatives, 3% were classed as false positives and 2% were classed as false negatives. In population of 1000 patients it would be possible to estimate these rates with the following 95% confidence intervals: true positive 71% (68-74%), true negative 24% (21-27%), false positive 3% (2-4%) and false negative 2% (1-3%). would be required to demonstrate a significant difference. Thus, our recruitment target of 1000 patients should be sufficient to answer the secondary outcomes provided recruitment is appropriately distributed across clinic settings and patient characteristic sub-groups.

DESCRIPTIVE STATISTICS
Descriptive statistics will be used to define the study population at baseline, given from the total population, and separately by those recruited from primary and secondary care. Patient characteristics will include demographic data on age, gender, ethnic origin and other baseline characteristics including medical history (past and present), current therapies, and mean clinic and daytime ambulatory blood pressure levels. Summary information regarding the clinic and ambulatory blood pressure monitoring conducted will be also be present, including arm used for monitoring, number of readings taken and reason for referral for ambulatory monitoring. A full list of variables to be presented is listed in the blank data tables/figures given in appendix 1. All data will be presented as means ± standard deviation or proportions of the total population, unless otherwise stated.

DEFINITION OF POPULATION FOR ANALYSIS
The population eligible for both primary and secondary analyses will be all those attending routine clinical practice for ambulatory blood pressure monitoring in both primary and secondary care. Patients will be excluded from the analysis if they do not have data relating to clinic (three readings from the same visit) or mean daytime ambulatory blood pressure, are aged <18 years, or lack the basic clinical information required to calculated an adjusted clinic blood pressure using the PROOF-BP prediction tool (age, sex, BMI, history of hypertension, cardiovascular disease and/or antihypertensive treatment). A comparison of characteristics from included and excluded patients will be given as an additional table in the online appendix.

DEFINITION OF VARIABLES FOR ANALYSIS
Clinic blood pressure will presented as the mean of the 2 nd and 3 rd readings in the baseline characteristics. For the primary analysis, clinic blood pressure will be defined as 1 st clinic reading taken, prior to fitting of the ambulatory blood pressure monitor, replicating the definitions used in the original PROOF-BP analyses.
Out-of-office blood pressure will be defined as the mean of all daytime ambulatory blood pressure readings recorded during ABPM. The daytime period will be defined as it was used for routine practice -i.e. defined by patient diary, and/or monitor settings pre-programmed by the treating clinician. Sensitivity analyses will explore the impact of limiting analyses to 'up-to-standard' ABPM readings. A minimum of 14 readings will be required to define an 'up-to-standard' daytime blood pressure. 3 Up-to-standard nighttime blood pressures will be defined as a minimum of 7 readings (if nighttime blood pressure is measured at 30 minute intervals) or 70% of attempted readings being present. 11 Mean 24hr blood pressure will be defined as all readings taken during the ABPM period.
All other variables will be defined as recorded from the participants' medical records. Sensitivity analyses will be conducted examining the primary outcomes using different definitions of clinic and ambulatory blood pressure (see section 5 below).

PRIMARY ANALYSIS
3.1 PRIMARY OUTCOME The primary outcome of this study will be to define the accuracy of the PROOF-BP prediction tool in terms of the proportion of true/false positive/negative results in the general population attending routine clinical practice. The PROOF-BP algorithm will be used to estimate an adjusted clinic blood pressure which will then be compared to the corresponding ambulatory blood pressure (the reference standard) to identify the proportion of patients classified as true positives (sustained hypertensives), false positives (white coat hypertensives), true negatives (normotensives) and false negatives (masked hypertensives). Thresholds of blood pressure level to define hypertension by each measurement method are defined in table 2 (below).
Where an adjusted clinic blood pressure would result in referral for ABPM (i.e. a blood pressure estimate of 130/80-145/90mmHg), hypertensive classification will be defined as correct (true positive/negative). Adjusted clinic and ambulatory blood pressures which are discordant from one another will be classified as false positive (white coat hypertensives) or false negatives (masked hypertensive), depending of the direction of disagreement.