Predicting Out-of-Office Blood Pressure in the Clinic (PROOF-BP)

Supplemental Digital Content is available in the text.

Abstract

Patients often have lower (white coat effect) or higher (masked effect) ambulatory/home blood pressure readings compared with clinic measurements, resulting in misdiagnosis of hypertension. The present study assessed whether blood pressure and patient characteristics from a single clinic visit can accurately predict the difference between ambulatory/home and clinic blood pressure readings (the home–clinic difference). A linear regression model predicting the home–clinic blood pressure difference was derived in 2 data sets measuring automated clinic and ambulatory/home blood pressure (n=991) using candidate predictors identified from a literature review. The model was validated in 4 further data sets (n=1172) using area under the receiver operator characteristic curve analysis. A masked effect was associated with male sex, a positive clinic blood pressure change (difference between consecutive measurements during a single visit), and a diagnosis of hypertension. Increasing age, clinic blood pressure level, and pulse pressure were associated with a white coat effect. The model showed good calibration across data sets (Pearson correlation, 0.48–0.80) and performed well-predicting ambulatory hypertension (area under the receiver operator characteristic curve, 0.75; 95% confidence interval, 0.72–0.79 [systolic]; 0.87; 0.85–0.89 [diastolic]). Used as a triaging tool for ambulatory monitoring, the model improved classification of a patient’s blood pressure status compared with other guideline recommended approaches (93% [92% to 95%] classified correctly; United States, 73% [70% to 75%]; Canada, 74% [71% to 77%]; United Kingdom, 78% [76% to 81%]). This study demonstrates that patient characteristics from a single clinic visit can accurately predict a patient’s ambulatory blood pressure. Usage of this prediction tool for triaging of ambulatory monitoring could result in more accurate diagnosis of hypertension and hence more appropriate treatment.

See Editorial Commentary, pp 834–835

High blood pressure (hypertension) is an important risk factor for cardiovascular disease, a significant cause of morbidity and mortality worldwide.¹ The diagnosis and management of hypertension depend on accurate measurement of blood pressure so that antihypertensive treatment can be targeted appropriately and unnecessary adverse effects and healthcare costs can be avoided.² Traditionally, blood pressure measurement has taken place in the primary care physician’s office or clinic using an electronic oscillometric or aneroid sphygmomanometer (clinic blood pressure; Table 1), but it has long been recognized that home or ambulatory (out-of-office) blood pressures provide more accurate estimates of a patient’s true mean blood pressure.³ This is in part because multiple readings are taken and it correlates with a range of cardiovascular outcomes and end-organ damage better than clinic blood pressure.^4,5

Table 1.

Definitions of Blood Pressure Measurements Described in the Present Study

Clinic blood pressure values are often different from out-of-office blood pressure and can lead to incorrect classification of blood pressure status and hence inappropriate management.^6,7 Patients with higher clinic blood pressure than the corresponding out-of-office pressure will have a negative home–clinic blood pressure difference (white coat effect) and are at risk of overtreatment (Figure S1 in the online-only Data Supplement).⁸ Conversely, patients with higher out-of-office blood pressures than the corresponding clinic blood pressure would be expected to have a positive home–clinic blood pressure difference (masked effect), and often remain unrecognized and therefore potentially undertreated (Figure S1).⁹ Such patients are at increased risk of target-organ damage¹⁰ and have cardiovascular morbidity and mortality not dissimilar to sustained hypertension.¹¹

Strategies to reduce these misclassifications are emerging and include the use of multiple automated clinic blood pressure readings which have been shown to reduce the white coat effect.^12,13 In the United States, the Preventive Services Task Force¹⁴ have recently released guidelines recommending that home or ambulatory blood pressure monitoring is used to confirm a diagnosis of hypertension, an approach which has already been adopted in the United Kingdom,¹⁵ where it is considered cost-effective because of a reduction in misdiagnosis caused by the white coat effect.² However, this approach will result in some patients with sustained hypertension identified by clinic blood pressure readings being sent for arguably unnecessary out-of-office monitoring, which some patients find uncomfortable, and importantly, this strategy will not capture those patients with masked hypertension.

Recent work by some of the authors has shown that the change in clinic blood pressure compared with multiple automated clinic readings from a single visit can predict the home–clinic blood pressure difference.¹⁶ This study aimed to use patient characteristics and details of repeated clinic blood pressure measurements to derive a model for predicting this home–clinic blood pressure difference. Furthermore, we aimed to validate this model and examine its application as a means to target ambulatory blood pressure monitoring more efficiently in routine clinical practice.

Methods

An extended version of the Methods is available in the online-only Data Supplement. Blood pressure definitions and terminology used are summarized in Table 1.

Study Design and Source Data

The present study was an individual patient data analysis of cohort studies conducted in a primary care setting.^12,17–21 A linear regression model predicting the home–clinic blood pressure difference was derived in 2 data sets using candidate predictors identified from a literature review. All included studies collected relevant data including clinic, home, and daytime ambulatory blood pressure readings, using a validated electronic oscillometric blood pressure monitor, and details of patient characteristics and medical history. The characteristics of patients from included studies are detailed in Table 2. Individual clinic readings were available in each study permitting estimation of a variety of different definitions of clinic blood pressure. Patients in the Conventional Versus Automated Measurement of Blood Pressure in the Office (CAMBO) study¹² had their clinic blood pressure measured with a BpTRU device with either the doctor or nurse taking the first reading and then leaving the room for the remaining measurements. In all other studies, multiple clinic readings were taken in the presence of a nurse or practice reception staff.^17–20 Because our study involved secondary analysis of existing data, it was not possible to standardize protocols for blood pressure measurement across studies, and specific protocols for the measurement of home and daytime ambulatory blood pressures did vary to some degree (Table S1).

Table 2.

Characteristics of Studies Used for Model Derivation and Validation

Patients were selected for the derivation cohort from the Blood Pressure in Ethnic groups (BP-Eth)¹⁷ and Telemonitoring and Self-Management in the Control of Hypertension (TASMINH2)¹⁹ studies (n=991) because these were considered to be sufficiently large and representative of the population likely to undergo blood pressure monitoring for diagnosis and management of hypertension. Patients from the remaining 4 studies^12,18,20,21 were used in the validation cohort (n=1172).

Statistical Analysis

Selection of Candidate Predictors

Candidate predictors considered for inclusion in the model were identified by literature review.²² Of the 60 identified, a total of 14 variables were considered for inclusion in the model, including age, sex, body mass index, diagnosis of hypertension and time from diagnosis, antihypertensive prescription, smoking status, alcohol consumption, diagnosis of cardiovascular disease, clinic blood pressure level (systolic/diastolic), and multiple clinic blood pressure characteristics defined as previously described.¹⁶ These characteristics were the difference between the first and last clinic blood pressure reading (referred to as the clinic blood pressure change [estimated from 3 or 6 readings]), the rate of the change in clinic blood pressure (referred to as the blood pressure slope [estimated from 3 or 6 readings]), and the curvature of this change in clinic blood pressure (referred to as the blood pressure quadratic [estimated from 6 readings]). Age, sex, and clinic blood pressure variables were included in the final model a priori because they were cited as significant predictors of white coat or masked hypertension in more than twice as many published studies compared with other predictors in a previous literature review.^22,23 Backward stepwise selection was used to select the remaining candidate predictors for the final model. Only predictors reaching a significance level of P<0.05 were included.

Model Derivation

A complete case analysis was conducted because of low amount of missing data. Two separate linear regression models were constructed examining factors that predict the systolic and diastolic home–clinic blood pressure difference (the difference between the first clinic blood pressure reading and mean out-of-office blood pressure). Out-of-office blood pressure was taken to be mean daytime ambulatory blood pressure where available; otherwise, mean home blood pressure was used (Table 1).^19,20 Because of colinearity between some of the candidate predictors listed above, separate models were compared in 3 stages using likelihood ratio tests. The best fitting model at each stage was considered in the next stage and where there was no significant difference in model fit, the most parsimonious model was selected. Stage 1 compared 5 different prediction models examining combinations of clinic blood pressure characteristic. Stage 2 compared the best fitting model from the first step using different definitions of clinic blood pressure (first clinic reading; mean of 1–3 readings; mean of 2–3 readings; mean of 1–6 readings; or the mean of 2–6 readings). The final stage explored prespecified interactions of all candidate predictors with age, sex, and diagnosis of hypertension, and the interaction between clinic blood pressure and the characteristics of the change in clinic blood pressure. Ninety-five percent confidence intervals (CIs) for model coefficients were estimated with bootstrap resampling (200 replications). Model coefficients are presented for centered continuous variables in the final model.

Model Validation and Performance

The agreement between predicted and actual home–clinic blood pressure differences was examined in both derivation and validation cohorts using Pearson correlation coefficient and Bland–Altman plots. A model-adjusted clinic blood pressure value was calculated by combining the original clinic pressure (first clinic reading) with the home–clinic blood pressure difference estimated from the model. The ability of the model-adjusted clinic blood pressure to predict out-of-office hypertension was assessed using area under the receiver operator characteristic (AUROC) curve statistics. High AUROC values (closer to 1) indicate better model discrimination.

Potential strategies for referral for out-of-office monitoring were explored in the derivation cohort (Tables S2 and S3), with the optimal strategy defined as a threshold which produced an overall classification error of ≤10% with the lowest proportion of patients referred for out-of-office monitoring. Model performance detecting true out-of-office hypertension was compared with existing strategies for blood pressure measurement described in international hypertension guidelines (Table S4).^15,24–26 The model was also applied to a nominal population from our validation cohort, with a comparable distribution of clinic blood pressures to that documented in the Health Survey for England. Using this nominal population, the number of patients being correctly diagnosed with hypertension per 1000 individuals was compared with the current National Institute for Health and Care Excellence (NICE) diagnostic algorithm¹⁵ (considered best of the rest).

Sensitivity analyses explored the model performance by individual study, predicting home or daytime ambulatory blood pressure, in patients with raised clinic blood pressure and those with controlled or normal clinic blood pressure. We also examined a revised diagnostic strategy which does not use the Predicting Out-of-Office Blood Pressure (PROOF-BP) algorithm, but in which patients with a clinic blood pressure between 130/85 and 160/100 mm Hg were referred for out-of-office monitoring. All sensitivity analyses were conducted in the validation cohort (except those by individual study which compared all available data).

All analyses were performed in STATA version 13.1 (MP parallel edition, StataCorp, TX). Data are presented as proportions of the total study population or means with SD or 95% CIs unless otherwise stated. Ethical approval was given for all individual studies contributing data, but approval for secondary analysis using anonymized data was not required.

Results

Characteristics of the Study Cohort

Of the 2470 patients with out-of-office blood pressure measurements enrolled across the 6 studies, relevant data were available for analysis in a total of 2163 patients (991 patients in the derivation cohort; 1172 patients in the validation cohort). Relevant data were not available in some patients enrolled in the CAMBO study¹² because not all centers recorded the individual automated clinic blood pressure readings required for this analysis (Table 2; Table S1). Characteristics of the derivation and validation cohorts were similar in terms of age, sex, the prevalence of systolic white coat hypertension and systolic masked hypertension (Table S5).

Model Derivation

Goodness-of-fit was similar between models examining 3 or 6 clinic blood pressure readings (derivation stage 1; adjusted R², 0.50–0.52) and those using different definitions of clinic blood pressure (derivation stage 2; adjusted R², 0.50–0.52). The most parsimonious model selected at each stage was that which used patient characteristics along with the clinic blood pressure change (estimated from 3 clinic readings), with the first clinic reading as an estimate of clinic blood pressure.

The systolic masked effect (a positive home–clinic difference) was associated with male sex and a positive clinic blood pressure change (Figure 1). The clinic blood pressure change and a history of hypertension were predictive of a diastolic masked effect. The systolic white coat effect (negative home–clinic blood pressure difference) was associated with increasing clinic blood pressure only (Figure 1). The diastolic white coat effect was associated with increasing clinic blood pressure, increasing age, and increasing pulse pressure. The final model included significant interactions between age, sex, clinic blood pressure, the clinic blood pressure change, pulse pressure, body mass index, history of cardiovascular disease, presence of an antihypertensive prescription, history of hypertension, and duration of hypertension. The final model (centered) coefficients are presented in Table 3 and the full equation is given in Figure S6.

Figure 1.

Coefficient plot showing predictors of the systolic and diastolic home–clinic blood pressure difference. A, Continuous predictors. B, Binary predictors. All coefficients are presented from the model before stepwise selection. Coefficients for continuous variables are presented as centered values per unit increase. BMI indicates body mass index; BP, blood pressure; and CVD, cardiovascular disease.

Table 3.

Linear Regression Model for Prediction of the Systolic/Diastolic Home–Clinic Blood Pressure Difference

Model Validation and Performance

The final model showed good calibration across all derivation and validation data sets (Pearson correlation, 0.62–0.80 [systolic]; 0.48–0.80 [diastolic]; P<0.001; Figures S2 and S3). At the extremes of home–clinic blood pressure difference, the model was less accurate, as evidenced by the slight skew observed in the Bland–Altman plots (Figures S4 and S5), suggesting that the model underpredicts those with a large masked effect and overpredicts those with a large white coat effect.

The model was good at discriminating out-of-office hypertension (masked or sustained hypertension) in the derivation cohort (AUROC, 0.80; 95% CI, 0.78–0.83 [systolic model]; 0.82; 95% CI, 0.80–0.85 [diastolic model]) and this discrimination was maintained in the validation cohort (AUROC, 0.75; 95% CI, 0.72–0.79 [systolic model]; 0.87; 95% CI, 0.85–0.89 [diastolic model]; Figure 2). Using the model-adjusted clinic blood pressure, the optimal thresholds for referral for out-of-office monitoring were ≥130/80 and <145/90 mm Hg. In other words, in a model-adjusted blood pressure of <130/80 mm Hg, patients were confidently predicted as normotensive and those with a model-adjusted blood pressure of ≥145/90 mm Hg were considered hypertensive. Anyone with a model-adjusted blood pressure between the 2 required out-of-office measurements. Using this model-adjusted blood pressure to triage patients for out-of-office monitoring, 93% of cases were correctly classified (Table 4). This was an improvement of ≤29% compared with strategies recommended in current clinical guidelines (PROOF-BP, 93% [92% to 95%]; American Heart Association, 73% [70% to 75%]; Canadian Hypertension Education Program, 74% [71% to 77%]; European Society of Hypertension, 73% [70% to 75%]; NICE, 78% [76% to 81%]) with similar usage of out-of-office monitoring (PROOF-BP, 58% referred [55% to 61%] versus NICE, 54% referred, [51% to 57%]; Table 4). In a nominal representative population, for every 1000 people aged from 45 to 74 years screened with the new algorithm, correct classification would be gained for 910 patients with 395 diagnosed as hypertensive, compared with the next best strategy (NICE algorithm) where 853 would be correctly classified and only 274 diagnosed as hypertensive. The additional 121 diagnoses of hypertension result from the detection of those patients with masked hypertension.

Figure 2.

Area under the receiver operator characteristic curve (AUROC) analysis showing performance of systolic/diastolic blood pressure (sBP/dBP) prediction models for discrimination of out-of-office hypertension. A, sBP prediction model. B, dBP prediction model.

Table 4.

Comparison of Model Performance With Current Clinical Practice in the Validation Cohort (n=1100)

Sensitivity Analyses

The results of the sensitivity analyses are detailed in the Appendix S1 (Table S6). Model performance was consistent across individual studies (AUROC, 0.61–0.78 [systolic model]; 0.74–0.91 [diastolic model]) and the new algorithm resulted in better targeting of out-of-office blood pressure compared with a revised diagnostic strategy using clinic blood pressure alone to triage patients for out-of-office monitoring. Using the new algorithm to triage only those patients with raised clinic blood pressure (ie, only considering those patients with a potential white coat effect) resulted in correct classification of 94% of patients with only 45% requiring out-of-office monitoring (Table S6).

Discussion

This study describes a clinical prediction model which combines patient characteristics (age, sex, body mass index, history of hypertension, cardiovascular disease, and antihypertensive treatment) and 3 clinic blood pressure measurements from a single visit to accurately predict a patient’s out-of-office blood pressure. Used as a triaging tool for out-of-office monitoring, detection of hypertension or uncontrolled blood pressure was markedly improved from existing diagnostic and management strategies, specifically including those with previously unrecognized masked hypertension.

Strengths and Limitations

This retrospective study used a large cohort of patients from 6 previous studies providing a population representative of individuals from the United Kingdom and North America undergoing blood pressure measurement in primary care. Sensitivity analyses revealed consistent model performance across individual studies, suggesting that it would be effective regardless of the electronic blood pressure monitoring device (BpTRU; Stabil-O-Graph) or measurement protocol used (rest period versus no rest period; nurse present during measurement versus nurse not present; automatic readings versus patient/nurse-initiated readings; 1-minute versus 2-minute intervals between readings). It is well known that blood pressure measurements made under controlled conditions in a research setting are not necessarily comparable with those made by a physician in routine clinical practice.^24,27,28 Differences occur for a variety of reasons, including the use of inadequate or uncalibrated devices^29–31 and suboptimal measurement techniques.^32–34 Indeed, the present algorithm requires 3 consecutive blood pressure readings to be taken at a single clinic visit and although this is recommended in most hypertension guidelines,^15,24–26 ensuring this approach is adhered to in routine practice may require some education of physicians and nurses. Therefore, although this prediction model is shown to be accurate in a research setting, it is not guaranteed to work in routine clinical practice, and prospective validation of the PROOF-BP prediction tool in a clinical setting warrants further investigation.

Bland–Altman plots demonstrated that the PROOF-BP prediction model tends to underestimate those with a large masked effect and overestimate those with a large white coat effect. This is likely to be explained by the underlying population in the derivation cohort which contained a higher proportion of hypertensive (defined by clinic readings) patients on treatment (704 [71%]), a population known to have an exaggerated white coat effect compared with normotensives (Figure S1).³⁵ Although the model was less accurate at extremes of home–clinic blood pressure difference, it showed good performance detecting out-of-office hypertension around the clinical threshold (140/90 mm Hg) where the average home–clinic blood pressure difference is smaller.³⁵

The present analyses used home blood pressure data to define out-of-office blood pressure, where daytime ambulatory measurements were not available (Table 1). Some argue that 24-hour ambulatory blood pressure should be used as the gold standard measure of blood pressure²⁶ because it contains information about night-time blood pressure which includes additional prognostic information. However, the recent review by the US Preventive Services Task Force found no apparent difference among 24-hour, daytime, and night-time ambulatory blood pressure measurement protocols for prediction of cardiovascular outcomes,¹⁴ and a sensitivity analysis in the present study demonstrated no change in the accuracy of the PROOF-BP prediction model using home or daytime ambulatory readings.

Comparison With Previous Literature

Many studies have examined the association between patient characteristics and the home–clinic blood pressure difference, noting age,^36–38 sex,^36–39 and clinic blood pressure level,^36–39 along with anxiety,^40,41 stress,⁴² and other factors²² as significant independent predictors of white coat or masked hypertension. The findings of the present study were consistent with previous literature, showing age, sex, clinic blood pressure, pulse pressure, and a history of hypertension as significant predictors of the home–clinic blood pressure difference. Interestingly, female sex was not a significant predictor of the white coat effect, although it was included a priori in the final prediction model because this association has been well-defined in the previous literature.^36,38

Few studies have suggested a strategy for targeted use of out-of-office blood pressure in routine clinical practice. Myers et al⁴³ and Godwin et al⁴⁴ have proposed the use of multiple (automated) office blood pressure readings taken using the BpTRU device to identify patients with high normal blood pressure (130/80 to 139/89 mm Hg) who could be referred for ambulatory blood pressure to confirm the presence of masked hypertension. Similarly, Viera et al⁴⁵ examined optimal automated clinic blood pressure levels for referral for out-of-office monitoring in patients with normal clinic pressure for detection of masked hypertension. However, they concluded that using clinic blood pressure alone was not sufficient because of high referral rates, and suggested that a combination of factors including patient characteristics might be more effective at targeting out-of-office blood pressure more efficiently.

Implications for Clinical Practice

The US Preventive Services Task Force¹⁴ recently released guidelines recommending that ambulatory blood pressure monitoring is used to confirm a diagnosis of hypertension. It is anticipated that these guidelines will follow a similar approach to that advocated in the United Kingdom by NICE, which does not capture those patients with masked hypertension. The present analyses propose a method for capturing nearly all patients with truly raised out-of-office blood pressure which is likely to result in a small increase in the amount of out-of-office monitoring required in routine practice but could still be cost-effective if it reduces the present best practice involving indiscriminate application of ambulatory monitoring.^2,14,15 Indeed, our sensitivity analyses show that in patients with raised clinic readings, the PROOF-BP prediction model could potentially reduce the proportion of referrals for daytime ambulatory monitoring by more than half to 285/629 (45%), with nearly all patients being accurately diagnosed (589/629 [94%] correctly classified) and acceptable false-positive and false-negative rates (6% and 0%, respectively). Importantly, this new method identifies patients with possible masked hypertension which is otherwise unsuspected unless there is evidence of unexpected end-organ damage.

An algorithm for using the PROOF-BP prediction tool in routine clinical practice is presented in Figure 3. Electronic blood pressure monitors which take 3 consecutive readings (at 1-minute intervals) are now cheap and routinely available, permitting use of this algorithm before, during, or after a standard physician consultation in primary care. The prediction model could easily be incorporated into general practice computer systems, accessed as an online calculator or even built into smartphones linked to blood pressure monitors to facilitate implementation in routine clinical practice. This novel approach to measurement and management would require buy in from both patients and practitioners and therefore some degree of education may be required during implementation.

Figure 3.

Algorithm for using the Predicting Out-of-Office Blood Pressure in clinic tool (PROOF-BP) prediction model to triage patients for out-of-office blood pressure monitoring. Existing strategies are based on the hypertension diagnostic pathway specified by the US Preventive Services Task Force and the National Institute for Health and Care Excellence. BP indicates blood pressure.

Perspectives

The present study shows that a combination of simple patient characteristics with 3 clinic blood pressure measurements from a single visit can accurately identify those patients requiring out-of-office blood pressure monitoring for suspected white coat hypertension and arguably most importantly masked hypertension. This prediction model has the potential to improve the accuracy of diagnosis and management of hypertension in primary care, and prospective validation in routine clinical practice along with analysis of cost-effectiveness are now warranted.

Acknowledgments

We would like to thank the researchers, patients, and practices who took part in the original Blood Pressure in Ethnic groups (BP-Eth), Conventional Versus Automated Measurement of Blood Pressure in the Office (CAMBO), Telemonitoring-Based Service Redesign for the Management of Uncontrolled Hypertension (HITS), Oxford self-monitoring study, Telemonitoring and Self-Management in the Control of Hypertension (TASMINH2), and Targets and Self-Management for the Control of Blood Pressure in Stroke and at Risk Groups (TASMINH-SR) studies without whom this work would have been impossible. We would also like to thank Roger Holder and David Yeomans for their assistance and support during the original conception, design, and dissemination of this study.

Sources of Funding

This work was funded by a Medical Research Council (MRC) Strategic Skills Post-doctoral Fellowship (MR/K022032/1) held by J.P. Sheppard, with support from a National Institute for Health Research (NIHR) Programme Grant (RP-PG-1209–10051). R.J. McManus holds an NIHR Professorship and leads the self-management theme of the NIHR Oxford Collaboration for Leadership in Applied Health Research and Care (CLAHRC). B. Williams is a NIHR Senior Investigator and his research is supported by the NIHR University College London Hospitals Biomedical Research Centre. F.D.R. Hobbs is in part funded as Director of the NIHR School for Primary Care Research, Theme Leader of the NIHR Oxford Biomedical Research Centre, and Director of the NIHR CLAHRC Oxford. The views and opinions expressed are of us and do not necessarily reflect those of the MRC, National Health Service, NIHR, or the Department of Health.

Disclosures

R.J. McManus has received equipment for research purposes from Omron and Lloyds Healthcare. C. Heneghan has received expenses and payments for his media work from Channel 4, British Broadcasting Corporation, FreshOne Television productions and the Guardian, and also expenses from the World Health Organization and the US Food and Drug Administration. He is also an expert witness in an ongoing medical device legal case, has received payment from British United Provident Association for analyzing and appraising guidelines and income from the publication of a series of tool kit books published by Blackwells. F.D.R. Hobbs has received limited research support in terms of blood pressure devices from Microlife and BpTRU. B. Williams works in academic collaboration with Healthstats, Singapore, in developing novel blood pressure–monitoring approaches. The other authors report no conflicts.

Supplementary Material

hyp-67-0941-s001.docx ^{(1.4MB, docx)}

Novelty and Significance

What Is New?

• This study shows that a simple linear regression model incorporating patient characteristics and 3 consecutive clinic blood pressure measurements from a single clinic visit can accurately identify those patients requiring out-of-office blood pressure monitoring for suspected white coat or masked hypertension.

• The model performance was consistent across populations, and in studies using different blood pressure monitors and measurement techniques, suggesting that the results are widely applicable across Primary Care.

What Is Relevant?

• This prediction model could be used as an online calculator or integrated into practice computer systems for triaging of out-of-office monitoring to permit detection of those patients with white coat or masked hypertension in routine clinical practice.

• The present analyses suggest such an approach would improve the detection of hypertension and provide a strategy for capturing patients with apparent masked hypertension for the first time.

Summary

Our findings suggest that it is possible to predict which patients are most likely to display a white coat or masked effect, using patient characteristics and multiple clinic blood pressure measurements from a single clinic visit.