Key Points
Question
What is the concordance between routine clinic blood pressure (BP) measurements and formal trial blood pressure measurements and does discordance vary by target trial blood pressure?
Findings
In this prognostic study of 3074 participants with 3 or more outpatient and trial BP measurements linking electronic health record (EHR) data with data from the Systolic Blood Pressure Intervention Trial, the mean systolic blood pressure recorded in outpatient electronic health records was higher than that measured in the trial. The difference between blood pressure recorded in electronic health records and trial blood pressure varied widely by site.
Meaning
These results highlight the importance of proper BP measurement technique and the potential shortcomings of comparing trial-measured BP with BP measurements from clinical EHRs.
Abstract
Importance
There are concerns with translating results from the Systolic Blood Pressure Intervention Trial (SPRINT) into clinical practice because the standardized protocol used to measure blood pressure (BP) may not be consistently applied in routine clinical practice.
Objectives
To evaluate the concordance between BPs obtained in routine clinical practice and those obtained using the SPRINT protocol and whether concordance varied by target trial BP.
Design, Setting, and Participants
This observational prognostic study linking outpatient vital sign information from electronic health records (EHRs) with data from 49 of the 102 SPRINT sites was conducted from November 8, 2010, to August 20, 2015, among 3074 adults 50 years or older with hypertension without diabetes or a history of stroke. Statistical analysis was performed from May 21, 2019, to March 20, 2020.
Main Outcomes and Measures
Blood pressures measured in routine clinical practice and SPRINT.
Results
Participant-level EHR data was obtained for 3074 participants (2482 men [80.7%]; mean [SD] age, 68.5 [9.1] years) with 3 or more outpatient and trial BP measurements. In the period from the 6-month study visit to the end of the study intervention, the mean systolic BP (SBP) in the intensive treatment group from outpatient BP recorded in the EHR was 7.3 mm Hg higher (95% CI, 7.0-7.6 mm Hg) than BP measured at trial visits; the mean difference between BP recorded in the outpatient EHR and trial SBP was smaller for participants in the standard treatment group (4.6 mm Hg [95% CI, 4.4-4.9 mm Hg]). Bland-Altman analyses demonstrated low agreement between outpatient BP recorded in the EHR and trial BP, with wide agreement intervals ranging from approximately −30 mm Hg to 45 mm Hg in both treatment groups. In addition, the difference between BP recorded in the EHR and trial BP varied widely by site.
Conclusions and Relevance
Outpatient BPs measured in routine clinical practice were generally higher than BP measurements taken in SPRINT, with greater mean SBP differences apparent in the intensive treatment group. There was a consistent high degree of heterogeneity between the BPs recorded in the EHR and trial BPs, with significant variability over time, between and within the participants, and across clinic sites. These results highlight the importance of proper BP measurement technique and an inability to apply 1 common correction factor (ie, approximately 10 mm Hg) to approximate research-quality BP estimates when BP is not measured appropriately in routine clinical practice.
Trial Registration
SPRINT ClinicalTrials.gov Identifier: NCT01206062
This prognostic study evaluates the concordance between blood pressures obtained in routine clinical practice and those obtained using the Systolic Blood Pressure Intervention Trial protocol and whether concordance varied by target trial blood pressure.
Introduction
In the Systolic Blood Pressure Intervention Trial (SPRINT), intensive blood pressure (BP) lowering to a target systolic BP (SBP) less than 120 mm Hg was associated with decreased risk for cardiovascular disease, all-cause mortality, and mild cognitive impairment.1,2 However, the method of BP measurement in the trial has raised concerns regarding translation of the intensive BP target into clinical practice, where BP measurement techniques vary widely.3,4 As in most landmark hypertension trials,5 BP was measured using a standardized protocol; in the case of SPRINT, American Heart Association (AHA) recommendations were followed.1 Blood pressure was measured after 5 minutes of quiet rest using an automated oscillometric device, proper participant positioning, appropriate cuff size and positioning, and a mean of 3 BP readings.6 Typically, these guidelines are not followed in routine clinical practice, resulting most commonly in an overestimation of BP level, with consequent overtreatment.5 This lack of concordance between routine clinic BP and trial BP measurements has led some to argue that the intensive SBP target of less than 120 mm Hg could correspond to a target as high as less than 140 mm Hg in routine practice.7 The discrepancy was also part of the rationale for the 2017 American College of Cardiology and AHA guideline recommendation for an SBP target of less than 130 mm Hg, a level 10 mm Hg higher than the intensive target in SPRINT.8
Studies that have compared “research quality” BP measurements with corresponding estimates in routine clinical practice are difficult to interpret. The BPs in these studies have typically not been measured on the same day and, even when measured on the same day, have rarely been obtained in a random order.5,9 Furthermore, the routine clinic BPs have often been obtained at the time of a referral for ambulatory BP monitoring or referral to a hypertension specialist, with weeks to months separating the routine measurements from the subsequent “research quality” readings.
SPRINT1 is one of the first large, randomized hypertension treatment trials conducted in the era of widespread use of electronic health records (EHRs), facilitating a direct comparison of trial BP measurements with corresponding, contemporaneous readings obtained in routine clinical practice. As part of a SPRINT ancillary study, we linked outpatient EHR data with trial data. We used this linked information to evaluate the concordance between BPs obtained in routine clinical practice with those obtained in the trial and to assess whether any discrepancies recognized would differ by target trial BP.
Methods
SPRINT was a randomized clinical open-label trial that was conducted at 102 clinical sites.1 A total of 9361 participants were randomized between November 8, 2010, and March 15, 2013, to either an intensive treatment (target study visit SBP<120 mm Hg) or standard treatment (target study visit SBP<140 mm Hg) group. The trial was terminated early after a median follow-up of 3.3 years owing to lower rates of cardiovascular disease events and all-cause mortality in the intensive treatment group. An institutional review board at all 102 participating sites approved the study protocol and this ancillary study was conducted under a waiver of informed consent owing to the use of existing trial and EHR data.
Baseline Data Collection
Data collected at baseline included sociodemographic information, with race/ethnicity assessed by self-report. Body mass index was calculated as weight in kilograms divided by height in meters squared. The estimated glomerular filtration rate was calculated using the CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) study equation.10 Trained study coordinators measured BP according to an AHA-adherent protocol; BP at all study visits was based on a mean of 3 automated readings using a properly sized cuff while the patient was seated taken 1 minute apart after 5 minutes of quiet rest.11 Appropriate cuff size was determined by measuring arm circumference at the midpoint of the upper arm. Adverse events were ascertained at quarterly visits and included participant reports of hospitalizations, emergency department visits, and other health outcomes of interest.12
EHR Data
To increase the likelihood that EHR data would be available, we selected sites for the SPRINT EHR ancillary study based on the following 3 criteria: availability of EHR information, a large number of SPRINT participants, and a high percentage of SPRINT participants who were patients within the primary affiliated health system. A total of 49 study sites (out of 102) provided EHR data for the current analyses, with 25 being Veterans Affairs (VA) medical centers. Sites identified trial participants within their EHR in 1 of 2 ways. A minority of sites used an internal listing of site participants linked to medical record numbers that they had maintained throughout the trial. For the other sites, the SPRINT data coordinating center provided identifiers as a series of cryptographic, 1-way hash codes constructed using secure hashing algorithms based on social security number, first name, last name, sex, and date of birth.13 For first and last name, only the first 6 positions were used after removing hyphens and apostrophes. The same identifiers were extracted from patient records at each site and transformed using an identical cryptographic procedure. Sites then linked SPRINT study identifiers to unique medical record numbers within the EHR based on the hash codes using an algorithm developed by Potosky et al14 and used by Weiner et al.15
Quality Control of EHR BP Readings
After linking participants to their medical record number within the EHR, each site extracted all vital signs, laboratory test results, billing and procedure codes, and health care encounters. With the exception of VA sites (where data were maintained on VA computing systems), these data were then transferred to the SPRINT data coordinating center and linked with trial data via study identifiers. Blood pressures were identified within the vital signs data via labels provided by each site. There was variability in the timing of the 6-month study visit. Outpatient BPs were defined as those that did not coincide with hospitalizations or emergency department visits reported as part of adverse event monitoring during the trial. Electronic health record and trial BPs could occur in any order during this time period and no temporal association was required (eg, 3 days apart).
Blood pressure measurements from the trial were in some instances entered into the EHR. Because there might have been a delay when a coordinator would transcribe a trial BP into the EHR, we excluded any EHR BP measurement if it reflected the same SBP and diastolic BP as a trial measurement and if the encounter date was within 1 week of the trial measurement. Finally, we calculated the mean of any outpatient EHR BP readings that occurred on the same day.
Statistical Analysis
Statistical analysis was performed from May 21, 2019, to March 20, 2020. We restricted analyses to participants who had outpatient measurements of BP in the EHR as well as BP readings from the trial on at least 3 separate days. We estimated mean differences in SBP between the groups (intensive treatment vs standard treatment) and between sources (trial measurements vs measurements in the outpatient EHR) using linear mixed models. Given the pragmatic nature of our study, our data are not linked like a traditional “agreement” study where one would compare multiple measurement modalities at the same point in time measuring the “same” quantity. Here, BP measurements from the EHR were generally obtained from different points in time as compared with the trial measurements. We are using the mixed models to examine the profile (or mean) SBP across follow-up. Models included random effects for participant and clinic site. Our primary models included a 3-way interaction between treatment group, source, and time since randomization, which was flexibly modeled using B splines. Subgroup analyses were conducted based on groups defined at baseline: age, sex, race/ethnicity, history of cardiovascular disease, kidney function, and clinical site (VA vs other), as well as the number of outpatient BP measurements available from the EHRs.
We used Bland-Altman methods to evaluate the agreement between BP measurements in the outpatient EHR and BP measurements taken during the trial, based on linear mixed models to account for repeated longitudinal measurements.16 For these analyses, we included only BPs recorded 135 days or more after randomization. This time frame was selected to account for variability in the timing of study visits to include BP readings from the visit window for the 6-month study visit (6 months ±45 days) to the end of the study intervention (August 20, 2015). We chose the 6-month study visit as mean SBP (at the group level) was generally stable by that point in follow-up.1 To evaluate within-participant variability, we also evaluated the correlation between the mean difference between SBP in the EHR and SBP in the trial during the window between the 6- and 18-month study visits (≥135 to <545 days after randomization) as compared with between the 18- and 36-month study visits (≥545 to <1136 days after randomization). All modeling was performed using proc glimmix for SAS, version 9.4 (SAS Institute Inc), with graphics generated using the R Statistical Computing Environment, version 3.6.1 (R Foundation for Statistical Computing). All P values were from 2-sided tests and results were deemed statistically significant at P < .05.
Results
Of the 4796 participants enrolled at participating sites, 3074 (64.1%) had at least 3 outpatient BPs available in the EHR and 3 BP readings from the trial (Figure 1). Baseline characteristics for the 3074 participants were similar between the intensive treatment group and standard treatment group (Table 1). The mean (SD) age was 68.5 (9.1) years, 592 participants (19.3%) were female, and 914 (29.7%) were Black, and the mean (SD) SBP was 137.7 (14.9) mm Hg and mean (SD) diastolic BP was 77.3 (11.5) mm Hg. Compared with the trial participants not included in these analyses, those included were less likely to be female and more likely to be White, had lower baseline BP, and were more likely to be treated with a statin or aspirin (eTable in the Supplement).
Figure 1. CONSORT Diagram for the Systolic Blood Pressure Intervention Trial.
BP indicates blood pressure; EHR, electronic health record; and SPB, systolic blood pressure.
Table 1. Baseline Characteristics of Systolic Blood Pressure Intervention Trial Participants Included in the Electronic Health Record Ancillary Study.
| Characteristic | Treatment, No. (%) | |
|---|---|---|
| Intensive (n = 1547) | Standard (n = 1527) | |
| Veterans Affairs clinic site | 929 (60.1) | 919 (60.2) |
| Age, y | ||
| Mean (SD) | 68.5 (9.1) | 68.5 (9.1) |
| ≥75 | 453 (29.3) | 460 (30.1) |
| Female | 306 (19.8) | 286 (18.7) |
| Race/ethnicity | ||
| White | 1011 (65.4) | 994 (65.1) |
| Black | 450 (29.1) | 464 (30.4) |
| Hispanic | 72 (4.7) | 55 (3.6) |
| Other | 14 (0.9) | 14 (0.9) |
| BMI, mean (SD) | 30.2 (5.7) | 30.1 (5.6) |
| History of cardiovascular disease | 367 (23.7) | 369 (24.2) |
| Blood pressure, mean (SD), mm Hg | ||
| Systolic | 137.5 (14.8) | 137.9 (15.0) |
| Diastolic | 77.4 (11.3) | 77.2 (11.7) |
| Orthostatic hypotension | 106/1542 (6.9) | 89/1521 (5.9) |
| Mean (SD) eGFR, mL/min/1.73 m2 | 71.0 (120.0) | 70.7 (19.7) |
| eGFR <60 mL/min/1.73 m2, No./total No. | 474/1543 (30.7) | 454/1520 (29.9) |
| Urine albumin to creatinine ratio, median (IQR), mg/g | 9.6 (5.7-24.2) | 9.5 (5.5-22.5) |
| HDL cholesterol, mean (SD), mg/dL | 51.1 (13.8) | 51.0 (13.6) |
| Triglycerides, median (IQR), mg/dL | 109 (77-149) | 109 (80-156) |
| Glucose, mean (SD), mg/dL | 99.4 (12.6) | 99.3 (12.2) |
| No. of antihypertensive agents, mean (SD) | 1.9 (1.0) | 1.9 (1.0) |
| Use, No./total No. | ||
| Statin | 761/1535 (49.6) | 795/1512 (52.6) |
| Aspirin | 875/1547 (56.6) | 873/1527 (57.2) |
Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); eGFR, estimated glomerular filtration rate based on the Chronic Kidney Disease Epidemiology Collaboration equation; HDL, high-density lipoprotein; IQR, interquartile range.
SI conversion factors: HDL cholesterol to millimoles per liter, multiply by 0.0259; triglycerides to millimoles per liter, multiply by 0.0113; and glucose to millimoles per liter, multiply by 0.0555.
During a median of 3.2 years of follow-up, the median number of outpatient EHR BP measurements was 17 (interquartile range, 9-27), with a median of 16 BP measurements (interquartile range, 14-18 BP measurements) taken as part of the trial. From the time of the visit window for the 6-month study visit (≥135 days after randomization) to the end of study intervention, the mean SBP in the intensive treatment group was 7.3 mm Hg (95% CI, 7.0-7.6 mm Hg) lower at trial visits (120.9 mm Hg) compared with the corresponding outpatient visits (128.2 mm Hg) (Figure 2; eFigure 1 in the Supplement; Table 2). The mean difference was 4.6 mm Hg (95% CI, 4.4-4.9 mm Hg) in the standard treatment group, with a mean SBP at the trial visits of 134.6 and a mean SBP of 139.3 mm Hg at the outpatient visits. Results were generally consistent across subgroups, with 2 exceptions (Table 2). In both treatment groups, the mean difference between outpatient EHR BP and measurements from the trial was larger in women as compared with men. In the intensive treatment group, outpatient EHR SBP was 9.5 mm Hg (95% CI, 8.8-10.1 mm Hg) higher than trial SBP for women, with a smaller mean difference of 6.7 mm Hg (95% CI, 6.4-7.0 mm Hg) for men. There was also a smaller mean difference for participants at VA clinical sites, although this difference was only apparent in the intensive treatment group. We also investigated whether the observed sex differences were associated with VA sites that predominantly included males. When the analysis was restricted to non-VA sites, the mean difference between SBP in the outpatient EHR and trial SBP was consistently smaller for men in both treatment groups (intensive treatment, 6.5 mm Hg [95% CI, 5.9-7.1 mm Hg]; and standard treatment, 3.6 mm Hg [95% CI, 3.1-4.2 mm Hg]) compared with women (intensive treatment, 9.5 mm Hg [95% CI, 8.8-10.2 mm Hg]; standard treatment, 6.2 mm Hg [95% CI, 5.5-6.9 mm Hg]).
Figure 2. Systolic Blood Pressure During Follow-up: Trial Measurements vs Outpatient Blood Pressures Extracted From the Electronic Health Record (EHR).
Estimates based on a linear mixed model with random intercepts for participant and clinic sites. Time since randomization was modeled using B-splines. Shaded areas denote 95% pointwise CIs.
Table 2. SBP Trial Measurements vs Outpatient Blood Pressures Extracted From the EHR, by Subgroups.
| Characteristic | Intensive treatment | Standard treatment | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Mean SBP, mm Hg | Mean difference between trial SBP and EHR SBP (95% CI), mm Hg | P value for interaction | Mean SBP, mm Hg | Mean difference between trial SBP and EHR SBP (95% CI), mm Hg | P value for interaction | |||||
| Trial | Outpatient, EHR | Trial | Outpatient, EHR | |||||||
| Overall | 120.9 | 128.2 | −7.3 (−7.6 to −7.0) | NA | 134.6 | 139.3 | −4.6 (−4.9 to −4.4) | NA | ||
| Age, y | ||||||||||
| <65 | 119.5 | 126.9 | −7.4 (−7.8 to −6.9) | .53 | 134.2 | 138.5 | −4.3 (−4.7 to −3.8) | .006 | ||
| 65 to <80 | 121.4 | 128.8 | −7.4 (−7.8 to −7.0) | 134.9 | 139.5 | −4.6 (−5.0 to −4.2) | ||||
| ≥80 | 122.6 | 129.4 | −6.9 (−7.8 to −6.0) | 134.9 | 140.8 | −5.8 (−6.7 to −4.9) | ||||
| Sex | ||||||||||
| Male | 120.7 | 127.5 | −6.7 (−7.0 to −6.4) | <.001 | 134.6 | 138.9 | −4.2 (−4.5 to −3.9) | <.001 | ||
| Female | 121.3 | 130.8 | −9.5 (−10.1 to −8.8) | 134.7 | 141.0 | −6.2 (−6.9 to −5.5) | ||||
| Race/ethnicity | ||||||||||
| White | 120.6 | 127.9 | −7.3 (−7.6 to −6.9) | .92 | 134.5 | 139.1 | −4.5 (−4.9 to −4.2) | .37 | ||
| Black | 121.7 | 129.1 | −7.4 (−7.9 to −6.8) | 135.2 | 140.1 | −4.9 (−5.4 to −4.4) | ||||
| Hispanic | 118.6 | 126.2 | −7.7 (−8.8 to −6.5) | 132.4 | 136.3 | −3.9 (−5.3 to −2.6) | ||||
| Other | 117.8 | 125.1 | −7.3 (−9.9 to −4.7) | 134.0 | 137.1 | −3.1 (−5.8 to −0.3) | ||||
| History of CVD | ||||||||||
| No | 120.6 | 128.1 | −7.5 (−7.8 to −7.2) | .02 | 134.9 | 139.6 | −4.8 (−5.1 to −4.4) | .21 | ||
| Yes | 121.8 | 128.5 | −6.7 (−7.3 to −6.1) | 133.9 | 138.2 | −4.3 (−4.9 to −3.7) | ||||
| eGFR, mL/min/1.73 m2 | ||||||||||
| ≥60 | 120.1 | 127.5 | −7.4 (−7.7 to −7.1) | .79 | 134.5 | 139.0 | −4.5 (−4.8 to −4.2) | .20 | ||
| <60 | 122.4 | 129.6 | −7.3 (−7.8 to −6.8) | 135.0 | 139.9 | −4.9 (−5.5 to −4.4) | ||||
| Clinical sites | ||||||||||
| Veterans Affairs | 120.6 | 127.5 | −6.9 (−7.2 to −6.5) | <.001 | 134.4 | 138.9 | −4.5 (−4.9 to −4.2) | .42 | ||
| Non–Veterans Affairs | 121.3 | 129.3 | −8.0 (−8.4 to −7.5) | 135.1 | 139.9 | −4.8 (−5.3 to −4.3) | ||||
| No. of outpatient SBP measurements in EHR | ||||||||||
| <17 | 120.9 | 128.5 | −7.6 (−8.0 to −7.2) | .10 | 134.8 | 140.0 | −5.2 (−5.7 to −4.8) | .008 | ||
| ≥17 | 121.0 | 128.1 | −7.2 (−7.5 to −6.8) | 134.7 | 139.0 | −4.4 (−4.7 to −4.0) | ||||
Abbreviations: CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; EHR, electronic health record; NA, not applicable; SBP, systolic blood pressure.
Although there were somewhat consistent group differences comparing SBP in the outpatient EHR with SBP measurements in the trial, the difference in SBP was highly variable. Bland-Altman analyses (Figure 3) generally demonstrated low agreement between BP in the outpatient EHR and trial BP, with wide agreement intervals ranging from approximately −30 to 45 mm Hg in both the intensive treatment and standard treatment groups. The mean difference between SBP in the EHR and trial SBP was between 5 and 15 mm Hg (the range typically cited as the expected “white coat effect” [when office BP is in the hypertensive range while out-of-office BP is not] with improper BP measurement) in only 662 of 1520 participants (43.6%) in the intensive treatment group and only 512 of 1512 participants (33.9%) in the standard treatment group. A total of 657 of 3032 participants (21.7%) had SBPs in the outpatient EHR that were, on average, lower than their trial SBP measurements. To assess whether the mean difference between SBP in the EHR and SBP in the trial was consistent within participants over time, we evaluated the correlation between the mean difference during months 6 to 18 and months 18 to 36. The correlation was low in both the intensive treatment group (Spearman ρ = 0.32) and the standard treatment group (ρ = 0.26) (eFigure 2 in the Supplement). Finally, the difference between routine clinical BPs and trial BPs varied by clinic site (eFigure 3 in the Supplement). In the intensive treatment group, 2 of 46 sites (4.3%) had a median mean SBP difference of 2 mm Hg or less, with 16 of 46 sites (34.8%) having a median mean SBP difference of 5 mm Hg or less. In the standard treatment group, 10 of 46 sites (21.7%) had a median mean SBP difference of 2 mm Hg or less, with 26 of 46 sites (56.5%) having a median mean SBP difference of 5 mm Hg or less.
Figure 3. Bland-Altman Plot Comparing Outpatient Systolic Blood Pressure (SBP) Readings From the Electronic Health Record (EHR) With Trial Measurements.
Size of data points is proportional to the number of outpatient EHR measurements of SBP. Dashed lines indicate 95% limits of agreement. Overall: 1176 (38.8%) mean EHR – trial SBP difference between 5 and 15 mm Hg. Intensive treatment: 663 (43.6%) mean EHR – trial SBP difference between 5 and 15 mm Hg. Standard treatment: 513 (33.9%) mean EHR – trial SBP difference between 5 and 15 mm Hg.
Discussion
Using linked outpatient EHR and trial data, we demonstrated that SBP measurements in the routine clinical setting were, on average, 5 to 8 mm Hg higher than SBP measured at SPRINT trial visits. With a lower target SBP, there was a greater mean difference between SBP measured in the routine clinical setting and SBP measured at trial visits. These results were generally consistent across subgroups. The mean difference in achieved SBP in the intensive and standard groups was similar whether using routine clinic or trial BPs. However, at an individual and clinic level, there was substantial variability in the difference between routine clinic and trial BPs. This discrepancy highlights the importance of considering concordance (differences within individuals) as opposed to population means when assessing BP measurement techniques.
To our knowledge, this is the first large, multicenter study to compare routine clinic BPs with BPs adherent to guideline and/or professional society recommendations. Our results are generally consistent with prior studies that have demonstrated that BPs measured using proper technique are typically 5 to 15 mm Hg lower than the corresponding BPs measured in the routine clinical practice settings.9 Improper BP measurement technique is likely the underlying cause of higher BPs in routine clinical practice compared with trial measurements. Nearly all of the protocol deviations in routine practice (eg, lack of rest period prior to readings, talking during measurement, and arm not supported) are associated with overestimation of BP.
Strengths and Limitations
This study has several strengths compared with prior studies, including a larger sample size, multiple centers, and a lack of referral bias (in many prior studies, the routine clinic BP was obtained at the time of referral for ambulatory BP monitoring or to a hypertension clinic).17,18,19,20 For example, in the study by Graves et al,20 the routine clinic BP was obtained at the time of referral for 6-hour ambulatory BP monitoring to evaluate an elevated casual BP. In addition, the present study assessed clinic and trial BPs over an extended period of time as opposed to onetime measurements in a clinic or study setting. Finally, this is the first study to evaluate the association of different BP targets with the difference between routine clinic BPs and research-quality BPs.
This study also has some limitations. Experts and guideline committees have argued that target BP in routine clinical practice should be higher (<130 or even <140 mm Hg) than the intensive treatment target in SPRINT.7,8 Although we did demonstrate that the mean outpatient SBP in the EHR for participants in the intensive treatment group was 128 mm Hg, the concordance between routine clinical practice BP and trial BP was highly variable. A sizable minority of participants (657 of 3032 [21.7%]) had outpatient SBPs in the EHR that were, on average, lower than their trial measurements. Variability in routine BP measurement practices and within-individual BP variability are potential explanations for why some individuals had higher BPs in routine practice than at SPRINT visits.
The difference between BPs in the EHR and trial BPs was not consistent over time (eFigure 2 in the Supplement); the difference between routine clinical BPs and trial BPs also varied by clinic site (eFigure 3 in the Supplement). Given this heterogeneity between participants, over time within the same participant, and across clinic sites, it is not possible to apply one common correction factor to translate routine clinical BPs to an equivalent AHA-adherent BP. This variability underscores the critical need to use proper BP measurement technique in routine clinical practice, which is particularly relevant because recent guidelines8 recommend lower target BPs than in the past. Our study indicates a potential for greater difference between routine clinic BPs and AHA-adherent BPs at a lower target BP.8,21 A similar trend was noted in the SPRINT Ambulatory Blood Pressure ancillary study in which daytime ambulatory SBP was 7 mm Hg higher than trial SBP in the intensive group but only 3 mm Hg higher in the standard group.22 Difficulty in implementing accurate BP measurement in routine practice may be considered a limitation of SPRINT; however, nearly all the observational studies that identified high BP as a cardiovascular risk factor and clinical trials that inform our current management of hypertension have used recommended approaches for measuring BP.5 Practice of evidence-based medicine requires the use of similar approaches in clinical practice.
To our knowledge, this is the first study linking EHR data collected as part of routine clinical care at multiple sites with data from a large randomized clinical trial. We demonstrate the feasibility of linking patient data with participant data using a secure hashing algorithm. Combining the 2 data sources allowed for analyses not possible within either source alone. Additional planned analyses include evaluation of outpatient acute kidney injury ascertained from EHR data and concordance between EHR-based outcomes and formally adjudicated trial outcomes. These studies will inform the future use of EHR data collected as part of routine care as an adjunct to prospective clinical trials and observational studies.
Our study is limited in that we included only a subset of trial sites and a subset of participants at those sites. Given limited resources, we obtained data from the larger sites and the main affiliated health system at each site. Despite this limitation, we were able to obtain outpatient BPs from the EHR of more than 3200 participants. Another limitation is the variability and unknown technique of BP measurement in the routine clinical setting. Most clinics likely use automated devices, but the type of device, duration of rest, number of readings, setting, patient and cuff positioning, and frequency of staff training are unknown. Some SPRINT sites measured BP while a clinician was present while other sites measured BP without a clinician present. There was no difference in follow-up BP or cardiovascular risk reduction between sites with a clinician present for BP measurement and those without a clinician present.11 Many of our sites were at VA medical centers, so our study had a higher percentage of men compared with the overall SPRINT cohort. We lack data to evaluate potential mechanisms explaining why the difference between routine clinic BP and trial BP is greater in women than men. Whether SBPs were obtained in the inpatient vs outpatient setting was not consistently available from the EHRs, so we used SPRINT adverse event data that captured emergency department visits and hospitalizations to define outpatient time periods. Finally, given the use of real-world data, outpatient BP measurements in the EHR were typically not coincident in time with measurements from the trial. Therefore, the differences we observed combine variability owing to measurement technique as well as within-person temporal variability.
Conclusions
Outpatient BPs from routine clinical practice were higher, on average, compared with BP measurements taken in SPRINT. Although the mean SBP difference was higher in the intensive treatment group compared with standard treatment group and in women compared with men, there was a consistently high degree of heterogeneity comparing outpatient BPs in the EHR with trial BPs. There was significant variability between participants, over time within the same participant, and across clinic sites. These results highlight the importance of proper BP measurement technique and an inability to apply one common correction factor (ie, approximately 10 mm Hg) to estimate research quality BP estimates when BP is not measured appropriately in routine clinical practice.
eTable. Baseline Characteristics of Participants Included in the Electronic Health Record Ancillary Study vs Remaining Trial Participants
eFigure 1. Mean Difference in Systolic Blood Pressure During Follow-up
eFigure 2. Correlation Between Outpatient Electronic Health Record (EHR)-Trial Systolic Blood Pressure Difference at 6-18 Months vs EHR-Trial Systolic Blood Pressure Difference at 18-36 Months
eFigure 3. Site Variability in Systolic Blood Pressure Concordance
References
- 1.Wright JT Jr, Williamson JD, Whelton PK, et al. ; SPRINT Research Group . A randomized trial of intensive versus standard blood-pressure control. N Engl J Med. 2015;373(22):2103-2116. doi: 10.1056/NEJMoa1511939 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Williamson JD, Pajewski NM, Auchus AP, et al. ; SPRINT MIND Investigators for the SPRINT Research Group . Effect of intensive vs standard blood pressure control on probable dementia: a randomized clinical trial. JAMA. 2019;321(6):553-561. doi: 10.1001/jama.2018.21442 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Schiffrin EL, Calhoun DA, Flack JM. SPRINT proves that lower is better for nondiabetic high-risk patients, but at a price. Am J Hypertens. 2016;29(1):2-4. doi: 10.1093/ajh/hpv190 [DOI] [PubMed] [Google Scholar]
- 4.Agarwal R. Hypertension: rest before blood pressure measurement: a lesson from SPRINT. Nat Rev Nephrol. 2016;12(3):127-128. doi: 10.1038/nrneph.2016.2 [DOI] [PubMed] [Google Scholar]
- 5.Drawz PE, Ix JH. BP measurement in clinical practice: time to SPRINT to guideline-recommended protocols. J Am Soc Nephrol. 2018;29(2):383-388. doi: 10.1681/ASN.2017070753 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Muntner P, Shimbo D, Carey RM, et al. Measurement of blood pressure in humans: a scientific statement from the American Heart Association. Hypertension. 2019;73(5):e35-e66. doi: 10.1161/HYP.0000000000000087 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Kjeldsen SE, Mancia G. Unobserved automated office blood pressure measurement in the Systolic Blood Pressure Intervention Trial (SPRINT): systolic blood pressure treatment target remains below 140 mmHg. Eur Heart J Cardiovasc Pharmacother. 2016;2(2):79-80. doi: 10.1093/ehjcvp/pvw002 [DOI] [PubMed] [Google Scholar]
- 8.Whelton PK, Carey RM, Aronow WS, et al. 2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults: a report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Hypertension. 2018;71(6):e13-e115. [DOI] [PubMed] [Google Scholar]
- 9.Drawz P. Clinical implications of different blood pressure measurement techniques. Curr Hypertens Rep. 2017;19(7):54. doi: 10.1007/s11906-017-0751-0 [DOI] [PubMed] [Google Scholar]
- 10.Levey AS, Stevens LA, Schmid CH, et al. ; CKD-EPI (Chronic Kidney Disease Epidemiology Collaboration) . A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150(9):604-612. doi: 10.7326/0003-4819-150-9-200905050-00006 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Johnson KC, Whelton PK, Cushman WC, et al. ; SPRINT Research Group . Blood pressure measurement in SPRINT (Systolic Blood Pressure Intervention Trial). Hypertension. 2018;71(5):848-857. doi: 10.1161/HYPERTENSIONAHA.117.10479 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Rocco MV, Sink KM, Lovato LC, et al. ; SPRINT Research Group . Effects of intensive blood pressure treatment on acute kidney injury events in the Systolic Blood Pressure Intervention Trial (SPRINT). Am J Kidney Dis. 2018;71(3):352-361. doi: 10.1053/j.ajkd.2017.08.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Dworkin MJ. SHA-3 standard: permutation-based hash and extendable-output functions. Published August 4, 2015. Accessed September 4, 2020. https://www.nist.gov/publications/sha-3-standard-permutation-based-hash-and-extendable-output-functions
- 14.Potosky AL, Riley GF, Lubitz JD, Mentnech RM, Kessler LG. Potential for cancer related health services research using a linked Medicare-tumor registry database. Med Care. 1993;31(8):732-748. doi: 10.1097/00005650-199308000-00006 [DOI] [PubMed] [Google Scholar]
- 15.Weiner M, Stump TE, Callahan CM, Lewis JN, McDonald CJ. A practical method of linking data from Medicare claims and a comprehensive electronic medical records system. Int J Med Inform. 2003;71(1):57-69. doi: 10.1016/S1386-5056(03)00089-3 [DOI] [PubMed] [Google Scholar]
- 16.Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res. 1999;8(2):135-160. doi: 10.1177/096228029900800204 [DOI] [PubMed] [Google Scholar]
- 17.Myers MG, Oh PI, Reeves RA, Joyner CD. Prevalence of white coat effect in treated hypertensive patients in the community. Am J Hypertens. 1995;8(6):591-597. doi: 10.1016/0895-7061(95)00049-U [DOI] [PubMed] [Google Scholar]
- 18.Brown MA, Buddle ML, Martin A. Is resistant hypertension really resistant? Am J Hypertens. 2001;14(12):1263-1269. doi: 10.1016/S0895-7061(01)02193-8 [DOI] [PubMed] [Google Scholar]
- 19.Head GA, Mihailidou AS, Duggan KA, et al. ; Ambulatory Blood Pressure Working Group of the High Blood Pressure Research Council of Australia . Definition of ambulatory blood pressure targets for diagnosis and treatment of hypertension in relation to clinic blood pressure: prospective cohort study. BMJ. 2010;340:c1104. doi: 10.1136/bmj.c1104 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Graves JW, Nash C, Burger K, Bailey K, Sheps SG. Clinical decision-making in hypertension using an automated (BpTRU) measurement device. J Hum Hypertens. 2003;17(12):823-827. doi: 10.1038/sj.jhh.1001626 [DOI] [PubMed] [Google Scholar]
- 21.Williams B, Mancia G, Spiering W, et al. ; ESC Scientific Document Group . 2018 ESC/ESH guidelines for the management of arterial hypertension. Eur Heart J. 2018;39(33):3021-3104. doi: 10.1093/eurheartj/ehy339 [DOI] [PubMed] [Google Scholar]
- 22.Drawz PE, Pajewski NM, Bates JT, et al. Effect of intensive versus standard clinic-based hypertension management on ambulatory blood pressure: results from the SPRINT (Systolic Blood Pressure Intervention Trial) ambulatory blood pressure study. Hypertension. 2017;69(1):42-50. doi: 10.1161/HYPERTENSIONAHA.116.08076 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
eTable. Baseline Characteristics of Participants Included in the Electronic Health Record Ancillary Study vs Remaining Trial Participants
eFigure 1. Mean Difference in Systolic Blood Pressure During Follow-up
eFigure 2. Correlation Between Outpatient Electronic Health Record (EHR)-Trial Systolic Blood Pressure Difference at 6-18 Months vs EHR-Trial Systolic Blood Pressure Difference at 18-36 Months
eFigure 3. Site Variability in Systolic Blood Pressure Concordance