Regression to the mean in home blood pressure: Analyses of the BP GUIDE study

Abstract

The aim of our study was to estimate the size of regression to the mean with home blood pressure (BP) monitoring and compare with that for office BP. Office and home BP measures were obtained from the BP GUIDE (value of central Blood Pressure for GUIDing managEment for hypertension) study, in which 286 patients had BP measured every 3 months for 12 months. Patients were categorized by 10 mm Hg strata of baseline BP, and regression to the mean measures was calculated for home and office BP. High baseline home BP readings tended to be lower on long‐term follow‐up, and low baseline readings tended to be higher. For example, patients in the group with mean baseline home systolic BP ≥ 150 mm Hg had a mean baseline systolic BP of 156 mm Hg, which fell to 143 mm Hg at 12 months; and patients in the group with mean baseline home systolic BP < 120 mm Hg had a mean baseline systolic BP of 113 mm Hg which rose to 120 mm Hg at 12 months. Similar patterns were seen in intervention and control groups, and for diastolic BP. The regression dilution ratio for home systolic BP and diastolic BP was 0.52 and 0.64, respectively, compared to 0.40 and 0.55 for office systolic BP and diastolic BP, respectively. Home BP is subject to regression to the mean to a similar degree as office BP. These findings have implications for the diagnosis and management of hypertension using home BP.

1. INTRODUCTION

Out of office BP monitoring with home and/or ambulatory measurements is now recommended in many international guidelines. ¹ , ² , ³ , ⁴ For example, the recent European Society of Cardiology (ESC) and European Society of Hypertension (ESH) guidelines recommend use of home BP over office BP because of greater reproducibility and correlation with hypertension‐mediated organ damage. ² Home BP has also been suggested as an important adjunct when there is suspicion of white coat or masked hypertension. ² Home BP monitoring values are a stronger predictor of cardiovascular risk than office BP and can lead to significantly improved BP control with appropriate titration of medical therapy. ⁵ , ⁶ , ⁷

BP has both short‐term fluctuations within a 24‐hour period and long‐term fluctuations over periods of days, weeks, seasons, and even years. ⁸ Prior studies have shown that office BP exhibits regression to the mean, where long‐term BP readings are less extreme than baseline values. ⁹ , ¹⁰ , ¹¹ Recently, ambulatory BP has also been shown to be subject to regression to the mean. ¹² This phenomenon has important implications for the diagnosis and management of hypertension. However, the degree of regression to the mean with home BP monitoring is uncertain. Therefore, our study aims to determine whether regression to the mean occurs with home BP monitoring, and to compare these effects to office BP monitoring.

2. METHODS

2.1. Study setting

This study is a post hoc analysis of the value for central Blood Pressure for GUIDing managEment for hypertension (BP GUIDE) study where home BP was measured according to a standardized protocol among 286 participants at baseline and each 3 months over 12 months (five visits). The study protocol and principal findings of BP GUIDE have been previously published. ¹³ , ¹⁴ In summary, BP GUIDE was a multicenter, randomized controlled trial that randomized patients with hypertension to usual care (control) or with addition of central BP measures to guide clinical decisions (intervention). Patients in the usual care group had BP management according to best practice. The intervention group received additional central aortic BP monitoring to guide care and had medications titrated to normalize seated central SBP. All participants underwent a variety of tests including office BP, 7‐day home BP, and other clinical tests as per absolute cardiovascular risk assessment.

Seven‐day home BP was recorded using a valid device owned by the patient or with an A&D UA 767 machine provided (A&D Mercury, Thebarton, SA, Australia). ¹⁵ Participants were asked to take duplicate BP readings 3 × per day (1, morning 6 am‐10 am; 2, midday; 3, evening 6 pm‐10 pm) after 5 minutes of seated rest, with the first reading discarded and the second reading recorded. Office brachial BP (Omron HEM 907; Omron Healthcare) was recorded in duplicate after ≈5 minutes of seated rest. The seven‐day home and office BP measurements were repeated at baseline and every 3 months for the 12‐month study period.

2.2. Participants

Participants with uncomplicated hypertension were recruited in three Australian primary care clinics (Brisbane, Queensland; Hobart, Tasmania, and Canberra, Australian Capital Territory). Inclusion criteria were as follows: 18‐75 years of age, non‐pregnant, receiving antihypertensive therapy for uncomplicated essential hypertension, and taking at least one but not more than three antihypertensive drugs (to exclude complicated/resistant hypertension). Exclusion criteria were as follows: severely abnormal LV mass index (>59 g/m² in women and >64g/m² in men), clinical history of coronary artery disease or renal disease, serum creatinine > 1.6 mg/dL, brachial BP > 180/100 mm Hg, aortic valve stenosis, or upper limb obstructive atherosclerosis.

Definitions for hypertension were SBP ≥ 140 mm Hg and/or DBP ≥ 90 mm Hg for office BP recordings and SBP ≥ 135 or DBP ≥ 85 mm Hg for home BP readings. ² , ¹⁶ White coat hypertension was defined as hypertension with office BP but not home BP measurements, whilst masked hypertension was defined as hypertension with home but not office BP measurements. Medication quantity was calculated by the daily defined dose as per World Health Organization. The daily defined dose allows for exact quantification of drug dose standardized across drug classes. ¹⁷ , ¹⁸

2.3. Statistical analysis

Data are reported as either mean and standard deviation or counts and percentages. Baseline and follow‐up mean office and home BP readings were calculated overall and for groups defined by the baseline SBP and DBP. The regression to the mean and dilution ratios was calculated using the baseline and 12‐month follow‐up BP readings. Regression to the mean and dilution ratios was estimated using the MacMahon method ⁹ , ¹⁹ and Rosner method. ²⁰ In the MacMahon method, regression dilution ratios were calculated by categorizing patients according 10 mm Hg strata of baseline BP, and using the range between means of highest and lowest baseline BP groups at 12 months divided by the range between means of highest and lowest baseline BP groups at baseline. The Rosner method is calculated as the slope of the regression line for 12‐month BP versus baseline BP. A comparison between regression dilution ratios was made between office and home BP through inclusion of an interaction term in the linear regression. Consistency between baseline and 12‐month follow‐up BP was assessed using single measures intraclass correlation coefficients (two‐way mixed with absolute agreement). Chi‐squared test was used to compare the proportion of hypertensive patients at baseline compared to 12‐month follow‐up. Separate analyses were performed for hypertensive and non‐hypertensive individuals. Sensitivity analyses were performed to assess the degree of regression to the mean in patients randomized to the intervention and usual care groups. Missing values were excluded on a pairwise basis such that all available BP measurements were included for each separate analysis (home SBP, home DBP, office SBP, office DBP). We also report the total number of patients that contributed to each BP measurement. All statistical analyses were performed with SPSS software version 23 (SPSS Inc).

3. RESULTS

3.1. Study population

The study population consisted of 286 patients, of whom 147 were randomized to the intervention and 139 to usual care (Table S1). The mean age was 64 ± 8 years, and 135 (47%) were female. The mean office SBP and DBP at baseline was 127 ± 14 mm Hg and 75 ± 10 mm Hg, respectively. The mean home SBP and DBP at baseline was 128 ± 13 and 74 ± 8 mm Hg, respectively. Most patients had home BP measurements at all follow‐up visits (93% with all SBP and 90% with all DBP). Similarly, most patients had clinic BP measurements at all follow‐up visits (91% with all SBP and 91% with all DBP). The absolute difference between 12‐month and baseline BP readings was SBP 8 ± 6 mm Hg and DBP 4 ± 4 mm Hg for home monitoring and SBP 11 ± 8 mm Hg and DBP 7 ± 6 mm Hg for office monitoring, respectively.

3.2. Regression to the mean

Figure 1 shows the mean home BP readings for patients stratified by BP groups. Patients with lower SBP readings at baseline had higher home SBP readings at 12 months. For example, patients with SBP < 120 mm Hg had a mean SBP of 113 mm Hg at baseline compared to 120 mm Hg at 12 months. Conversely, patients in the home SBP ≥ 150 mm Hg group had a mean baseline SBP of 156 mm Hg compared to 143 mm Hg at 12 months. The regression dilution ratio for home SBP was 0.52. Similarly, regression to the mean was observed with home DBP measurements, although to a lesser extent (regression dilution ratio 0.64). We observed a similar pattern for office BP readings for both SBP and DBP (Figure 1). The regression dilution ratio for office SBP was 0.43 and for office DBP was 0.51.

FIGURE 1.

Regression to the mean for home and office BP measurements. A, Home systolic, B, home diastolic, C, office systolic, and D, office diastolic BP measurements. SBP levels are stratified at baseline and for each follow‐up visit according to the baseline SBP levels < 120, 120‐129, 130‐139, 140‐149, and ≥150 mm Hg. DBP levels are stratified at baseline and for each follow‐up visit according to the baseline DBP levels < 70, 70‐79, 80‐89, and ≥90 mm Hg

Most of the regression occurred within the first 3 months. For example, patients with home SBP ≥ 150 mm Hg at baseline had a mean decrease in home SBP of 11 ± 9 mm Hg by 3 months, compared to a further 2 ± 12 mm Hg decrease in SBP from months 3 to 12. Similar regressions were seen for home DBP, clinic SBP, and clinic DBP (Table S2). Regression to the mean was not a phenomenon restricted to baseline visit. For any visit, people with extreme BP values tended to have less extreme values at past and future timepoints. For example, if patients were grouped by 12‐month BP level (rather than baseline BP level), those with the highest and lowest BP levels at that timepoint had considerably less extreme BP levels at previous visits (Figure 2).

FIGURE 2.

“Reverse” regression to the mean, with groups classified by twelve‐month blood pressure categories. A, Home systolic, B, home diastolic, C, office systolic, and D, office diastolic blood pressure measurements. SBP levels are stratified at baseline and for each follow‐up visit according to the 12‐month SBP levels < 120, 120‐129, 130‐139, 140‐149, and ≥150 mm Hg. DBP levels are stratified at baseline and for each follow‐up visit according to the 12‐mo DBP levels < 70, 70‐79, 80‐89, and ≥90 mm Hg

Figure 3 shows the correlation between baseline and 12‐month BP measurements for both home and office recordings. Office BP exhibited greater regression towards the mean compared to home BP (Figure 3). The regression dilution ratios calculated by the Rosner method for office and home SBP were 0.41 and 0.55, respectively (P = .024), and for office and home DBP were 0.50 and 0.75, respectively (P < .001). The interclass correlation coefficient was high between baseline and 12‐month BP readings (Figure 3). The slope of a simple linear regression line was less than 45 degrees for both home and office SBP/DBP measurements, suggesting low baseline measures tended to increase at 12 months, while high baseline measures tended to decrease at 12 months.

FIGURE 3.

Correlation between baseline and 12‐mo follow‐up home and office blood pressure. Home (A, B) and office (C, D) blood pressure measures. The solid line represents a linear regression trend line, and the slope of this line is indicated—the lower the value, the greater the degree of regression to the mean. The dotted line indicates the line of identity, that is, the slope of 1.0. ICC, Intra‐class correlation coefficient

3.3. Impact of regression to the mean on hypertension category

Table 1 shows the mean BP levels at baseline and 12 months for patients according to hypertension category for office and home recordings. There was a significant change in the proportion of patients diagnosed as hypertensive on home and/or clinic BP because of regression to the mean. At baseline, there were 16 patients classified as hypertensive according to both home and office BP. These patients exhibited an average home and office BP decreases of 5/3 mm Hg and 12/7 mm Hg, respectively. Compared to 16 patients at baseline, only 2 (13%) patients remained categorized as hypertensive according to both home and office BP measurements at 12 months (P < .001). Across all visits, there were a similar number of patients with concordant hypertension with both home and office BP at baseline (6%), 3 months (7%), 6 months (5%), 9 months (7%), and 12 months (6%). However, this comprised markedly different people at each visit. There were 55 (19%) patients with concordant hypertension at one or more visit, 19 (7%) patients having concordant high BP at two or more visits, and no patients having concordant high BP at all five visits.

TABLE 1.

Mean BP levels at baseline and 12‐mo follow‐up defined by baseline office and home BP levels

Baseline BP category			Classification at baseline	Classification at baseline and 12 mo	Mean office BP		Mean home BP
	Office BP ≥140/90 mm Hg	Home BP ≥135/85 mm Hg	N	N (%)	Baseline	12 mo	Baseline	12 mo
Concordant hypertension	Yes	Yes	16	2 (13%)	149/83	136/76	147/83	142/80
White coat hypertension	Yes	No	38	7 (18%)	145/86	130/78	120/71	122/72
Masked hypertension	No	Yes	54	21 (39%)	124/74	127/74	143/79	137/76
Concordant normotension	No	No	166	110 (66%)	122/72	124/73	123/72	127/74

Abbreviations: BP, blood pressure; N, number.

Thirty‐eight patients were categorized as white coat hypertension at baseline, who exhibited an average office BP decrease of 14/9 mm Hg but a small home BP increase of 2/1 mm Hg. Compared to 38 patients at baseline, 7 (18%) patients would have been defined as white coat hypertension at 12 months (P < .001). Although a similar proportion of patients had white coat hypertension during each visit (7%‐14%), this comprised different people at each visit. Across the five visits, 86 (30%) patients had white coat hypertension during one or more visits, 34 (12%) patients in two or more visits, and one patient at all five visits.

Among the 54 patients with apparent masked hypertension at baseline, there was an average home SBP/DBP reduction of 6/3 mm Hg but a small increase in office SBP/DBP of 2/0 mm Hg. Compared to 54 patients at baseline, only 21 (39%) patients had masked hypertension at 12 months (P < .001). Across the five visits, 148 (52%) patients had masked hypertension at one or more visit, 94 (33%) patients had masked hypertension at two or more visits, and 8 (3%) patients had masked hypertension at all five visits.

There were 166 patients who were normotensive according to both home and office BP recordings at baseline. These patients had an average increase in home SBP/DBP of 4/2 mm Hg and average increase in office SBP/DBP of 1/1 mm Hg. At 12 months, 110 (66%) patients remained normotensive at follow‐up (P < .001). Across the five visits, 235 (82%) patients had concordant normotension on both home and office BP at one or more visits, 197 (69%) patients at two or more visits, and 60 (21%) patients at all five visits.

3.4. Sensitivity analyses

Sensitivity analyses showed similar regressions to the mean for home BP among patients randomized to intervention versus usual care (Figure S1). Regression dilution ratios for the intervention group were 0.40 for home SBP and 0.67 for home DBP compared to 0.

65 for home SBP and 0.64 for home DBP in the usual care group. There was significant reclassification of hypertension (concordant, white coat, masked, or nomotension) in both the intervention and usual care groups (Tables S3 and S4).

The mean daily defined doses do not explain the regressions to the mean seen for home and office BP (Figure S2). For example, for patients with home SBP ≥ 150 mm Hg at baseline, there was an 11 mm Hg fall in SBP between months 0 and 3 and only a 2.4 mm Hg fall between months 3 and 12, while sum of daily doses prescribed were 2.1, 2.2, and 2.5 at months 0, 3, and 12, respectively.

4. DISCUSSION

Although prior studies have demonstrated regression to the mean for home BP ²¹ , ²² , ²³ and ambulatory BP, ¹² this is the first analysis to quantify the degree of regression for home BP over a prolonged follow‐up. These findings demonstrate that regression to the mean occurs in both home and office BP readings but occurs to a smaller degree in home BP compared to office BP monitoring. This has important implications regarding the use of home BP monitoring for the diagnosis and management of hypertension.

There is general consensus from current guidelines that home BP monitoring is a more reproducible measurement compared to office BP and is more closely related to hypertension mediated organ damage and better predicts cardiovascular risk. ² , ¹⁶ However, the findings of the current study indicate that regression to the mean still exists for home BP and home BP measurements over a seven‐day period do not necessarily reflect BP recordings in the preceding 12 months, nor predict future BP recordings in the following 12 months. In the present analyses, 72% of patients who were diagnosed as hypertensive on office and/or home BP monitoring were recategorized as normotensive at 12 months. Conversely, 34% of patients who were categorized as normotensive on both office and home BP readings at baseline were recategorized as hypertensive at 12 months. Many patients who are diagnosed as normotensive may in fact have high BP on follow‐up and therefore benefit from antihypertensive therapy. ²⁴ In patients with hypertension who are initiated on antihypertensive therapy, some of the decrease in home BP will be due to regression to the mean. This could lead to an overestimation of the efficacy of BP lowering therapy and reassurance about medication compliance. Failure to correct for regression to the mean in any risk factor will lead to underestimation of the true association of that risk factor with disease, in direct proportion to the amount of regression to the mean. ²⁵

Practically, there is a need for clinicians to account for regression the mean in the diagnosis and management of hypertension. Given regression to the mean is the statistical phenomenon where repeated measurements are observed within a degree of random error around the true mean, ²⁶ one would expect a reduction in the variability by averaging repeated measurements. This is consistent with the smaller degree of regression to the mean we observed with home BP compared to clinic BP monitoring, since home BP monitoring is calculated from measurements across 7 days. As per the current BP guidelines, clinicians should be encouraged to use home or ambulatory BP measurements when available, to minimize the impact of regression to the mean. ² , ¹⁶ However, since regression to the mean is still evident in both home and ambulatory BP, there is a need to account for this expected variability. ¹² This can be achieved by subtracting the observed BP change from the expected degree of regression obtained from published studies of similar populations of interest. For example, a patient with a home SBP of 113 mm Hg (mean home SBP in the <120 mm Hg group) may expect an increase in SBP of 5.4 mm Hg by 3 months. Any additional increase in SBP above 5.4 mm Hg is less likely to be accounted for by regression to the mean alone.

These results should be interpreted in the context of the study's strengths and weaknesses of the analysis. By stratifying the baseline BP values over a wide range, we were able to show a consistent relationship between an increase in BP over time in patients in the lower BP groups, and a decrease in BP in patients in the higher BP groups. We were also able to quantify the degree of BP change in both the office and home BP groups, providing regression dilution ratios, which measure the change in spread of values, that is, the degree to which frequency distributions of average values are narrower than those of baseline values. Since home BP has a narrower frequency distribution than office BP, the absolute change in BP due to regression to the mean will be less with home BP than office BP measurement, and so regression dilution ratios offer greater comparability between measures. The study protocol also consisted of repeated measures of home BP at three‐month intervals, which shows that regression to the mean is observed across a one‐year interval, although most of the regression occurs within the first 3 months. Given the home BP device recorded the measurements automatically, this reduced the likelihood of reporting bias. The main limitation of this study was the relatively small sample size, particularly evident in the small numbers of patients with concordant hypertension. More large‐scale studies are needed to quantify the degree of expected regression to the mean in different patient subgroups.

Some of the changes in BP during follow‐up may be due to other factors besides regression to the mean. We also acknowledge that these findings are observational, particularly the relationships between mean BP and daily defined doses of antihypertensives. However, it is unlikely that changes in antihypertensive treatment explained much of the changes in BP. Firstly, there is little correlation between the changes in BP and daily defined doses at each visit. For example, most of the regression in BP occurred within the first 3 months, whereas the medication dose changes were not particularly greater, nor restricted to the first 3 months. Secondly, the same pattern of regression existed for both SBP and DBP, whilst medication dose changes were related to SBP only. Thirdly, the intervention group, which received central BP monitoring, had much larger changes in the daily defined dose during follow‐up compared to usual care. However, regression to the mean existed in both the intervention and usual care groups (Figure S1).

In conclusion, we found that home BP exhibits considerable regression to the mean and classifications of hypertension based on one‐off measures of office and home BP have a high probability of changing over time. This suggests the need for repeated home BP measurements, particularly during the first 3 months, where the greatest degree of regression tends to occur. There should also be more reliance given to calculated absolute cardiovascular risk in making treatment decisions, since this directly determines the benefits of BP lowering ²⁷ and is not substantially affected by regression to the mean. ²⁸ These findings also indicate that pre‐ and post‐measures of home BP may not be reliable estimates of treatment efficacy, since some of the BP reductions following antihypertensive therapy initiation will be due to regression to the mean. Therefore, treatment regimens should be chosen more empirically. Finally, there will be significant underestimation of home BP and risk of cardiovascular disease if analyses are not adjusted for regression to the mean. ²⁵

CONFLICT OF INTEREST

The authors have no conflicts of interest to disclose.

AUTHOR CONTRIBUTIONS

AR involved in the conception and design of the study, and administrative support. JS involved in the provision of study materials or patients, and collection and assembly of data. AS, MN, ER, and NW involved in the data analysis and interpretation of the manuscript. NW, EA, AS, MN, JS, and AR wrote the manuscript. All authors finally approved the manuscript.

Supporting information

Appendix S1

Wang N, Atkins ER, Salam A, Moore MN, Sharman JE, Rodgers A. Regression to the mean in home blood pressure: Analyses of the BP GUIDE study. J Clin Hypertens. 2020;22:1184–1191. 10.1111/jch.13933