Age‐Specific Associations of Usual Blood Pressure Variability With Cardiovascular Disease and Mortality: 10‐Year Diabetes Mellitus Cohort Study

Background The detrimental effects of increased variability in systolic blood pressure (SBP) on cardiovascular disease (CVD) and mortality risk in patients with diabetes mellitus remains unclear. This study evaluated age‐specific association of usual SBP visit‐to‐visit variability with CVD and mortality in patients with type 2 diabetes mellitus. Methods and Results A retrospective cohort study investigated 155 982 patients with diabetes mellitus aged 45 to 84 years without CVD at baseline (2008–2010). Usual SBP variability was estimated using SBP SD obtained from a mixed‐effects model. Age‐specific associations (45–54, 55–64, 65–74, 75–84 years) between usual SBP variability, CVD, and mortality risk were assessed by Cox regression adjusted for patient characteristics. After a median follow‐up of 9.7 years, 49 816 events (including 34 039 CVD events and 29 211 mortalities) were identified. Elevated SBP variability was independently, positively, and log‐linearly associated with higher CVD and mortality risk among all age groups, with no evidence of any threshold effects. The excess CVD and mortality risk per 5 mm Hg increase in SBP variability within the 45 to 54 age group is >3 times higher than the 70 to 79 age group (hazard ratio, 1.66; 95% CI, 1.49–1.85 versus hazard ratio, 1.19; 95% CI, 1.15–1.23). The significant associations remained consistent among all subgroups. Patients with younger age had a higher association of SBP variability with event outcomes. Conclusions The findings suggest that SBP visit‐to‐visit variability was strongly associated with CVD and mortality with no evidence of a threshold effect in a population with diabetes mellitus. As well as controlling overall blood pressure levels, SBP visit‐to‐visit variability should be monitored and evaluated in routine practice, in particular for younger patients.


Supplemental Methods
Usual variability SBP using mixed effects model Given a dataset with N individuals and ni SBP measurements from the i th individual, i = 1, … ,N, let Yi j, j = 1, … , ni,be the j th measurement of individual i taken at measurement time tij.
Consider a standard linear mixed effects model where Xi j is a covariate vector for the fixed effects and Zi j is a covariate vector for the random effects bi, assumed normally distributed bi ∼ N(0, Σb). The residual errors i j a re a ssumed independent and normally distributed, i j ∼ N(0, 2 ). We can allow variability in the repeated measurements to differ between individuals by replacing the residual SD with an individualspecific residual SD a nd a ssuming t hat t he are randomly distributed. We assume a lognormal distribution for the residual SD distribution, ensuring positivity of the SDs, ∼ logN( , 2 ). T he choi ce of l og-normal dist ribution also allo ws a na tu ral exte nsion of t he model to incorporate correlation between the usual level and the residual SD by assuming a multivariate normal distribut ion for the random effects and log residual SD where Σb is a vector of covariances between the random effects and the random residual errors. The details of the models have been described elsewhere in the literature [12]. For this study, the model was For the Bayesian estimation, we used diffuse uniform prior distributions U[0, 100] for SDs, uniform U[−1, 1] prior distributions for correlation parameters, and diffuse normal prior distributions N(0, 100 2 ) for all other parameters. Priors were specified for the bivariate and trivariate normal distributions by expressing them as two and three conditional univariate normal distributions, respectively. In the current study, we used a burn-in of 1000 Markov Chain Monte Carlo updates for the mixed effects models. The posterior means (95% credible interval) for usual SBP and SBP variability were 137.1 (112.9, 160.3) and 12.7 (7.1, 20.9), respectively.

Rosner's regression method
Rosner's regression method is used to calculate the regression dilution ratio to evaluate the association between systolic blood pressure (SBP) variability and event outcomes in sensitivity analysis 1. The detailed method is shown as below.
First, the standard deviation (SD) of SBP as the variability measurement was calculated based on the measurements of SBP at 9 visits (baseline, every 3 months in the past). To deal with random errors in this SBP variability measurement, regression dilution ratio was applied to the analyses based on Rosner's regression method [34], employing the measurements of SBP at 8 visits unit 2 years after baseline (every 3 months after baseline). The timeline for the measurement of SBP and outcome ascertainment of this method shown in the Figure below.
Regression dilution ratio was calculated as the coefficient relating to the post measurement to the baseline measurement. Finally, continuous hazard ratios for baseline measurement were multiplied by the ratio to estimate the association for SBP variability. For example, regression dilution ratio of 4 for SD of SBP. If an hazard ratio for baseline SD of SBP of 1.1 was calculated, the hazard ratio for SBP was calculated as e (4 × ln(1.1)) = 1.5.

Figure.
Study design for sensitivity analysis 1.
In current study, the mean of SD of SBP at post SD of SBP was 12.1mmHg (SD: 4.8mmHg). Applying the Rosner's regression method, the regression dilution ratio was 4.13. To evaluate the association between SBP variability and event outcomes, multivariable Cox proportional hazards regressions adjusted with baseline characteristics, usual SBP and regression dilution ratio. The results were shown in Figure S5.    Figure S1. Study design for the investigation of the association visit-to-visit in systolic blood pressure (SBP) and cardiovascular diseases and all-cause mortality. The measurements of SBP at 9 visits (baseline, each 3 months in the past) were used to calculate usual mean and variability of SBP. The median follow-up period was 9.7 years after baseline Usual SBP variability (mmHg) All composite events Figure S2. Adjusted Hazard ratio for incidence of CVD, CHD, Stroke, Heart failure, all−cause mortality, CVD mortality, non−CVD mortality and their composite with increasing usual SBP variability by multivariable Cox regressions. Figure S3. Adjusted Hazard ratio for incidence of CVD, CHD, Stroke, Heart failure, all-cause mortality, CVD mortality, non-CVD mortality and their composite with increasing usual SBP variability by multivariable Cox regressions with restricted cubic spline.
Hazard ratio was adjusted by age at risk, sex, smoking status, body mass index, SBP, diastolic blood pressure, haemoglobin A1c, low-density lipoproteincholesterol, estimated glomerular filtration rate, the usages of oral anti-diabetic drugs, insulin, angiotensin converting enzyme inhibitor/angiotensin receptor blocker, β-blocker, calcium channel blocker, diuretic, other anti-hypertensive drugs, lipid-lowering agent, Charlson's index and usual SBP. Shaded region represents 95% confidence intervals. SBP=Systolic blood pressure; CVD=Cardiovascular disease; CHD=Coronary heart disease;  . Age-specific adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, CVD mortality and non-CVD mortality with increasing usual SBP variability by multivariable Cox re-gressions.
HR was adjusted by age at risk, sex, smoking status, body mass index, SBP, diastolic blood pressure, haemoglobin A1c, low-density lipoprotein-cholesterol, estimated glomerular filtration rate, the usages of oral antidiabetic drugs, insulin, angiotensin converting enzyme inhibitor/angiotensin receptor blocker,  Figure S5. Adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, all cause mortality, CVD mortality, non-CVD mortality and their composite with each 5mmHg increasing usual SBP variability using multivariable Cox regressions based on Rosner's regression method (Sensitivity analysis 1).
Applying the Rosner's regression method, the regression dilution ratio was 4.13. HR was adjusted by age at risk, gender, smoking status, body mass index, SBP, diastolic blood pressure, haemoglobin A1c, low-density lipoprotein-cholesterol, estimated glomerular filtration rate, the usages of oral anti-diabetic drugs, insulin, angiotensin converting enzyme inhibitor/angiotensin receptor blocker, β-blocker, calcium channel blocker, diuretic, other anti-hypertensive drugs, lipid-lowering agent, Charlson's index, usual SBP and regression dilution ratio. SD=Standard deviation; SBP=Systolic blood pressure; CVD=Cardiovascular disease; HR=Hazard ratio.

Figure S6. Adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, all cause mortality, CVD mortality, non-CVD mortality and their composite with each 5mmHg increasing usual SBP variability using multivariable Cox regressions based on complete case analysis (Sensitivity analysis 2).
HR was adjusted by age at risk, sex, smoking status, body mass index, SBP, diastolic blood pressure, haemoglobin A1c, low-density lipoprotein-cholesterol, estimated glomerular filtration rate, the usages of oral anti-diabetic drugs, insulin, angiotensin converting enzyme inhibitor/angiotensin receptor blocker, β-blocker, calcium channel blocker, diuretic, other anti-hypertensive drugs, lipid-lowering agent, Charlson's index and usual SBP. SD=Standard deviation; SBP=Systolic blood pressure; CVD=Cardiovascular disease; HR=Hazard ratio.  Figure S7. Adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, all cause mortality, CVD mortality, non-CVD mortality and their composite with each 5mmHg increasing usual SBP variability using multivariable Cox regressions based on patients with at least 12 months follow-up period (Sensitivity analysis 3).
HR was adjusted by age at risk, sex, smoking status, body mass index, SBP, diastolic blood pressure, haemoglobin A1c, low-density lipoprotein-cholesterol, estimated glomerular filtration rate, the usages of oral anti-diabetic drugs, insulin, angiotensin converting enzyme inhibitor/angiotensin receptor blocker, β-blocker, calcium channel blocker, diuretic, other anti-hypertensive drugs, lipid-lowering agent, Charlson's index and usual SBP. SD=Standard deviation; SBP=Systolic blood pressure; CVD=Cardiovascular disease; HR=Hazard ratio.  Figure S8. Adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, all cause mortality, CVD mortality, non-CVD mortality and their composite with each 5mmHg increasing usual SBP variability using multivariable Cox regressions based on patients with at least 2 SBP measurements on or before baseline (Sensitivity analysis 4).
HR was adjusted by age at risk, sex, smoking status, body mass index, SBP, diastolic blood pressure, haemoglobin A1c, low-density lipoprotein-cholesterol, estimated glomerular filtration rate, the usages of oral anti-diabetic drugs, insulin, angiotensin converting enzyme inhibitor/angiotensin receptor blocker, β-blocker, calcium channel blocker, diuretic, other anti-hypertensive drugs, lipid-lowering agent, Charlson's index and usual SBP. SD=Standard deviation; SBP=Systolic blood pressure; CVD=Cardiovascular disease; HR=Hazard ratio.  Figure S9. Adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, all cause mortality, CVD mortality, non-CVD mortality and their composite with each 5mmHg increasing usual SBP variability using multivariable Cox regressions based on patients with at least 7 SBP measurements on or before baseline (Sensitivity analysis 4).

49,816
Figure S10. Adjusted hazard ratios for the risk of CVD, coronary heart disease, stroke, heart failure, all cause mortality, CVD mortality, non-CVD mortality and their composite with each 5mmHg increasing usual SBP variability using multivariable Cox regressions with additional adjustment for patients with the usage of aspirin on or before baseline (Sensitivity analysis 5).