Abstract
In a multiethnic cohort of older people with diabetes (n=1178), we assessed whether ambulatory blood pressure (BP) monitoring improves prediction of all-cause mortality and cardiovascular mortality when added to baseline covariates, including office BP and heart rate (HR). Secondary analyses assessed whether albuminuria may mediate the association of pulse pressure with mortality. The ambulatory arterial stiffness index was calculated as “1-slope” from the within-person regression of diastolic-on-systolic ambulatory BP readings. Mean follow-up was 6.6±0.4 years. There were 287 deaths; death certificates were available for 215 deaths (75%), and 110 of them were deemed of cardiovascular cause. Cox models were built incrementally. First, models using clinical and laboratory variables selected albuminuria and office HRs as independent predictors of all-cause and cardiovascular mortality. When ambulatory monitoring data were added, sleep:wake HR ratio and ambulatory arterial stiffness index added significantly to the prediction of all-cause mortality, but only sleep:wake HR ratio added to the prediction of cardiovascular mortality. Office HR and albuminuria retained significance as predictors of both types of mortality. Secondary analyses without adjustment for albuminuria confirmed the predictive value of office HR and sleep/wake HR, whereas 24-hour pulse pressure and sleep systolic BP were also independently predictive of all-cause and cardiovascular mortality, respectively. In conclusion, office HR and albuminuria were strong predictors of mortality. Ambulatory monitoring improved the prediction of risk through its assessment of sleep HR dipping and of ambulatory arterial stiffness index, a measure of the dynamic relationship between systolic and diastolic BPs. Albuminuria may mediate the association between BP and mortality.
Keywords: mortality, ambulatory blood pressure, diabetes mellitus, arterial stiffness
People with diabetes mellitus exhibit a consistently higher mortality risk, as compared with people without diabetes mellitus, and this excess mortality is particularly important in the elderly.1 Thus, there is a strong interest in identifying markers of higher mortality risk in elderly people with diabetes mellitus and determining whether those markers add to well-characterized predictors, such as albuminuria2 and a history of congestive heart failure (CHF).3
Since the publication of the study by Perloff et al in 1983,4 ambulatory blood pressure (BP) monitoring (ABPM) has been shown in several studies to be superior to office BP measurements in predicting mortality in different patient populations. Over the years, investigators have identified a number of BP variables measured by ABPM that provide valuable information when assessing mortality risks. Those variables include mean 24-hour systolic and diastolic BPs, 24-hour pulse pressure (PP), abnormal sleep dipping patterns, and the ambulatory arterial stiffness index (AASI). In addition, ABPM allows an evaluation of heart rate (HR) over 24 hours, including an assessment of sleep dipping patterns, and the 24-hour HR data may also add to the predictive information provided by office HR measurements.5
There is a significant gap in the literature with regard to the value of ABPM to improve the prediction of mortality in people with diabetes mellitus. There have been only 4 prospective studies; they were limited to white and Japanese populations and had significant methodologic limitations.6–9 Three studies had a relatively small sample size and a small number of events, limiting the reproducibility of the multivariate analyses.6,8,9 In addition, 2 studies failed to include albuminuria in their predictive models,7,9 and 1 of those 2 assessed ABPM performed in hospitalized patients, thereby causing a selection bias and reducing the applicability of its findings.7 Finally, none of them assessed the predictive value of the AASI or the HR dipping pattern during sleep, both of which have been reported to be independent predictors of mortality in people without diabetes mellitus.10–13
There is widespread consensus about the predictive value of both albuminuria and PP in the elderly. Higher PP is also an independent predictor of the progression of albuminuria in elderly people with diabetes mellitus, and the association is stronger for 24-hour (as compared with office) PP.14 However, to the best of our knowledge, data examining the hypothesis that albuminuria may mediate the association between PP and mortality in elderly people with diabetes mellitus have not been reported.
We studied prospectively a large (n=1178), multiethnic sample of elderly people with diabetes mellitus. Our primary goal was to assess the incremental value of ABPM, when added to a clinical assessment that included office BP and HR measurements, to predict all-cause mortality. We also performed secondary analyses to determine whether albuminuria could be a mediator in the association between PP and mortality.
Methods
Study Participants
We studied 1178 participants enrolled in the Informatics for Diabetes Education And Telemedicine (IDEATel) Study.15 We examined the relationship between data obtained at the baseline IDEATel examination, conducted during the years 2000 and 2001, and mortality during follow-up, through February 28, 2008. The components of the baseline examination, and the telemedicine intervention in IDEATel are described in the online data supplement (available at http://hyper.ahajournals.org). Albuminuria was measured as the spot urinary albumin:creatinine ratio (uACR). All of the IDEATel participants signed informed consent, and the protocol was approved by the institutional review boards at all of the participating institutions.
Office BP Measurement
Office BP was measured at the baseline visit using the Dinamap Pro 100 (Critikon) automated oscillometric device. Participants were instructed to take their antihypertensive medication(s) as usual the morning of the examination. Three measurements were obtained at 1-minute intervals in a seated position after 5 minutes of rest in a quiet room, using a standardized protocol.16 The average of the second and third measurements was recorded as the resting BP. Office PP was defined as the difference between systolic and diastolic resting BPs.
Ambulatory BP Monitoring
ABPM was performed at the baseline visit using a SpaceLabs 90207 monitor (SpaceLabs) following a published protocol.17 BP was recorded every 20 minutes for a 24-hour period. Sleep and wake intervals were defined from diary entries and confirmed by a telephone interview on the morning when monitoring ended. A minimum of 6 valid awake readings and 4 valid sleep readings were required for the computation of wake and sleep averages. A reading was accepted as valid if it was nonartifactual and within physiological range. The mean (SD) number of measurements per participant was 64.5 (8.3), whereas the minimum number was 32 (in 1 participant). Ambulatory 24-hour PP was defined as the mean difference between all of the systolic and diastolic BP readings. Ambulatory sleep:wake ratios were calculated for systolic BP and HR and categorized as dipping (ratio: ≤0.9), nondipping pattern (0.9 < ratio ≤ 1), or sleep rise (ratio: >1). AASI was calculated from unedited recordings as follows: the regression of diastolic on systolic BP was estimated for each participant (not forcing the regression line through 0), and AASI=(1−regression slope).
Vital Status Ascertainment
Vital status of the participants was determined through queries of the Center for Medicare and Medicaid Services database, which contains data updated weekly by automated cross-referencing with the Social Security Administration database. We last queried the Center for Medicare and Medicaid Services vital status database for this cohort on February 28, 2008.
Death Certificate Data
Death certificate queries were performed through the New York State Department of Health (last query performed August 26, 2008) and the New York City Bureau of Vital Statistics (last query performed January 7, 2008). Based on the death certificate data, 2 investigators (W.P. and A.M.) independently classified deaths as cardiovascular or noncardiovascular using the National Center for Health Statistics definitions (cardiovascular: International Classification of Diseases, 10th Revision, codes I00 to I99 for underlying cause of death; noncardiovascular: all other International Classification of Diseases, 10th Revision, codes for underlying cause of death).18 They achieved 99% agreement, and the disagreements were resolved by consensus in open review.
Statistical Analysis
Variables that were positively skewed, including uACR, were log10 transformed before the multivariate analyses to better approximate a normal distribution. Comparisons of baseline characteristics according to vital status were made using χ2 or Fisher’s exact tests (when any expected cell frequency was <5) for categorical variables, Student t test for continuous variables approximating a normal distribution, and the Mann–Whitney U test for continuous variables that were not normally distributed. Correlations were assessed using Pearson’s or Spearman’s methods, as appropriate.
The goal of the multivariate Cox proportional hazards models was to test the independent association of ABPM with mortality after adjustment for other covariates, including office BP measurements. Thus, the models were built in a hierarchical fashion using the likelihood ratio test for hypothesis testing. Separate models were built for all-cause and cardiovascular mortality. Time from the baseline examination to the event (or censoring) was the outcome variable, and all of the models were adjusted for age at enrollment (modeling time from age at baseline to age at time of event/censoring rendered virtually identical results). First, we built the optimal predictive models that used clinical, laboratory, and office BP data, using a backward selection process with a P<0.10 for removal of a variable from the equation. The variables considered for inclusion, based on well-established risk factors, biological plausibility, and availability in the data, were age; gender; race/ethnicity; body mass index; duration of diabetes mellitus; smoking; history of CHF; previous myocardial infarction; previous stroke; use of β-blockers, angiotensin-converting enzyme inhibitors, or angiotensin receptor blockers; diuretics; calcium channel blockers; hemoglobin A1c; uACR (log10 transformed); high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, office HR, office systolic BP, diastolic BP, and PP. The variables selected through that process composed model 1. Subsequently, ABPM variables were added to model 1 in a stepwise fashion, starting with the most informative and stopping when the likelihood ratio test failed to show a significant improvement of the model (P<0.05). AASI was categorized into tertiles, because preliminary analyses showed that this provided a better fit than entering AASI as a continuous variable. All of the predictor variables were considered fixed at baseline (ie, none was treated as time dependent). Multicollinearity for BP variables was assessed by calculating the tolerance for each of them in full models. None of them exhibited tolerance values <0.20, which would have indicated excessive multicollinearity.19 Correctness of the proportional hazards assumption was verified using the Harrell and Lee modification of the Schoenfeld goodness-of-fit test.20 The Cox models were compared in terms of calibration, discrimination, and added predictive ability.21,22 Calibration was measured using the −2 log-likelihood statistic and compared using the likelihood ratio test, whereas discrimination of each model was evaluated using the c-statistic and its 95% CI.23 The integrated discrimination improvement method, which does not require predefined risk categories,22 assessed the increment in predictive ability of the model when ambulatory monitoring variables were added to the clinical characteristics. We also performed secondary analyses examining the potential role of uACR as a mediator of the relationship of ambulatory BP to mortality. We did this because PP has known predictive value in the elderly24,25 but is also clearly associated with levels of uACR26 and the progression of albuminuria.27,28 Those analyses used the same approach described above but excluded uACR as a covariate. Statistical analyses were performed using SPSS 15.0 (SPSS Inc) and SAS 9.1 (SAS Institute Inc).
Results
There were 1178 IDEATel participants with complete baseline data, including ABPM recordings. Mean follow-up was 6.6±0.4 years. There were 287 deaths; death certificates were available for 215 deaths (75%), and 110 of those were deemed to be attributed to a cardiovascular cause. According to the American Heart Association definitions,29 75 of the cardiovascular deaths were classified as cardiac, 15 as cere-brovascular, and 20 as death from other cardiovascular disease. To preserve the statistical power of the study, all of the cardiovascular deaths were analyzed together. Table 1 summarizes the baseline characteristics of the participants, categorized by vital status at the end of follow-up. Most participants in both groups (81%) were taking antihypertensive medications. Participants who died tended to be older, male, and to have had a history of CHF, myocardial infarction, or stroke; a longer duration of diabetes mellitus; and higher triglyceride levels at baseline. They also tended to have had higher office PP and HR, higher 24-hour PP and AASI, and an abnormal sleep-dipping pattern of HR and systolic BP. Survivors were more frequently Hispanic and had higher HDL cholesterol levels.
Table 1.
Selected Baseline Characteristics of 1178 Participants Categorized by Vital Status at the End of Follow-Up
| Characteristic | Alive (n=891) | Deceased (n=287) | P Value |
|---|---|---|---|
| Age, y | 70.3±6.1 | 73.1±6.9 | <0.001 |
| Female, % | 63.3 | 47.4 | <0.001 |
| Race/ethnicity, % | 0.003 | ||
| White | 45.6 | 55.4 | |
| Hispanic | 41.0 | 30.0 | |
| Black | 13.4 | 14.6 | |
| Randomization to telemedicine, % | 50.3 | 49.1 | 0.74 |
| CHF, % | 9.5 | 25.4 | <0.001 |
| Previous myocardial infarction, % | 14.6 | 31.7 | <0.001 |
| Previous stroke, % | 8.6 | 14.6 | 0.003 |
| Duration of diabetes mellitus, y | 10.4±8.9 | 13.2±10 | <0.001 |
| Currently smoking, % | 7.1 | 11.5 | 0.014 |
| Taking ACE-I/ARB, % | 63.0 | 59.6 | 0.17 |
| Taking β-blocker, % | 25.7 | 29.8 | 0.10 |
| Taking calcium channel blocker, % | 29.1 | 28.2 | 0.82 |
| Taking diuretic, % | 25.6 | 22.3 | 0.27 |
| Taking any antihypertensive medication, % | 81.1 | 81.1 | 0.98 |
| Body mass index, kg/m2 | 31.4±6.2 | 30.6±6.3 | 0.06 |
| Albumin:creatinine ratio, mg/g | 15.6 (8.1 to 40.9) | 54.9 (15.7 to 224.4) | <0.001 |
| Hemoglobin A1c | 7.4±1.4 | 7.5±1.7 | 0.07 |
| HDL cholesterol, mg/dL | 47.8±13.2 | 43.4±14.9 | <0.001 |
| LDL cholesterol, mg/dL | 106.6±33.4 | 107.7±40.6 | 0.83 |
| Triglycerides, mg/dL | 165.9±93.4 | 182.6±118.4 | 0.015 |
| Office systolic BP, mm Hg | 140.7±22.2 | 143.5±24.7 | 0.07 |
| Office diastolic BP, mm Hg | 70.8±10.7 | 70.7±10.7 | 0.91 |
| Office PP, mm Hg | 69.9±17.7 | 72.8±10.7 | 0.019 |
| Office HR, bpm | 69.7±10.8 | 72.8±10.9 | <0.001 |
| 24-hour systolic BP, mm Hg | 132.5±13.9 | 136.4±16.9 | <0.001 |
| 24-hour diastolic BP, mm Hg | 69.2±8.2 | 69.2±8.9 | 0.97 |
| 24-hour PP, mm Hg | 63.3±13.2 | 67.2±13.7 | <0.001 |
| Sleep blood pressure pattern, %* | 0.006 | ||
| Dipping | 32.4 | 23.3 | |
| Nondipping | 48.8 | 51.6 | |
| Rise | 18.8 | 25.1 | |
| Sleep heart rate pattern, %* | <0.001 | ||
| Dipping | 50.5 | 34.5 | |
| Nondipping | 39.6 | 47.4 | |
| Rise | 9.9 | 18.1 | |
| AASI (×10) | 5.1±1.4 | 5.5±1.5 | <0.001 |
Data are from the IDEATel Study, New York, 2000–2008. Data are means±SDs, median (interquartile range), or percentages. ACE inhibitors indicates angiotensin-converting enzyme inhibitors; ARB, angiotensin receptor blockers. Comparisons were made, as appropriate, using the χ2 or Fisher’s exact tests for categorical variables and Student t or Mann–Whitney U test for continuous variables. AASI has no units.
P for linear trend.
Ambulatory HR variables measured through ABPM showed a strong correlation with office HR (correlation coefficients ranging from 0.64 to 0.70), with the exception of the sleep:wake HR ratio (correlation coefficient: −0.08). These relationships are summarized in Table S1.
The proportion of people who died in different BP categories is summarized in Table 2. There was a dose-response relationship of mortality to tertiles of office HR, 24-hour PP, AASI, and systolic BP and HR sleep-dipping patterns.
Table 2.
Cumulative Percentage of Mortality Within Selected BP and HR Categories
| BP Variable | Percentage of Mortality |
P Value |
|---|---|---|
| BP control, office* | 0.17 | |
| Controlled (n=382) | 22.5 | |
| Uncontrolled (n=796) | 25.3 | |
| Office HR | 0.008‡ | |
| First tertile (42 to 65 bpm; n=393) | 19.1 | |
| Second tertile (66 to 74 bpm; n=403) | 26.6 | |
| Third tertile (75 to 118 bpm; n=382) | 27.3 | |
| Office PP | 0.058‡ | |
| First tertile (25 to 61 mm Hg; n=392) | 21.9 | |
| Second tertile (62 to 77 mm Hg; n=397) | 23.4 | |
| Third tertile (78 to 158 mm Hg; n=389) | 27.8 | |
| BP control, ABPM† | 0.09 | |
| Controlled (n=196) | 20.4 | |
| Uncontrolled (n=982) | 25.2 | |
| 24-h PP | <0.001‡ | |
| First tertile (32 to 58 mm Hg; n=393) | 20.6 | |
| Second tertile (59 to 69 mm Hg; n=392) | 19.4 | |
| Third tertile (69 to 115 mm Hg; n=393) | 33.1 | |
| Sleep BP pattern | 0.006‡ | |
| Dipping (n=355) | 18.9 | |
| Nondipping (n=583) | 25.4 | |
| Rise (n=240) | 30.0 | |
| Sleep HR pattern, % | <0.001‡ | |
| Dipping (n=549) | 18.0 | |
| Nondipping (n=489) | 27.8 | |
| Rise (n=140) | 37.1 | |
| AASI | <0.001‡ | |
| First tertile (n=394) | 18.8 | |
| Second tertile (n=392) | 21.7 | |
| Third tertile (n=392) | 32.4 |
Data are from the IDEATel Study, New York, 2000–2008. AASI has no units.
Uncontrolled office BP was defined as systolic BP >130 mm Hg or diastolic BP >80 mm Hg.
Uncontrolled 24-hour BP was defined as mean systolic BP >120 mm Hg or mean diastolic BP >70 mm Hg. Comparisons were made with the χ2 or Fisher exact test, as appropriate.
P for linear trend.
Cox proportional hazards models were built in a hierarchical fashion that assessed information from clinical and laboratory variables first and then assessed the incremental predictive value of ABPM variables. Models were built for all-cause and cardiovascular mortality. Table 3 summarizes our findings for all-cause mortality. First, we built the best predictive model using non-ABPM variables, model 1. The variables retained in model 1 for the prediction of all-cause mortality were age, gender, duration of diabetes mellitus, smoking, history of CHF, previous myocardial infarction, use of angiotensin-converting enzyme inhibitor/angiotensin receptor blocker, uACR (log10 transformed), HDL cholesterol, and office HR. No office BP variable was selected. The hazard ratios (95% CIs) for the second and third tertiles of office HR were 1.48 (1.09 to 1.99) and 1.56 (1.15 to 2.12), respectively. The adjusted hazard ratios estimated for each ABPM variable when added individually to models 1 are summarized in Table S2.
Table 3.
Cox Proportional Hazards Models for All-Cause Mortality
| Model 1: −2 Log Likelihood=3640.188 |
Model 2: −2 Log Likelihood=3606.117 |
|||
|---|---|---|---|---|
|
|
|
|||
| Variable | Hazard Ratio (95% CI) | P Value | Hazard Ratio (95% CI) | P Value |
| Age, y | 1.05 (1.04 to 1.07) | <0.001 | 1.05 (1.03 to 1.07) | <0.001 |
| Female gender | 0.67 (0.52 to 0.86) | 0.002 | 0.66 (0.52 to 0.85) | 0.001 |
| Duration of diabetes mellitus, y | 1.02 (1.01 to 1.03) | 0.013 | 1.01 (0.99 to 1.02) | 0.088 |
| Current smoker | 1.79 (1.22 to 2.64) | 0.003 | 1.78 (1.21 to 2.62) | 0.004 |
| CHF | 1.92 (1.41 to 2.61) | <0.001 | 1.74 (1.27 to 2.37) | <0.001 |
| Previous myocardial infarction | 1.61 (1.21 to 2.15) | 0.001 | 1.54 (1.15 to 2.08) | 0.004 |
| ACE inhibitor/ARB use | 0.80 (0.63 to 1.03) | 0.081 | 0.82 (0.65 to 1.05) | 0.123 |
| uACR (log10 transformed), mg/g | 2.18 (1.77 to 2.68) | <0.001 | 2.14 (1.74 to 2.64) | <0.001 |
| HDL cholesterol, mg/dL | 0.99 (0.98 to 1.00) | 0.041 | 0.99 (0.98 to 1.00) | 0.097 |
| Office HR (per 10 bpm) | 1.24 (1.12 to 1.38) | <0.001 | 1.26 (1.13 to 1.39) | <0.001 |
| Sleep:wake heart rate ratio (per SD) | 1.36 (1.21 to 1.54) | <0.001 | ||
| AASI (tertiles) | 0.025 | |||
| First (reference) | ||||
| Second | 0.95 (0.69 to 1.30) | |||
| Third | 1.36 (1.01 to 1.83) | |||
Model 1 shows selected baseline characteristics, including office HR. Model 2 includes model 1 plus sleep:wake heart rate ratio. Data are from the IDEATel Study, New York, 2000–2008. ARB indicates angiotensin receptor blocker. The −2 log-likelihood statistic measures the calibration of the models (smaller values reflect better calibration). All of the variables were measured at baseline.
At the next step, we assessed the incremental predictive value of ABPM variables when added to model 1 in a forward selection fashion, starting with the most informative variable and stopping when the likelihood ratio test failed to show a significant improvement of the model. The ABPM variables that added significant predictive information to model 1 for the prediction of all-cause mortality were the sleep:wake HR ratio and the AASI (entered as tertiles; Table 3, model 2). There was no evidence of an interaction between AASI and the sleep:wake HR ratio (P for the interaction term=0.48). Sleep:wake HR ratio, as changes per SD, provided a better fit than a categorical variable. However, this may be difficult to interpret, and, therefore, the hazard ratios for categories of nocturnal dipping were also calculated, for descriptive purposes. The hazard ratios (95% CIs) for a nondipping sleep HR pattern or a sleep rise, compared with the normal pattern, were 1.47 (1.12 to 1.92) and 1.97 (1.38 to 2.79), respectively. The hazard ratios for the second and third AASI tertiles, compared with the first, were 0.95 (0.69 to 1.30) and 1.37 (1.02 to 1.85), respectively. Higher office HR remained an independent predictor of mortality in the fully adjusted model (P<0.001). Models 1 and 2 exhibited similar discrimination; their c-statistics (95% CIs) were 0.73 (0.70 to 0.76) and 0.74 (0.71 to 0.77), respectively. The predictive ability of model 2, as assessed by the integrated discrimination improvement,22 was significantly better than that of model 1 (P<0.001). Both models had significantly better discrimination than a model limited to age and gender, which exhibited a c-statistic of 0.65 (0.61 to 0.69). Of note, the forced inclusion of β-blocker use into the models did not change our findings. Moreover, the addition of systolic BP sleep dipping patterns failed to improve model 2, and there was a loss in calibration if 24-hour PP was added.
Table 4 summarizes our findings for cardiovascular mortality, which were similar to those noted above for all-cause mortality, and differed only in that a smaller number of variables were selected. Model 1 included age, duration of diabetes mellitus, history of CHF, HDL cholesterol levels, uACR (log10 transformed), and office HR. Model 2 identified the sleep:wake HR ratio as the only ambulatory monitoring variable that added incremental predictive value to model 1. The hazard ratios (95% CIs) for a nondipping sleep HR pattern or a sleep rise, compared with the dipping pattern, were 1.42 (0.92 to 2.21) and 2.29 (1.35 to 3.87), respectively. Models 1 and 2 exhibited similar discrimination; their c-statistics (95% CIs) were 0.77 (0.73 to 0.82) and 0.78 (0.74 to 0.82), respectively. Addition of the sleep:wake HR ratio increased significantly the predictive ability of the model, as measured by the integrated discrimination improvement (P<0.001).22
Table 4.
Cox Proportional Hazards Models for Cardiovascular Mortality
| Model 1: −2 Log Likelihood=1380.275 |
Model 2: −2 Log Likelihood=1370.177 |
|||
|---|---|---|---|---|
|
|
|
|||
| Variable | Hazard Ratio (95% CI) | P Value | Hazard Ratio (95% CI) | P Value |
| Age, y | 1.06 (1.03 to 1.09) | <0.001 | 1.06 (1.03 to 1.09) | <0.001 |
| Duration of diabetes mellitus, y | 1.02 (1.00 to 1.04) | 0.023 | 1.02 (1.00 to 1.04) | 0.040 |
| Congestive heart failure | 2.81 (1.85 to 4.28) | <0.001 | 2.60 (1.71 to 2.37) | <0.001 |
| uACR (log10 transformed), mg/g | 2.54 (1.83 to 3.52) | <0.001 | 2.14 (1.74 to 2.64) | <0.001 |
| HDL cholesterol, mg/dL | 0.98 (0.97 to 0.99) | 0.036 | 0.99 (0.98 to 1.00) | 0.043 |
| Office HR (per 10 bpm) | 1.25 (1.05 to 1.48) | 0.011 | 1.29 (1.09 to 1.53) | 0.004 |
| Sleep:wake heart rate ratio (per SD) | 1.37 (1.13 to 1.66) | 0.001 | ||
Model 1 shows selected baseline characteristics, including office HR. Model 2 includes model 1 plus sleep:wake heart rate ratio. Data are from the IDEATel Study, New York, 2000–2008. ARB indicates angiotensin receptor blocker. The −2 log-likelihood statistic measures the calibration of the models (smaller values reflect better calibration). All of the variables were measured at baseline.
Results of the secondary analyses, in which Cox models were built without uACR as a covariate, are described in Tables S3 and S4. They rendered similar results as compared with the primary analyses, including the selection of sleep: wake HR ratio, but differed in that PP was an independent predictor of all-cause mortality, whereas AASI was not. In the fully adjusted model, the all-cause mortality hazard ratio per each 10-mm Hg increase in 24-hour PP was 1.20 (1.04 to 1.40). With regard to cardiovascular mortality, sleep:wake HR ratio was once again an independent predictor in the fully adjusted model, and sleep systolic BP was the other ambulatory monitoring variable that added incremental predictive value. The hazard ratio for cardiovascular mortality per 10-mm Hg increase in sleep systolic BP was 1.18 (1.05 to 1.32).
Discussion
We found in a sample of older people with diabetes mellitus followed for a mean of 6.6 years that, in addition to established risk factors: (1) office HR was a strong predictor of mortality, outperforming office BP measurements; (2) ABPM added significantly to the prediction of mortality: the variables that contributed independent predictive information were the sleep HR dipping ratio (all-cause and cardiovascular mortality) and the AASI (all-cause mortality only); and (3) albuminuria was also a strong predictor, and it appeared to play a role in mediating in the relationship between PP and mortality. Of note, the predictive variables that we identified, namely, loss of nocturnal HR decline, arterial stiffness, and albuminuria, reflect cardiovascular or renal end-organ damage, possibly caused in part by poor control, long duration, or both, of diabetes mellitus. These findings are novel and merit further discussion.
The strong predictive value of higher office HR that we have observed in this cohort is in agreement with previous reports in the general population. The association of office HR with mortality risks in the general population was described 2 decades ago by the Framingham investigators,30 and their findings have been replicated on numerous occasions. However, the relative predictive value of office HR, as compared with ambulatory HR measurements, remains largely unknown in people with diabetes mellitus. The data from other studies, which included a minority of people with diabetes mellitus, are somewhat contradictory. Verdecchia et al10 reported that office HR did not have independent value to predict mortality in a cohort of people with hypertension who underwent ABPM. In contrast, in the placebo arm of the Systolic Hypertension in Europe Trial, a significant independent association with mortality was found for office HR but not for ambulatory HR.5 However, the sleep:wake ratio was not evaluated.5 Of note, the number of events in both reports was small (74 and 39 deaths, respectively), severely limiting the statistical power of their analyses. There were 287 deaths in our study. Finally, in a population-based study in Milan, Italy, neither office nor ambulatory HR was an independent predictor of mortality, but participants were both younger and healthier than in our study.31
When assessing the incremental predictive value of ABPM, the ratio between sleep and wake HR added significant information to the fully adjusted models for the prediction of all-cause and cardiovascular mortality. In contrast, it is likely that 24-hour HR did not add to the prediction of mortality because it was highly correlated with office HR (Pearson R=0.70). The sleep:wake HR ratio, which was not correlated with office HR, has not been examined previously as a predictor of mortality in people with diabetes mellitus. Previous studies in hypertensives have reported an independent association of the sleep-dipping HR patterns with mortality risk: one of them was the aforementioned study by Verdecchia et al.10 Another study, by Ben-Dov et al,11 also included a small number of participants with diabetes mellitus (9%) in the sample. In an International Database on Ambulatory blood pressure monitoring in relation to Cardiovascular Outcomes (IDACO) collaborative analysis of pooled data by Hansen et al,32 night:day HR ratio was also an independent predictor of all-cause mortality, but it did not reach statistical significance to predict cardiovascular mortality (hazard ratio=1.12; 95% CI: 0.99 to 1.27). Their sample size was large (n=6928), but only 7% of the subjects were people with diabetes mellitus. Another notable difference with our study was that uACR and office HR were not included in their Cox models. Of note, we found that the sleep:wake HR ratio, as a continuous variable, provided a better fit in multivariate models than when categorized as dipping, nondipping, and riser. However, the increment in risk associated with those categories was also reported, for descriptive purposes.
The putative mechanisms underlying the association of office HR and sleep:wake heart ratio with higher mortality are several and may be shared to a large extent. However, the fact that those variables had independent predictive value in this cohort suggests that the overlap is not complete. Cardiovascular autonomic neuropathy may play an important role in our elderly cohort with a long-standing history of diabetes mellitus, resulting in an impairment in the normal sympathetic withdrawal that should occur while resting. Importantly, low HR variability has been clearly associated with a higher incidence of fatal cardiovascular events.33
The other ABPM variable with independent predictive value in our primary analysis was the AASI, which was first described by Li et al34 in 2006. AASI is elevated in people with diabetes mellitus, probably as a result of cardiovascular end-organ damage.35 However, previous publications assessing the use of AASI to predict mortality did not provide estimates for people with diabetes mellitus and had contradictory findings: AASI was an independent predictor in 2 studies12,13 but not in a third.36 The exact physiological meaning of the AASI remains controversial.37 Originally intended to assess arterial stiffness, its use for that purpose has been questioned.38 Alternatively, it has been proposed that the AASI reflects dynamic arterial stiffening, the relative increment in stiffness from diastolic to systolic values,39 or the relationship between the pulsatile and steady components of BP.40 Regardless of the exact pathophysiologic meaning of the AASI, our findings suggest that it may contribute significantly to the risk stratification of people with diabetes mellitus.
PP has been shown to be particularly useful to predict mortality in the elderly, in whom it is believed to reflect arterial stiffness.24,25 Interestingly, we found that office and 24-hour PPs were independently predictive of all-cause mortality only in models that did not adjust for albuminuria. In turn, sleep systolic BP added to the prediction of cardiovascular deaths in models that did not include albuminuria as a covariate, as reported previously by Fagard et al41 in hypertensives. This suggests that albuminuria, or a yet-unidentified variable closely associated with it, may be in the causal pathway of the association between higher BP levels and mortality. This is biologically plausible because increments in albuminuria over time have been directly associated with higher mortality,42 and 24-hour PP is associated with the progression of albuminuria in elderly people with diabetes mellitus.14,27 Previous studies assessing the predictive value of ABPM in diabetes mellitus failed to consider albuminuria as a covariate.7,9 From a clinical standpoint, that was a significant deficiency, because periodic uACR measurements are now part of the standard of care for people with diabetes mellitus.
It should be noted that most participants in our study (81%) were taking antihypertensive medications, which were not stopped for the baseline examination. This may have played a role in decreasing the predictive information provided by office blood pressure levels in this study. Interestingly, office HR was an independent predictor of events, although it was also measured under the influence of antihypertensive medications.
Several limitations are noteworthy. First, our sample was composed of older subjects, with long-standing diabetes mellitus. Thus, our findings may not generalize to younger patients and particularly to those with type 1 diabetes mellitus. Second, like all observational studies, ours was subject to the risk of residual confounding because of measurement error or unmeasured confounders. Third, the IDEATel baseline examination did not include an assessment of renal function, such as serum creatinine. Patients with advanced renal failure were excluded, but we do not know whether the addition of a serum creatinine measurement would have substantially changed our results. Finally, we did not perform sleep studies to determine whether abnormal sleep patterns of HR and BP were because of sleep apnea. The strengths of this study include its prospective design; a large sample that was well-characterized, elderly, multiethnic, and had adequate representation of women; and a larger number of deaths than in previous studies of ABPM and mortality prediction in people with diabetes mellitus. ABPM was performed using a well-validated methodology.43 Moreover, we analyzed the incremental value of ABPM when added to clinical variables in a hierarchical approach that aims to resemble the use of risk prediction tools in the clinical setting.44
Perspectives
Office HR was an independent predictor of all-cause mortality in this cohort, whereas office BP was not. The predictive value of office HR persisted after incorporating the information provided by ABPM. This highlights the clinical relevance of HR when risk-stratifying elderly people with diabetes mellitus. ABPM added significant information to the prediction of mortality risk. Specifically, the sleep:wake HR ratio and the AASI, which reflects the dynamic relationship between diastolic and systolic BPs, were independent predictors of all-cause mortality, whereas only the sleep:wake HR ratio added to the prediction of cardiovascular mortality. In addition, we found evidence suggesting that albuminuria, a powerful marker of increased risk, may mediate the association between elevated BP and higher mortality. Overall, we believe that our findings highlight the relationship between diabetic end-organ damage, as reflected by changes in HR, BP, and albuminuria, and the mortality risks in elderly people with diabetes mellitus.
Supplementary Material
Acknowledgments
Source of Funding
This research was supported by Cooperative Agreement 95-C-90998 from the Centers for Medicare and Medicaid Services.
Footnotes
Disclosures
None.
References
- 1.Bertoni AG, Kirk JK, Goff DC, Jr, Wagenknecht LE. Excess mortality related to diabetes mellitus in elderly Medicare beneficiaries. Ann Epidemiol. 2004;14:362–367. doi: 10.1016/j.annepidem.2003.09.004. [DOI] [PubMed] [Google Scholar]
- 2.Gerstein HC, Mann JF, Yi Q, Zinman B, Dinneen SF, Hoogwerf B, Halle JP, Young J, Rashkow A, Joyce C, Nawaz S, Yusuf S. Albuminuria and risk of cardiovascular events, death, and heart failure in diabetic and nondiabetic individuals. JAMA. 2001;286:421–426. doi: 10.1001/jama.286.4.421. [DOI] [PubMed] [Google Scholar]
- 3.Bertoni AG, Hundley WG, Massing MW, Bonds DE, Burke GL, Goff DC., Jr Heart failure prevalence, incidence, and mortality in the elderly with diabetes. Diabetes Care. 2004;27:699–703. doi: 10.2337/diacare.27.3.699. [DOI] [PubMed] [Google Scholar]
- 4.Perloff D, Sokolow M, Cowan R. The prognostic value of ambulatory blood pressures. JAMA. 1983;249:2792–2798. [PubMed] [Google Scholar]
- 5.Palatini P, Thijs L, Staessen JA, Fagard RH, Bulpitt CJ, Clement DL, de Leeuw PW, Jaaskivi M, Leonetti G, Nachev C, O’Brien ET, Parati G, Rodicio JL, Roman E, Sarti C, Tuomilehto J. Predictive value of clinic and ambulatory heart rate for mortality in elderly subjects with systolic hypertension. Arch Intern Med. 2002;162:2313–2321. doi: 10.1001/archinte.162.20.2313. [DOI] [PubMed] [Google Scholar]
- 6.Astrup AS, Nielsen FS, Rossing P, Ali S, Kastrup J, Smidt UM, Parving HH. Predictors of mortality in patients with type 2 diabetes with or without diabetic nephropathy: a follow-up study. J Hypertens. 2007;25:2479–2485. doi: 10.1097/HJH.0b013e3282f06428. [DOI] [PubMed] [Google Scholar]
- 7.Nakano S, Ito T, Furuya K, Tsuda S, Konishi K, Nishizawa M, Nakagawa A, Kigoshi T, Uchida K. Ambulatory blood pressure level rather than dipper/nondipper status predicts vascular events in type 2 diabetic subjects. Hypertens Res. 2004;27:647–656. doi: 10.1291/hypres.27.647. [DOI] [PubMed] [Google Scholar]
- 8.Mannucci E, Lambertucci L, Monami M, Fedeli A, Chiasserini V, Marchionni N, Masotti G, Ungar A. Pulse pressure and mortality in hypertensive type 2 diabetic patients. A cohort study. Diabetes Metab Res Rev. 2006;22:172–175. doi: 10.1002/dmrr.598. [DOI] [PubMed] [Google Scholar]
- 9.Eguchi K, Pickering TG, Hoshide S, Ishikawa J, Ishikawa S, Schwartz JE, Shimada K, Kario K. Ambulatory blood pressure is a better marker than clinic blood pressure in predicting cardiovascular events in patients with/without type 2 diabetes. Am J Hypertens. 2008;21:443–450. doi: 10.1038/ajh.2008.4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Verdecchia P, Schillaci G, Borgioni C, Ciucci A, Telera MP, Pede S, Gattobigio R, Porcellati C. Adverse prognostic value of a blunted circadian rhythm of heart rate in essential hypertension. J Hypertens. 1998;16:1335–1343. doi: 10.1097/00004872-199816090-00015. [DOI] [PubMed] [Google Scholar]
- 11.Ben-Dov IZ, Kark JD, Ben-Ishay D, Mekler J, Ben-Arie L, Bursztyn M. Blunted heart rate dip during sleep and all-cause mortality. Arch Intern Med. 2007;167:2116–2121. doi: 10.1001/archinte.167.19.2116. [DOI] [PubMed] [Google Scholar]
- 12.Dolan E, Thijs L, Li Y, Atkins N, McCormack P, McClory S, O’Brien E, Staessen JA, Stanton AV. Ambulatory arterial stiffness index as a predictor of cardiovascular mortality in the Dublin Outcome Study. Hypertension. 2006;47:365–370. doi: 10.1161/01.HYP.0000200699.74641.c5. [DOI] [PubMed] [Google Scholar]
- 13.Kikuya M, Staessen JA, Ohkubo T, Thijs L, Metoki H, Asayama K, Obara T, Inoue R, Li Y, Dolan E, Hoshi H, Hashimoto J, Totsune K, Satoh H, Wang JG, O’Brien E, Imai Y. Ambulatory arterial stiffness index and 24-hour ambulatory pulse pressure as predictors of mortality in Ohasama, Japan. Stroke. 2007;38:1161–1166. doi: 10.1161/01.STR.0000259604.67283.69. [DOI] [PubMed] [Google Scholar]
- 14.Palmas W, Moran A, Pickering T, Eimicke JP, Teresi J, Schwartz JE, Field L, Weinstock RS, Shea S. Ambulatory pulse pressure and progression of urinary albumin excretion in older patients with type 2 diabetes mellitus. Hypertension. 2006;48:301–308. doi: 10.1161/01.HYP.0000232644.98208.65. [DOI] [PubMed] [Google Scholar]
- 15.Shea S, Starren J, Weinstock RS, Knudson PE, Teresi J, Holmes D, Palmas W, Field L, Goland R, Tuck C, Hripcsak G, Capps L, Liss D. Columbia University’s Informatics for Diabetes Education and Telemedicine (IDEATel) Project: rationale and design. J Am Med Inform Assoc. 2002;9:49–62. doi: 10.1136/jamia.2002.0090049. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Bild DE, Bluemke DA, Burke GL, Detrano R, Diez Roux AV, Folsom AR, Greenland P, Jacob DR, Jr, Kronmal R, Liu K, Nelson JC, O’Leary D, Saad MF, Shea S, Szklo M, Tracy RP. Multi-ethnic study of atherosclerosis: objectives and design. Am J Epidemiol. 2002;156:871–881. doi: 10.1093/aje/kwf113. [DOI] [PubMed] [Google Scholar]
- 17.Pickering TG. Ambulatory Blood Pressure. Redmond, WA: Spacelabs Medical; 1994. [Google Scholar]
- 18.Arias E, Anderson RN, Kung HC, Murphy SL, Kochanek KD. Deaths: final data for 2001. Natl Vital Stat Rep. 2003;52:1–115. [PubMed] [Google Scholar]
- 19.Kleinbaum D. Applied Regression Analysis and Other Multivariable Methods. 3. Pacific Grove, CA: Brooks/Cole; 1998. p. 241. [Google Scholar]
- 20.Kleinbaum DG, Klein M. Survival Analysis. 2. New York, NY: Springer; 2005. [Google Scholar]
- 21.Cook NR. Use and misuse of the receiver operating characteristic curve in risk prediction. Circulation. 2007;115:928–935. doi: 10.1161/CIRCULATIONAHA.106.672402. [DOI] [PubMed] [Google Scholar]
- 22.Pencina MJ, D’Agostino RB, Sr, D’Agostino RB, Jr, Vasan RS. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med. 2008;27:157–172. doi: 10.1002/sim.2929. [DOI] [PubMed] [Google Scholar]
- 23.Pencina MJ, D’Agostino RB. Overall C as a measure of discrimination in survival analysis: model specific population value and confidence interval estimation. Stat Med. 2004;23:2109–2123. doi: 10.1002/sim.1802. [DOI] [PubMed] [Google Scholar]
- 24.Mazza A, Pessina AC, Gianluca P, Tikhonoff V, Pavei A, Casiglia E. Pulse pressure: an independent predictor of coronary and stroke mortality in elderly females from the general population. Blood Press. 2001;10:205–211. doi: 10.1080/08037050152669710. [DOI] [PubMed] [Google Scholar]
- 25.Staessen JA, Thijs L, O’Brien ET, Bulpitt CJ, de Leeuw PW, Fagard RH, Nachev C, Palatini P, Parati G, Tuomilehto J, Webster J, Safar ME. Ambulatory pulse pressure as predictor of outcome in older patients with systolic hypertension. Am J Hypertens. 2002;15:835–843. doi: 10.1016/s0895-7061(02)02987-4. [DOI] [PubMed] [Google Scholar]
- 26.Moran A, Palmas W, Pickering TG, Schwartz JE, Field L, Weinstock RS, Shea S. Office and ambulatory blood pressure are independently associated with albuminuria in older subjects with type 2 diabetes. Hypertension. 2006;47:955–961. doi: 10.1161/01.HYP.0000216634.73504.7d. [DOI] [PubMed] [Google Scholar]
- 27.Palmas W, Pickering T, Eimicke J, Moran A, Teresi J, Schwartz J, Field L, Weinstock R, Shea S. Value of ambulatory arterial stiffness index and 24-hour pulse pressure to predict progression of albuminuria in elderly people with diabetes mellitus. Am J Hypertens. 2007;20:493–500. doi: 10.1016/j.amjhyper.2006.11.005. [DOI] [PubMed] [Google Scholar]
- 28.Palmas W, Pickering T, Teresi J, Schwartz J, Eguchi K, Field L, Weinstock R, Shea S. Nocturnal blood pressure elevation predicts progression of albuminuria in elderly people with type-2 diabetes. J Clin Hypertens. 2007;10:12–20. doi: 10.1111/j.1524-6175.2007.07170.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Luepker RV, Apple FS, Christenson RH, Crow RS, Fortmann SP, Goff D, Goldberg RJ, Hand MM, Jaffe AS, Julian DG, Levy D, Manolio T, Mendis S, Mensah G, Pajak A, Prineas RJ, Reddy KS, Roger VL, Rosamond WD, Shahar E, Sharrett AR, Sorlie P, Tunstall-Pedoe H. Case definitions for acute coronary heart disease in epidemiology and clinical research studies: a statement from the AHA Council on Epidemiology and Prevention; AHA Statistics Committee; World Heart Federation Council on Epidemiology and Prevention; the European Society of Cardiology Working Group on Epidemiology and Prevention; Centers for Disease Control and Prevention; and the National Heart, Lung, and Blood Institute. Circulation. 2003;108:2543–2549. doi: 10.1161/01.CIR.0000100560.46946.EA. [DOI] [PubMed] [Google Scholar]
- 30.Kannel WB, Kannel C, Paffenbarger RS, Jr, Cupples LA. Heart rate and cardiovascular mortality: the Framingham Study. Am Heart J. 1987;113:1489–1494. doi: 10.1016/0002-8703(87)90666-1. [DOI] [PubMed] [Google Scholar]
- 31.Sega R, Facchetti R, Bombelli M, Cesana G, Corrao G, Grassi G, Mancia G. Prognostic value of ambulatory and home blood pressures compared with office blood pressure in the general population: follow-up results from the Pressioni Arteriose Monitorate e Loro Associazioni (PAMELA) Study. Circulation. 2005;111:1777–1783. doi: 10.1161/01.CIR.0000160923.04524.5B. [DOI] [PubMed] [Google Scholar]
- 32.Hansen TW, Thijs L, Boggia J, Li Y, Kikuya M, Bjorklund-Bodegard K, Richart T, Ohkubo T, Jeppesen J, Torp-Pedersen C, Lind L, Sandoya E, Imai Y, Wang J, Ibsen H, O’Brien E, Staessen JA. Prognostic value of ambulatory heart rate revisited in 6928 subjects from 6 populations. Hypertension. 2008;52:229–235. doi: 10.1161/HYPERTENSIONAHA.108.113191. [DOI] [PubMed] [Google Scholar]
- 33.Tsuji H, Venditti FJ, Jr, Manders ES, Evans JC, Larson MG, Feldman CL, Levy D. Reduced heart rate variability and mortality risk in an elderly cohort. The Framingham Heart Study. Circulation. 1994;90:878–883. doi: 10.1161/01.cir.90.2.878. [DOI] [PubMed] [Google Scholar]
- 34.Li Y, Dolan E, Wang JG, Thijs L, Zhu DL, Staessen JA, O’Brien E, Stanton A. Ambulatory arterial stiffness index: determinants and outcome. Blood Press Monit. 2006;11:107–110. doi: 10.1097/01.mbp.0000189791.90488.a1. [DOI] [PubMed] [Google Scholar]
- 35.Gavish B, Ben-Dov IZ, Bursztyn M. Putting a spin on the ambulatory arterial stiffness index. J Hypertens. 2008;26:1268–1269. doi: 10.1097/HJH.0b013e3282fbf608. [DOI] [PubMed] [Google Scholar]
- 36.Hansen TW, Staessen JA, Torp-Pedersen C, Rasmussen S, Li Y, Dolan E, Thijs L, Wang JG, O’Brien E, Ibsen H, Jeppesen J. Ambulatory arterial stiffness index predicts stroke in a general population. J Hypertens. 2006;24:2247–2253. doi: 10.1097/01.hjh.0000249703.57478.78. [DOI] [PubMed] [Google Scholar]
- 37.Schillaci G, Parati G, Pirro M, Pucci G, Mannarino MR, Sperandini L, Mannarino E. Ambulatory arterial stiffness index is not a specific marker of reduced arterial compliance. Hypertension. 2007;49:986–991. doi: 10.1161/HYPERTENSIONAHA.106.082248. [DOI] [PubMed] [Google Scholar]
- 38.Li Y, Wang JG, Dolan E, Gao PJ, Guo HF, Nawrot T, Stanton AV, Zhu DL, O’Brien E, Staessen JA. Ambulatory arterial stiffness index derived from 24-hour ambulatory blood pressure monitoring. Hypertension. 2006;47:359–364. doi: 10.1161/01.HYP.0000200695.34024.4c. [DOI] [PubMed] [Google Scholar]
- 39.Gavish B, Ben-Dov IZ, Bursztyn M. Linear relationship between systolic and diastolic blood pressure monitored over 24 h: assessment and correlates. J Hypertens. 2008;26:199–209. doi: 10.1097/HJH.0b013e3282f25b5a. [DOI] [PubMed] [Google Scholar]
- 40.Benetos A, Lacolley P. From 24-hour blood pressure measurements to arterial stiffness: a valid short cut? Hypertension. 2006;47:327–328. doi: 10.1161/01.HYP.0000200705.61571.95. [DOI] [PubMed] [Google Scholar]
- 41.Fagard RH, Celis H, Thijs L, Staessen JA, Clement DL, De Buyzere ML, De Bacquer DA. Daytime and nighttime blood pressure as predictors of death and cause-specific cardiovascular events in hypertension. Hypertension. 2008;51:55–61. doi: 10.1161/HYPERTENSIONAHA.107.100727. [DOI] [PubMed] [Google Scholar]
- 42.Spoelstra-de Man AM, Brouwer CB, Stehouwer CD, Smulders YM. Rapid progression of albumin excretion is an independent predictor of cardiovascular mortality in patients with type 2 diabetes and microalbuminuria. Diabetes Care. 2001;24:2097–2101. doi: 10.2337/diacare.24.12.2097. [DOI] [PubMed] [Google Scholar]
- 43.Pang TC, Brown MA. Accuracy of ambulatory blood pressure monitors in routine clinical practice. Am J Hypertens. 2006;19:801–809. doi: 10.1016/j.amjhyper.2005.12.012. [DOI] [PubMed] [Google Scholar]
- 44.Hachamovitch R, Berman DS, Kiat H, Cohen I, Friedman JD, Shaw LJ. Value of stress myocardial perfusion single photon emission computed tomography in patients with normal resting electrocardiograms: an evaluation of incremental prognostic value and cost-effectiveness. Circulation. 2002;105:823–829. doi: 10.1161/hc0702.103973. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.