Hematocrit Predicts Long-Term Mortality in a Nonlinear and Sex-Specific Manner in Hypertensive AdultsNovelty and Significance
Hematocrit has been inconsistently reported to be a risk marker of cardiovascular morbidity and mortality. The Glasgow Blood Pressure Clinic Study cohort included 10951 hypertensive patients, who had hematocrit measured at their initial clinic visit and followed for ≤35 years. Cox proportional hazards models were used to estimate hazard ratios for all-cause, cardiovascular, ischemic heart disease, stroke, and noncardiovascular mortality. There were 3484 deaths over a follow-up period of 173245 person-years. Hematocrit was higher in men (median, 0.44; interquartile range, 0.42–0.47) than in women (median, 0.41; interquartile range, 0.38–0.43). The lowest risk for all-cause mortality was seen in quartile 2 for men (range, 0.421–0.440) and women (range, 0.381–0.400). Compared with quartile 2, the adjusted hazard ratios for quartiles 1, 3, and 4 were, respectively, 1.11 (range, 0.97–1.28), 1.19 (range, 1.04–1.37), and 1.22 (range, 1.06–1.39) in men and 1.17 (range, 1.01–1.36), 0.97 (range, 0.83–1.13), and 1.19 (range, 1.04–1.37) in women. Men showed a J-shaped pattern for cardiovascular mortality and a linear pattern for noncardiovascular mortality in cause-specific analysis, whereas in women a U-shaped pattern was observed for noncardiovascular mortality only. Higher baseline hematocrit was associated with higher on-treatment blood pressure during follow-up. Baseline hematocrit did not affect the time to reach target blood pressure. The increased risk of death attributed to higher hematocrit was seen in men and women irrespective of their achievement of target blood pressure, indicating that the risk is independent of the effect of hematocrit on blood pressure. Hypertensive patients with hematocrit levels outside of the sex-specific reference ranges identified in this study should be targeted for more aggressive blood pressure and cardiovascular risk reduction treatment.
Hematocrit (Hct), the proportion of blood volume occupied by red blood cells, is a major determinant of blood viscosity, blood pressure (BP), venous return, cardiac output, and platelet adhesiveness.1–5 Several prospective studies have demonstrated associations between Hct and risk of cardiovascular disease, including coronary heart disease (CHD) and stroke.6–10 The association between Hct and mortality has been described as “J-“ or “U” shaped in women and inverse-linear in men.5,8 In contrast, a population-based 17-year follow-up study reported a U-shaped relationship between Hct and CHD mortality (with the lowest CHD mortality rates being seen among those with Hct in the middle tertile) in men and a positive linear relationship in women (with linear increase in CHD mortality from the lowest to highest tertile of Hct).6 More recently, in a 28-year follow-up study among men aged 55 years, borderline polycythemia (Hct >0.50) was associated with 1.8-fold increase in CHD death even after adjustment for established coronary risk factors.9 The observed association between Hct and BP 5,11 and the sex differences in blood viscosity and Hct levels12 may be important factors to consider in the management of hypertensive patients, especially if Hct is an independent predictor of long-term mortality. We examined the association between Hct and long-term mortality in a large hypertensive cohort to dissect the sex-specific differences in risk posed by Hct after the initiation of antihypertensive treatment.
Study Setting and Study Population
The Glasgow Blood Pressure Clinic provides secondary and tertiary level service to individuals with hypertension from the west of Scotland. All of the patients referred to the BP clinic had a diagnosis of hypertension made in primary care using definitions of hypertension based on contemporaneous guidelines and if appropriate treatment commenced in primary care. All of the patients were treated at Glasgow Blood Pressure Clinic until their BP control was stabilized with continuing follow-up at the BP clinic or in primary care. The frequency of visits to Glasgow Blood Pressure Clinic mainly depended on individual patient BP levels and presence of other comorbidities.
Blood samples collected at first visit and during follow-up at the BP clinic were analyzed by the hospital laboratory auto-analyzer, which calculates Hct as the product of red blood cell count and the erythrocyte mean cell volume. BP measurements at the clinic were taken manually 3 times using standardized sphygmomanometers at each visit by specialist hypertension nurses; the mean of the last 2 measurements was recorded at each visit. Patients attending the clinic were advised to take their regular medications as usual. Only clinic BP readings during routine outpatient visits were recorded and used for analysis. Blood samples were collected at baseline and at regular intervals for estimation of routine hematologic and biochemical indices. Estimated glomerular filtration rate was calculated from the baseline serum creatinine values using the Modification of Diet in Renal Disease Study Group equation.13 A structured format was used to assess the presence of existing cardiovascular disease, tobacco (any versus none), and alcohol use (any versus none).
Record linkage with the General Register Office for Scotland ensured notification of a subject's death (provided that it occurred in the United Kingdom) together with the primary cause of death according to the International Classification of Diseases, 10th Revision, version for 2007 (ICD-10) codes. We considered cardiovascular deaths (CVD mortality; ICD-10 codes I00 to I99), ischemic heart disease (IHD) deaths (IHD mortality; ICD-10 codes 120 to I25), and stroke deaths (stroke mortality; ICD-10 codes I60 to I69) in the analysis. Deaths other than those attributed to cardiovascular causes are classified as non-CVD deaths. Mortality data were collected up to April 2011, allowing a maximum of 35 years for participants who had been under follow-up for the longest time.
We divided the study population into 4 different groups based on Hct quartiles. Men and women have different reference ranges of Hct, hence we derived 2 groups of quartiles separately for men and women. These were ≤0.420, 0.421 to 0.440, 0.441 to 0.460, and ≥0.461 for men and ≤0.380, 0.381 to 0.400, 0.401 to 0.420, and ≥0.421 for women. To explore potential causal associations among BP, Hct, and mortality we assessed the response to treatment for a subgroup of our population. Participants who had been followed for a minimum of 5 years were selected. We identified 2 groups, responders and nonresponders. Responders were people in whom BP was lowered to <140/90 mmHg measured ≥30 days after the first visit and remained below that level for at least a year, and nonresponders were subjects who did not meet those criteria.
The baseline characteristics of study subjects in different Hct quartiles were compared using ANOVA for continuous variables and χ2 test for categorical variables. Cox proportional hazards models were set up to analyze the influence of baseline Hct on all-cause, CVD, ischemic heart disease (IHD), stroke, and non-CVD mortality. The covariates included were baseline age, sex, body mass index, smoking status, systolic BP (SBP), diastolic BP, alcohol use, tobacco use, estimated glomerular filtration rate, and cardiovascular comorbidity. A variable on year of first visit strata (epochs) was used to adjust the secular trend in mortality and was divided into 5 categories (first visit 1977 or before, between years 1978 and 1985, 1986 and 1993, 1994 and 2001, or 2002 and after). Initially Hct was assessed as a continuous variable in the overall population (model 1). Because Hct range was different in men and women along with the observation of a nonlinear trend in mortality, further analyses were performed using Hct as a categorical variable based on sex-specific quartiles (model 2 in men and model 3 in women). The final models included all of the covariates that were significant (P<0.10) in the multivariate analysis. A similar analysis protocol was followed for responder and nonresponder subgroups. To model efficiently the nonlinear relationship observed for Hct and mortality, multivariable Cox regression models with restricted spline functions were used.14 All of the analyses were performed using STATA, version 12 (Stata Corp).
Generalized estimating equation regression models were used to study the change in SBP and diastolic BP over the initial 5 years of follow-up across Hct quartiles. The generalized estimating equations model was adjusted for age, sex, smoking, alcohol use, body mass index, and estimated glomerular filtration rate. The association between Hct quartiles, the time to reach target BP, and the number of antihypertensive drugs prescribed were assessed using log-rank test in a bivariate model and Cox proportional hazards regression in a multivariate model after adjustment for all of the covariates that were significant (P<0.1). Data were censored after 5 years of follow-up and analyses performed in men and women separately.
A total of 10951 patients with baseline Hct values were included in the current analysis. The full demographic and clinical characteristics for men and women are given in Table 1. More than half of the study patients were women (51.3%; n=5722). Men were younger than women (49.6 versus 51.3 years), and both were similarly overweight (body mass index, 27.5). Compared with women, men had significantly higher values for BP and estimated glomerular filtration rate, and a significantly higher proportion of men reported alcohol and tobacco use. A total of 192 and 503 people had missing data for tobacco use and alcohol use, respectively.
Hct was higher in men (median, 0.44; interquartile range, 0.42–0.47) than in women (median, 0.41; interquartile range, 0.38–0.43). The distributions of Hct in males and females are shown as probability density plots in Figure S1 (available in the online-only Data Supplement). The baseline characteristics of the study population stratified by sex and sex-specific Hct quartiles are shown in Table 2. The lowest Hct quartile in men was ≤0.420 (range, 0.280–0.420), whereas for women it was ≤0.380 (range, 0.220–0.380). The highest quartile was also different for men and women (≥0.461 [range, 0.461–0.670] and 0.421 [range, 0.421–0.550], respectively).
Follow-Up Period and Event Rates
Complete follow-up event data with information on all of the baseline risk variables were available for 10007 participants. During the 173245.83 person-years of follow-up, there were 3484 (34.8%) all-cause deaths; 52.6% of these deaths were in men. There were 1920 cardiovascular deaths (19.2%), 56.1% in men; 1077 IHD deaths (10.8%), 60.4% in men; 458 stroke deaths (4.6%), 50.7% in men; and 1379 non-CVD deaths (13.8%), 48.1% in men. Unadjusted mortality rates in deaths per 1000 person-years are presented in Table 3. A J-shaped pattern in men for all-cause and CVD mortality and a U-shaped pattern in women for all-cause and non-CVD mortality were observed (Table 3 and Figure S1). The lowest mortality rate in both men and women for all-cause mortality was quartile 2, corresponding with a Hct range of 0.421 to 0.440 and 0.381 to 0.400, respectively.
All-cause mortality rates for men and women by Hct quartiles are presented as Kaplan-Meier plots (Figure S2). The median survival rates for each quartile 1 to 4, in men and women, were 24.10/29.94/26.56/24.93 and 34.08/34.76/32.79/28.38 years, respectively (log-rank P value for both men and women <0.001). A similar pattern was observed for cardiovascular and noncardiovascular mortalities.
Using Hct as a continuous variable, a 1% increase in Hct augmented the overall adjusted event rates for all-cause mortality (hazard ratio [HR], 1.15 [95% CI, 1.05–1.26]), CVD mortality (HR, 1.21 [95% CI, 1.08–1.37]), IHD (HR, 1.3 [95% CI, 1.10–1.53]), and borderline for stroke mortality (HR, 1.28 [95% CI, 1.00–1.65]) but not non-CVD mortality (HR, 1.05 [95% CI, 0.91–1.21]; Table 4). Considering Hct as a categorical variable, the relationship between Hct quartiles and all-cause mortality showed a U-shaped pattern in women and a J-shaped pattern in men after adjusting for all baseline risk factors (P=0.02 and 0.001 in men and women, respectively; Table 4 and Figure S1). In men, compared with quartile 2, the lowest quartile was associated with a 25% (HR, 1.25 [95% CI, 1.04–1.50]) increased risk of CVD mortality, and higher quartiles showed a linear increase in risk of CVD mortality of 21% (HR, 1.21 [95% CI, 1.01–1.46]) and 28% (HR, 1.28 [95% CI, 1.08–1.53]) higher risk for quartiles 3 and 4, respectively (P=0.029). This reflected the nonlinear pattern for all-cause mortality (J-shaped) in men but not for non-CVD mortality (Table 4). In women, quartiles 2 and 3 showed no difference in risk for any of the causes of mortality (U-shaped). However, the highest and the lowest quartiles showed, respectively, 17% (HR, 1.17 [95% CI, 1.01–1.36]) and 19% (HR, 1.19 [95% CI, 1.04–1.37]) increased risk of all-cause mortality (P=0.001) and 47% (HR, 1.47 [95% CI, 1.18–1.84]) and 25% (HR, 1.25 [95% CI, 1.00–1.55]) higher risk of non-CVD mortality (P<0.001). Hct did not predict CVD mortality in women. To fully examine the nonlinear relationship between Hct and mortality, we examined the association of Hct, represented as a cubic spline, with CVD and non-CVD mortality in men and women and show results consistent with the primary analyses as demonstrated in Figure 1. To address reverse causality and to also demonstrate that the baseline Hct of our study population was stable, we repeated the analysis in a subset of the population who had ≥3 Hct readings in the first 5 years of follow-up (n=4477). The coefficient of variation of Hct in 94% (n=4122) of this subgroup was ≤0.1. The same pattern of results was seen in the subgroup overall and in those with coefficient of variation ≤0.1 (Tables S1 and S2).
Responders and Nonresponders
In our study population, 2297 and 2218 subjects were classified as nonresponders and responders, respectively. There was a longitudinal increase in BP across Hct quartiles 3 and 4 in men and women over the first 5-year follow-up period. Compared with Hct quartile 2, the mean increase in SBP in men was 1.29 mmHg (95% CI, 0.35–2.23) and 3.22 mmHg (95% CI, 2.30–4.14), respectively, in Hct quartiles 3 and 4. In women, the mean SBP increased by 3.79 mmHg (95% CI, 2.85–4.74) in the Hct quartile 4 in comparison with quartile 2 (Figure 2). Among male responders and nonresponders, quartile 2 was associated with the lowest mortality and quartile 4 associated with the highest mortality risk. In women, the relationship was more linear, with the quartiles 1 and 4 associated with the lowest and the highest risks, respectively. The Kaplan-Meier plots illustrating this for cardiovascular mortality are presented in Figure 3, and the multivariate-adjusted survival analyses are presented in Table S3. The time to reach target BP among responders was similar across Hct quartiles on bivariate analysis (log rank test P=0.072 and P=0.654 for men and women, respectively) and multivariate analysis adjusting for covariates significant at P<0.1 (P=0.204 and P=0.280 for men and women, respectively).
Our study among hypertensive patients attending a secondary BP clinic suggests that Hct is an independent predictor of all-cause, CVD, IHD, and non-CVD mortality. The Hct ranges with the lowest mortality risk in men and women are 0.421 to 0.44 (quartile 2; J-shaped) and 0.381 to 0.420 (quartiles 2,3; U-shaped), respectively. In absolute Hct values, Hct >0.44 in either men or women is associated with increased risk of death. However, at the lower end of the scale, in men risk of death increases at approximate Hct <0.42, whereas in women the risk of death increases only at a lower threshold of Hct <0.38.
Hct, a major determinant of whole blood viscosity, is associated with an increased risk of IHD, stroke, and all-cause mortality in subjects with and without preexisting CVD.7,15–18 However, in some studies, after adjustment for multiple risk factors, Hct failed to significantly predict all-cause or CVD mortality events.15,17 We show a nonlinear relationship between Hct and mortality in our hypertensive cohort, and this is consistent with several other reports of nonlinear relationships between Hct and mortality outcomes in more general populations.6,8
Furthermore, we show that men and women also show differences in causes of Hct-related mortality. Quartiles of Hct predict CVD mortality in men and non-CVD mortality in women. The reason for this apparent dichotomy is unclear. There is an abundance of evidence linking Hct with cardiovascular events.3,7,9,17,18 Relatively high effect sizes in the association between CVD mortality and Hct observed in our study are consistent with previous reports.3,9 In women, one can speculate that a generally lower Hct in women may result in reduced plasma viscosity and, thus, a lower risk for IHD and stroke. Estrogen has been linked with platelet aggregation19 and smooth muscle cell proliferation20,21 and in combination with a high Hct may predispose to increased morbidity. As observed previously, Hct is a more potent risk factor in younger women,6,8 a group who typically have higher estrogen levels. A lower fibrinogen and plasma viscosity were observed in older women (aged 52–65 years) taking hormone replacement therapy, which would suggest that estrogen is only partly responsible for mediating the effects of Hct.22 We show that the different reference ranges of Hct in men and women influence mortality in a complex manner and can explain the variable results in previous studies. Further studies in larger cohorts will help disentangle the relationship between sex and CVD versus non-CVD mortality in men and women.
Our lack of significant association between stroke death and Hct may be attributable to a relative lack of deaths in this group, although >450 people did experience a stroke in our cohort. Inconsistent findings have been reported previously. In an analysis of 2077 people who had experienced a stroke,23 high mean Hct predicted lacunar stroke in people with an SBP >150 mmHg. However, in the Framingham Heart Study participants, women in the age group of 65 to 94 years showed an approximately U-shaped relationship between Hct quartiles and stroke mortality.6 In another prospective study of 7735 men aged 40 to 59 years at baseline, an Hct of >0.51 was found to be associated with fatal and nonfatal stroke.10 However, these studies failed to include diastolic BP, a strong independent predictor of stroke, as a covariate in their analysis models in comparison with our study.
Epidemiological studies have shown a relationship between Hct and BP in both normotensive and hypertensive subjects.11,24,25 Hct was positively correlated with SBP (r=0.085, P<0.01 and r=0.264, P<0.001 for men and women, respectively) and diastolic BP (r=0.214, P<0.001 and r=0.266, P<0.001) in a mixed hypertensive and normotensive cohort.24 Because blood viscosity (and, thus, Hct) and vascular resistance affect total peripheral resistance to blood flow, which is abnormally high in essential hypertension, these results are not unexpected.11,26–28 We show that higher baseline Hct is associated with higher on-treatment BP during follow-up in our hypertensive cohort, and this may partly explain the higher risk associated with higher Hct. However, despite this relationship, baseline Hct did not affect the time to reach target SBP or diastolic BP in our cohort. The increased risk of death because of higher Hct was seen in men and women irrespective of their achievement of target BP, indicating that the risk is independent of the effect of Hct on BP. We are limited by the scope of our data to determine whether more aggressive management of BP in the higher Hct groups with earlier time-to-target BP would have reduced the risk.
The main strengths of our study are the prospective design, long-term follow-up, high event rates, almost equal numbers of men and women, the outcomes based on fatal end points, and the effect sizes adjusted for multiple confounding factors, including baseline preexisting disease conditions. However, our study participants were taken from a hypertensive, overwhelmingly white population, and, therefore, the findings may not be generalizable to other populations. In addition, although we have attempted to adjust for changes in management of hypertensive individuals, by stratifying according to epochs of attendance, bias may still be present. Although we have detailed follow-up BP data, adherence data are lacking in our cohort, and this is a limitation that will need to be addressed in future studies. Furthermore, a stated limitation must be to highlight that inferences about mortality relating to Hct measures taken many years before mortality occurred may be subject to bias. However, the variation in Hct observed in a subgroup of population with repeat Hct measurements was relatively small, and the results were consistent even in a subgroup of individuals with coefficients of variation of Hct ≤0.1.
Our findings suggest that, in the assessment and management of newly diagnosed hypertensive patients, Hct levels should be taken into consideration as an important risk predictor, and further research is needed to elucidate the mechanism of this risk and to define management strategies. In the meantime, patients with Hct levels outside of the sex-specific reference ranges identified in this study should be targeted for more aggressive BP and cardiovascular risk reduction treatment.
Baseline Hct is a routinely measured inexpensive biomarker that influences follow-up BP and is an independent predictor of mortality in the hypertensive population. We identify differences in mortality between men and women that have not previously been known. Hypertension and cardiovascular risk factor management should be tailored based on Hct to improve outcomes.
We thank the patients and staff at the Glasgow Blood Pressure Clinic at the Western Infirmary in Glasgow and National Health Service Greater Glasgow and Clyde. P.J. is supported by a Wellcome Trust Capacity Strengthening Strategic Award to the Public Health Foundation of India and a consortium of United Kingdom universities.
S.P. and L.P. conceived, designed, analyzed, and wrote the article. P.J. analyzed and wrote the article. L.M. and J.H. wrote the article. J.M. provided statistical advice. P.H., M.W., J.D., P.M., G.C.J., S.M., A.F.D., G.L., and G.T.M. critically reviewed the article. All of the authors read and approved the final version of the article.
This paper was sent to Gerald F. DiBona, Consulting editor, for review by expert referees, editorial decision, and final disposition.
The online-only Data Supplement is available with this article at http://hyper.ahajournals.org/lookup/suppl/doi:10.1161/HYPERTENSIONAHA.112.191510/-/DC1.
- Received January 17, 2012.
- Revision received February 6, 2012.
- Accepted June 16, 2012.
- © 2012 American Heart Association, Inc.
- Guyton A,
- Richardson T
- Lowe G
- Vazquez BY,
- Martini J,
- Tsai AG,
- Johnson PC,
- Cabrales P,
- Intaglietta M
- Danesh J,
- Collins R,
- Peto R,
- Lowe GD
- Fowkes FG,
- Lowe GD,
- Rumley A,
- Lennie SE,
- Smith FB,
- Donnan PT
- Rosenson RS,
- McCormick A,
- Uretz EF
- Tzoulaki I,
- Murray GD,
- Lee AJ,
- Rumley A,
- Lowe GD,
- Fowkes FG
- Miller VM
- Fitzpatrick LA,
- Ruan M,
- Anderson J,
- Moraghan T,
- Miller V
- LaRue L,
- Alter M,
- Lai SM,
- Friday G,
- Sobel E,
- Levitt L,
- McCoy R,
- Isack T
- Cirillo M,
- Laurenzi M,
- Trevisan M,
- Stamler J
- Sharp DS,
- Curb JD,
- Schatz IJ,
- Meiselman HJ,
- Fisher TC,
- Burchfiel CM,
- Rodriguez BL,
- Yano K
Novelty and Significance
What Is New?
In a large hypertensive population, both men and women demonstrated a nonlinear association between Hct and mortality.
Men showed a J-shaped association between Hct and cardiovascular mortality.
Women showed a U-shaped association between Hct and noncardiovascular mortality.
What Is Relevant?
Hematocrit may be a useful prognostic marker of future death in hypertensive populations.
Future studies should investigate both causation and the effect of treatment in hypertensive people in relation to their Hct.
Hematocrit should be measured in hypertensive people, and those likely to be at high risk of future death should be identified as high-risk individuals, potentially warranting more aggressive treatment.