Leukocyte count and incidence of subarachnoid haemorrhage: a prospective cohort study

Between 1974 and 1992, a screening program with the aim to identify individuals at high risk for cardiovascular disease, the Malmö preventive project, was conducted in Malmö, Sweden [16]. Complete birth cohorts, born between 1921 and 1949, were invited to a health examination including a physical examination, a panel of laboratory tests and a self-administered questionnaire. A total of 22,444 men were examined from 1974 to 1984 and 10,902 women were examined between 1977 and 1992. The Health Service Authority of Malmö approved the screening program, and all participants gave informed consent. Linkage with the national hospital discharge register and the causes of death register was approved by the ethics committee at Lund University.

In the Malmö preventive project, leukocyte count was measured in participants invited between 1974 and 1981. During this period, men born in 1921, 1926–1940, 1942, 1944, 1946 and 1949 were invited, and women born in 1926, 1931, 1938, and 1949. A total of 19,930 subjects, 2,732 women and 17,198 men, participated. Participation rate was 71% both for men and women.

Subjects with a history of stroke (n = 10), and those with missing information about systolic blood pressure (n = 14), total cholesterol (n = 28) and diabetes (n = 81) were excluded. We also excluded seven individuals with leukocyte counts >20.0 × 10⁹ cells/L as extreme outliers from the population distribution, since they may have had a subclinical hematologic disease or laboratory error at baseline [12]. A total of 19,794 participants (2,711 women and 17,083 men) with a mean age ± SD of 44 ± 6 years remained for the analysis. Mean age in women was 42 ± 9 years (range 28–54) and in men 44 ± 5 years (range 27–61).

Blood pressure (mm Hg) was measured twice in the right arm after 10 minutes of rest. The average of two measurements was used. A sphygmomanometer and a rubber cuff of appropriate size were used. Height (cm) and weight (kg) were measured under standardized conditions and body mass index was calculated as weight/height² (kg/m²). Current smoking habits, alcohol consumption and use of antihypertensive medication were evaluated in a questionnaire. Subjects were categorized in terms of smoking as never smokers, former smokers (i.e. non-smokers who previously had been smoking daily for at least 6 months) and current smokers (i.e. daily smoking).

Alcohol abuse was assessed by means of the modified shortened version of the Michigan Alcoholism Screening Test [17]. Subjects with more than two (of nine) affirmative answers were considered to have high alcohol consumption. Information about previous myocardial infarction (MI) before screening was based on the questionnaire and data from the Swedish hospital discharge register. Subjects who confirmed a doctor’s diagnosis of angina pectoris or who used nitrates were considered to have angina pectoris (AP). In the present study prevalent MI and AP was combined into one variable; history of MI or AP. Physical inactivity in spare time was assessed using the question: ‘Are you mostly engaged in sedentary activities in spare time, for example, watching TV, reading, going to the movies?’.

Blood samples were taken after an overnight fast. Leukocyte count was determined with automatic counters in accordance with standard methods at the hospital laboratory. Serum cholesterol and blood glucose were analysed with standard methods. Diabetes was defined as fasting whole blood glucose > 6.1 mmol/L or self-reported treatment for diabetes. Forced expiratory volume in one second (FEV1) was measured in a subsample (n = 18,607, 94%) of the present cohort, using spirometry. The methods for the lung function assessment and standardization of FEV1 to percentages of the individuals predicted value have been described elsewhere [18].

All subjects were followed from the baseline examination until first SAH, death, emigration from Sweden or December 31, 2008, whichever came first. SAH cases were identified with the Malmö stroke register, the Swedish hospital discharge register or the causes of death register (ICD 8 and 9 code 430, and ICD 10 code I60), as previously described [18]. These registers provide complete coverage of hospital discharges and deaths in Sweden. All diagnoses are made by a physician at hospital discharge or death of the patient, and contra-signed by board-certified specialists [19].

Leukocyte count was categorized into sex-specific quartiles with equal proportions of women in each quartile, in order to adjust for the relationship between sex and SAH. One-way analysis of variance (for continuous variables) and logistic regression (for categorical variables) were used for cross-sectional analyses of the risk factor distributions at baseline across quartiles of leukocytes. Cox proportional hazards regression was used to calculate risk factor-adjusted hazard ratios (HR) and 95% confidence intervals (CI). Leukocyte concentration was evaluated as a continuous variable (per 1-SD increase) and in sex-specific quartiles. Unadjusted results and results from a multivariable model adjusted for the categorical variables sex, smoking status, antihypertensive medication, high alcohol consumption, diabetes and physical inactivity, and the continuous age, systolic blood pressure, body mass index, cholesterol, are presented. To explore whether FEV1 or a history of MI/AP, explained the relationship between leukocytes and SAH, these covariates were entered separately to the multivariable model. Covariates were selected because they have shown associations with SAH [5] and systemic inflammation [20], respectively, and thus are potential confounders of the association between SAH and leukocytes. The fit of the proportional hazards model was checked visually by plotting the incidence rates over time, and by plotting Schoenfeld residuals as a function of time and statistically testing for a non-zero slope. Possible interactions between leukocyte count and each covariate, respectively, were checked by interaction terms added to the full model. The Cox regression analyses were also performed separately in current, former and never smokers, and in women and men. Harrell’s C statistics was used to estimate the ability of the risk factor adjusted Cox model to predict SAH.

Baseline characteristics by quartiles of leukocytes are described in Table 1. Leukocyte count was significantly associated with systolic blood pressure, current smoking, high alcohol intake, body mass index, total cholesterol, diabetes and physical inactivity.

During the mean follow-up of 26.8 ± 6.9 years (530,562 person years), ninety-five subjects (78 men and 17 women) had a SAH, corresponding to a crude incidence of 17.9 per 100,000 person years (17.2 in men and 22.1 in women). Mean age at SAH was 58.1 ± 11 years and mean time from screening to SAH was 15.1 ± 8.4 years.

Higher leukocyte concentration was significantly associated with higher incidence of SAH. Per 1 SD (2.01 × 10⁹ cells/L) increase of leukocytes, the crude HR (95% CI) for SAH was 1.30 (1.09-1.54), p = 0.003, and the HR after adjustment in the multivariable model 1.26 (1.05-1.53), p = 0.014 (Table 2). After additionally adjusting for FEV1 (in % of the predicted values) HR was 1.25 (1.04-1.52), p = 0.020, per SD increase of leukocytes. Further adjusting for history of MI/AP did not substantially change the result.

To explore if the relationship between leukocytes and SAH changed with time, the follow-up was divided at the median follow-up time for those with SAH (14.6 years) and separate models was fitted for the two intervals. The HR was 1.33 (1.03-1.67) and 1.28 (1.00-1.64), respectively, in the first and second time period (p for interaction with time 0.9).

There were no statistically significant interactions between leukocytes and other variables in the analysis. In subgroup-analysis, the association between SAH incidence and leukocyte concentration (per 1 SD) remained significant in men (p = 0.021) and in smokers (p = 0.011) after full adjustment. The full multivariate model could not be applied in all subgroups due to the limited number of events (Table 2).

The association between leukocyte count and incident SAH was significant in a dose–response manner when leukocyte concentration was analysed in sex-specific quartiles (unadjusted p for trend = 0.010), Table 2 and Figure 1.

Adding leukocyte concentration (continuous/per 1 SD) to the other variables in the full multivariable model increased the SAH predictive ability of the model (c statistics 0.6248 vs. 0.6439).