Article Text
Abstract
Background Previous studies, all of <20 years of follow-up, have suggested an association between lung function and the risk of fatal stroke. This study investigates the stability of this association in a cohort followed for 4 decades.
Methods The Bergen Clinical Blood Pressure Survey was conducted in Norway in 1964–1971. The risk of fatal stroke associated with forced expiratory volume after one second (FEV1) was estimated with Cox proportional hazards regression, making progressive adjustment for potential confounders.
Results Of 5617 (84%) participants with recorded baseline FEV1, 462 died from stroke over 152 786 subsequent person-years of follow-up according to mortality statistics of 2005; mean (SD) follow-up was 27 (12) years. An association between baseline FEV1 (L) and fatal stroke was observed; HR=1.38 (95% CI 1.11 to 1.71) and HR=1.62 (95% CI 1.22 to 2.15) for men and women, respectively (adjusted for age and height). The findings were not explained by smoking, hypertension, diabetes, atherosclerosis, socioeconomic status, obstructive lung disease, physical inactivity, cholesterol or body mass index and persisted in subgroups of never-smokers, subgroups without respiratory symptoms and survivors of the first 20 years of follow-up. For male survivors with a valid FEV1 at follow-up (1988–1990) (n=953), baseline FEV1 (L) indicated a possible strong and independent association to the risk of fatal stroke after adjustments for individual changes in FEV1 (ml/year) (HR 1.95 (95% CI 0.98 to 3.86)).
Conclusion There is a consistent, independent and long-lasting association between lung function and fatal stroke, probably irrespective of changes during adult life.
- Cerebrovascular disease
- epidemiology
- long-term studies
- lung function
- risk factors
- geriatrics
- lifestyle
- mortality
Statistics from Altmetric.com
- Cerebrovascular disease
- epidemiology
- long-term studies
- lung function
- risk factors
- geriatrics
- lifestyle
- mortality
Introduction
Cerebral stroke is the second most frequent cause of death in the industrialised world, as it is in Norway,1 ,2 and it is the leading cause of acquired and permanent disabilities. Adequate risk assessment for stroke and prophylactic interventions are essential to reduce the incidence and mortality of this devastating condition.
The associations between lung function, respiratory symptoms and stroke are still under discussion.3 ,4 The relationship between lower lung function and risk of stroke has previously been reported for both stroke incidence and stroke mortality,5 ,6 and the association is independent of other risk factors including smoking.7 Asthma has been related to stroke and macrovascular disease.8 ,9 Some of the previous studies in this field were not population based and included men only.5 ,10 The largest population-based study reported a response rate of 73.6%. None of the studies published so far have been able to study the robustness of the association after more than 20 years of follow-up, and all previous studies on this relationship are based on one baseline measurement of lung function only.11 ,12 The robust relationship between lung function and vascular events may be due to a common, unrecognised offending factor (fetal or lifetime exposure) that affects both forced expiratory volume after one second (FEV1) and the vascular system, or it may be due to a causally linked process.11 If the risk of stroke was more strongly related to the rate of declining lung function than with the baseline lung function levels, this would suggest that the responsible mechanisms are mainly operating in adult life.12
The aims of the present study are, in a general Norwegian population with both sexes represented, a long-term follow-up and a high response rate, to assess the independent association between baseline lung function as well as that of longitudinal changes in lung function and the risk of fatal stroke.
Methods
Study population
The study cohort and methods have been extensively described elsewhere.13 ,14 A random population-based sample of 6811 subjects aged 22–75 years was invited to the Bergen Clinical Blood Pressure Survey from 1965 to 1971. Sixty-seven persons were excluded because they had died, and six were excluded because they had emigrated before the screening took place, leaving 6738 eligible subjects. The study participants and non-participants have been compared in a previous publication.14 In total, 5653 subjects participated in the study, of whom, 3119 (85%) were women and 2534 (83%) were men (p<0.05). The mean age was 47.0 and 47.5 years in the participants and non-participants, respectively (p=0.26), but non-participants died at younger age for both genders. The number of years of life lost at median survival for non-participants compared with participants was six for men and four for women (both p<0.001, log-rank test, Kaplan–Meier plot). In 1988–1990, the survivors of men initially aged 22–54 years and who still lived in Bergen (n=1154) were invited to participate in a follow-up survey.15 The cohort is assumed to consist exclusively of Caucasians. The aim of the baseline survey was to examine the prevalence of smoking and cardiovascular risk factors in a general population and the long-term effects from these. The local committee of medical ethics and the institutional review board has approved the study.
Independent variables
Lung function
Details of the forced expiration manoeuvres have been described elsewhere.15 Briefly, dry-wedge bellow spirometers (Vitalograph) were used both at baseline (P-model)16 in 1965–1971 and at follow-up in 1988–1990 (S-model). Trained technicians recorded the highest values of FEV1 and forced vital capacity (FVC) from at least two acceptable attempts. All values used in the analyses were corrected to body temperature and pressure-saturated conditions.15 Baseline levels of FEV1 and FVC were expressed in litres, whereas the (longitudinal) changes in FEV1 and FVC between baseline and follow-up in 1988–1990 were expressed as millilitres per year. The predicted values of FEV1 and FVC were calculated using prediction equations recently derived and published from the same geographic area.17 As a supplement, we performed standard adjustments for FEV1 and FVC by expressing these variables as percentages of their respective predicted values (FEV1% and FVC%, respectively) and dichotomised around the median. Information regarding bronchitis and asthma was obtained from self-reports. Chronic obstructive pulmonary disease (COPD) was defined from the spirometric measurements as FEV1/FVC<70%.
Variables for adjustment
An extensive array of variables was recorded for each participant at baseline. We first tested the association between lung function and stroke for confounding from the following baseline variables: sex, age, body height, hypertension, diabetes, pre-existing atherosclerosis, self-reported bronchitis and/or asthma, self-reported dyspnoea, smoking habits, socioeconomic status, physical inactivity, body mass index and cholesterol. Information on exposure to passive smoking in childhood or ‘in utero’ was not available.
Blood pressure (BP) at baseline was measured using a Mercury Sphygmomanometer, model Mark3, designed by Rose et al.18
Hypertension was defined as a systolic BP of 140 mm Hg or more or a diastolic BP of 90 mm Hg or more.
The diagnosis of atherosclerosis was based on a combination of the interviewing medical officer's conclusions (after interviewing the respondents regarding their medical history and symptoms of intermittent claudication, angina pectoris and myocardial infarction) and information from the questionnaires. Information on previous transient ischaemic attacks was not available.
Current smoking was defined as the consumption of at least one cigarette per day, 10 g of pipe tobacco per week or 1 cigarette per week for at least 1 year. A smoker who had stopped smoking more than 1 month before the baseline examination was regarded as a former smoker. The smoking habits were divided into five groups: lifetime non-smokers, former smokers, current smokers of pipe/cigars, current smokers of 1–9 cigarettes per day and current smokers of 10 or more cigarettes per day.19
Socioeconomic status was categorised according to the British Registrar General's classification of occupations20: white collar 1 and 2 (upper), blue collar 1 and 2 (lower) and ‘others’ (including students, housewives, retired etc.).
Physical activity was dichotomised into active or inactive in which inactivity was defined as physical activity less than Sunday strolls or gardening.
Bronchitis and/or asthma only present before 15 years of age was defined as childhood asthma/bronchitis (yes/no) and asthma and/or bronchitis present after 15 years of age was defined as adult asthma/bronchitis (yes/no).
Dyspnoea was recorded on a scale from 0= no respiratory difficulties to 7= constant dyspnoea, based on self-report and physician interview conclusion. We dichotomised the dyspnoea information into no (0) and any kind of dyspnoea (1).
Serum (s) glucose and cholesterol were measured according to the Lipid Research Clinics Programme, using a colorimetric three-channel auto analyser (Technicon Auto Analyzer II).21 Diabetes was defined as either being known from an individual's medical history or being indicated by an incidental blood glucose level above 11.0 mmol/l.
Ascertainment of outcome
The outcome of interest was fatal stroke reported as the underlying cause of death from the death certificates. When a registered Norwegian dies inside Norwegian borders, a death certificate is issued by a physician and sent to Statistics Norway. The registered underlying cause of death is the disease (or injury) that initiates the series of events that eventually leads to death.22 The physician issuing the death certificate uses available information from hospital records, clinical postmortem examination or other sources to complete the form. Statistics Norway (the country's compiler and keeper of all national statistics) has linked information on emigration, time and the underlying cause of death from September 1965 until December 2005 to the data file using the National Identity Numbers for all the invited subjects. The causes of death were grouped into the 65 causes in the ‘European shortlist’ (Eurocodes) which are used in the European official mortality statistics.23 Eurocode 36 denotes fatal stroke and includes all stroke subtypes. The codes from International Classifications of Diseases (ICD codes) used to identify fatal strokes are listed in table 1.
The validity of fatal cerebral strokes reported as the underlying cause of death in mortality statistics for this cohort had a Cohen's κ coefficient of 0.79, sensitivity of 0.75 and a positive predictive value of 0.87 by comparison to autopsy findings.24 Data on non-fatal strokes were not available.
Statistical analyses
The strength of the association between lung function and fatal cerebral stroke was assessed using Cox proportional hazards models. We split the baseline FEV1 and the change in FEV1 (longitudinal) into quartiles and carried out univariate analyses. Both these analyses showed an approximately linear trend in the association between lung function measurement and fatal stoke, enabling us to investigate the FEV1 measures further as continuous variables (the HRs calculated are per litre lower level in FEV1 for baseline data and per millilitre per year decrease in FEV1 across longitudinal data).
The Cox assumption of proportional hazards over time was assessed by plotting partial residuals from the Cox analysis of the FEV1-stroke association against time of follow-up. The association was also studied in the first and second halves of the follow-up time separately.
Interactions between FEV1 and other explanatory variables were included one by one and excluded from the model if not statistically significant at the 5% level.
We then performed univariate analyses of multiple variables potentially affecting the association between lung function and fatal stroke and additionally univariate analyses of a FEV1% (<median vs ≥median), FVC% (<median vs ≥median) and FEV1/FVC <0.70 (yes/no), with χ2 tests and t tests used as appropriate.
Age, body height and sex are the variables included in the standard prediction models of both FEV1 and FVC.17 These variables were therefore chosen to be included in the adjusted Cox proportional hazard model. The assessment of confounding beyond age, body height and sex was accomplished by comparing the effect estimate obtained from calculations including just FEV1, sex, age and height with the estimate obtained after the additional inclusion of each of the selected potential confounders.25 Variables that were significant on a 5% level and that changed the estimate of the effect of FEV1 by more than 10% were included in the model.
Subgroup analyses were also performed in lifetime non-smokers (never/ever), in a subgroup of participants reporting no subjective dyspnoea (yes/no) and in a subgroup without diabetes, hypertension or history of atherosclerosis (neither/any), adjusting for sex, age and body height only (the main model).
A multiple regression model was finally carried out to illustrate the effect of further adjustments, including FEV1, sex, age at baseline, body height, smoking habits (five categories), hypertension (yes/no), diabetes (yes/no), serum cholesterol (cut point of median), presence of atherosclerosis (yes/no) and physical inactivity (yes/no).
Finally, we studied the subgroup of participants with recorded measurements of FEV1 both at baseline and in 1988–1990. This enabled us to compare the strength of the association for baseline data and for longitudinal data, respectively, and the effect of corresponding adjustments. Score tests were used to assess model fit for baseline data compared with change over time. Analyses were computed with SPSS (V.15.0) software.
Results
Of the 6738 eligible subjects, 5653 (84%) participated in the baseline survey and 5617 (83%) provided valid FEV1 measurements and hence were analysed further in this study. The study participants lived longer, and the proportion of fatal stroke was (467/3597) 13.0% among the study participants and (81/790) 10.3% in the non-participants who had died during follow-up (OR=1.31 (95% CI 1.02 to 1.69), adjusted for age at baseline and sex and OR=1.14 (95% CI 0.88 to 1.48), adjusted for age at death and sex).
Among the participants, the percentage of cases with recorded data on FEV1, FVC, sex, age, body height, BP, s-cholesterol, smoking habits, self-reported breathlessness, atherosclerosis, socioeconomic status, history of bronchitis and asthma, diabetes and physical inactivity varied between 96% and 100%. Blood glucose measurements were done for 3509 participants (62%). Of the male survivors initially aged 22–54 years (n=1316), altogether 134 men had moved out of the study area by follow-up in 1988–1990. Therefore, 1154 men were invited at follow-up and 1032 (89%) participated and 953 (83%) provided valid spirometric measurements at both baseline and follow-up.
The total follow-up time for the cohort with recorded data on FEV1 was 152 786 person-years, and the mean±SD follow-up time was 25±12 and 29±11 years for men and women, respectively. By the end of 2005, 1714 (68%) men and 1850 (60%) women had died. From these, 186 (10.9%) men and 276 (14.9%) women had died from a stroke (Eurocode 36).23 The characteristics of the study cohort are presented in table 2.
Since FEV1 showed a closer association with fatal stroke than FVC in the preliminary univariate analyses, we chose to continue the calculations with FEV1 only (table 3). The association between stroke and level of FEV1 was initially calculated separately for men and women using Cox proportional hazards regression models with progressive adjustments for age, body height and different potential confounders. Age, body height and sex are the variables used in the prediction models for FEV1.17 In the association between lung function and risk of fatal stroke, the unadjusted HR (95% CI) was 2.6 (2.3 to 2.9), the height-adjusted HR (95% CI) was 3.3 (2.9 to 3.8), the age-adjusted HR (95% CI) was 1.1 (1.0 to 1.3) and the sex-adjusted HR (95% CI) was 3.5 (3.1 to 3.9). Since the age adjustment resulted in a marked drop in the HR for FEV1, we also studied the lungstroke association in different strata of age at baseline (tertiles). The association between FEV1 and risk of fatal stroke did not differ significantly between age groups. We found HR=2.41 (95% CI 1.18 to 4.91) in the age group <40 years at baseline, HR=1.54 (1.13 to 2.09) in 40–55 years, and HR=1.41 (1.14 to 1.77) in 56+ years (adjusted for sex and height). The HRs of the FEV1 × age (both continuous) and FEV1 × sex interaction terms tested separately in the main model were not statistically significant. We adjusted for sex instead of carrying out sex-stratified analyses in order to maintain power. After adjustments for sex, age and body height, the other confounding variables changed the FEV1-related HR for a fatal stroke <10% and thus were not included in the main model.
The calculations of the main model were also carried out in a subgroup of lifetime non-smokers (n=2317), in order to eliminate potentially unrecognised residual confounding from smoking, resulting in HR=1.88, 95% CI (1.39 to 2.55) per litre lower level in FEV1. We also performed subgroup analyses in participants without dyspnoea (n=4366) resulting in HR=1.47, 95% CI (1.16 to 1.86) per litre lower level in FEV1. This was done in order to investigate whether the lung function measurement would add information to stroke risk assessment in subjects free from respiratory symptoms. Finally, we did subgroup analyses of the main model in participants without diabetes, hypertension and a medical history suggesting atherosclerotic symptoms (angina pectoris, intermittent claudication or myocardial infarction) at baseline (n=3507), HR=1.43, 95% CI (1.07 to 1.90). This was done in order to assess whether the relationship between poor lung function and the risk of stroke precedes the presence of atherosclerosis. The HRs associated with FEV1 remained remarkably stable irrespective of the subgroups studied.
A multiple regression analysis including variables associated with fatal stroke in the univariate analyses was finally carried out to illustrate that none of the confounding factors reported in other research could explain the association of interest.
To minimise any bias from disease prior to baseline that could possibly lead to both reduced lung function and increased mortality from stroke, we performed a sensitivity analysis where events from the first 5 years of the follow-up period were excluded. This changed the HR, associated with 1 litre lower level in FEV1 across baseline data, from HR=1.38, 95% CI (1.11 to 1.71) to HR=1.30 (1.02 to 1.66) and from HR=1.62 (1.22 to 2.15) to HR=1.61 (1.20 to 2.15) for men and women, respectively, indicating that this type of bias has not affected the results.
Previous studies have shown that the impact of risk factors may differ between successive time periods.26 We calculated the hazard related to the early (0–19 years) and the late (20–40 years) of the follow-up period and found that these were comparable (table 4). (The corresponding results for the entire follow-up period are given in table 3, Main model.)
Finally, we explored the association between FEV1 and stroke risk, after adjustments for changes in lung function during adult life. The final analyses were performed on a subgroup of 953 men who had recordings on FEV1 at both baseline and the follow-up examination in 1988–1990. The mean FEV1 (SD) at baseline in this subgroup was 4.42 (0.83) l. The mean decline in lung function per year between the baseline examination and the follow-up examination in 1988–1990 was 53 (19) ml/year. Of the 953 men who had complete sets of data on FEV1 both at baseline and in 1988–1990, 23 fatal strokes were recorded in the mortality statistics in the follow-up period running from the second FEV1 measurement (1988–1990) until end of follow-up (2005). The association between the two different lung function measures and fatal stroke was explored using four different models. Model 1: unadjusted; Model 2: including only baseline FEV1, age and body height; Model 3: including only change in FEV1 per year between baseline and the follow-up survey in 1988–1990, age and body height and Model 4: including both baseline FEV1, change in FEV1 per year between baseline and the follow-up survey in 1988–1990, age and body height (table 5). We found a better model fit with the baseline level of lung function than with the change in FEV1 (likelihood ratio test).
Discussion
We found a robust and long-lasting association between lung function and fatal stroke in both men and women. Our study adds further information on this issue because of the very long and complete follow-up period, high response rate, adjustments for a wide range of possible confounding factors, the minimisation of bias from ill-health and the subgroup analyses that show robust persistent associations in lifetime non-smokers, in participants reporting no respiratory difficulties and in those who had no signs of atherosclerosis at baseline.
Our analyses shed light on some possible explanations for the association between lung function and the risk of vascular events. The results weigh against a hypothesis that reduced FEV1 may serve as an epiphenomenon for a common environmental offending factor that affects both the pulmonary and the vascular systems, as adjustments for the available possible confounding factors and ill-health bias resulted only in minor changes on the risk estimates associated with FEV1.
Another possible explanation for the association is that an inflammatory link exists between lung processes and cardiovascular disease. Inducing airway inflammation in rabbits can incite and propagate systemic inflammation, which in turn may contribute to the progression of atherosclerosis.27 Low-grade systemic inflammation is a major risk factor for plaque genesis, progression and rupture.28 FEV1 is known to be associated with carotid artery intimae-medial thickness29 and is possibly a marker for smooth muscle hyperplasia. COPD is associated with a systemic inflammatory response, with elevated white blood cell count and levels of C reactive protein, fibrinogen and cytokines, which have the potential to activate the vascular endothelium.11 ,30 The prevalence of COPD (FEV1/FVC<0.7) in our population was, however, too low to explain the findings, and the results were robust in a subgroup of participants who did not report any complaints of dyspnoea. Thus, this explanation is likely only if a systemic inflammatory response is initiated from a minor airway obstruction not even recognised by the individual suffering from it.
Interestingly, there seemed to be a sex difference in the association between FEV1 and early and late follow-up strokes, but the FEV1 × sex interaction term failed to reach statistical significance. Such a difference, if real, might reflect a survivor effect and could hypothetically explain the increased overall stroke rate in women compared with men. A stronger association between FEV1 and stroke in women may be partly explained by sex differences in lung growth and structure and hormonal determinants of airway behaviour.31 For example, women are more susceptible to the effects of smoking than men.32
The apparent lung-stroke association may be due to an unrecognised common offending factor which influences both stroke risk and lung function. In that case, adjustments for longitudinal changes in lung function would be expected to reduce the stroke hazard associated with the baseline lung function. In the subgroup analysis performed in the male survivors in average 23 years after baseline, the risk of fatal stroke was associated both with changes in lung function over time and with the baseline level. The adjusted effect estimate HR=1.95 (p<0.05) associated with 1 litre reduction in FEV1 across baseline data indicated a possible strong and independent association between baseline FEV1 and the risk of fatal stroke (though not formally statistically significant, probably due to lack of power).
The association between baseline lung function and fatal stroke persisted after adjustments for decline in lung function during adult life, which suggests that the lung-stroke association could not be solely explained by confounding mechanisms acting during adulthood.
Strengths and limitations
A major strength of this study is the very long follow-up period and the access to longitudinal data on individual FEV1 changes over time. To our knowledge, this is the first study that reports on the association between lung function measurements and the risk of fatal stroke with a follow-up period longer than 20 years and it is the only study with adjustments for individual FEV1 changes over time. The cohort was randomly selected from a general population of both sexes with an extraordinarily high participation rate (84%). The clinical characteristics of the study population generally resemble those found in the county services in the 1970s.33 The study recorded an extensive number of clinical variables for each of the participants that were used to adjust for potential confounding factors. Our results support those from other recently published population-based studies including both sexes,7 ,34 but in contrast to these, we have used the unadjusted FEV1 measure. Traditionally, FEV1 is adjusted for age and body height to express FEV1 as a percentage of a predicted value. This adjustment hides the independent association between these three variables and stroke, and therefore, we examined the FEV1-stroke association with progressive adjustments for age, height and other variables.12
Despite the very high attendance rate at the initial survey in 1965–1971 (84%), all the participants had to be capable of attending the clinic in person, so frail and ill members of the target population are potentially under-represented. This selection is, however, more likely to deflate than to exaggerate the association between level of lung function and excess risk of fatal stroke.
Information on previous strokes or transient ischaemic attacks was not available from the data records. The generated variable ‘atherosclerosis’ was therefore a surrogate variable based on ischaemic symptoms from the heart and legs only.
There might be some residual confounding due to imprecise measurements of lifetime exposure to tobacco, but our analyses support the view that the association between lung function and stroke is independent of smoking.7 ,34 Air pollution might represent further confounding,35 but the degree of atmospheric pollution in Bergen at this time was very low.36 Occupational exposure to gas and dust was not measured, but socioeconomic status (which is related to occupational exposure) did not influence the association. Residual confounding from undiagnosed diabetes at baseline may be possible, but it seems less probable as the effect estimate (HR) of the lung-stroke association was the same in diabetic and non-diabetic participants, respectively (results not shown). Obstructive sleep apnoea is increasingly described as an independent risk factor for stroke and might also confound the association.37 Seventy-five per cent of the participants who smoked 10+ cigarettes per day were men and 77% were in the lower two tertiles of age at baseline (<56 years). The change in the association between smoking and fatal stroke from the univariate analyses to the multiple regression model is due to the fact that those smoking 10+ cigarettes per day died from other causes and at younger ages than those who died from stroke (competing risks). As an example, the HR of death from ischaemic heart disease (Eurocode 34) among the smokers of 10+ cigarettes per day ischaemic heart disease was 1.39 (1.16 to 1.68), unadjusted. It is conceivable that FEV1 reductions may be due to childhood bronchiolitis, bronchopulmonary dysplasia (although uncommon) or lung growth problems related to intrauterine insults including maternal smoking. Unfortunately, information on these variables was not available.
Finally, our outcome measure is based on fatal stroke diagnoses from the death certificates and not on measures of total strokes. Previous studies have compared risk factors for stroke incidence and stroke mortality.6 ,38 These have concluded that studies with information on stroke mortality are likely to give results applicable to stroke incidence.
We conclude that lung function is consistently, independently and persistently associated with the risk of fatal stroke for both men and women. The mechanism of the association is still unknown, and it might be neither causal nor reversible, but the findings indicate that the association is apparent also prior to changes in FEV1 during adult life. FEV1 as a percentage of a predicted value might therefore be useful in identifying individuals at particular risk of future fatal strokes.
What is already known on this subject
-
An inverse association between lung function and the risk of fatal stroke has been suggested.
What this study adds
-
The association cannot be explained by confounding from smoking, hypertension, diabetes, atherosclerosis, socioeconomic status, obstructive lung disease, physical inactivity, serum cholesterol or body mass index. The association also persists in a subgroup of never-smokers, in a subgroup without respiratory symptoms and in survivors of the first 20 years of follow-up.
-
The association between baseline lung function and fatal stroke persists after adjustments for the changes in lung function during 23 years of adult life, which suggests that the lung-stroke association may be due to a common offending factors acting during fetal or childhood life or a possible causal mechanism.
Acknowledgments
We wish to thank Dr Olav Sulheim for his extensive work of collecting and recording all the clinical variables at baseline examinations.
References
Footnotes
-
Funding The University of Oslo supported the research reported in this paper. The Norwegian Council for Cardiovascular Disease, the WHO and the Research Foundation for Thoracic Medicine, University of Bergen, Norway, gave financial support for the Clinical Survey in Bergen in 1964–1971, data management and quality controls of the files.
-
Competing interests None.
-
Ethics approval Ethics approval was provided by Oslo, Norway.
-
Provenance and peer review Not commissioned; externally peer reviewed.