a Department of Public Health, Erasmus University Rotterdam, The Netherlands.
b Department of Epidemiology and Biostatistics, Erasmus University Medical School Rotterdam, The Netherlands.
c International Centre for Health and Society, Department of Epidemiology and Public Health, University College London, UK.
d Department of Research & Development, Kensington & Chelsea and Westminster Health Authority, London, UK.
e Julius Center for Patient Oriented Research, Utrecht University, Medical School, The Netherlands.
Professor MG Marmot, International Center for Health and Society, Department of Epidemiology and Public Health, University College London Medical School, 119 Torrington Place, London WC1E 6BT, UK. E-mail: m.marmot{at}public-health.ucl.ac.uk
Abstract
Objectives To determine the seasonal effect on all-cause and cause-specific mortality and to identify high-risk groups.
Methods A 25-year follow-up of 19 019 male civil servants aged 4069 years.
Results All-cause mortality was seasonal (ratio of highest mortality rate during winter versus lowest rate during summer 1.22, 95% CI : 1.11.3), largely due to the seasonal nature of ischaemic heart disease. Participants at high risk based on age, employment grade, blood pressure, cholesterol, forced expiratory volume, smoking and diabetes did not have higher seasonal mortality, although participants with ischaemic heart disease at baseline did have a higher seasonality effect (1.38, 95% CI : 1.21.6) than those without (1.18, 95% CI : 1.11.3) (P = 0.03).
Conclusions Seasonal mortality differences were greater among those with prevalent ischaemic heart disease and at older ages, but were not greater in individuals of lower socioeconomic status or with a high multivariate risk score. Since seasonal differences showed no evidence of declining over time, elucidating their causes and preventive strategies remains a public health challenge.
Mortality rates show strong seasonal effects, with all-cause mortality rates highest in the winter.15 Over half of the excess is due to cardiovascular disease with much of the remainder due to respiratory diseases.1,2,4 The mechanisms underlying seasonal variation in mortality are not yet completely elucidated, but may include outside and inside air temperature, wind chill factors, snowfall, sunlight exposure, air pollution, activity pattern, influenza incidence, psychological condition and/or food intake, and their effects on pathophysiological mechanisms related to disease.69
Identification of groups who are at high risk for a seasonal death offers the opportunity to both elucidate potential mechanisms and to design preventive strategies. Previous studies have suggested that the winter excess may be greater amongst people of lower social class10 (who may for example be less able to afford housing insulation or central heating), older people and those with pre-existing health problems.11 The seasonal variation in blood pressure may be greater among smokers compared to non-smokers,12 but few studies have examined whether risk factors for premature morality identify groups with greater seasonal effects. Most previous analyses have examined routinely available mortality data and have therefore had little information characterizing individual risk.
We sought therefore to determine in the Whitehall cohort study of British civil servants the effect of season on all-cause and cause-specific mortality. Furthermore we determined whether seasonality effects were greater in high-risk groups defined on the basis of age, employment grade, pre-existing disease, or multivariate combinations of risk factors.
Subjects and Methods
A total of 19 019 male civil servants aged 4069 years attended the screening examination of the Whitehall study between September 1967 and January 1970. More details regarding design and methods are provided elsewhere.13 In short, each participant filled in a standard questionnaire that included age, self-reported smoking, civil servants' employment grade and symptoms of chest pain and chronic bronchitis. At the screening examination a single blood pressure reading was obtained with the participant seated, blood was drawn for plasma cholesterol estimation, a glucose tolerance test was conducted and a forced expiratory volume in one second (FEV1) was measured.
High-risk groups were defined on the basis of age at death, employment grade, prevalent ischaemic heart disease, prevalent chronic bronchitis, or multivariate combinations of risk factors. Age was categorized as 4064 years, 6574 years and 75 years. Employment grade, a measure of socioeconomic status, was categorized as high grades (administrative, professional and executive) and low grades (clerical and other grades, e.g. messengers and other unskilled workers). For the analyses using employment grade, 886 men from the Diplomatic Service and the British Council were excluded, as their employment status was not comparable with the employment grades above. Prevalent ischaemic heart disease was defined by self-reported angina, prolonged chest pain (pain of possible myocardial infarction'), previous admission to hospital for ischaemic heart disease or an electrocardiogram suggesting ischaemia (any of Minnesota codes 1.13, 4.14, 5.13 or 7.1). Chronic bronchitis was defined as the presence of persistent phlegm (MRC questionnaire). Finally, the participants were classified into groups with differing degrees of risks. For each cause of death a multiple logistic regression model was fitted to our data using the following terms: age, employment grade, systolic blood pressure, plasma cholesterol concentration, FEV1, (adjusted for age and height), smoking habits and the presence of diabetes or glucose intolerance. Four separate risk scores (for all-cause, ischaemic heart disease, cerebrovascular diseases, and respiratory diseases) were computed from the coefficients of these regressions. The participants were ranked according to each of these scores as either low risk (<60th percentile), moderate risk (6080th percentile) or high risk (>80th percentile).
Records from 99.3% men were flagged at the National Health Service Central Registry, which notified us of all deaths up to the end of January 1995. Causes were classified according to the International Classification of Disease, Eighth Revision (ICD-8). The following codes were analyzed: ischaemic heart disease (410414), cerebrovascular disease (430438), other cardiovascular diseases (390404, 420429, 440448), malignancy (140239) and respiratory disease (460519). For 28 people, cause of death was missing and these people were excluded from all analyses. In total 18 841 men were followed up for at least 25 years with 8347 having a known cause of death.
Data analyses
We created an expanded dataset for these analyses in which, for each subject and each individual month of follow-up, a new record was created giving the total days of follow-up during that month. Deaths were allocated to the appropriate month, current age group, calendar year and risk group. This analysis allows for the fact that recruitment into the study took just over 2 years and also for the differing lengths of the months. This expanded dataset was then summarized by computing the total number of deaths from each specific cause and the total person time at risk in these separate categories. Creation of the summary dataset was done using the statistical package SAS.14
Seasonal variation in mortality was modelled assuming that the outcome of interest followed a sinusoidal curve with a period of one year. This curve can be described mathematically using just two parameters: a sine and cosine term. The test of seasonality was computed using a likelihood ratio test with two degrees of freedom by comparing two models, with and without the seasonality terms. The models with the seasonal components were also compared with models showing overall heterogeneity between the 12 months to assess whether the seasonal model described the month-to-month variation adequately. The sinusoidal variations in mortality rates can be summarized using two useful terms; one showing the month of peak incidence and the other showing the estimated ratio of the highest (winter) to lowest (summer) incidence rates. Both these terms can be derived using the coefficients of the sine and cosine parameters and have been used to describe the seasonal effects previously.15
For the analyses of the seasonal effect by age, employment grade, multivariate risk score group and prevalent disease status, the highest (winter) to lowest (summer) mortality rate ratios were assessed for all calendar years combined. Tests for differences in the magnitude of the seasonality effect between risk groups were computed using the more conservative test of heterogeneity, rather than test of trend. In cases where the seasonality effect actually changes monotonically across risk groups, a test for trend would have given a more extreme P-value.
All models for mortality were fitted using Poisson regression with the statistical package GLIM, which was also used to compute the mortality rate ratios and 95% CI.
Results
Figure 1 shows the seasonal variation in all-cause and cause-specific mortality rates. The number of deaths, test of seasonality, estimated month of the peak incidence and the highest: lowest ratio by cause of death are shown in Table 1
. A strong seasonal variation was seen for all-cause mortality. The pattern was mainly due to seasonal variation in ischaemic heart diseases, cerebrovascular diseases and respiratory diseases with a slight effect due to other cardiovascular'. No seasonal fluctuation was seen for neoplasm and other' deaths. All models containing seasonality terms showed adequate fits to the observed month-by-month mortality rates. For most causes of death showing seasonal effects, the winter peak was in January. The largest relative fluctuation of the mortality rates with season was seen for respiratory diseases. During the winter peak the respiratory disease mortality rate was nearly twice that of the lowest rate (1.98, 95% CI : 1.642.40). However, ischaemic heart disease, the commonest cause of death, contributed the greatest part to the absolute difference between the lowest (summer) and the highest (winter) rates in all-cause mortality and, together with respiratory disease accounted for over three-quarters of this difference.
|
|
|
However, this variation in all-cause mortality could be biased by the pattern of causes of death. For that reason, we assessed the amplitude of the seasonal fluctuation in those causes of death that showed a seasonal pattern, i.e. cardiovascular diseases and respiratory diseases. Table 3 shows that, after stratifying and controlling for age at death, in the last three decades no clear decreasing effect of season on causes with a seasonal pattern exists. However, the proportion of deaths due to seasonally-related causes decreased with calendar period and, within each calendar period the proportion of deaths that were from seasonally sensitive causes increased with age at death. The month of the peak mortality rates were confined to a 3-month period (DecemberFebruary) for 17 of the 21 age-at-death and calendar-period cells.
|
Table 4 shows the amplitudes of the seasonal effect by groups based on a multivariate cause-specific mortality risk score. For all-cause mortality, the seasonality effect did not differ by risk group. However for stroke mortality, the rate ratio was highest in the high-risk group (1.61, 95% CI : 1.22.2) and lowest in the low risk group (1.13, 95% CI : 0.71.7).
|
|
In a 25-year follow-up of over 18 000 men, the winter excesses in ischaemic heart disease and respiratory disease together explained more than three-quarters of the winter excess in all-cause mortality. The burden of winter mortality excess was greater among elderly people, since a larger proportion of deaths in the elderly is from seasonally sensitive causes. Men with prevalent ischaemic heart disease at baseline showed significantly greater seasonality for all-cause mortality than those without. Participants at high multivariate risk of all-cause, coronary, stroke or respiratory mortality did not have greater seasonality effects than those at lower risk. This suggests the importance of other, as yet unidentified, factors in characterizing groups in whom seasonal effects may be greater. Given that the magnitude of seasonal differences in mortality did not decline between 1967 and 1995, better understanding of the causes of seasonal differences in mortality in order to inform preventive intervention is urgent.17
An important strength in this study was the ability to test whether the seasonal variation in mortality was larger among high-risk groups, characterized by older age, low socioeconomic status, prior disease or multivariate disease risk. It has been hypothesized that elderly people are more sensitive to seasonal effects, as it is likely that among elderly people influenza epidemics are more frequent or that their body response to the outside temperature is less adequate. Furthermore, blood pressure may vary more among elderly people between winter and summer.18,19 Our finding that the excess of deaths in winter is larger among elderly people is consistent with this hypothesis. However, in contrast to other studies which also reported age-gradients for specific causes of death,2,4,10,20,21 in our study the ratio of the mortality rate for specific causes of death during winter and during summer did not increase significantly with age. Thus, the difference between the seasonal variation in all-cause mortality of the younger and older age groups was not caused by a larger amplitude of the seasonal variation among the elderly, but was due to the different pattern of causes of death in these age groups. In other words, elderly people are not more sensitive to seasonal effects, but they are more likely to die from causes with a seasonal pattern.
It has been proposed that lower socioeconomic status may be associated with a larger winter excess in mortality.4 This may arise if lower grade employees were less able to protect themselves from the effects of temperature (for example because of poorer housing insulation or less central heating) or, independent of temperature, were more prone to other seasonal precipitants of mortality, such as infection. Indeed it has further been proposed that differences in domestic microclimate might contribute to socioeconomic status differences in coronary mortality.22 However, we found no evidence for greater seasonal effects among those of lower social status, consistent with other recent studies.23,24
We found no evidence that conventional risk factors identified groups in whom the seasonality effect was greater. The seasonality effects were found consistently across disease-specific risk groups defined from a multiple logistic regression using age, employment grade, systolic blood pressure, cholesterol, FEV1, smoking and diabetes. We further tested whether high-risk groups defined by the Framingham risk equation25 for cardiovascular disease were associated with greater seasonality, but they were not (data not shown). This suggests the importance of additional, as yet unidentified, factors in characterizing groups in whom seasonality effects are greater. Possible factors, which were not included in our analyses, are genetic factors and/or environmental factors, such as food habits or climate factors.
The importance of identifying such factors is underscored by the lack of decline in seasonality effects between 19671995. The marked increases in availability of domestic measures to control ambient temperature as well as the increasing use of influenza vaccination had no discernible effect on the degree of seasonal mortality. In absolute terms, the number of people dying from seasonally sensitive diseases is decreasing. Curwen et al., McDowall et al. and Seretakis et al. in the US reported a decline in the seasonal variation by calendar period.10,16,22 McDowall et al. suggested that the decline in the seasonal variation from the 1960s to the 1980s was due to increased use of a central heating system and an enormous fall in air pollution period.10
A winter excess in coronary, stroke and respiratory mortality is reported in several other studies.1,20,2633 The mechanisms explaining the seasonal variation in coronary, stroke and respiratory mortality are poorly understood. Climatic factors are clearly important, with both outdoor and indoor air temperature exerting biological effects on haemostasis, blood viscosity, lipids, the sympathetic nervous system and vasoconstriction.18,34,35 Other climatic factors may also be important, such as sunlight, wind speed and air pollution. The rise in respiratory diseases during winter might be explained by influenza epidemics.16,21 Kunst et al. have reported that influenza may explain 34% of the cold related mortality in The Netherlands.36 The rise in influenza might also cause a rise in cardiovascular diseases during winter.37 Coronary disease is an inflammatory process in which acute (or chronic) infections may stimulate inflammatory pathways. Variation in the severity of the influenza epidemics from year to year may have affected our results. During years with a more severe influenza epidemic, the seasonal variation in cardiovascular diseases might be expected to be larger.
Humans, like other organisms, display seasonal behaviour; food habits, activity patterns, smoking habits (more smoking indoors during winter) and psychosocial factors such as loneliness may all differ between winter and summer. Several psychosocial stressors show a larger effect on blood pressure during winter.38 Mundal et al. showed that the seasonal fluctuation in physical fitness may provide an explanation for the seasonal variation in blood pressure.39 Several other cardiovascular risk factors have been shown to exhibit seasonal variation; e.g. cholesterol,35,40,41 haemostatic factors, fibrinogen,15,35,37,4244 and blood pressure.12,34,45
A further strength of the study is its population basis. In several studies the winter excess in mortality is examined in hospital-based studies.34,38,46 Rothwell reported that the widely varying winter excess in mortality in hospital-based studies might be an artifact of a variation in the likelihood of hospital admissions.27 Our study is free from this possible bias since we studied seasonal variation in a non-hospitalized prospective cohort study.
Our method to assess the amplitude of the seasonal variation (rate ratio of the highest versus the lowest) provides a simple, statistically powerful model for the true variation in mortality rate during the season. There was no evidence, in our data, for any departure from this model. However, it has been found that besides a winter excess, there may also be a heat-related excess in deaths during summer.11 Our data would need to be augmented with climate data for the whole of the follow-up period and require more deaths to be able to investigate this possibility.
Understanding and tackling seasonality in mortality is a public health challenge which has not been met. In relative terms, the seasonal effects showed no evidence of decline between 1967 and 1995; older age, low socioeconomic status or high multivariate risk did no identify a group more prone to seasonal, cause-specific mortality effects. However, seasonal effects were greater among those with prevalent ischaemic heart disease, suggesting, in this group at least, the potential for preventive strategies.
Acknowledgments
This study is a collaboration between the International Center for Health & Society in London and the research program of the Erasmus Center for Research on Aging (a collaboration of the faculties of Economics, Law, Sociology, Medicine and Health Policy and Management of the Erasmus University Rotterdam and the University Hospital Rotterdam Dijkzigt, The Netherlands. MJS is supported by a grant from the British Heart Foundation and CTMvR is supported for this project by the Dutch Heart Foundation. HH is supported by a Department of Health Public Health Career Scientist Award. MM is supported by an MRC Research Professorship.
References
1 Mackenbach JP, Kunst AE, Looman CW. Seasonal variation in mortality in The Netherlands. J Epidemiol Community Health 1992; 46:26165.[Abstract]
2 Crombie DL, Fleming DM, Cross KW, Lancashire RJ. Concurrence of monthly variations of mortality related to underlying cause in Europe. J Epidemiol Community Health 1995;49:37378.[Abstract]
3 Sakamoto-Momiyama M. Changes in the seasonality of human mortality: a medico-geographical study. Soc Sci Med 1978;12:2942.
4 Curwen M. Excess winter mortality: a British phenomenon. Health Trends 1990/91;4:16975.
5 McKee CM. Deaths in winter: can British learn from Europe? Eur J Epidemiol 1989;5:17882.[ISI][Medline]
6 Kunst AE, Groenhof F, Mackenbach JP. The association between two windchill indices and daily morality variation in the Netherlands. Am J Public Health 1994;84:173842.[Abstract]
7 Gorjanc ML, Flanders WD, VanDerslice J, Hersch J, Malilay J. Effects of temperature and snowfall on mortality in Pennsylvania. Am J Epidemiol 1999;149:115260.[Abstract]
8 Eurowinter Group. Cold exposure and winter mortality from ischaemic heart disease, cerebrovascular disease, respiratory disease, and all causes in warm and cold regions of Europe. Lancet 1997; 349:134146.[ISI][Medline]
9
Kloner RA, Poole WK, Perritt RL. When throughout the year is coronary death most likely to occur? A 12-year population-based analysis of more than 220 000 cases. Circulation 1999;100:163034.
10 McDowall M. Long term trends in seasonal mortality. Popul Trends 1981;26:169.
11 Mackenbach JP, Borst V, Schols JMGA. Heat-related mortality among nursing-home patients. Lancet 1997;349:129798.[ISI][Medline]
12
Kristal Boneh E, Harari G, Green MS. Seasonal change in 24-hour blood pressure and heart rate is greater among smokers than nonsmokers. Hypertension 1997;30:43641.
13 Reid DD, Brett GZ, Hamilton PJ, Jarrett RJ, Keen H, Rose G. Cardiorespiratory disease and diabetes among middle-aged male Civil Servants. A study of screening and intervention. Lancet 1974; i:46973.
14 SAS Institute. SAS User's Guide: Carry NC, SAS Institute Inc; 1985.
15 van der Bom JG, de Maat MP, Bots ML, Hofman A, Kluft C, Grobbee DE. Seasonal variation in fibrinogen in the Rotterdam Study. Thromb Haemost 1997;78:105962.[ISI][Medline]
16 Curwen M, Devis T. Winter mortality, temperature and influenza: has the relationship changed in recent years? Popul Trends 1988;53:1720.
17 Ross G. The Strategy of Preventive Medicine. Oxford: Oxford University Press, 1992.
18 Ballester F, Corella D, Perez-Hoyos S, Saez M, Hervas A. Mortality as a function of temperature. A study in Valencia, Spain, 19911993. Int J Epidemiol 1997;26:55161.[Abstract]
19 Woodhouse PR, Khaw KT, Plummer M. Seasonal variation of blood pressure and its relationship to ambient temperature in an elderly population. J Hypertens 1993;11:126774.[ISI][Medline]
20 Bainton D, Moore F, Sweetnam P. Temperature and deaths from ischaemic heart disease. Br J Prev Soc Med 1977;31:4953.[ISI][Medline]
21 Marshall RJ, Scragg R, Bourke P. An analysis of the seasonal variation of coronary heart disease and respiratory disease mortality in New Zealand. Int J Epidemiol 1988;17:32531.[Abstract]
22 Seretakis D, Lagiou P, Lipworth L, Signorello LB, Rothman KJ, Trichopoulos D. Changing seasonality of mortality from coronary heart disease. JAMA 1997;278:101214.[Abstract]
23
Gemmell I, McLoone P, Boddy FA, Dickinson GJ, Watt GCM. Seasonal variation in mortality in Scotland. Int J Epidemiol 2000;29:27479.
24 Shah S, Peacock J. Deprivation and excess winter mortality. J Epidemiol Community Health 1999;53:499502.[Abstract]
25 Anderson KM, Wilson PW, Odell PM, Kannel WB. An updated coronary risk profile. A statement for health professionals. Circulation 1991;83:35662.[ISI][Medline]
26 Ballaro A, Cortina-Borja M, Collin J. A seasonal variation in the incidence of ruptured abdominal aortic aneurysms. Eur J Vasc Endovasc Surg 1998;15:42931.[ISI][Medline]
27 Rothwell PM, Wroe SJ, Slattery J, Warlow CP. Is stroke incidence related to season or temperature? The Oxfordshire Community Stroke Project. Lancet 1996;347:93436.[ISI][Medline]
28 Anonymous. Cold at heart. Lancet 1989;ii:254.
29 Anderson TW, Le Riche WH. Cold weather and myocardial infarction. Lancet 1970;i:29196.
30 McDowall M. The mortality of agricultural workers: using the thirteenth decennial occupational mortality. Popul Trends 1986;45:1417.
31 Rose G. Cold weather and ischaemic heart disease. Br J Prev Soc Med 1966;20:97100.[ISI][Medline]
32 West R. Seasonal variation in CHD mortality. Int J Epidemiol 1989; 18:46365.[ISI][Medline]
33 Rogot E, Padgett SJ. Associations of coronary and stroke mortality with temperature and snowfall in selected areas of the United States, 19621966. Am J Epidemiol 1976;103:56575.[Abstract]
34 Kunes J, Tremblay J, Bellavance F, Hamet P. Influence of environmental temperature on the blood pressure of hypertensive patients in Montreal. Am J Hypertens 1991;4:42226.[ISI][Medline]
35 Stout RW, Crawford V. Seasonal variations in fibrinogen concentrations among elderly people. Lancet 1991;338:913.[ISI][Medline]
36 Kunst AE, Looman CW, Mackenbach JP. Outdoor air temperature and mortality in The Netherlands: a time-series analysis. Am J Epidemiol 1993;137:33141.[Abstract]
37 Woodhouse PR, Khaw KT, Plummer M, Foley A, Meade TW. Seasonal variations of plasma fibrinogen and factor VII activity in the elderly: winter infections and death from cardiovascular disease. Lancet 1994;343:43539.[ISI][Medline]
38 James GD, Yee LS, Pickering TG. Winter-summer differences in the effects of emotion, posture and place of measurement on blood pressure. Soc Sci Med 1990;31:121317.[ISI][Medline]
39 Mundal R, Kjeldsen SE, Sandvik L, Erikssen G, Thaulow E, Erikssen J. Seasonal covariation in physical fitness and blood pressure at rest and during exercise in healthy middle-aged men. Blood Press 1997;6:26973.[Medline]
40 Woodhouse PR, Khaw KT, Plummer M. Seasonal variation of serum lipids in an elderly population. Age Ageing 1993;22:27378.[Abstract]
41 Gordon DJ, Hyde J, Trost DC et al. Cyclic seasonal variation in plasma lipid and lipoprotein levels: the Lipid Research Clinics Coronary Primary Prevention Trial Placebo Group. J Clin Epidemiol 1988; 41:67989.[ISI][Medline]
42 Keatinge WR, Coleshaw SR, Cotter F, Mattock M, Murphy M, Chelliah R. Increases in platelet and red cell counts, blood viscosity, and arterial pressure during mild surface cooling: factors in mortality from coronary and cerebral thrombosis in winter. Br Med J 1984;289: 140508.[ISI][Medline]
43 Stout RW, Crawford VL, McDermott MJ, Rocks MJ, Morris TC. Seasonal changes in haemostatic factors in young and elderly subjects. Age Ageing 1996;25:25658.[Abstract]
44 Brennan PJ, Greenberg G, Miall WE, Thompson SG. Seasonal variation in arterial blood pressure. Br Med J 1982;285:91923.[ISI][Medline]
45 Rose G. Seasonal variation in blood pressure in man. Nature 1961; 189:235.
46 Douglas AS, Dunnigan MG, Allan TM, Rawles JM. Seasonal variation in coronary heart disease in Scotland. J Epidemiol Community Health 1995;49:57582.[Abstract]