Background Meteorological factors like cold temperatures and heavy snowfalls have been reported to increase myocardial infarction (MI) incidence, but there are inconsistencies in results as well as in methodology in previous studies. The objective of this study was to examine the impact of meteorological factors on incidence of MI in a population-based study in Tromsø, Norway (69°39′N).
Methods A total of 32 110 participants from the Tromsø Study enrolled between 1974 and 2001 were followed throughout 2004. Each incident case of MI was validated by the review of medical records and death certificates. Meteorological data from the Tromsø Weather Station were collected from the Norwegian Meteorological Institute database. Poisson regression models were applied to analyse the impact of meteorological factors on MI incidence. All analyses were stratified by sex and age.
Results A total of 1882 first-ever MIs were registered. The main finding was an increase in MI incidence among persons older than 65 years with decreasing temperatures (p=0.016) and increasing snowfall (p=0.030). When comparing the lower and upper limits of the temperature distribution (−10°C with 20°C), the MI risk increased by 47% (RR=1.47, 95% CI 1.09 to 2.13). Comparing limits of the snowfall distribution (10 with 0 mm), the MI risk increased by 44% (RR=1.44, 95% CI 1.07 to 1.94).
Conclusions In this subarctic population, MI incidence was little affected by the weather, probably due to behavioural protection. However, cold weather and heavy snowfall may be associated with increased risk of MI among older people.
- myocardial infarction/heart disease
- seasons/seasonal variation
Statistics from Altmetric.com
The impact of weather on human health, described as early as 430 BC by Hippocrates,1 has gained renewed interest with the recent years' focus on global climate change.2 The relationship between meteorological variables and incidence of myocardial infarction (MI) has been studied since 1938,3 but studies based on prospective cohorts with adjudicated MIs and long follow-up are few. Cold temperatures have been associated with increased MI incidence,4 ,5 but the effect of temperature is smaller among populations living in colder climates6 ,7 or in years with cold winters,5 suggesting an adaption to regional6 ,7 and also seasonal5 mean temperatures. Meteorological factors other than atmospheric temperature are less frequently studied, and such research is wanted.8 Heavy snowfall has been reported to increase the risk of MI, but the literature is scarce and often based on one extreme event, few cases9 ,10 or fatal cases only.11 ,12 The concern of climate changes causing more extreme weather conditions that can affect human health,2 and previous inconsistencies in methodology concerning quality of data and statistics,2 ,8 makes it important to investigate the relationship between several meteorological variables and adjudicated MIs in a prospective cohort design. The objective of this study was to investigate the impact of daily meteorological variables on first-ever MI in a well-defined subarctic population in northern Norway, exposed to a harsh climate with constantly changing weather conditions like heavy snowfalls and extreme seasonal variations in daylight. This population has previously been investigated for seasonal variation on risk of MI,13 which showed a small winter excess of MI incidence. The current in-depth study of possible effects of meteorological factors is a natural extension of our previous work.
Tromsø is the largest city of northern Norway with approximately 67 000 inhabitants. Tromsø is situated above the Arctic Circle at 69°39′N, the same latitude as Siberia and northern Alaska. The sun is below the horizon from mid-November to mid-January and does not set between mid-May and mid-July, giving Tromsø a dark winter season and a summer season with 24 h daylight. The Tromsø Study is a single-centre, population-based prospective study carried out by the University of Tromsø in cooperation with the former National Health Screening Service. It was initiated in 1974 to investigate the high mortality of cardiovascular disease in northern Norway. The study consists of six different surveys to which total birth cohorts and random samples of Tromsø inhabitants were invited. Included in this analysis were those who attended the first five surveys conducted in 1974, 1979–1980, 1986–1987, 1994–1995 and 2001 (table 1). The participation rate was above 77% in all surveys. Approvals were obtained from the Regional Board of Research Ethics, the Data Inspectorate and the Directorate of Health and Social Affairs. A standardised examination protocol including physical examination, blood sampling and administration of questionnaires was followed at each survey. The questionnaires included questions on previous MI. A total of 38 164 men and women participated at least once. For this analysis, we excluded those subjects younger than 35 years at the end of follow-up (n=5282), subjects with a validated MI before start of follow-up (n=390), subjects who migrated from Tromsø before their assigned date of examination (n=160) and subjects who did not consent to research (n=222). This left 32 110 participants for inclusion. They were followed from the date of examination until 31 December 2004, first MI, death or migration from Tromsø.
Case identification and definition
The Tromsø Study has collected standardised data on the incident cases of acute MI in the study population since 1974. A broad search strategy is used. Hospitalised cases are identified by searching the discharge diagnosis register at the University Hospital of Tromsø, the only local hospital. In the period 1969–1979, the International Classification of Diseases (ICD) version 8 codes 410–414, 427, 795–796, from 1980 to 1998, ICD version 9 codes 410–414, 798, 427.5 and thereafter ICD version 10 codes I20–I25, R96, R98, R99, I46 were used. In 2006, the Tromsø Study participant list was linked with the Causes of Death Registry at Statistics Norway for the period 1974–2004, and the death certificate was retrieved for those with an underlying or contributing diagnosis of cardiovascular disease or sudden unexpected death and who had not already been registered through the hospital search. This procedure identified out-of-hospital incident MI cases including nursing home residents and cases of sudden cardiac death outside the institutions. Data were censored for date of registered emigration, obtained from the Population Registry of Norway, or death from causes other than MI. Norway has a unique personal identification system that allows exact matching of population register data.
A final adjudication of MI events was performed by an endpoint committee consisting of experienced physicians. Events were defined as first-ever non-fatal or fatal MI. For each incident case, a full medical record including available records from prehospital care (ambulance service, general practitioners, nursing homes) and/or death certificate were searched for diagnostic criteria based on clinical presentation, levels of myocardial biomarkers, electrocardiogram and autopsy, if applicable. Event ascertainment followed a detailed protocol according to a slightly modified version of the WHO MONICA/MORGAM protocol.14
Data from the Tromsø Weather Station were collected from the official Norwegian Meteorology Institute web page15 for the period 14 February 1974 (when the first participant entered the study) to 31 December 2004. The Tromsø Weather Station is situated 100 m above sea level in immediate vicinity of the Tromsø City Centre. Daily data on minimum, maximum and daily mean temperature (in degree Celsius), mean and maximum wind speed (in metres per second), hours of sunlight (in hours), mean atmospheric pressure (in hectopascals), mean relative humidity (in per cent), total precipitation (in millimetres), precipitation as snow given in snow water equivalent (in millimetres) and largest snow cover on ground (in centimetres) were complete for the whole period, except for data on hours of sunlight where 1.1% observations were missing. Daily mean temperature is calculated as t=N−k×(N−min), where N is the mean of measured temperatures at 06:00, 12:00 and 18:00, k is a local factor adjusting for the lack of night temperatures and min is the daily minimum temperature, a method giving approximately the same value as continuously measured day and night temperatures.16 We used the Norwegian Meteorological Institute's geographically dependent definition of meteorological season for the area of Tromsø16: winter (6 November to 13 April), spring (14 April to 22 June), summer (23 June to 18 August) and autumn (19 August to 5 November). There is virtually no air pollution in Tromsø, and such data were not included in this study.
All analyses were produced using STATA V.11 (Stata Corp LP). Weather data for each calendar day during follow-up were linked to person-years at risk and number of MI events. Meteorological exposure variables were estimated as a 3-day average (the date of the MI event and the two previous days). The associations between MI incidence and the meteorological variables were assessed using Poisson regression models. In order to accommodate linear and curvilinear associations, the meteorological independent variables were included as a linear term and fractional polynomial terms in separate models. In the linear models, the percentage change in risk of MI was estimated per SD increase in the meteorological variables. Scatter plots of MI incidence rates (per 1000 person-years) in intervals of daily mean temperature (1°C interval length), mean wind speed (1 m/s), mean atmospheric pressure (5 hPa), mean relative humidity (5%) and precipitation as snow (1 mm) were created, and smoothed curves were drawn using the estimated coefficients from the fractional polynomial regression models. RR estimates with 95% CIs were estimated between upper and lower limits of the distribution of the meteorological variables using the fractional polynomial models. For the variable snowfall, a separate analysis for winter season only was performed. Sensitivity analyses with month added to the model as a categorical variable were performed to adjust for season. Possible two-way interactions between the meteorological variables and calendar time in days, gender or age were assessed separately by adding product terms to the Poisson regression models.
The effect of lagged exposure of the meteorological variables was assessed by comparing different models of lagged exposure data. Ten different models were compared using the Akaike Information Criterion: model 1 included the date of the event (lag 0) only, model 2 included the date of the event and the previous day (lag 0 and lag 1), model 3 included lag 0 to lag 2, and so on, up to model 10 including lag 0 to lag 9. The best-lagged exposure model did not give a better fit nor showed a more clear association with MI incidence than the model where 3-day average was used. Results from the lagged models are therefore not shown.
In order to examine the season-specific effect of unusual weather on MI for all meteorological variables except snowfall, z-scores for the meteorological variables were calculated for each week of the year. The effect of z-score on MI risk was assessed using the same models as for the unstandardised meteorological variables above. All tests were two-sided using a 5% significance level. The analyses were stratified by sex and age (35–64 and ≥65 years).
Study population and meteorology
A total of 1882 incident cases of MI occurred between 1974 and 2004. The mean age at first MI was 61 years in men and 72 years in women. The median length of follow-up was 17 years. There was no interaction between meteorological variables' effect on MI incidence and time period. Data from the period 1974 to 2004 were therefore pooled. Summary statistics of daily values of meteorological variables for the whole time period of 61 102 days are presented in table 2.
Meteorological variables and first-ever MI
Percentage changes in risk of MI incidence per SD of various meteorological variables (3-day average values) are presented in table 3. Temperature was inversely related with risk of MI, but the association was statistically significant only in the older age group (table 3). There was no significant linear relationship between MI and mean wind speed, relative humidity, atmospheric pressure or precipitation as snow (table 3). The risk associated with snowfall was similar in an analysis using just winter compared with an analysis using the whole year.
Incidence rates of MI by intervals of meteorological factors (3-day average) and fitted model curves for the total material are presented in figure 1. Mean wind speed had an inverted U-shaped curve (p=0.004) and mean relative humidity had a U-shaped curve (p=0.004), while daily mean temperature, mean atmospheric pressure and precipitation as snow showed non-significant flatter curves (figure 1). Sensitivity analysis including month as a covariate showed agreement with the main analysis with regard to the coefficients and the p values (results not shown).
In analyses stratified by sex and age, we found no difference in the association between MI and mean wind speed, relative humidity or atmospheric pressure (results not shown). For daily mean temperature and precipitation as snow, we observed age differences, and for snowfall also sex differences, in the relationship with MI (figure 2). In those aged ≥65 years, a significant inverse association was observed for temperature (p=0.016), and a positive association was observed for snowfall (p=0.030) (figure 2). When comparing the lower and upper limits of the temperature distribution (−10°C with 20°C), a 47% increase in MI risk was estimated (RR=1.47, 95% CI 1.09 to 2.13). Comparing limits of the snowfall distribution (10 with 0 mm), a 44% increased MI risk was estimated (RR=1.44, 95% CI 1.07 to 1.94). In those aged 35–64 years, temperature was not associated (p=0.21, RR=1.05, 95% CI 0.79 to 1.41) for lower versus upper limits of the distribution, while snowfall showed an inverse association with MI (p=0.048, RR=0.67, 95% CI 0.43 to 1.05) for upper versus lower limits of the distribution (figure 2). p Value for interaction with age was 0.15 for temperature and 0.0057 for snowfall. In women, a positive association was observed between MI and precipitation as snow (p=0.035), while no significant association was observed in men (p=0.12) (figure 2). In women, the RR comparing 10 with 0 mm snowfall was 1.39 (95% CI 1.08 to 1.80). The corresponding figure in men was 0.77 (95% CI 0.55 to 1.10). p Value for interaction between sex and snowfall was 0.023.
Unusual weather and MI incidence
The association between MI and week-specific z-scores of the meteorological variables was in agreement with the associations observed for the unstandardised meteorological variables (results not shown).
This study presents the impact of weather on MI incidence in a population-based cohort with >30 years of follow-up. Our main findings were an increased risk of MI among older people at colder temperatures and heavy snowfall and among women at heavy snowfall. The use of first-ever MI validated according to standardised definitions as the outcome, the Norwegian unique person number to search hospital and national registries and the complete data from the Tromsø Weather Station gave the possibility to link accurately validated MI incidence rates to daily meteorological variables.
An increased risk of MI in cold temperatures in the oldest age group has previously been reported by some population-based studies,4 ,6 but not all.5 ,17 Cold temperatures are associated with an increase in blood pressure18–20 and higher blood levels of epinephrine, norepinephrine,21 fibrinogen, C reactive protein levels,22 clotting factors and platelet count, as well as an increase in whole-blood viscosity and a decrease in plasma volume.20 Such influences may contribute to trigger coronary events. Cold-related changes in blood pressure,19 plasma cholesterol and fibrinogen23 have been reported to be more evident in older people, which may result in a greater risk of MI with a decrease in temperature in older people as compared with young people.
We found an increased risk of MI with increasing snowfall in the older age group, while an opposite trend was seen in the younger. Heavy snowfall has been reported to increase the risk of MI, but the literature is mainly based on case reports9 ,10 or fatal cases only.11 ,12 To our best knowledge, the study by Gerber and colleagues17 from the Olmsted County in Minnesota is the only population-based study investigating the effect of snowfall on incident MI. Gerber and colleagues17 found no association between MI and snowfall, and there was no interaction with age. After large snowfalls, snow shovelling can be a physically heavy exertion. Snow shovelling has been reported to cause increased heart rate and blood pressure.24 However, when pacing themselves, old and young men, as well as men with known coronary heart disease, performed the snow removal in relation to their individual maximal work capacity.25 This study25 was a controlled experiment, and none of the participants experienced chest pain during the snow removal. Reports from actual cases show that snow removal is being continued despite symptoms of chest pain.10 It might be that older people are more susceptible to physiological responses like elevated blood pressure and tachycardia during snow shovelling due to normal ageing processes with reduced organ function and comorbid diseases. The older people have more atypical MI symptoms,26 which may increase the risk of ignoring or not sensing the development of an MI during snow removal.
Women, but not men, had higher risk of MI with precipitation as snow. Gender differences in risk of MI with meteorological factors have been reported by some population-based studies,6 ,17 but not all,5 and results are inconsistent.7 Barnett et al6 found in a study of populations worldwide that women showed greater odds of having a coronary event in cold periods compared with men, and this difference in odds was greater in populations in warmer climates. In this subarctic population, the association between MI incidence and temperature, wind speed, atmospheric pressure or relative humidity was not modified by sex.
We found a U-shaped relationship for relative humidity and MI in men and women in the older age group and in an analysis where all subgroups were included, while two previous studies from colder areas found no relation between relative humidity and MI incidence,27 ,28 implying that the impact of humidity on MI is unclear. The inverse U-shaped curve between wind speed and MI incidence may be explained by a reduction of outdoor activity when it is windy, which has been suggested by others finding an inverse relationship between wind speed and mortality.29 Comparisons of MI incidence in populations in several geographical areas around the world show that populations living in colder climates are less affected by meteorological factors like cold.6 ,7 This may be explained by an adaption to the climate based on protective clothing,30–32 staying indoors in well-insulated and adequately heated houses,31–33 as well as having a higher living standard with socioeconomic equality,34 and little outdoor work.32
The actual outdoor exposure to the weather conditions for each individual case of MI is not known. Epidemiologists use meteorological factors like outdoor temperature as a surrogate for personal exposure35 since measuring the individual daily dynamic exposure of representative population samples is impractical and expensive. Weather extremes can reduce the susceptible pool for the following season36 because the most vulnerable people are affected, a phenomenon introducing extra variance. Our analyses include a follow-up time of 31 years, which should average out these effects.
The study is based on a prospective cohort, whereas many studies in this research field used retrospective or routinely collected data. Other strengths are the large sample size, extensive follow-up time, validated endpoints and high-quality weather data.
In this subarctic population, the incidence of MI was marginally affected by the weather, probably due to behavioural protection. Older people should take extra precautions at cold temperatures and after heavy snowfall.
What is already known on this subject
Global climate changes may involve more extreme weather.
The effect of weather on incidence of heart diseases might be dependent of the local climate.
Existing inconsistencies in research results and methodology require the use of validated endpoints in population-based studies to investigate the effect of meteorological factors on myocardial infarction incidence.
What this study adds
The incidence of myocardial infarction in this subarctic population was marginally affected by the weather, probably due to behavioural adaption to the climate.
Older people had increased risk of MI with cold temperatures and heavy snowfall.
Competing interests None.
Ethics approval This study was conducted with the approval of the Regional Board of Research Ethics, the Data Inspectorate and the Directorate of Health and Social Affairs.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.