Temporal variation in case fatality of acute myocardial infarction in Finland

Abstract

Background. Previous studies have suggested that seasonal variation and weather conditions have an influence on the incidence and mortality of acute myocardial infarction (AMI). The influence of these factors on AMI case fatality is less studied.

Aims. The aim of this study was to examine the temporal variation of AMI case fatality and the effect of daily weather conditions on it.

Methods. We analysed death registry and hospital discharge data from all men and women (n=7328) with their first AMI occurrence in the seven largest cities in Finland in the years 1983, 1988, and 1993, aged 25 to 74 years.

Results. The mean annual 28-day case fatality was 44%. We found significant weekly and monthly variation of case fatality (P<0.001). The December holiday season had the highest case fatality throughout the year in women and men aged 65–74 years (P<0.05). The highest weekly case fatality was on Sundays; it differed significantly from the rest of the weekdays only for the oldest age-group (64–74) (P<0.01).

Conclusions. There is significant weekly and monthly variation in case fatality of AMI. The highest case fatality risk for AMI is during the Christmas season and on Sundays. Weather conditions were not found to have an effect on the case fatality.

• Case fatality rate
• meteorological factors
• myocardial infarction
• seasons

Introduction

It is well recognized that human health is influenced by environmental factors such as weather conditions and season 1–3. A large number of studies have shown the impact of these factors on mortality, particularly due to respiratory and cardiovascular diseases 2, 4–8. In developed countries, coronary heart disease (CHD) is the leading cause of death and disability, measured in disability-adjusted life years (DALYs) 9, 10. The World Health Organization (WHO) predicted an even further increase in CHD prevalence by the year 2030, mainly due to the expected population ageing, growing world affluence, increase in the prevalence of type 2 diabetes and obesity, as well as climate change 9, 11–13. The major clinical manifestations of CHD are acute myocardial infarction (AMI) and sudden death.

The influence of weather conditions and season on the incidence of AMI has been widely studied primarily in association with temperature and air pressure changes and wind activity 1, 3, 14–17. The influence of season and weather conditions on AMI mortality, on the other hand, has been focused upon only in a small number of studies 18–21. However, the results are difficult to interpret since it is impossible to distinguish whether the observed pattern of variation is due to incidence or case fatality. The only published study, to our knowledge, on case fatality and season divided the data into four seasons and found a winter peak in case fatality 22.

The climate in Finland is primarily influenced by its geographical position, being one of the northernmost countries in the world. There are both maritime and continental influences. Due to this continental influence and the Gulf Stream, the average temperature is several degrees higher than in other areas of the same latitude. Winters are rather cold and summers comparatively hot, and weather types can change quickly, as Finland lies in the transition zone from tropical to polar air masses. For this study, the seven largest Finnish city areas have been chosen: Helsinki, Turku, Jyväskylä, Tampere, Kuopio, Oulu, and Rovaniemi. The climate characteristics for these city areas are similar but differ in certain aspects like weekly variation of temperature and wind strength. The mean temperature follows approximately the geographical position of the cities, being coldest in the north and warmest in the south. The mean winter temperatures differ more from city to city than the summer ones.

In this study we examine the effect of season and weather conditions on the 28-day case fatality of AMI by calculating the case fatality rates of the years 1983, 1988, and 1993 by month and testing them for seasonality. In addition, we compare five statistical models: one with and one without a linear trend, one with a weather component, one with a periodic seasonal component, and one with both a weather and seasonal component. Both the goodness-of-fit and the predictive properties of these models are compared.

Key messages

• There is significant weekly and monthly variation in case fatality of acute myocardial infarction (AMI).

• There is no statistically significant effect of daily weather conditions on case fatality of AMI.

• The highest case fatality risk for AMI is during the Christmas season and on Sundays.

Abbreviations

Table

AMI	acute myocardial infarction
CDS	Cause of Death Statistics
CHD	Coronary heart disease
DALY	Disability-adjusted life year
HDR	Hospital Discharge Registry
ICD	International Classification of Diseases
MSE	mean squared error
WHO	World Health Organization

Regression models:

Table

NT	incl. age and sex only
LT	incl. age, sex, and linear trend
FS	incl. age, sex, and seasonal Fourier terms
W0	incl. age, sex, and weather covariates
WS	incl. age, sex, seasonal Fourier terms, and weather covariates

Patients and materials

The data consist of all AMI patients in Finland in 1983, 1988 and 1993. The data were obtained from two nationwide registries: the Cause of Death Statistics (CDS) maintained by Statistics Finland and the Hospital Discharge Registry (HDR) maintained by the National Research and Development Centre for Welfare and Health. The computerized record linkage of the two data sets was performed by using the personal identification number assigned to all Finnish residents, in order to obtain the deaths and hospitalizations due to AMI in Finland for those years (International Classification of Diseases ICD-9 codes 410–414). The place of residence of the AMI cases was set by the municipality of the patient at time of diagnosis, and the sex-age-municipality-specific population data were obtained from Statistics Finland. The linkage also made it possible to distinguish between first and recurrent events of AMI: in order to do this the HDR was checked for any previous mention of AMI. For this study, we only considered patients with the first recorded occurrence of AMI. The event was considered fatal if the death had occurred either out of hospital or within 28 days of arrival in the hospital. The overall sensitivity of the ICD codes for MI in the combined HDR and CDS was earlier found to be 83% and the positive predictive values 90% for certain areas of Finland 23. Mähönen et al. 24 found the corresponding agreement in diagnosis to vary from 87% to 100% for the ICD codes 410–414.

The weather data were obtained from the European Centre for Medium-Range Weather Forecasts ERA-40 project 25, 26. It consists of interpolated measurements for air temperature, pressure, and wind speed projected at 6-h intervals for a 2.5°×2.5° grid worldwide. Comparison with locally available meteorological data for the cities Helsinki and Oulu showed high correlation.

The geographical positions for the seven cities selected for this study are the following: Helsinki (60.2N, 24.9E), Kuopio (62.5N, 27.4E), Oulu (65.0N, 25.3E), Rovaniemi (66.3N, 25.4E), Tampere (61.3N, 23.5E), Turku (60.3N, 22.2E), and Jyväskylä (62.1N, 25.4E) (Figure 1). The population at risk for the three study years was in total 1,319,964 men and 1,508,634 women.

Figure 1.  The study area includes the seven largest cities of Finland.

Methods

In the analysis, the time of event was taken to be either the time of hospital admission (if the case data came from the HDR) or the time of death (if the case data came from the CDS). The weather variables were evaluated relative to the time of event.

The monthly and weekly 28-day case fatality rates were tested for seasonality, by comparing the fit of binomial regression with a logit link, adjusted by sex and age, with and without month and weekday factors, respectively.

Five regression models of increasing complexity have been fitted: a model with covariates on sex and age only (NT), one with a linear trend (LT), a model with seasonal Fourier terms 27 which includes up to four harmonics and an additional weekday harmonic (FS), a model with daily weather covariates and no seasonal effects (W0), and a model with both Fourier terms and weather covariates (FW).

The daily case fatality was assumed to follow a binomial distribution, and only the days with at least one case of AMI were included in the analysis.

The preliminary graphical analysis indicated that case fatality increases with age in an almost linear fashion. The cross-term for age and sex was also included since it proved to be significant (P = 0.03). On the other hand, the linear time trend term was included in the LT model only, since it proved to be non-significant (P = 0.71). The optimum number of harmonics was assessed using the method described in Hunsberger et al. 28, and all five harmonics were thus included in the model. The weather covariates finally included in the W0 and FW models were the daily temperature and its square.

The predictive properties of the model were assessed through a one (day)-out cross-validation 29, for which the mean squared errors (MSE) were evaluated. A smaller MSE indicates better predictive power for the model.

The statistical analysis was performed using R-programming environment 30.

Results

A total of 7328 cases of the first AMI occurrence were registered in the years 1983, 1988, and 1993. The overall 28-day case fatality was 44%, because 3196 subjects died within 28 days of the diagnosis. The case fatality increased with age but did not differ significantly between the years under study (see Table I). Variation by month and by day of the week was statistically significant with P < 0.001. The observed case fatality percentages are shown in Figure 2 and Figure 3. The corresponding distributions of first AMI incidence are given in the background for comparison.

Figure 2.  Monthly variation of first acute myocardial infarction (AMI) incidence and first AMI case fatality rates by age-group and sex for the years 1983, 1988, and 1993.

Figure 3.  Weekly variation of first acute myocardial infarction (AMI) incidence and first AMI case fatality rates by age-group and sex for the years 1983, 1988, and 1993.

Table I.  First acute myocardial infarction (AMI) case fatality rates by sex, year, and age-group in Finland.

	25–54 years old		55–64 years old		65–74 years old
Year	Death/cases	Rate	Death/cases	Rate	Deaths/cases	Rate
Men
1983	165/521	32% (28%; 36%)	280/608	46% (42%; 50%)	378/675	43% (52%; 60%)
1988	140/462	30% (26%; 35%)	246/575	43% (39%; 47%)	318/574	55% (51%; 60%)
1993	136/403	34% (29%; 38%)	183/470	39% (35%; 43%)	335/598	56% (52%; 60%)
Women
1983	23/95	24% (16%; 33%)	75/224	34% (27%; 40%)	269/593	45% (41%; 49%)
1988	17/87	20% (11%; 28%)	78/207	38% (31%; 44%)	267/545	49% (45%; 53%)
1993	13/56	23% (12%; 34%)	52/175	30% (23%; 37%)	221/460	48% (43%; 53%)

Only the average weekly temperature and its square were finally included in the weather model based on the stepwise selection algorithm, starting with the model which included wind speed and atmospheric pressure rate, both linear and squared. The coefficients for the five fitted models are shown in Table II. Since the binomial regression model requires a logit link, these coefficients are not directly interpretable beyond the direction of the effect.

Table II.  Coefficients and the mean squared errors (MSE) obtained from a 1-out cross-validation for the five different models of increasing complexity.

Model	No trend (NT)		Linear trend (LT)		Seasonal Fourier terms (FS)		Weather (W0)		W0 + FS (WF)
	Coefficient	P-value	Coefficient	P-value	Coefficient	P-value	Coefficient	P-value	Coefficient	P-value
Sex	−1.357	0.002	−1.358	0.002	−1.301	0.003	−1.348	0.002	−1.290	0.003
Age	0.043	<0.000	0.044	<0.000	0.044	<0.000	0.044	<0.000	0.044	<0.000
Sex×age	0.015	0.027	0.015	0.027	0.014	0.038	0.014	0.029	0.014	0.041
Temp ^a	–	–	–	–	–	–	−0.008	0.066	−0.001	0.930
Temp2 ^b	–	–	–	–	–	–	0.001	0.06	0.001	0.205
Time, t ^c	–	–	−0.002	0.710	–	–	–	–	–	–
SinWeek ^d	–	–	–	–	0.065	0.050	–	–	0.066	0.048
CosWeek ^d	–	–	–	–	0.141	<0.000	–	–	0.142	<0.000
MSE	0.2360		0.2360		0.2354		0.2362		0.2357

^a Temperature.

^b Temperature squared.

^c Change over time (linear trend).

^d Fourier terms, functions of the day of the week.

The cross-validation analysis indicated that the best prediction of case fatality is provided by a model with temporal effects but without weather covariates (model FS, Table II). However, the differences in mean squared errors (MSE) of all the non-trivial models were small compared to the trivial no trend (NT) model. Therefore, neither the temporal variation nor information on weather is of use in the prediction of case fatality.

December proved to be the month with the highest observed fatality for women and for men 65–74 years of age (see Table III and Figure 2). Although the case fatality rates for the month with the highest risk did not differ significantly from the next highest risk month in any sex-age-group, they were significantly higher (P < 0.05) than the average for the rest of the year for all except the middle-aged men. Furthermore, the Christmas season, defined each year as 23rd to 31st of December, proved the most risky of all, with significantly higher case fatality rates for all except young men (25–54) and older women (65–74) (P < 0.05).

Table III.  Monthly case fatality rates of acute myocardial infarction in Finland 1983, 1988, and 1993 by sex and age.

	Men						Women
Age	25–54 years		55–64 years		65–74 years		25–54 years		55–64 years		65–74 years
Month	Deaths/cases	Rate	Deaths/cases	Rate	Deaths/cases	Rate	Deaths/cases	Rate	Deaths/cases	Rate	Deaths/cases	Rate
Jan	32/117	27%	54/139	39%	73/150	49%	7/23	30%	19/51	37%	68/147	46%
Feb	33/95	35%	63/136	46%	80/147	54%	2/19	11%	18/47	38%	65/127	51%
Mar	40/119	34%	73/150	49%	90/169	53%	3/22	14%	15/48	31%	58/147	39%
Apr	41/123	33%	66/143	46%	92/149	62%	7/25	28%	19/57	33%	61/145	42%
May	35/121	29%	41/124	33%	91/150	61%	3/19	16%	17/48	35%	58/120	48%
Jun	41/133	31%	65/156	42%	96/168	57%	8/24	33%	17/51	33%	59/112	53%
Jul	34/123	28%	67/137	49%	88/147	60%	1/14	7%	12/36	33%	74/136	54%
Aug	44/108	41%	69/157	44%	78/150	52%	2/19	11%	13/48	27%	57/137	42%
Sep	30/106	28%	54/132	41%	71/151	47%	5/22	23%	15/55	27%	56/128	44%
Oct	42/122	34%	52/124	42%	87/174	50%	3/18	17%	17/57	30%	67/131	51%
Nov	32/127	25%	48/127	38%	89/156	57%	5/21	24%	22/63	35%	59/141	42%
Dec	37/92	40%	57/128	45%	96/136	71%	7/12	58%	21/45	47%	75/127	59%

The weekday distribution was more uniform. Sunday was found to be the highest case fatality day of the week. Sunday estimates were significantly higher than the average for the rest of the week only for the age-group 65–74 for both sexes (P < 0.1).

Discussion

We studied the temporal variation in case fatality of AMI with weather conditions and season using 7328 Finnish cases of first AMI occurrence, of both sexes, aged 25–74 years in the years 1983, 1988, and 1993. We found significant weekly and monthly variation of case fatality. The weather components used did not add significantly to the model.

Due to the lack of data we could not investigate case fatality among patients over 75 years old. Therefore the results cannot be generalized for older age-groups.

The temporal variation of case fatality due to AMI is not well studied. Most of the studies focus on the seasonality of AMI mortality 18, 19. The seasonal variation of AMI mortality is hard to interpret. It is difficult to say whether higher mortality originates from a higher incidence of AMI or higher case fatality, or both. On the other hand the investigation of case fatality results in clearer policy implications in terms of ready availability of transport, medical personnel, and necessary treatment during the periods of expectedly high case fatality.

The AMI mortality has been declining steadily in the last decades. The reason for that is first and foremost the change in the incidence of AMI due to better primary and secondary prevention. However, the case fatality does not add much to this decline. One reason is that most of the deaths occur out of hospital, thus the improvement and wider availability of hospital treatment methods over time do not have an impact 31–33. This aligns with our results.

The highest case fatality of AMI in our study was in December. An earlier study focusing on seasonality of case fatality 22 also observed a higher level of case fatality in the winter months. Since data in that study included only in-hospital deaths the case fatality percentages were generally lower than found in our study.

The highest AMI case fatality occurred on Sundays. A previous study carried out in New Jersey, US, in 1987–2002 34 found significantly higher mortality in patients admitted to the hospital during week-ends compared to those admitted on other weekdays. Patients were less likely to undergo invasive treatments, and the time between admission and performance of procedures was longer during week-ends than on weekdays. The reason for that might be the specialized hospital staffing, which is usually lower during week-ends than on weekdays. The same shortage relate to December holidays, where we found increased AMI case fatality as well.

Several studies have reported a significant Monday peak in AMI onset 35, 36 and sudden death of presumably cardiac origin 37. The Monday incidence peak is also in evidence in our data. However, there is no reason for the incidence and case fatality to correlate unless high incidence becomes a cause of competition for resources (i.e. ambulances, hospital beds, medical staff, and medication), which is not the case in Finland.

Possible reasons for temporal or seasonal variation of case fatality of AMI may originate from several factors. One is the direct physiological impact of weather and season through e.g. temperature, air pollution, or pollen concentration 1, 3, 7, 17, 38–40. However, in our study we did not find a statistically significant effect of the temperature on the case fatality. Concomitant infectious diseases such influenza, pneumonia, or other respiratory infections also underlie seasonality and could add to the case fatality 22, 41, 42. Also blood pressure and peripheral vasoconstriction vary with the season 43, 44. Although the incidence of infectious diseases, high blood pressure, and peripheral vasoconstriction are higher in the winter months, this cannot explain the peak in December. Time to treatment can also underlie seasonal fluctuations, e.g. extended time to reach the hospital in winter due to bad traffic conditions. However, these conditions are typically not extremely difficult in December. Kloner et al. 45 found a peak in coronary deaths during the Christmas holidays. They suggest a superimposing of respiratory infections, behavioural changes in the consumption of food, salt, and alcohol, and the emotional and psychological stresses of those holidays. Those reasons could be the same for the AMI case fatality peak during the December holidays found in this study, possibly with an additional factor of low specialized hospital staffing.

In conclusion, there is a statistically significant weekly and monthly variation in case fatality of AMI in Finland. The highest case fatality risk for AMI is during the Christmas season and on Sundays. Weather conditions were not found to have an effect on the case fatality.

Acknowledgements

This study was funded by the EU project GEO-BENE.