International Journal of Biometeorology

, Volume 58, Issue 5, pp 799–808

The influence of meteorological and geomagnetic factors on acute myocardial infarction and brain stroke in Moscow, Russia

  • Dmitry Shaposhnikov
  • Boris Revich
  • Yuri Gurfinkel
  • Elena Naumova
Original Paper

DOI: 10.1007/s00484-013-0660-0

Cite this article as:
Shaposhnikov, D., Revich, B., Gurfinkel, Y. et al. Int J Biometeorol (2014) 58: 799. doi:10.1007/s00484-013-0660-0


Evidence of the impact of air temperature and pressure on cardiovascular morbidity is still quite limited and controversial, and even less is known about the potential influence of geomagnetic activity. The objective of this study was to assess impacts of air temperature, barometric pressure and geomagnetic activity on hospitalizations with myocardial infarctions and brain strokes. We studied 2,833 myocardial infarctions and 1,096 brain strokes registered in two Moscow hospitals between 1992 and 2005. Daily event rates were linked with meteorological and geomagnetic conditions, using generalized linear model with controls for day of the week, seasonal and long-term trends. The number of myocardial infarctions decreased with temperature, displayed a U-shaped relationship with pressure and variations in pressure, and increased with geomagnetic activity. The number of strokes increased with temperature, daily temperature range and geomagnetic activity. Detrimental effects on strokes of low pressure and falling pressure were observed. Relative risks of infarctions and strokes during geomagnetic storms were 1.29 (95 % CI 1.19–1.40) and 1.25 (1.10–1.42), respectively. The number of strokes doubled during cold spells. The influence of barometric pressure on hospitalizations was relatively greater than the influence of geomagnetic activity, and the influence of temperature was greater than the influence of pressure. Brain strokes were more sensitive to inclement weather than myocardial infarctions. This paper provides quantitative estimates of the expected increases in hospital admissions on the worst days and can help to develop preventive health plans for cardiovascular diseases.


Geomagnetic storm Barometric pressure Daily temperature range Cardiovascular disease Hospitalization Risk factors 


It is believed that external triggers of acute cardiovascular events play as important a role as the known long-term risk factors of cardiac disease. The relative risk of the onset of myocardial infarction attributed to certain triggers may be two- to three-fold (Willich 1999). These triggers include morning activities after awakening and arising, physical exertion, sexual activity, emotional upset, overeating, excessive use of salt and alcohol, lack of sleep, holidays such as Christmas and New Year, as well as certain environmental factors like earthquakes, blizzards, adverse weather conditions or exposure to airborne particulates (Kloner 2006). The pathophysiological links between external triggers and the onset of myocardial infarction include stimulation of the sympathetic nervous system, catecholamine release, increase in arterial pressure, heart rate, blood viscosity, cholesterol and fibrinogen concentrations, red blood cell and platelet counts (Bhaskaran et al. 2009). Brain stroke can also be triggered by a variety of behavioral (e.g., bursts of anger or sudden change in body posture) and environmental (Koton 2003) factors. For example, such important physiological predictors of stroke as temporal increase in blood pressure, fibrinogen concentrations and serum lipids have been linked to changes in air temperature and barometric pressure within a few hours prior to stroke onset (Vlak et al. 2011; Jimenez-Conde et al. 2008; Landers et al. 1997; Hori et al. 2012; McArthur et al. 2010).

The available evidence of the impact of climatic conditions on acute cardiovascular morbidity is more limited and controversial than the findings regarding cardiovascular mortality. On the one hand, the rise in hospital admissions due to circulatory causes attributed to inclement weather is smaller than the rise in respective mortality (Linares and Diaz 2008). On the other hand, daily mortality data is generally more ready available and more reliable than statistics of hospital admissions (HA). In the absence of a national registry of cardiovascular morbidity in Russia, a population-based HA study was not feasible. The present study analyzed clinically validated HA with myocardial infarction (MI) and brain stroke (BS), recorded continuously in cardiovascular departments of two Moscow hospitals from 1992 to 2005.

Three environmental variables were selected for this study: ambient temperature, barometric pressure and geomagnetic activity; their impacts on cardiovascular outcomes were analyzed in a uniform fashion within a single statistical model. These three environmental factors have not been studied previously with equal attention. Ambient temperature and pressure have been studied more thoroughly; fewer publications have considered the health effects of geomagnetic activity (GMA). Attempting to rectify this, we place more emphasis on the latter in this paper.

The effects of daily variations in temperature may be statistically independent from the effects of cold or heat. Some researchers have studied day-to-day changes in temperature (which could be both positive and negative), while others studied diurnal temperature range (which is always positive). A U-shaped relationship between day-to-day variations in temperature and ischemic stroke risk was reported by Kyobutungi et al. (2005), while Hong et al. (2003) reported that only decreasing temperature affected the incidence of stroke; Morabito et al. (2011) reported that increasing temperature was a risk factor for stroke hospitalizations, while Messner et al. (2002) observed the same effect for non-fatal MI. Several studies reported positive associations between daily temperature range (DTR) and the incidence of stroke (Coelho et al. 2010; Sung et al. 2011; Lim and Kim 2011).

Changes in barometric pressure can also modulate the occurrence of vascular events. Decreased atmospheric pressure from the previous day was found to be associated with oxygen saturation—a screening test for early detection of heart obstructive disease—and congestive heart failure (Goldberg et al. 2008), the occurrence of acute myocardial infarction (Houck et al. 2005), ischemic stroke (Morabito et al. 2011), non-lacunar stroke (Jimenez-Conde et al. 2008) and hemorrhagic stroke (Dawson et al. 2008).

Geomagnetic field is a physical phenomenon resulting from the different rotation speeds of different layers of our planet. Incoming high-speed solar-wind streams and solar coronal mass ejections typically strike the Earth’s magnetic field 3 days after the solar flare, causing large-scale disturbances in the Earth’s magnetosphere—geomagnetic storms (Kavanagh and Denton 2007; Liemohn et al. 2010). These can be especially important for those living at high geomagnetic latitudes where the geomagnetic fluctuations have larger amplitudes (Oinuma et al. 2002; Palmer et al. 2006). Otsuka et al. (2001a) hypothesized that human magnetoreception is modulated by latitude and alternating light–darkness. Researchers have reported statistical associations between GMA and several cardiovascular health endpoints, including variations in capillary blood flow (Gurfinkel et al. 1995), paroxysmal atrial fibrillation (Stoupel et al. 1994), heart rate variability (HRV) (Gmitrov and Ohkubo 1999; Otsuka et al. 2001b; Oinuma et al. 2002), blood pressure (Stoupel et al. 1995a; Watanabe et al. 1994; Ghione et al. 1998; Watanabe et al. 2001), endocrine function (Stoupel et al. 1983) and MI-related deaths (Cornelissen et al. 2002; Stoupel et al. 1995b). An overview of recent results in heliobiology can be found in Palmer et al. (2006).

The goal of the present study was to find out how air temperature, atmospheric pressure, and geomagnetic activity influenced hospitalizations with MI and BS. The relative risks associated with these environmental factors were then compared to identify the principal drivers of acute cardiovascular events.

Materials and methods

Health outcomes and exposure characteristics

We obtained the dates of all physician-confirmed and valid cases of acute MI and BS in two large hospitals in Moscow, Russian Federation. Both hospitals reported daily counts of MI for the period between 1 August 1994 and 31 March 2002. Only one hospital reported daily counts of BS for the period 1 January 1992 to 28 December 2005. After elimination of hospital records with uncertain dates, the assembled data set consisted of 2,833 cases of MI (approximately 1 event per day) and 1,096 cases of BS (approximately 1 event per 5 days). No distinction was made for age and sex of the patients. Of course, our study covered only a small fraction of daily HA in Moscow. The average daily number of MI hospitalizations during the study period in Moscow, a city with a population of 10 million, varies between 30 and 40 (Interfax 2012) and the number of BS hospitalizations varies between 100 and 120 (Fedin et al. 2004). Therefore, our study covered about 3 % of HA with MI and 0.2 % of HA with BS in the city. Therefore, one cannot exclude possible selection bias in our study caused, for example, by varying patient admission policies of the two hospitals. It is of note that one of these (which supplied only MI data) is called “World War II Veterans Hospital” and treats mostly very senior patients.

Daily average measures of ambient air temperature and barometric pressure were calculated on the basis of 3-h averages reported by the weather station of Moscow State University. Describing geomagnetic disturbance levels, a researcher can use either local or global (planetary) indices. For this study, local GMA data were obtained from the observatory of Institute of Terrestrial Magnetism, Ionosphere and Radio Wave Propagation of Russian Academy of Sciences (IZMIRAN) located some 40 km south of Moscow.

Local disturbance levels are determined by measuring the range (difference between the highest and lowest values) during 3-h time intervals for the most disturbed of the two horizontal magnetic field components. First, however, the quiet-day variation pattern has to be removed from the magnetogram—a somewhat subjective procedure (Menvielle and Berthelier 1991). The range is then converted into a local K index taking the values 0 to 9 according to a quasi-logarithmic scale, which is station specific, with 0 representing an unusually quiet period, 1–4 representing the most typical conditions, and 5–9 marking geomagnetic storms (Gmitrov and Gmitrova 2004). The following definition of K variations was given by Siebert in 1971:

K variations are all irregular disturbances of the geomagnetic field caused by solar particle radiation within the 3-h interval concerned. All other regular and irregular disturbances are non K variations. Geomagnetic activity is the occurrence of K variations” (Siebert and Meyer 1996).

In this paper, we used the local K-indexes to construct a daily measure of local geomagnetic activity (A-index) similar to the planetary Ap index, but calculated at Moscow time (GMT + 4.00). Because of the non-linear relationship of the K-scale to magnetometer fluctuations, it is not meaningful to take averages of a set of K indices. What is done instead is to convert each K back into a linear scale called the equivalent 3-h amplitude or equivalent range a, using the middle class amplitude (MCA) table as described by Berthelier and Menvielle (1993). Then, the daily A-index is the average of the eight a-indices measured during that day. As a guide, the A-index is graded as follows: 0–7 quiet, 8–15 unsettled, 16–30 active, and > 30 stormy.

Discrete weather events

Discrete weather events were defined as several consecutive days with extreme weather conditions, for example, geomagnetic storms, heat waves and cold spells. A geomagnetic storm is a temporary disturbance of the Earth’s magnetosphere, which lasts typically for 24–48 h, but some storms may last for many days. A formal definition requires that during a geomagnetic storm Kp ≥ 5 (NOAA 2005). During the study period 1992–2005, the IZMIRAN Institute reported 363 geomagnetic storms, each lasting between 1 and 6 days, adding up to 1,337 stormy days.

In the absence of a standard definition of a cold spell or a heat wave (Meehl and Tebaldi 2004), we adopted the following symmetrical definitions in this paper: A cold spell is a succession of at least 5 days with daily mean temperatures below the 3rd centile of the long-term distribution of daily temperatures (−13.8 °C during 1992–2005), and a heat wave is as a succession of at least 5 days with daily mean temperatures above the 97th centile (+22.7 °C). During the 14-year period 1992–2005 we identified 6 cold spells and 11 heat waves, with average duration of 7.7 and 7.3 days, respectively. Summing up, we estimated total number of “extreme temperature event days” as 127 per 14 years. This is roughly 10 times less than the number of days with geomagnetic storms.

Statistical analysis

We first checked that the daily morbidity counts closely followed Poisson distributions without overdispersion (μ = σ2). This provided justification for applying Poisson generalized linear model to study relationships between the environmental factors and health outcome:
$$ ln\left[ {E(M)} \right]=\mathop{\sum}\nolimits{g_i}\left( {{X_i}} \right)+\mathop{\sum}\nolimits{\beta_j}{Y_j}+f(t) $$
where ln[E(M)] is the natural logarithm of expected daily counts of hospital admissions. The set of continuous predictors Xi included daily mean temperature T, temperature change from the previous day ΔT and diurnal temperature range DTR; daily mean barometric pressure P and its change from the previous day ΔP; and the local 24-h index of geomagnetic activity, A. Functions gi were modeled as either linear functions or restricted cubic spline functions with 3–5 knots. Linear functions were used to calculate the constant log-rates, where appropriate, while splines described the associations with continuous covariates in a more flexible way. We used a Wald-type test for nonlinearity of gi beyond the first knot. Categorical variables Yj included different ordinate terms for the days of the week (DOW), and binary variables indicating days with geomagnetic storms, cold spells and heat waves; regression coefficients βj were interpreted as log relative risks of morbidity during such events. We also allowed for a time lag of up to 1 week after exposure to each predictor and selected the lag at which the effect on hospitalizations was the greatest (peak lag). Due to strong collinearity, each model could contain either an A-index or a geomagnetic storm indicator, but not both (Pearson’s rstorm,A = 0.58).
A function of calendar time f(t) contained linear and quadratic terms and two sinusoidal terms with periods of 12 and 6 months to account for long-term (secular) trends and possible seasonal quasi-periodicity of health outcomes:
$$ f(t)=Const+{\beta_{lin }}t+{\beta_{quad\ }}{t^2}+{\beta_1}\cos \frac{{2\pi \left( {t-{\theta_1}} \right)}}{365 }+{\beta_2} \cos \frac{{4\pi \left( {t-{\theta_2}} \right)}}{365 } $$

In Eq. (2), thetas denote the phases of periodical oscillations. This representation of f(t) was used to test statistical hypotheses about the seasonal behavior of health outcomes. In the sensitivity analysis, we alternatively modeled f(t) as a restricted cubic spline function of time with regularly spaced knots every 120 days. The choice of deseasonalization algorithm had a negligible influence on the established relationships between the environmental variables and health outcomes.



Table 1 presents the summary statistics of continuous and categorical covariates separately for the MI and BS models. Analysis of the seasonal patterns of MI revealed a highly significant half-year periodicity (P = 0.003) with maximums on 21 October and 21 April, and minimums on 20 January and 21 July. The relative amplitude of these oscillations (measured as \( \frac{{{M_{\max }}-{M_{\min }}}}{{{M_{\min }}}}\times 100\% \) after correction for long-term trends) was 20 %. It is of note that MI oscillation followed GMA oscillations with a 16-day time lag; the relative amplitude of GMA oscillations was 39 % (P < 0.001). Seasonal variations of MI also coincided with seasonal variations in frequency of geomagnetic storms. The frequency of the latter may be measured by mathematical expectation, which is the ratio of average number of days with storms to average number of quiet days. This ratio was 0.30 in autumn and 0.25 in spring, versus 0.18 during winter and summer. Annual winter to summer oscillations of MI were not significant. In contrast with MI, BS displayed a significant (P = 0.023) annual periodicity, with a maximum on 18 January and a relative amplitude of seasonal oscillations of 8.5 %.
Table 1

Summary statistics on meteorology, geomagnetic data and outcome variables. MI myocardial infarctions, BS brain strokes


Model for MI

Model for BS

Study period

1 August 1994–31 March 2002

1 January 1992–28 December 2005

Number of validated hospitalizations



Number of cold spell days



Number of heat wave days



Number of geomagnetic storm days



Percentiles of distributions

1 %

50 %

99 %

1 %

50 %

99 %

Daily mean temperature (°C)







Diurnal temperature range (°C)







Barometric pressure (mm Hg)







Change in pressure from previous day (mm Hg)







Local 24-h A-index







Effects of ambient temperature

The expected MI rate decreased with temperature both in winter and in summer (Fig. 1a); the effect was delayed, with a peak lag of 6 days. No detrimental effect of heat was established. Conversely, the decrease in MI per each degree Centigrade was even greater at higher temperatures. For example, a 10 °C fall in daily mean T between −7 and −17 °C was associated with an increase in MI of only 2 %, while the same decrement from 25 to 15 °C was associated with a 21 % increase in MI. The expected BS rate increased with temperature in a broadly linear fashion across the entire range of annual temperatures (not shown in Fig. 1). Each 10 °C increment in daily temperatures led to a 16 % (95 % CI 2 % to 30 %) increase in BS, with a 1-day lag.
Fig. 1

Relationship of myocardial infarction (MI) hospitalizations to air temperature (a), barometric pressure (b), change in pressure from previous day (c), and geomagnetic activity index (d). Relationship of brain stroke (BS) hospitalizations to diurnal temperature difference (e), barometric pressure (f), change in pressure from previous day (g), and geomagnetic activity index (h). For a given value of the predictor, each curve allows outcome to be compared to the average of the outcomes by considering all values of this predictor, while taking into account the confounding effects of other environmental variables included in the model, the seasons, long-term trends and days of the week (e.g., a relative risk of 1.3 at a given value of the predictor indicates that we expect 30 % more hospitalizations than on average)

The relationships of MI and BS with temperature change from day to day ΔT were not statistically significant. A statistically significant relationship with DTR was established only for BS, i.e., a J-shape with the minimum near DTR = 4 °C (Fig. 1e). A 10 °C increase in DTR from 4 to 14 °C was associated with a 26 % (2 % to 57 %) increase in BS, with a peak lag of 5 days.

Analysis of heat waves and cold spells showed two significant health effects: relative risk of MI diminished during heat waves with 1-day peak lag: RR = 0.68 (0.48–0.95), P = 0.023; relative risk of BS almost doubled during cold spells with 2-day peak lag: RR = 1.91 (1.07–3.41), P = 0.029. In the subsequent sensitivity analysis, we relaxed the definition of heat waves and cold spells, and required that three or more successive days would constitute a heat wave or a cold spell (using the same temperature thresholds). Despite an increase in statistical power of test samples, the resultant health risks lost statistical significance.

Effects of barometric pressure

The observed relationships with daily mean pressure P and day-to-day change in pressure ΔP had a U-shape for MI (Fig. 1b, c), and descended steadily for BS (Fig. 1f, g). The relationship between MI and P (Fig. 1b) had a minimum at 746 mm Hg; a decrease in pressure from 746 to 724 mm Hg (the 1st centile) was associated with a relative risk RR = 1.27 (1.02–1.58); an increase from 746 to 763 mm Hg (the 99th centile) was associated with RR = 1.21 (1.00–1.47), with a peak lag of 4 days. The relationship with day-to-day change in pressure ΔP had minimum at ΔP = −2 mm Hg (Fig. 1c). A decrease in ΔP from −2 to −13 mm Hg (the 1st centile) was associated with RR = 1.11 (0.90–1.37); an increase from −2 to +12 mm Hg (the 99th centile) was associated with RR = 1.25 (1.04–1.50); this relationship was immediate (lag = 0).

Because the relationships BS(P) and BS(ΔP) decreased monotonously, they can be described by relative risks associated with variation in the predictor between the 99th and the 1st centiles. A decrease in pressure from 764 to 725 mm Hg was associated with RR = 1.57 (1.03–2.40), with a peak lag of 3 days (Fig. 1f). Similarly, a decrease in ΔP from 12 to −13 mm Hg was associated with RR = 1.65 (1.06–2.56), with a lag of 5 days (Fig. 1g). The variation in BS caused by changing air pressure is thus much greater than the variation in MI.

Effects of geomagnetic field

The number of MI and BS increased steadily with A-index (Fig. 1d, h). Both relationships were close to log-linear but the relationship for BS flattened out at A ≥ 20, or above the 80th centile of the long-term distribution. The corresponding log rates per unit increase in A-index were 0.38 % (0.05 % to 0.70 %) for MI (with zero lag) and 0.9 % (0.1 % to 1.7 %) for BS (with 1-day lag). The impacts of geomagnetic storms on MI and BS were both immediate and highly significant: RR = 1.29 (1.19–1.40); P < 0.001 for MI and RR = 1.25 (1.10–1.42); P = 0.001 for BS. According to the NOAA space weather scale, about 62 % of “stormy” days are “minor” geomagnetic storms (NOAA 2005). Can exclusion of minor storms from the test sample result in greater health risks? We did not observe this: after exclusion of days with minor storms from the test sample, the resultant health risks lost statistical significance.

Analysis of confounding

All the results reported above were derived from a “full model” (1), which contained all continuous variables Xi mentioned in Materials and methods. Therefore, the reported risk estimates have been adjusted for all covariates. However, certain environmental variables appear to be highly collinear, which can partly explain the differences in effect estimates reported in different studies. There are two pairs of variables that deserve special attention because of moderate collinearity: T, DTR and P, ΔP. In our study, Pearson’s rT,DTR = 0.45 and rP,ΔP = 0.28; most researchers suggest |r| > 0.5–0.7 as a threshold of high collinearity beyond which it begins to severely distort model estimation and subsequent prediction (Dormann et al. 2012). Exclusion of a collinear variable from the full model and observing the degree of discrepancy between the crude and adjusted estimates helps to evaluate the magnitude of confounding. Let us use BS as an outcome variable to illustrate this:
  • Temperature effect: As follows from Table 2, temperature effect can be measured by log rate: relative increases in BS per 1 °C were βadj =1.6 % vs βcrude = 1.8 %.

  • DTR effect: an increase in DTR from 4 to 14 °C is associated with a 26 % increase in BS in the adjusted model vs a 29 % increase in the crude model.

  • Pressure effect: a decrease in pressure from 764 to 725 mm Hg is associated with a 57 % increase in BS in the adjusted model vs a 73 % increase in the crude model.

  • ΔP effect: a decrease in ΔP from 12 to –13 mm Hg is associated with a 65 % increase in BS in the adjusted model vs a 86 % increase in the crude model.

Table 2

Overview of established dose–response relationships. GMA Geomagnetic activity

Meteorological or geomagnetic parameter




Relative amplitudea

Peak lag, days


Relative amplitude

Peak lag, days

Air temperature

Steady decrease

48 %


Linear increase

97 %


Diurnal temperature range



30 %


Barometric pressure


27 %


Steady decrease

57 %


Change in pressure from previous day


27 %


Steady decrease

65 %


24-h local GMA index

Linear increase

18 %


Steady increase

19 %


Geomagnetic storms

RR = 1.29 (1.19; 1.40) P < 0.001



RR = 1.25 (1.10; 1.42) P = 0.001



Cold spells


RR = 1.91 (1.07; 3.41) P = 0.029



Heat waves

RR = 0.68 (0.48; 0.95) P = 0.023




aMaximal variation in morbidity established along the dose–response curve, 100 % × (MmaxMmin)/Mmin; only relationships significant at 0.05 level are shown

We conclude that the magnitude of confounding between T and DTR is still quite small. However, our results indicate significant confounding between P and ΔP : exclusion of a collinear variable leads to overestimation of risk associated with the remaining variable by roughly one-third. Although bivariate correlations among some other predictors in our model were also statistically significant, collinearity was small (|r| < 0.15) and we checked that the difference in the results derived from unadjusted and adjusted models was negligible.


This paper describes the relationships between the onset of acute cardiovascular events, weather and geomagnetic activity, taking into account temporal confounding by DOW, seasons, and longer-term trends, presumably caused by changing patient admission policies of the hospitals.

Concerning the effects of temperature, we established that the temperature–morbidity curves for MI and BS run in opposite directions across the entire range of annual temperatures: MI risk decreased with temperature, while BS risk increased. We also established a significant decrease in MI during heat waves. It is interesting that the estimated effects of heat in Moscow (each 10°C increment in daily temperatures led to a 17 % decrease in MI and a 13 % increase in BS) have been confirmed during the anomalous heat wave in Moscow in July 2010. The total number of hospital admissions in Moscow in July of 2010 due to MI decreased by 20 % in comparison with July of 2009, while total number of BS increased by 10 % (RIA 2010). Although a detrimental effect of cold was not established for BS along the temperature–morbidity curve, we observed a statistically significant increase in BS during cold spells. These seemingly conflicting findings can be explained if the detrimental effect of cold on BS is observed only at extremely low temperatures where temperature–morbidity curve loses its statistical significance.

How do these findings fall within an international context? Despite the growing number of air temperature/morbidity studies, there is disagreement about the direction of the effect of temperature on risk of MI and BS. A systematic survey of daily time-series MI studies showed that only three out of nine site-specific studies reported a detrimental effect of heat on non-fatal MI events, whereas five out of nine studies (with data from winter season) reported detrimental effect of cold (Bhaskaran et al. 2009). In light of Bhaskaran’s findings, a detrimental effect of cold on non-fatal MI events seems more likely than a detrimental effect of heat, which agrees with our results. The inverse association between MI and temperature all year (including warm period) has been also confirmed in a worldwide 17-country study based on monthly time series (Chang et al. 2004), and more recently in Denmark (Wichmann et al. 2012), in Italy (Abrignani et al. 2009), in Korea (Lee et al. 2010).

The association of BS with temperature is even more complex because different types of stroke may respond differently to changes in ambient temperature (Wang et al. 2009; Feigin et al. 2000). About 80 % of all brain strokes are ischemic strokes, about 12 % are hemorrhagic strokes, and other types make up the rest. Giua et al. (2010) cited five international studies that reported a detrimental effect of heat on hospitalizations with various types of strokes, similar to our findings for total stroke (we could not differentiate particular types of stroke). On the other hand, Magalhães et al. (2011); Wang et al. (2009) and Feigin et al. (2000) reported detrimental effect of cold on ischemic stroke admissions, and the same effect on hemorrhagic stroke was reported by Morabito et al. (2011). The controversial findings of international studies on stroke led Giua et al. (2010) to conclude that the underlying relationships may indeed differ in different climatic areas.

Our findings regarding the influence of air pressure and variations in air pressure on MI and BS seem to agree with most results obtained in other countries (Dawson et al. 2008; Abe et al. 2008). For example, one important result of our study is that low pressure and falling pressure can present risk factors for both MI and stroke onset. Concerning possible pathophysiological mechanisms, Jimenez-Conde et al. (2008) noted that hypobaric conditions activate thrombosis in air travel. Changing air pressure may directly influence vessel walls, triggering endogenous inflammatory mechanisms and changing their endothelial function. On the other hand, Morabito et al. (2008) discovered that sudden changes from stable anticyclonic weather to cyclonic weather, accompanied by large drops in pressure, were associated with the highest winter blood pressure. Among all meteorological variables, fluctuations of atmospheric pressure (FAP) were found to be correlated most closely with HRV (Delyukov et al. 2001). Impaired HRV is one of the most powerful predictor of acute vascular events. At the same time, FAP accompany magnetic storms (Rostoker and Sharma 1980) and changes in solar activity (Shindell et al. 1999), and could be one of the intermediaries between large-scale geomagnetic phenomena and the biosphere (Delyukov et al. 2001).

Disturbances of the Earth’s geomagnetic field are planetary in character and are distributed evenly over almost the whole hemisphere. Comparison of the local A-index with the planetary Ap index (available from the World Data Center for Geomagnetism, Kyoto at showed nearly perfect correlation between them (Pearson’s r = 0.993 during the period 1992–2005). One source of the difference between the two indices is that each observatory has its individual annual cycle of daily K-variations according to its geographic and geomagnetic coordinates. Although most researchers use the planetary Ap index in epidemiological studies, we preferred using the local GMA index in this study, because the difference in time zones (Ap is measured at UTC/GMT) could be essential for immediate effects. If the relationship between MI and GMA is indeed immediate (as Table 2 suggests), then the association between daily MI outcomes and local 3-h a-indexes should have a peak in the morning, because myocardial infarctions occur mostly in the morning hours: for example, ISIS-2 (1992) reported a marked circadian pattern of MI with maximum occurrence between 8:00 and 11:00 a.m. We observed this peak when we regressed MI against the eight 3-h a-indices, one at a time: MI hospitalizations were most sensitive to a9-12 (Fig. 2).
Fig. 2

Predicted percentage increases in daily number of hospitalizations with MI as a function of local 3-h geomagnetic activity index a (the equivalent range) measured during that day

This finding confirmed the existence of morning MI peaks in Moscow and an immediate link between GMA and infarctions. This link may be explained by experimentally proved instant hemodynamic effects of geomagnetic disturbances. Red blood cells (RBS) are extremely sensitive to electromagnetic forces. Previous research has revealed the relationships between GMA and platelet aggregation, blood coagulation and microcirculation disorders (Gurfinkel et al. 1995; Pikin et al. 1998; Gurfinkel and Voeikov 2006). It was hypothesized that human magnetoreception may regulate production of adrenal hormones responsible for activation of the clotting system, and the increase in aggregation and spasm in the afferent vessels of the microcirculatory network. Assumption of upright posture in the morning triggers MI in much the same way, causing hemodynamic stress by increasing platelet aggregation ability (Tofler et al. 1987; Brezinski et al. 1988).

Because of its immediate effects on cardiovascular health, GMA is somewhat similar to the conventional triggers with the induction time (hazardous period) of only 1–2 h before the onset of an acute cardiovascular event (Koton 2003). However, the lagged effects of changing air temperature and pressure observed in this study cannot be explained in the same manner, because of the considerable delay between the “environmental trigger” and the response variables. Although many other studies have reported similar results with respect to time lags, e.g., Wolf et al. (2009) reported a 3-day lag between non-fatal MI and temperature decrease, the true effect is likely to have a more complex lag pattern distributed over longer time periods. Bhaskaran et al. (2009) studied lags of up to 28 days and reported that the strongest effect of decreasing temperature on MI was observed at lags of 8–14 days. Obviously, lags do not have to be the same for heat and cold, which our model failed to take into account.

Table 2 summarizes the established dose–response relationships between hospitalizations and meteorological variables and reports maximal variations in morbidity, attributed to each environmental factor. This measure is called the ‘relative amplitude’ and is defined similarly to the amplitude of seasonal oscillations \( \frac{{{M_{\max }}-{M_{\min }}}}{{{M_{\min }}}}\times 100\% \). For linear dose–response relationships, it corresponds to variation of independent variable between the 1st and the 99th centiles. For J-shaped or U-shaped relationships, spread is measured between the minimum and maximum of the dose–response curve. Thus, relative amplitude provides a flexible measure to compare relative variations in morbidity along monotonous and non-monotonous dose–response curves. In our study, this measure was used to compare potential influences of variations in air temperature, pressure and GMA on hospital admissions. Only statistically significant health effects were considered. It is important to emphasize that all relative amplitudes were calculated from the same model adjusted for all covariates. Table 2 shows that the influence of pressure is greater than the influence of GMA, and that the influence of temperature is greater than the influence of pressure. This ranking remains the same for hospital admissions with myocardial infarctions and brain strokes. One may also see that BS hospitalizations are generally more sensitive to weather variables than MI hospitalizations.

The findings presented in this paper provide quantitative estimates of the expected increases in hospital admissions on the worst days, which can be used by decision makers in the area of public health protection.


The authors thank Dr. Julia Shugai from the Skobeltsyn Institute of Nuclear Physics for her consultations on geomagnetic field issues. This research was supported by the grant program of Presidium of Russian Academy of Sciences “Fundamental Sciences for Medicine”.

Copyright information

© ISB 2013

Authors and Affiliations

  • Dmitry Shaposhnikov
    • 1
    • 4
  • Boris Revich
    • 1
  • Yuri Gurfinkel
    • 2
  • Elena Naumova
    • 3
  1. 1.Environmental Health Laboratory, Institute of ForecastingRussian Academy of SciencesMoscowRussian Federation
  2. 2.Central Hospital of Joint-Stock Company Russian RailwaysMoscowRussian Federation
  3. 3.Tufts University School of MedicineBostonUSA
  4. 4.Institute of ForecastingMoscowRussia

Personalised recommendations