Circadian Variations in the Occurrence of Cardiac Arrests
Initial and Repeat Episodes
Background—Patterns of temporal variation of cardiac arrests may be important for understanding mechanisms leading to the onset of acute cardiovascular disorders. Previous studies reported diurnal variation of the onset of cardiac arrests, with high incidence in the morning and in the evening, lack of daily variation during the week, and some seasonal variation. The association between the time of day and recurrent cardiac arrests has not been previously examined.
Methods and Results—We explored temporal variation in 6603 out-of-hospital cardiac arrests attended by the Seattle Fire Department. The data exhibit diurnal variation, with a low incidence at night and two peaks of approximately the same size (at 8 to 11 am and 4 to 7 pm). The evening peak is attributed primarily to the patients found in ventricular fibrillation, whereas arrests that show other rhythms exhibit mainly a morning peak. Cardiac arrests associated with survival have more pronounced diurnal variation than episodes in which survival did not occur. This difference persists after adjustment for rhythm. For 597 patients who had at least two separate cardiac arrests, we found no overall association between the times of day of the recurrent arrests. For women, however, the times of day of the first and second arrests were closer to each other than one would expect if the times were entirely unrelated.
Conclusions—Cardiac arrests do not occur randomly during the day, but rather follow certain periodic patterns. These patterns are probably associated with patterns of daily activities. The hypothesis that cardiac arrests are triggered by a person’s activity rather than by some underlying characteristics of his or her disease is supported by the lack of association between the times of the first and second arrests in the patients with recurrent arrests.
Many studies in the past decade have demonstrated diurnal variation in the onset of acute cardiovascular disorders, such as myocardial ischemia,1 2 3 acute myocardial infarction,4 5 6 7 8 9 10 11 12 cardiac arrhythmias,13 14 15 and sudden cardiac death.16 17 18 19 20 21 22 The results consistently show an increased incidence of acute cardiac events in the morning hours (6 am to noon) and a low incidence at night. Some data suggest another late afternoon or evening peak of incidence.
Circadian variation may differ according to certain characteristics of the patients. Non–Q-wave infarctions6 23 and recurrent infarctions11 are reported to exhibit less distinctive or no circadian variation. Drugs may suppress diurnal variation, specifically β-blockers,4 6 7 12 13 antiarrhythmic drugs,13 and aspirin.9 These findings have potentially significant implications for understanding of mechanisms leading to acute cardiovascular disorders and consequently for prevention and therapy.24 25 26 27
Studies describing circadian variation of sudden cardiac deaths use a variety of data sources: death certificates and witness interviews,16 17 18 populations defined by the availability of information,21 22 and information provided by EMS.18 20 28 Reviewing death certificates would capture all cardiac deaths; however, the diagnosis and information on death certificates are often imprecise. Furthermore, these studies would miss all survivors of cardiac arrest. EMS data usually provide precise time of day. However, their data may not include all events, because some of the patients are not attended by EMS. Specific populations, such as residents of nursing homes or patients with a particular diagnosis who may be followed by some registry, do not offer widely generalizable results.
This study of temporal variations of cardiac arrests uses a database of cardiac arrests attended by the SFD, which responds to all emergency medical calls in the city of Seattle. Using this database of 6603 cases of out-of-hospital cardiac arrest permitted the study of circadian variation for several patient subgroups. Particularly unique is the opportunity to study the time of the recurrent cardiac arrest in relation to the time of the initial cardiac arrest in 597 patients who were resuscitated from a previous arrest.
The database of cardiac arrests was started in 1970 and has continued to the present without interruption. Cases are identified by personal review of all run reports submitted by SFD paramedics, and survivors have been followed up through hospitalization and at least annually thereafter.29 Among a case series of 6603 out-of-hospital cardiac arrests attended by SFD from March 1985 through February 1993, there were 183 cardiac arrests in children (≤15 years old); 166 arrests in patients whose age was unknown; 13 arrests with unknown time of EMS dispatch; 993 arrests with presumed noncardiac etiology (eg, trauma, drug overdose, respiratory arrest); and 1558 arrests for which time of collapse could not be estimated because these were untreated events, ie, pronounced dead on EMS arrival, and either were unwitnessed or witness status was unknown. The remaining 3690 episodes represent out-of-hospital adult cardiac arrests of cardiac etiology that were either witnessed (1750) or treated with ALS. The structure of the data is shown in Table 1⇓. We were interested in the circadian variation of the 5248 adult cardiac arrests of cardiac etiology.
Time of Cardiac Arrest
For witnessed or treated cases, the time of EMS dispatch was taken to be the time of cardiac arrest. In these cases, the delay in requesting aid is highly unlikely to have exceeded 20 minutes.
For unwitnessed arrests and untreated patients, there is no good surrogate for the time of collapse. Most such patients (99%) are found in asystole, and there was no means to reliably estimate how long the patient had been dead. These deaths generally occurred during the night. In some cases, an interval during which the arrest occurred might have been estimated, but this is not routinely recorded.
For the analysis of circadian patterns in subgroups, we chose to restrict analysis to the 3690 witnessed or treated episodes in which we were reasonably confident of the time of arrest. These represented ≈70% of the episodes attributed to cardiac causes (Table 1⇑). Because others have examined overall circadian patterns, our primary interest was in how the patterns might differ by subgroup.
For the analysis involving repeated arrests, the EMS reports, particularly the narrative, were reviewed, and the time of collapse was estimated by the center of the interval if one was available and otherwise, by a random time between 10 pm and 6 am for those who died at night.17
To explore the circadian pattern, we used three techniques: the histogram (dividing the day into 24 categories of 1-hour length with centers at whole hours), a kernel estimate (a nonparametric technique to fit a smooth curve30), and harmonic polynomials (fitted by the maximum-likelihood method). The χ2 test was used to test for the presence of circadian variation against the hypothesis of uniform distribution of arrests.
We compared circadian variation in subgroups according to sex, age (16 to 40 years, 41 to 65 years, and ≥66 years old), days of the week (working days versus weekend), rhythm on arrival of emergency personnel (VF, asystole, PEA), race (white, black, Asian, native American), and survivor status (resuscitated and discharged alive from hospital, died in hospital, died with no hospital admission) by the χ2 test (dividing the day into six 4-hour intervals), the Kolmogorov-Smirnov test,31 and a likelihood ratio test32 associated with the fitted harmonic polynomials. Log-linear models33 were used to explore the influence of each covariate after adjustment for the effect of other covariates. For the log-linear models, the day was divided into six 4-hours intervals.
We assumed that missing data were missing at random, and patients with missing covariate values were omitted from the analyses in which these covariates were needed. Because certain categories were rare (specifically the race categories of native Americans and Asians), some analyses were performed excluding these subgroups.
We identified 597 patients who experienced ≥2 separate out-of-hospital cardiac arrests between March 1970 (when the database was started) and March 1993. For the patients who had more than one recurrent arrest, we used only the data about the first recurrence.
To investigate whether the time of the initial arrest was similar to that of the second arrest, we used absolute differences between the two times of day. We compared the mean, median, and overall distribution (Kolmogorov-Smirnov test) of these differences to the expected mean, median, and distribution of the differences under the hypothesis that there was no association between the times. The expected distribution was obtained by sampling random pairs from the times of the first and the second arrests. The set of pairs was resampled 500 times.
We also tested the null hypothesis in specific subgroups; men, women, 3 age categories (16 to 40 years, 41 to 65 years, and ≥66 years old), and categories according to the length of period between the arrests (0 to 5 months, 5 to 18 months, 18 to 54 months, and ≥54 months) based approximately on the quartiles.
In addition, (1) patients having both cardiac arrests at approximately the same time of day (differences up to 2 and 4 hours) were compared with other patients as to their characteristics (sex, age, months between arrests, medical history, and medications as prescribed at the time of the second arrest) and (2) a regression model for the absolute difference between the times of arrests was built to investigate independent influence of these characteristics.
Figure 1⇓ shows the histogram, the kernel estimate, and the maximum-likelihood harmonic polynomial estimate of the distribution of the arrests that were witnessed or that were treated with ALS (3690). All 3 methods result in a similar pattern: a low incidence at night, a sharp increase between 8 am and noon with peaking at ≈10 am, a relatively high incidence during the day, and another peak between 5 and 8 pm. Both morning and evening peaks are of a similar magnitude. The hypothesis of uniform distribution of the cardiac arrests over all 24 hours was rejected (χ2=252.5, df=24, P<0.0001).
The pattern of dispatch times of the unwitnessed and untreated cases exhibits a strong morning peak, between 8 am and noon, probably because many people who died at night were discovered in the morning hours (Figure 2⇓).
The distribution of subgroups among the 3690 witnessed or treated events is shown in Table 1⇑. No difference in circadian variation of cardiac arrests was found between men and women (all P>0.36 for the χ2, maximum-likelihood, and Kolmogorov-Smirnov tests); between whites, blacks, and Asians (all P>0.12); and between working days and weekends (all P>0.11). Results for age were equivocal (for the χ2 and the maximum-likelihood, P>0.23; but for the Kolmogorov-Smirnov test, P=0.0258) and suggested that elderly patients (≥66 years old) have different circadian variation than middle-aged patients.41–65 Circadian variation differed significantly according to heart rhythm on EMS arrival (P<0.0001) and survivor status (P<0.0001).
Figure 3⇓ shows the harmonic polynomials fitted to the distribution of arrests in three age groups. The scale of the fitted polynomials was chosen so that the area under the curve is 1. Young patients had a relatively higher incidence at night than older patients and also exhibited a higher incidence in the morning than in the evening. Patients in the age category 41 to 65 years had a higher evening peak. Elderly patients maintained a rather high incidence during the daytime hours, with less pronounced peaks.
The distribution of VF events follows the general pattern except that the evening peak is substantially higher than the morning peak (Figure 4⇓). Asystole episodes exhibit primarily a morning peak beginning earlier than for other rhythms. Cases with PEA follow a bimodal density with a prominent morning peak.
Baseline covariates are highly correlated (Table 2⇓): rhythm with age (middle-aged patients are more likely to be found in VF, whereas the elderly are more often found in asystole, P<0.0001), rhythm with race (whites have a higher probability of being found in VF than blacks and Asians, P=0.007), and age with race (blacks tend to have cardiac arrests at a younger age, P<0.0001). Among these characteristics, only rhythm was significantly related to the circadian distribution of arrests after adjustment for the other factors (P<0.0001 from the log-linear model).
Although the distributions of arrests in the groups determined by survival status appear similar (Figure 5⇓), they are significantly different. If patient characteristics, including rhythm, are adjusted for, the circadian pattern of arrests still differs according to survivor status (P<0.0001 from the log-linear model). If analysis is restricted to cases with VF on EMS arrival (Figure 6⇓), patients who died before they could be admitted to the hospital had less pronounced circadian variation, with a relatively high incidence at night and no morning peak (P<0.0001).
Between March 1970 and March 1993, 597 cases of recurrent events were observed in the survivors of a prior cardiac arrest. The time of EMS dispatch can be considered to be within 1 to 20 minutes of the time of arrest for all resuscitated patients; hence, time of the onset of the first arrest can be estimated by the time of EMS dispatch. The time of the second out-of-hospital arrest could be estimated closely for 325 cases: 318 were either witnessed or treated with ALS (251 by SFD and 67 by other medical personnel); the remaining 7 cases were not attended by medical personnel, but time of collapse was obtained from witnesses. Table 3⇓ displays the distribution of covariates for these 325 repeat cardiac arrests.
Of the 272 episodes for which the time of onset of second arrest could not be closely approximated by dispatch time, an interval of <6 hours is available for 110 episodes, and 59 arrests are known to have occurred at night. Figure 7⇓ (top) shows the distribution of the first and second cardiac arrests in the sample of 325 patients whose time of second arrest is accurately known. Compared with the times of first arrest, the distribution of the second arrest has a higher incidence during the night, with two small peaks of similar size in the morning and in the evening. The middle panel shows the distribution for the 435 patients whose time of second arrest was known either precisely or imprecisely (within 6 hours). For the latter, the center of the interval was assigned as the time of arrest. The bottom panel corresponds to 494 patients whose second arrest time was known either precisely or imprecisely or who died at night. For those who died at night, a random time between 10 pm and 6 am was assigned as the time of the second arrest. Circadian variation of the second arrest when imprecise time arrests were added was still present (P<0.0001) but was less pronounced than the circadian variation of the precisely timed second arrests.
When the mean and the median absolute differences of the arrest times were compared with those expected under the hypothesis of no association, no evidence of a significant association between the times of the arrests was found (P=0.12 and P=0.19 for the mean and median, respectively). The Kolmogorov-Smirnov test comparing the distribution of the absolute differences with the expected distribution also showed no evidence of association (P=0.64). Subgroups based on sex, age at first arrest, and length of the period between arrests were separately examined. Women formed the only subgroup having the times of the first and second arrests significantly closer together (P=0.029, P=0.054, and P=0.020 for comparison of the mean, median, and overall distribution, respectively) than would occur by chance.
Grouping patients according to whether the times of their two episodes differed by <4 or >4 hours, we found that only sex (P=0.0006) and β-blocker use (P=0.0206) differed significantly between the groups: women constituted 22.8% and 8.6% of those with and without arrests within a 4-hour window, respectively; β-blockers were taken by 12.7% and 27.3%, respectively. Of those having arrests within a 2-hour window, 42.6% took antiarrhythmic drugs (other than amiodarone) compared with 57.4% in those with arrests outside a 2-hour window (P=0.0450).
We built a regression model with the absolute difference between the arrest times as a response and used sex, β-blocker use, and antiarrhythmic use (other than amiodarone) as covariates. Because of missing values in β-blocker and antiarrhythmic usage, only 207 patients could be included in the model. In the multivariate model, only sex (P=0.0238) and β-blockers (P=0.0488) were significant: the absolute difference between the two arrest times is estimated by the model to be 1.75 hours shorter for women than for men, and prescription of β-blockers is estimated to increase the difference between the times of arrest by 1.14 hours.
Most studies of circadian variation of sudden cardiac death have found a low incidence of events at night and a substantial incidence in the morning (6 am to noon).16 17 18 19 20 Some studies reported a secondary peak in the late afternoon or evening hours, usually around 5 to 7 pm,16 18 20 or sometimes later.22 Our data support these earlier findings but suggest the evening peak to be of approximately the same magnitude as the morning peak of incidence. This discrepancy may be caused partly by the exclusion of unwitnessed and untreated events from our analysis. We felt that exclusion of these patients could not be avoided because, if the patient is pronounced dead on arrival of EMS, the time of onset cannot be known precisely and may substantially precede the time of call. Excluded arrests with unknown time of onset (unwitnessed and untreated) were similar to the witnessed or treated arrests as to age, race, and day of the week. However, a higher proportion of the excluded arrests occurred in women (Table 1⇑). Circadian variation of unwitnessed but treated arrests was very similar to circadian variation of witnessed arrests, confirming our belief that the influence of uncertainty of the precise time of onset in unwitnessed but treated events is minimal.
The EMS is not called in an unknown, but probably small, proportion of cardiac arrests. Evaluation of the recurrent arrests (Figure 7⇑) suggests how the incidence at night may be underestimated if unattended arrests are omitted. However, recurrent arrest patients are a specific subgroup that may differ from the general population of cardiac arrest patients. Interestingly, the first arrest in the recurrent arrest patients does not exhibit an evening peak, whereas in general, recurrences do exhibit an evening peak of about the same size as the morning peak.
Most of the studies reported to date have not had enough cases or available information to detect differences between the circadian variation of subgroups with reasonable power. Some studies noted that no difference was observed between the circadian variation of men and women16 17 20 and different age groups.16 17 22 A study using emergency data from Berlin20 showed that the characteristic bimodal pattern of the distribution of cardiac arrests was attributed mainly to the patients who had VF on arrival of EMS. Other rhythms were also found less frequently at night, but the overall circadian variation was less pronounced. In the same study, younger patients (<65 years old) had a low incidence of cardiac arrests at night and a high incidence during the day, with moderate peaks in the morning and in the evening. Elderly patients (>65 years old) showed a very prominent morning peak and almost no evening peak.
In accord with the Berlin study,20 we found that the bimodal pattern is mainly an attribute of VF and that patients with asystole or PEA had a less pronounced evening peak. We also observed no difference in circadian variation between sexes, races, or days of the week. Our analysis suggests that differences between the age categories are probably explicable by different rhythm distributions and thus are more likely due to sociodemographic factors than to biological differences (eg, older persons are more likely to be living alone and thus less likely to have their cardiac arrest directly witnessed with a corresponding prompt call to 911, resulting in a higher likelihood of being found in asystole compared with VF).
We found a significant difference between circadian variation of arrests according to whether the victims survived. We suspect that this may reflect a relation between survival and activity of the patient before collapse and is not due to any circadian variation in EMS (which might occur, eg, from out-of-service times). Interestingly, a difference between survivor and nonsurvivor circadian variation persists even after rhythm on EMS arrival is adjusted for and is present if we restrict to VF episodes only.
The concept of circadian variation in arrest times due to biological or behavioral circadian patterns might be tested in cases in which a patient serves as his or her own control. However, to the best of our knowledge, no prior study has analyzed recurrent arrests in patients who were resuscitated from a first arrest. Because survivors of cardiac arrests in Seattle are regularly followed up and information about subsequent arrests is collected, we were able to identify a relatively large population of patients with recurrent cardiac arrests. We tested the hypothesis that a given patient is likely to experience recurrent cardiac arrests at approximately the same time of the day against the hypothesis that the times of cardiac arrests of a patient are entirely random, ie, unrelated. Overall, we found no evidence that the patients had subsequent cardiac arrests at a time of day similar to their first arrests. In women, however, the second arrest occurred closer to the time of the first arrest than might be expected by chance if the times were truly unrelated. This finding did not seem to be entirely explicable by the difference between prescribed medications but must be considered hypothesis-generating only because it may well be a chance result of multiple comparisons. The fact that β-blockers suppress circadian variation of myocardial infarction was already noted in several studies.4 6 7 12 If β-blockers also suppress circadian variation of cardiac arrests, we would expect that the interval between the times of day of the two arrests would widen, because the expected difference between the arrest times if no circadian variation is present is larger than in the presence of circadian variation. Hence, our finding supports the hypothesis of suppression of circadian variation of cardiac arrests by β-blockers.
We do not believe that the overall lack of association between the time of day of initial and recurrent arrests is due to inadequate sample size or lack of comparability of the data between different time periods. The EMS system and associated data capture and management system in Seattle has been remarkably stable since 1970. Although a sample size of 325 is not large, it should be adequate for detection of modest effects; however, we did not find even a trend toward an association.
Tools for analyzing circadian variation data are imperfect. Although histograms and harmonic polynomials are useful, one has to be cautious when interpreting the graphs of circadian variation. Impressions from histograms often depend on the width and the starting point of the bars. Harmonic polynomials, although often used to describe circadian variation, may sometimes exhibit too much of their harmonic nature. Log-linear models and χ2 tests may depend on particular partitioning of the time scale. Nevertheless, we believe that our results are reliable, because they were essentially confirmed by several analyses. One must be cautious, however, about the interpretation of subgroup analyses in repeated-arrest data, because some covariates (especially medication use) had a substantial number of missing values, and therefore some results are based on a limited number of observations.
Using the Seattle database of out-of-hospital cardiac arrests, we confirmed the diurnal variation of the onset of cardiac arrests reported in other studies. The pattern of diurnal variation differs depending on the rhythm on arrival of the EMS. The circadian variation of arrests in those who survived differs from those who did not survive, even after adjustment for differences in rhythm. This finding is not explicable by varying availability of emergency care at different times of day.
Overall, we found no association between the time of the first and the time of the second arrest for patients with recurrent events. In the subset of women, however, we found some indication of such association. The general lack of association between the times of recurrent arrests may indicate that there is little relationship between the characteristics of the patient and the time of his or her arrest. Rather, it may be the patient’s activity and/or the environmental influences that trigger the arrest.
Selected Abbreviations and Acronyms
|ALS||=||advanced life support|
|EMS||=||Emergency Medical Services|
|PEA||=||pulseless electrical activity|
|SFD||=||Seattle Fire Department|
This work was supported in part by the Medic One Foundation, grant R01-HS-08197-03 from the Agency for Health Care Policy and Research, and contract N01-HC-65042 from the National Heart, Lung, and Blood Institute.
- Received December 1, 1997.
- Revision received February 25, 1998.
- Accepted March 1, 1998.
- Copyright © 1998 by American Heart Association
Rocco MB, Barry J, Campbell S, Nabel E, Cook EF, Goldman L, Selwyn AP. Circadian variation of transient myocardial ischemia in patients with coronary artery disease. Circulation. 1987;75:395–400.
Andrews TC, Fenton T, Toyosaki N, Glasser SP, Young PM, MacCallum G, Gibson RS, Shook TL, Stone PH. Subsets of ambulatory myocardial ischemia based on heart rate activity: circadian distribution and response to anti-ischemic medication. Circulation. 1993;88:92–100.
Muller JE, Stone PH, Turi ZG, Rutherford JD, Czeisler CA, Parker C, Poole WK, Passamani E, Roberts R, Robertson T, Sobel BE, Willerson JT, Braunwald E, and the MILIS study group. Circadian variation in the frequency of onset of acute myocardial infarction. N Engl J Med. 1985;313:1315–1322.
Hjalmarson A, Gilpin EA, Nicod P, Dittrich H, Henning H, Engler R, Blacky AR, Smith SC, Ricou F, Ross J. Differing circadian patterns of symptom onset in subgroups of patients with acute myocardial infarction. Circulation. 1989;80:267–275.
Willich SN, Linderer T, Wegscheider K, Leizorovicz A, Alamercery I, Schroder R, and the ISAM Study Group. Increased morning incidence of myocardial infarction in the ISAM study: absence with prior β-adrenergic blockade. Circulation. 1989;80:853–858.
Ridker PM, Manson JE, Buring JE, Muller JE, Hennekens CH. Circadian variation of acute myocardial infarction and the effect of low-dose aspirin in a randomized trial of physicians. Circulation. 1990;82:897–902.
Willich SN, Lowel H, Lewis M, Arntz R, Baur R, Winther K, Keil U, Schroder R, and the TRIMM study group. Association of wake time and the onset of myocardial infarction: triggers and mechanisms of myocardial infarction (TRIMM) pilot study. Circulation. 1991;84(suppl VI):VI-62–VI-67.
Tofler GH, Muller JE, Stone PH, Forman S, Solomon RE, Knatterud GL, Braunwald E, and the TIMI Research Group. Modifiers of timing and possible triggers of acute myocardial infarction in the Thrombolysis in Myocardial Infarction phase II (TIMI II) study group. J Am Coll Cardiol. 1992;20:1049–1055.
Lampert R, Rosenfeld L, Batsford W, Lee F, McPherson C. Circadian variation of sustained ventricular tachycardia in patients with coronary artery disease and implantable cardioverter-defibrillators. Circulation. 1994;90:241–247.
Muller JE, Ludmer PL, Willich SN, Tofler GH, Aylmer G, Klangos I, Stone PH. Circadian variation in the frequency of sudden cardiac death. Circulation. 1987;75:131–138.
Arntz HR, Willich SN, Oeff M, Bruggemann T, Stern R, Heinzmann A, Matenaer B, Schroder R. Circadian variation of sudden cardiac death reflects age-related variability in ventricular fibrillation. Circulation. 1993;88:2284–2289.
Kleiman NS, Schechtman KB, Young PM, Goodman DA, Boden WE, Pratt CM, Roberts R, and the Diltiazem Reinfarction Study Investigators. Lack of diurnal variation in the onset of non–Q wave infarction. Circulation. 1990;81:548–555.
Muller JE, Tofler GH, Stone PH. Circadian variation and triggers of onset of acute cardiovascular disease. Circulation. 1989;79:733–743.
Cobb LA, Weaver WD, Fahrenbruch CE, Hallstrom AP, Copass MK. Community-based interventions for sudden cardiac death: impact, limitations, and changes. Circulation. 1992;85(suppl I):I-98-I-102.
Silverman BW. Density Estimation. New York, NY: Chapman & Hall; 1986.
Lehmann EL. Nonparametrics. San Francisco, Calif: Holden Day; 1975.
Lehmann EL. Testing Statistical Hypotheses. 2nd ed. New York, NY: John Wiley; 1986.
Agresti A. Categorical Data Analysis. New York, NY: John Wiley; 1990.