Article Text

Download PDFPDF

Original research
Heart rate and premature atrial contractions at 24hECG independently predict atrial fibrillation in a population-based study
  1. Anders Paul Persson1,2,
  2. Artur Fedorowski1,3,
  3. Bo Hedblad1,
  4. Margaretha Persson1,4,
  5. Steen Juul-Möller1,
  6. Gunnar Engström1,
  7. Linda S B Johnson1,2
  1. 1 Department of Clinical Sciences, Lund University, Malmö, Sweden
  2. 2 Department of Clinical Physiology, Skånes universitetssjukhus Malmö, Malmö, Sweden
  3. 3 Department of Cardiology, Skånes universitetssjukhus Malmö, Malmö, Sweden
  4. 4 Department of Medicine, Skånes universitetssjukhus Malmö, Malmö, Sweden
  1. Correspondence to Dr Anders Paul Persson, Department of Clinical Sciences, Lund University, Malmö 214 28, Sweden; anders_p.persson{at}


Background Low resting heart rate and premature atrial contractions (PACs) predict incident atrial fibrillation (AF) and could be interdependent, since PACs occur in the gaps between normal beats.

Objective To study the association between low heart rate at 24hECG, PACs and incident AF in a prospective population-based cohort.

Methods In the Malmö Diet and Cancer study, 24hECGs were performed in 377 AF-free subjects. The endpoint was clinical AF retrieved from national hospital (mean follow-up 17 years). The interaction between increased supraventricular activity (SVA) top quartile of either PACs/hour or supraventricular tachycardias/hour) and mean heart rate (mHR) as regards AF risk was assessed in multivariable Cox regression analyses adjusted for age, sex, height, BMI, systolic blood pressure, antihypertensive medication, smoking and homeostasis model assessment of insulin resistance.

Results There were 80 (21%) incident cases of AF. Below median mHR (80 bpm/75 bpm for women/men) was associated with increased AF incidence (HR: 1.89, 95% CI 1.18 to 3.02, p=0.008). There was no correlation between mHR and SVA (p=0.6) or evidence of a multiplicative interaction between these factors for AF risk (p for interaction=0.6) In the group with both increased SVA and below median mHR (17% of the population) the relative risk of AF was very high (HR 4.5, 95% CI 2.2 to 9.1, p=0.001).

Conclusion Low mHR at 24hECG independently predicts AF, but there is no association between mHR and SVA, and these factors are independent as regards AF risk. Subjects with both low mHR and increased SVA have high AF risk.

  • atrial fibrillation
  • 24h-electrocardiogram
  • heartrate
  • population
  • supraventricular ectopy, premature atrial contraction

Statistics from

Request Permissions

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.


Low resting heart rate is a known predictor of atrial fibrillation (AF).1–3 The mechanistic links are not clear, but increased atrial stretch during low heart rate leading to structural and electrical remodelling is a possible cause,4 5 and low resting heart rate could also be an early sign of autonomic imbalance or atrial disease. However, other possible predictors of AF, such as mean or minimum heart rate over the course of 24 hours, have not been sufficiently studied.

Premature atrial contractions (PACs) have also been shown to be useful in AF prediction.6–8 It is conceivable that low heart rate increases the probability for ectopic beats to occur due to the longer interval between normal heart beats. Low heart rate is also associated with increased vagal tone that also shortens the atrial refractory period,9 thereby possibly promoting both PACs and AF.

AF is very common in the elderly population and has serious consequences such as stroke, heart failure and dementia.10 11 Currently, there are no validated screening or prevention programmes for AF, and despite optimal treatment, many patients eventually have to accept permanent AF. Recent studies have shown promise for the possibility of AF prevention and even regression.12 However, the best validated AF prediction model, the CHARGE-AF score, has a C-statistic of around 0.7,13 which can be improved with the addition of biomarkers,14 but more accurate predictive models are needed in order to develop effective prevention programmes.

The aim of this study was to assess low heart rate at 24hECG as predictor of incident AF in the population-based Malmö Diet and Cancer Study (MDCS) cohort and to evaluate whether there was a correlation between PACs and low heart rate and if low heart rate and PACs interact as regards AF risk.


Study population

The study population was derived from the large population-based MDCS cohort and has been described in more detail elsewhere.8 Briefly, the MDCS cohort enrolled 30 447 individuals (40% men) in the years 1991–1996. A random sample (6103 individuals) underwent additional screening activities in a cardiovascular subcohort, and of these 909 were re-examined in the years 1998–2000 after random inclusion weighted by homeostasis model assessment for insulin resistance (HOMA-IR). Higher HOMA-IR was slightly oversampled (15% included from each of Q1 and Q2, 30% from Q3 and 40% from Q4). A random sample of 389 re-examinations included 24hECG screening, and the present study is derived from these subjects. We excluded subjects with prevalent AF (n=7) and inadequate 24hECG recordings (n=5). The final study population consisted of 377 subjects. The derivation of the study population is described in figure 1.

Figure 1

Derivation of the study population. HOMA, homeostasis model assessment for insulin resistance; MDCS, Malmö Diet and Cancer Study.

Data collection

All study participants underwent physical examination. Blood tests were drawn after overnight fast. Height and weight were measured standing, without shoes and in light indoor clothing. Blood pressure was measured manually, after 10 min supine rest, with a modifiable cuff sphygmomanometer.

Smoking and the use of antihypertensive medication were derived from a self-administered questionnaire. Leisure time physical activity was assessed by a previously described modification of the Minnesota Leisure Time Physical Activity Questionnaire,15 16 which was administered at the baseline in the full MDCS cohort. Low physical activity was defined as the lowest sex-specific quartile of self-reported physical activity.

24hECGs were recorded with X,Y,Z coupling and a 256 Hz sampling rate using a LifeCard CF Digital Holter recorder and the Pathfinder tool for analysis (both Spacelabs Healthcare, Issaquah,Washington, USA). Mean analysis time was 23.5 hours (SD 1.2 hours). Beta-blockers were discontinued 3 days before the recording. Mean heart rate (mHR) was defined as the average heart rate during the entire registration, and minimum heart rate was defined as the lowest heart rate during one recorded minute. PACs were defined as a premature beat with an RR interval ≤80% of the previous RR interval. Three or more consecutive PACs were considered a supraventricular tachycardia (SVT).

End-point retrieval

Subjects were followed until AF diagnosis (diagnosis codes 427D for the 9th revision of the International Classification of Diseases (ICD-9), and I48 for the 10thth revision (ICD-10)), death (n=117) or censoring by emigration (n=2, after 15 and 16 years of follow-up, respectively). Atrial flutter was not distinguished from AF, due to the similarities between these diagnoses.17 All endpoints were retrieved from the Swedish National Hospital Discharge Register administered by the Swedish National Board of Health and Welfare, which covers all inpatient diagnoses since the year 1987 and all hospital outpatient diagnoses since the year 2000. The AF in MDCS has been validated and has an accuracy of 95%.18 The conventional duration for diagnosis of AF in Sweden is 30 s. The study conforms to the declaration of Helsinki.

Statistical methods

All analyses were performed using Stata for Macintosh, V.15.1. Normality was assessed visually, and non-normal variables (HOMA-IR, SVTs and PACs) were transformed by the natural logarithm, after the addition of a small constant (1) to PAC and SVT counts. Univariate analyses were performed with the χ2 test for categorical variables, Student’s t-test for normally distributed continuous variables and Mann-Whitney’s U-test for non-normal continuous variables (HOMA-IR, PACs/hour and SVTs/hour). Cox regression analyses, with follow-up time as time-scale, were used to study the association between mean and minimum heart rate and incident AF. Three prespecified models were used. Model 1 adjusted for age at baseline and sex, model 2 adjusted for model 1+height, body mass index (BMI), systolic blood pressure, antihypertensive medication, smoking and ln-HOMA. Model 3 additionally adjusted for low physical activity. In order to assess the effect of attained age and period of calendar time, we also performed a Lexis expansion on age and calendar time in 5-year strata. A Cox regression model using the expanded dataset and adjusting for attained age and calendar time yielded similar results. In light of this the simpler model using baseline age for adjustment was chosen. The proportional hazards assumption was assessed using –log-log plots and found to be satisfied.

We analysed mHR both as a continuous variable and stratified into sex-specific quartiles. Exploratory Cox regression analyses of incident AF across quartiles of mHR suggested that above/below median was the most appropriate cut-off point for dichotomisation. The dichotomised mHR variable was used for tests of interaction between mHR and PACs as regards AF risk. The median value of mHR in our cohort was similar to those reported in previous studies that have evaluated relationship between heart rate and CV morbidity and mortality.19 Increased supraventricular activity (SVA) was defined as the top quartile of PACs per hour or the top quartile of SVT per hour.

The association between increased SVA and low heart rate was analysed using χ2 test. The likelihood ratio test was used to assess goodness of fit after inclusion of mHR. A p value below 0.05 was considered statistically significant.


The age range at examination was 53–74 years (mean age 65±6 years). The median mHR was 80 beats per minute in women and 75 beats per minute in men. Baseline characteristics are presented in more detail in table 1. During a mean follow-up time of 17 years, there were 80 incident AF events (cumulative incidence 21%). The incidence of AF (95% CI) per 1000 person-years was 20 (13 to 31) in the lowest quartile, 18 (13 to 26) in quartile 2, 11 (6 to 18) in quartile 3 and 10 (6 to 17) in quartile 4. table 1 also presents baseline characteristics by for subject with and without AF during follow-up, with univariate analyses of differences. Results from the Cox regression models assessing the heart rate and risk of incident AF are summarised in table 2. Low mHR was associated with incident AF in all three models. The addition of low mHR to model 2 resulted in a significantly likelier model (p=0.007)

Table 1

Population characteristics

Table 2

Cox regression analyses of heart rate at 24hECG and incidence of atrial fibrillation

The HRs for the combinations of low heart rate and SVA suggested an additive effect between the two variables (table 2, and figure 2), and there was no significant interaction between mHR and SVA (p=0.6), with respect to incidence of AF. In the subgroup with both low mHR and increased SVA (17% of the population), 44% had incident AF. There was no significant univariate association between low mHR and SVA (χ2=0.3, df=1, p=0.6).

Figure 2

Kaplan-Meier survival functions by groups of mean heart rate (mHR) and increased supraventricular activity (SVA) (top quartile of either premature atrial contractions (≥5.5 per hour) or supraventricular tachycardias (≥0.13 per hour)).

There was no statistically significant association between minimum heart rate and incident AF (table 2). There was, however, a strong correlation between mHR and minimum heart rate (Pearson correlation coefficient=0.83).

Beta-blockers were discontinued 3 days before the 24hECG registration, and subjects who had been taking beta-blockers had a higher mHR than subjects who had not (81 compared with 77 beats per min). In order to account for the possible effect of beta-blocker withdrawal on the registered heart rate, we performed a subanalysis where these subjects (n=38) were excluded. In this subgroup, the association between mHR and incident AF was similar to the results of the entire cohort (HR per 1 beat decrease 1.04, 95% CI 1.01 to 1.07, p=0.01 in model 2).

In order to better account for the possible effect of high physical activity on the association between mHR and incident AF, we also performed subanalyses where this variable, defined as either the top quartile or top decile of reported physical activity, were added to model 2. Neither of these adjustments altered results in a substantial way (data not shown).


AF is a major population health problem and improved AF prediction may be useful in attempts to construct primary prevention programme.20 This study has attempted to assess the relationship between heart rate and PACs, two known predictors of AF, as regards AF risk, in order to assess a plausible mechanistic link between the two and to determine whether these variables may be useful in AF prediction. For this purpose, we have used the MDCS cohort, which is population-based and has a long follow-up via national registers that have been shown to be of high quality for the AF diagnosis. We found low mHR at 24hECG to be associated with incident AF, independently of traditional AF risk factors. There was a correlation between low mHR and minimum heart rate, an indication that mHR at 24hECG could reflect the same mechanisms as low resting heart rate. Our results are therefore in line with previous findings of an association between resting heart rate and AF in the Cardiovascular Health Study, and the Norwegian cohorts, the Tromsø study and the Age 40-programme, but extend these by showing that mHR at 24hECG can be used to predict AF as well.1–3 The association between low mHR and incident AF was independent of the presence of premature atrial beats, and there was no evidence of an interaction between mHR and SVA. There was an association between low mHR and AF among subjects without increased SVA. Finally, we have shown that the subgroup with low mHR and increased SVA were at high risk of AF.

There are several plausible mechanistic links between heart rate and AF. Low heart rate may be associated with larger stroke volumes, and thereby atrial stretch or strain, which may lead to structural and electrical remodelling, a known precursor to AF.5 Lower heart rate could also be a measure of autonomic imbalance, which is known to be associated with AF,21 22 or a consequence of sinus node fibrosis, perhaps as part of the emerging concept atrial myopathy.23

Another plausible mechanism that could link low heart rate to incident AF is through longer time between normal heart beats, wherein the electrical activity characteristic of AF can occur.

Based on exploratory analyses, we dichotomised mHR at the median. The risk increase associated with low mHR was rather high, and thus applies to half of the study population. It may be, therefore, that mHR could improve predictive models for AF, which in turn could be useful in order to design large-scale prevention or screening. At the same time, the relatively high cut-off values applied in this study may not be useful on their own in clinical risk assessments, since such a large proportion of the population would be considered at risk. The sample size of the present study does not allow for more detailed assessments of the association between mHR and incident AF, but larger studies could possibly find a more precise cut-off, with better discrimination of individuals at risk.

The group of subjects with increased SVA and low mHR had a highly elevated risk of AF, with a more than quadrupled risk. Recent studies have shown that risk factor interventions among subjects with AF and BMI >27 kg/m2 can lead to regression of AF type.8 Primary prevention studies for AF among high-risk individuals have not yet been conducted, but the present study indicates that suitable populations for such interventions could perhaps be identifiable based on 24hECG results.

We did not find a significant association between low minimum heart rate and incident AF. Other studies have reported low resting heart rate as a risk factor for AF development, and we believe the non-significant finding is a likely beta error due to the small study population, especially considering the correlation between low minimum heart rate and low mHR, as well as the fact that measurement error is likely to be smaller for mHR since this represents an average heart rate over the full 24 hours of recording. However, low minimum heart rate at 24hECG represents a somewhat different measure of heart rate than resting heart rate assessed during the day, usually occurring during the night-time, and we cannot rule out that this measure is different compared with resting heart rate as regards AF risk.

We also assessed whether the association of low mHR and incident AF was influenced by physical activity habits, since low mHR could be a marker for degree of habitual physical activity. However, adjustment for either low or high physical activity did not result in any substantial difference in the effect size for the association between mHR and AF. This finding is in line with previous studies of low resting heart rates.1 2

Strengths and limitations

The population-based design, long follow-up time and virtually no loss to follow-up are major strengths of the study. The size of the sample is limited, however, and precise estimation of appropriate cut-off points for the association between mHR and AF was not possible. We have adjusted for many known AF risk factors and found the association between low mHR and incident AF to be independent of these. Some residual confounding may exist, especially in regard to physical activity, which was self-reported and not measured at the time of the 24hECG, although we do not believe that it would be large enough to affect our conclusions. The number of events in the study population is limited, and the sensitivity analyses have included more than the common rule of thumb of one covariate per 10 events, in order to control for all necessary confounding, which has been considered appropriate in this setting. Any overadjustment bias introduced should have been towards null, and we do not consider overadjustment to be likely to have affected results of the study in a substantial way. This fact should be considered when interpreting these results, however. We had no means of adjusting for the activity level during the day when the 24hECG was recorded.


Low mHR at 24hECG independently predicts AF, and there was no evidence of a multiplicative interaction between SVA and mHR. Individuals with both low mHR and increased SVA are at high risk of incident AF and could perhaps benefit from preventive measures.

Key messages

What is already known on this subject?

  • It is known that low resting heart rate and premature atrial contractions both predict atrial fibrillation, but not how these factors interact.

What might this study add?

  • Those with both low heart rate and increased premature atrial contractions have a high risk of atrial fibrillation, but mean heart rate at ambulatory ECG and premature atrial contractions likely cause atrial fibrillation through separate mechanisms.

How might this impact on clinical practice?

  • Those with both low mean heart rate and frequent premature atrial contractions have a high risk of future atrial fibrillation. Repeated or longer lasting ECG recordings in patients at high risk of developing AF could be of value, and future trials will be needed to confirm to which extent.



  • Contributors APP has performed analyses and drafted the manuscript. LSBJ has conceived of the study, supervised analyses and critically revised manuscript. GE, BH and AF have supervised analyses and critically revised manuscript. SJ-M and MP have participated in data collection and revised the manuscript.

  • Funding LSBJ is supported by governmental funding within the Swedish National Health Services. AF, GE and LSBJ are supported by the Swedish Heart and Lung Foundation.

  • Competing interests SJ-M is a previous medical director of Cardiome Inc.

  • Patient consent for publication Not required.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement No data are available.