Association between accelerometer-derived physical activity and incident cardiac arrest

Abstract

Aims

Studies on objectively measured physical activity (PA) have investigated acute cardiovascular outcomes but not cardiac arrest (CA). Our study aimed to investigate the dose–response relationship between accelerometer-measured PA and CA by intensity of PA.

Methods and results

This prospective cohort study included 98 893 UK Biobank participants whose PA data were measured using wrist-worn accelerometers. Total PA volume was measured using the average overall acceleration. Minutes per week of light PA (LPA), moderate PA (MPA), and vigorous PA (VPA) were recorded. The incident CA was identified using diagnostic codes linked to hospital encounters and death records. Cox proportional hazard models with restricted cubic splines were used to study the associations, including sex differences. During the follow-up period (median: 7.31 years; interquartile range: 6.78–7.82 years), 282 incident CAs (0.39 per 1000 person-years) occurred. Total PA was inversely related to CA risk. The CA risk decreased sharply until the time spent in MPA or VPA reached ∼360 min or 20 min per week, respectively, after which it was relatively flat. The LPA was not associated with CA risk. Subgroup analyses showed a more pronounced association between PA and a reduced risk of CA in women compared to men.

Conclusion

Accelerometer-measured PA, particularly MPA and VPA, was associated with a lower CA risk. Furthermore, a stronger association was observed in women than men.

Graphical Abstract

Open in new tabDownload slide

Introduction

Regular physical activity (PA) has numerous health benefits, including improved cardiac and pulmonary function and lower mortality.^1,2 The World Health Organization (WHO)/European Society of Cardiology (ESC)/American Heart Association (AHA) recommends that adults undertake at least 150–300 min of moderate PA (MPA) or 75–150 min of vigorous PA (VPA) weekly. Despite all the health benefits of PA, cardiac arrest (CA) incidence is transiently increased during PA, which is known as the paradox of PA.^3–5 Cardiac arrest is disproportionately more likely to occur during PA, accounting for 6% to 17% of all CA.^4–8

Previous studies such as the Nurses’ Health Study and Physicians’ Health Study have explored the correlation between PA levels and CA risk⁹; however, limitations such as inadequate adjustment for clinical variables and imprecise PA measurement methods have hindered a comprehensive understanding. These previous studies primarily focused on MPA and VPA, leaving a gap in understanding the effect of light PA (LPA) on CA.⁷ A case–control study found that women have a lower risk of developing PA-related CA than men.¹⁰ However, there is a paucity of large-scale prospective cohort studies that assess gender-based differences in CA risk associated with PA.

To address these limitations and advance understanding, this study leverages data from the UK Biobank, representing the largest study to date utilizing accelerometer-measured PA. Using an accelerometer can provide a more accurate estimation of PA levels than self-reported questionnaires.¹¹ Herein, we explore the dose–response relationship between accelerometer-measured PA and CA, with a particular focus on different intensities of PA. An important aspect of this study is the utilization of wearables to determine PA levels, a novel approach that adds a unique dimension to the assessment of PA-CA association.

Methods

The Biobank study was a population-based prospective cohort study conducted across the UK between 2006 and 2010 and design of the was previously described.¹² Baseline data on the participants’ demographic and clinical characteristics, including lifestyle and health data, medical history, and biological samples, were obtained through physical examinations, interviews, and laboratory tests. The Biobank study was approved by the North West Multi-Center Research Ethics Committee (299116), and all participants provided informed consent. Detailed information about this study can be found at www.ukbiobank.ac.uk.

Device-measured physical activity

AX Objective PA data were collected from 103 686 UK Biobank participants using Axivity 3 wrist-worn triaxial accelerometers between 2013 and 2015 (see Supplementary material online, Figure S1). All participants were invited randomly through e-mail (n = 236 519) and asked to wear accelerometers on their dominant wrist for 7 days at all times, including while working, walking, and sleeping. All invited participants had three or more valid monitoring days, including at least one weekend. Each accelerometer was calibrated for each participant to minimize deviation or bias. Further details regarding the data collection and processing can be found in previous studies.^1,13 Participants with previous CA (n = 84) before PA measurement were excluded from the present study. A total of 4709 participants were excluded due to problems with PA data, and 98 893 participants were included in the present study (see Supplementary material online, Figure S2). Total PA volume was measured using the average overall acceleration.¹⁴ Minutes per week of LPA, MPA, and VPA were defined as the time spent in 30–125 mg, 125–400 mg, and >400 mg intensity activity, respectively.¹⁴

Outcome ascertainment

Cardiac arrest was diagnosed using diagnostic codes linked to hospital encounters and death records. Baseline and incident CA cases were identified through the ‘first occurrence of health outcomes’ defined by a three-character International Statistical Classification of Diseases and Related Health Problems 10th Revision code (field ID: 131347). The follow-up of each participant was determined from the date of completing the measurement of PA data to the date of CA development, death, or censoring (31 May 2022; field ID: 131346), whichever occurred first.

Covariates

Age, sex, race, Townsend deprivation index (TDI), smoking and alcohol consumption, diet, employment status (with, without, or retired), mobility limitations (yes or no), history of myocardial infarction, and diabetes, cardiorespiratory fitness, grip strength, and left ventricular ejection fraction, were identified using questionnaires, interviews, and medical records at recruitment. Race was categorized as white and non-white. The TDI was calculated based on specific areas of output from previous national censuses that allow the identification of the most- and least-deprived areas of the country and provide information on the issues faced by people living in different parts of the country.^12,13,15 Smoking and alcohol status was categorized as current, quit, or never. Systolic blood pressure (SBP), diastolic blood pressure (DBP), heart rate (HR), weight, cardiorespiratory fitness,¹⁶ grip strength,¹⁷ and left ventricular ejection fraction were measured by registered nurses and doctors at the assessment centre. Cardiorespiratory fitness was assessed through net oxygen consumption, calculated from individuals’ body weight and maximum workload during cycle ergometry on a stationary bike (eBike, Firmware v1.7) using the following equation: net oxygen consumption = 7 + 10.8 (workload)/weight.¹⁶ Grip strength was defined as an average of measurements of the right and left hands divided by weight.¹⁷ The diets included fresh fruit (pieces/day), processed meat (times/week), vegetables (tablespoons/day), and fish (times/week). Low density lipoprotein (LDL) and high density lipoprotein (HDL) were measured according to local laboratory.

Ethics approval and consent to participate

No participants were involved in setting the research questions, outcome measures, or in the design and implementation of the study. No plans exist to involve patients in the dissemination.

Consent for publication

All listed authors have approved the manuscript and have agreed to its publication.

Statistical analysis

The baseline characteristics of the included participants were presented as numbers, proportions, or mean ± standard deviation (SD). Continuous variables were compared using Student’s t-test or Mann–Whitney U test according to the distribution type, and categorical variables were compared using the χ² test. Missing data were coded as missing indicator categories for categorical variables such as race and mean values for continuous variables such as BMI. Cox proportional hazard models with three models were used to assess the relationship between accelerometer-measured PA (continuous or categorical variables) and CA risk by intensity of PA. We also studied whether MPA or VPA meeting the WHO/ESC/AHA recommendations was associated with a lower CA incidence. Model 1 was adjusted for age, sex, and race. Model 2 was further adjusted for history of MI, HF, and diabetes, BMI, smoking, alcohol consumption, employment status, and TDI. Model 3 was further adjusted for SBP, DBP, HR, LDL, HDL, and diet, in addition to Model 2. We found no evidence of violation of the proportional hazard assumption based on tests using Schoenfeld residuals. Additionally, we used restricted cubic splines with four knots to flexibly determine the dose–response association between PA and CA risk. We also used a two-piecewise linear regression model to examine the threshold effect of PA on CA incidence. We determined the threshold value using a trial method by moving the trial turning point along a predefined interval and selecting the value that gave the maximum model likelihood. A previous study found that habitual PA reduced VPA-related CA.^4,6 We studied the joint associations between total PA and VPA and CA incidence. Interaction and stratified analyses were conducted according to sex, as a previous study reported that men were more likely to develop CA than women.⁷ We performed several sensitivity analyses by excluding CA that occurred early in the follow-up period (two years) or participants with MI and right censoring the criteria for age (>70 years). We also performed analyses using tertiles of total PA, LPA, MPA, and VPA rather than the quantiles of these measures. Previous studies found that cardiorespiratory fitness, grip strength, and left ventricular ejection fraction protect against CA. Finally, we further adjusted for cardiorespiratory fitness, grip strength, and left ventricular ejection fraction in addition to Model 3.^17–20 All statistical analyses were two-sided, and P-values of <0.05 were considered statistically significant. All analyses were performed using the STATA 17 software (StataCorp, College Station, TX, USA).

Results

A total of 98 893 participants were included in this study after excluding PA data with problem or CA survivors (see Supplementary material online, Figure S2). During the follow-up period (median, 7.31 years; interquartile range, 6.78–7.82 years; 722 332 person-years at risk), 282 incident CAs occurred (0.39 per 1000 person-years). Participant characteristics according to the incidence of CA are presented in Table 1. Participants with CA were more likely to be older, male, have a history of MI, HF, and T2DM, and less total PA, LPA, MPA, and VPA time than their counterparts. They also had higher SBP and lower HDL and LDL levels.

We found no evidence of violation of the proportional hazards assumption. Total PA and time spent on MPA and VPA were inversely related to CA risk when adjusted for sociodemographic factors, lifestyle factors, and clinical characteristics (Table 2). Each 1 SD increase in total PA was associated with a 27% lower risk of CA incidence (HR,0.73,95%CI,0.59–0.89, Model 3, Table 2). The association was almost linear; with increased total PA, the risk of CA decreased linearly (Figure 1A). However, we did not find an association between the time spent on LPA and CA risk (Table 2). With increased time spent on LPA, CA risk was relatively flat or decreased slowly (Figure 1B).

Increase in time spent on VPA (HR 0.80 per 1 SD,95%CI,0.65–0.99, Model 3) was associated with lower risk of CA incidence (Table 2). Compared with the participants who spent 0–25 min in VPA, those who spent ≥ 75 min did not have a lower risk of CA incidence (HR,0.61,95%CI,0.33–0.1.14, Model 3, Table 2). Meeting the WHO/ESC/AHA standard did not show a lower CA incidence than those without (HR,0.65,95%CI,0.35–1.21, Model 3). The association between the time spent on VPA and CA risk was non-linear (P < 0.01). Short time spent on VPA (<20 min/week) effectively reduced CA incidence, following which the CA risk was relatively flat or decreased slowly (Figure 1D).

Similarly, compared with the participants who spent 0–150 min in MPA, those who spent ≥ 600 min had a 49% lower risk of CA incidence (HR,0.51,95%CI,0.32–0.81, Model 3, Table 2). Increased time spent on MPA (HR 0.76 per 1 SD,95%CI,0.66–0.89, Model 3) and activity meeting the WHO/ESC/AHA recommendations (HR,0.68,95%CI,0.49–0.96, Model 3) were associated with a lower risk of CA incidence (Table 2). The association between the time spent on MPA and CA risk was non-linear (P < 0.01). The CA risk decreased sharply until the time spent in MPA reached ∼360 min, after which it became relatively flat or decreased slowly. (Figure 1C).

The risk matrix illustrating the joint associations between total PA, VPA, and CA is presented in Supplementary material online, Figure S3 and is based on the HRs in the Supplementary material online, Table S1. Higher levels of total PA and time spent engaging in VPA were associated with a lower incidence of CA. Meanwhile, participants with low levels of total PA but who spent a long time on VPA did not have a high CA risk.

Subgroup analyses revealed that sex played an interactive role in the association between total PA, time spent in MPA and VPA, and CA incidence (P for interaction < 0.05). A more pronounced association between PA (total PA, MPA, and VPA) and a reduced risk of CA was observed in women compared to men (see Supplementary material online, Table S2). The sex specific dose–response relationship between accelerometer-measured PA and CA was presented in Supplementary material online, Figure S4. A stronger association between PA and CA was observed in women than men. Several sensitivity analyses showed that our findings were robust, with the exception of VPA. The results remained unchanged, excluding CA in the first two years of follow-up, right censoring criteria for age (>70 years), exclusion of participants with MI, use of tertiles of total PA, LAP, and MPA, and further adjustment for cardiorespiratory fitness, grip strength, and left ventricular ejection fraction except for VPA (see Supplementary material online, Tables S3 and S4). Time spent on VPA had a marginal association with lower CA incidence but without statistical significance because of the insufficiently large sample size.

Discussion

We investigated the association between PA, especially MPA and VPA, and CA incidence in the UK Biobank, the largest device-measured PA study to date. We found that the total PA was associated with a lower incidence of CA. Intensity plays a role in the prospective association between PA and CA incidence over and above total PA volume. Moderate PA and VPA were associated with lower the CA rate, even in participants with low total PA; however, LPA was not. A more pronounced association between PA and a reduced risk of CA was observed in women compared to men. We also found that the association between the time spent in MPA and VPA and the incidence of CA was not linear. Increased time spent in MPA and VPA did not further reduce the incidence of CA.

Previous studies have found that PA measured by daily step counts was associated with lower cardiac death or all-cause mortality.^21,22 However, CA is a specific event that can be one of the causes of cardiac death. Cardiac arrest can result from various factors, including but not limited to cardiac death, drug toxicity, respiratory cause neurological event gastrointestinal case, endocrine/metabolic causes, and etc. It is clear that cardiac death only accounts for 55% of CA cases,⁵ highlighting the need for a more comprehensive understanding of the various contributing factors. Furthermore, accelerometer-derived measurements are generally considered more accurate than daily step counts for several reasons, including their ability to capture not only the quantity of PA but also the intensity and duration. Daily step counts are easy to understand and can serve as a motivational tool for increasing daily activity. However, when a more detailed analysis of PA is required, especially in a research or clinical setting, accelerometers provide a more accurate and comprehensive picture of PA levels.

To our knowledge, this study is the first to prospectively assess the association between device-measured PA and the long-term risk of CA among the general population, providing more robust results than previous studies.^4,9 Disagreeing with previous studies, we found that MPA and VPA were associated with lower CA risk, which was not intermediated by sociodemographic, lifestyle factors, and clinical characteristics. Although participants with higher PA levels tended to be younger and healthier, PA, mainly MPA and VPA, was directly associated with lower CA even after further adjusting for these confounding factors.⁹ We did not find that VPA can increase the risk of CA, as found in previous studies^3,4 but might reduce the risk of CA. The reasons for this disagreement are manifold. First, the absolute risk of CA was extremely low, and the sample size was relatively small. Secondly, using the frequency of VPA or simple questionnaires to evaluate PA is insufficient to evaluate the intensity and volume of PA.^4,7 Intermittent VPA has always been ignored in questionnaires and has multiple potential health benefits.²³ In contrast to simple questionnaires, accelerometers continuously record PA at high resolution, allowing us to measure VPA more accurately.^11,13 Previous studies mainly focused on long-time VPA because of lack of methods to accurately measure intermittent or short-time VPA, which might draw the conclusion that VPA increases CA incidence.^4,24 Thirdly, although our present study has much large sample size and more accurate measurement of PA, the VPA on CA incidence was not robust in the sensitivity analyses. Further studies on the association between VPA and the incidence of CA are required.

We further studied the dose–response relationship between total PA or PA intensity and CA risk, which has been ignored in previous studies. The association between MPA, VPA, and CA incidence was not linear; therefore, more time spent on MPA and VPA did not further reduce the incidence of CA. For example, 20 min/week of VPA effectively reduced the risk of CA, and a longer duration did not provide more benefit. The Physicians’ Health Study using the frequency of VPA was inappropriate; one VPA session might have exceeded 20 min.⁴ Therefore, the authors concluded that VPA increased CA incidence.⁴ The effect of VPA on CA risk was not affected by total PA volume; participants with low total PA could also benefit from VPA. Therefore, intermittent VPA may be a suitable choice for individuals unable or willing to exercise. The effect of LPA on CA has not been investigated in previous studies that have mainly focused on VPA or MPA.^4,6 We found that LPA did not reduce the risk of CA, and then the beneficial effect of PA on CA reduction depends on MPA and VPA. There may be several reasons. Light PA typically involves lower energy expenditure than MPA or VPA. Lower energy expenditure in LPA may contribute to a lesser impact on physiological and metabolic parameters that are directly linked to CA risk reduction. Furthermore, given that the total time available for PA is relatively constant for an individual, the increased time in LPA may mean less time available for higher-intensity activities.

We also found that women benefited more from PA in reducing CA, which cannot be fully explained by the fact that men had a 9–19-fold higher risk of exertion-related CA than women.^6,10,25,26 Women still benefit from MPA that might not be intense enough to trigger CA. Further studies are required to unveil this phenomenon. These findings indicate that individuals can choose individualized strategies to benefit from PA based on their willingness.

Limitations

First, PA data and circumstances at the time of CA were unavailable; therefore, we could not determine whether CA was triggered by PA or whether there was a significant increase in PA, especially VPA, that may have triggered CA. Furthermore, as all patients with accelerometer data for at least 3 days were included, this may have been a relevant limitation bias for the results shown. Moreover, as the incidence of CA was collected using death records and diagnostic codes, we may have missed events or events may have been coded the wrong way. Secondly, although this was the largest prospective study of PA-related CA in the general population, the sample size was relatively small because of the rarity of CA. This may have limited the accuracy of the results. Thirdly, the acceleration data and baseline characteristics were not simultaneously collected. Acceleration data were collected between 2013 and 2015, with a median of 5.7 years later. However, a previous study found that the baseline characteristics were generally unchanged when studying participants who had one more visit.¹ Fourthly, residual confounding factors cannot be avoided in observational studies. However, residual confounding was a minor issue in the present study as the relative confounding of CA incidence was fully adjusted. Furthermore, the adjustments had minimal effect on effect size. Fifthly, all included participants were from the UK, indicating that these results may not be conclusive across other ethnicities (e.g. Asians) with different characteristics and lifestyles. Sixthly, we acknowledge that accelerometers may not capture all types of PAs, especially those that are not conducive to typical accelerometer placement or motion detection. For instance, activities such as indoor cycling or other specialized exercises may not be accurately recorded by accelerometers. The placement and design of accelerometers may not be optimal for capturing specific movements, leading to potential underestimations or inaccuracies in the assessment of PA levels, particularly during these specialized activities.

Conclusions

Accelerometer-measured PA, particularly MPA and VPA, was associated with a lower CA risk. Furthermore, a stronger association was observed in women than men.

Supplementary material

Supplementary material is available at Europace online.

Authors’ contributions

Z.X. collected the data, conducted statistical analysis, and provided critical input for the manuscript (including figures and tables). S.Q. wrote this manuscript and revised under the supervision of Z.X. All authors have read, provided critical feedback on, and approved the final manuscript.

Funding

This work was supported in part by National Natural Science Foundation of China project 82000298 and Natural Science Foundation of Hunan Province 2021JJ40883 to Z.X. The funders have played no role in the research.

Data availability

All data are available in the UK Biobank data (www.ukbiobank.ac.uk). UK Biobank data are publicly accessible for research use by application.

References

Author notes