Data from the UK Biobank demonstrates that increased brief, sporadic moderate-to-vigorous physical activity reduces mortality

Abstract

Background

Current physical activity guidelines recommend 150-300 min/week of moderate-to-vigorous physical activity (MVPA) to improve health. However, whether brief, sporadic MVPA can be included within this recommended dose remains unclear.

Methods

In a cohort study, we analyzed data from 96,054 UK Biobank participants whose MVPA was objectively measured by accelerometers at baseline between 2013 and 2016. The primary outcome was all-cause mortality, with follow-up through 2022. A machine learning algorithm, trained on ground-truth camera data, was employed to differentiate between sporadic and bouted MVPA.

Results

During a follow-up period of 8.0 years, 3586 deaths and 4948 incident cardiovascular disease events are identified. Compared to the least active individuals (65 min/week of sporadic MVPA at 10^th percentile), those accumulating 150 and 300 min/week of sporadic MVPA have 48% (95% confidence interval [CI]: 44% to 52%) and 50% (95% confidence interval [CI]: 41% to 57%) lower all-cause mortality, respectively. For bouted MVPA, compared to the least active individuals (38 min/week of bouted at 10^th percentile), those accumulating 150 and 300 min/week of bouted MVPA have 33% (95% CI: 29% to 37%) and 49% (95% CI: 44% to 53%) lower all-cause mortality, respectively. In joint analysis, optimal health benefits are observed in individuals who incorporated both bouted and sporadic MVPA, rather than merely accumulating additional sporadic MVPA beyond 150 min/week.

Conclusions

In conclusion, meeting the minimum physical activity recommendation through accumulating brief, sporadic MVPA is supported. Incorporating bouted MVPA, rather than accumulating additional sporadic MVPA beyond 150 min/week, may confer further health benefits.

Plain language summary

The health benefits of brief, sporadic physical activity are increasingly recognized. However, it is still unclear how many minutes of sporadic physical activity per week are needed for health benefits. In this study, we analyzed a large cohort of 96,054 participants from the United Kingdom. To align with current physical activity guidelines, we focused on activity at or above moderate intensity, known as moderate-to-vigorous physical activity (MVPA). We found that higher levels of sporadic MVPA were associated with lower risks of death from all causes and lower risk of cardiovascular diseases. Maximal benefits of sporadic MVPA are gained when MVPA is 150 min/week. Further increases in sporadic MVPA did not provide appreciable additional benefits.

Cai et al. analyze accelerometer data from UK Biobank participants using machine learning to separate sporadic from bouted moderate-to-vigorous physical activity (MVPA) and link them to mortality and cardiovascular outcomes. They find that 150 min/week of sporadic MVPA yields maximal mortality benefit, and adding bouted MVPA offers further gains.

Introduction

As the global epidemic of chronic diseases and premature deaths continues to be exacerbated by physical inactivity and sedentary lifestyles, the promotion of physical activity (PA) has become an urgent public health priority and a crucial focus in clinical practice¹. Compelling evidence reveals numerous health benefits of regular PA, such as improved fitness, reduced cardiovascular disease (CVD) risk, and enhanced longevity². Current PA guidelines recommend 150–300 min of moderate intensity or 75–150 min of vigorous-intensity PA per week, or an equivalent combination of both (i.e., moderate-to-vigorous physical activity [MVPA]) to achieve substantial health benefits^2,3.

However, the guideline recommendations on MVPA volume have been primarily built on evidence from self-reported MVPA^4,5. Longitudinal epidemiological studies of self-reported leisure-time MVPA support the justification of 150–300 min/week by revealing the dose-response association between MVPA and mortality. Achieving the guideline-recommended volume of MVPA was associated with at least 70% of the potential maximal benefit of MVPA on all-cause mortality⁶. The additional benefits of performing high amounts of MVPA (>300 min/week) tended to be marginal or plateaued^4–7. Additionally, the landmark clinical trials that established the cause-and-effect relationship between PA and chronic disease prevention and management further justified the guideline recommendations^8–11. For example, lifestyle modifications that incorporated 150 min/week of MVPA, comprising a minimum of two supervised sessions per week, resulted in a 58% reduction in diabetes incidence¹¹.

Over the past decade, there has been a surge of research exploring the health benefits of accumulating brief, sporadic MVPA, such as opting for stairs over elevators or interrupting sedentary time with brief walks. Short-term clinical trials have demonstrated the positive impact of interrupting sedentary behavior with brief bouts of MVPA on cardiometabolic biomarkers^12,13. Using accelerometer-measured PA, two cohort studies found that the volume of MVPA, but not the bout length of MVPA, was important in predicting mortality outcomes^14,15, suggesting that the health benefits of brief, sporadic MVPA are potentially comparable to bouted MVPA. The growing evidence from wearable devices has paved the way for more strategies to promote active lifestyles and empower more individuals to meet the PA guidelines via various daily life activities that were previously not captured by self-reported leisure-time MVPA.

However, the dose-response association between sporadic MVPA and health benefits has not been adequately addressed. Importantly, there is minimal evidence regarding the applicability of the recommended dose of 150–300 min to sporadic MVPA. It is imperative to determine if achieving the guidelines through brief, sporadic MVPA is associated with similar health benefits as compared to achieving it through leisure-time MVPA, and if accumulating high levels of sporadic MVPA beyond the guideline-recommended volume can further improve health. Another gap in current research is the lack of evidence concerning the role of sporadic MVPA in the context of bouted MVPA. Specifically, it is unclear whether sporadic MVPA should be promoted for all individuals, regardless of their PA level, or only for those who are not engaging in sufficient bouted MVPA. Furthermore, if an individual has already achieved the recommended volume through sporadic MVPA, it remains largely unexplored to what extent bouted MVPA will provide “add-on” health benefits. Answering those questions is pivotal for advancing PA research and informing clinical practice.

Traditionally, MVPA bouts are identified using a fixed rule, such as at least 8 min above the moderate threshold within 10 consecutive minutes¹⁶. Although this method allows for up to 2 min of breaks to introduce some tolerance for rest, it is inadequate for accommodating the complex movement patterns present in free-living MVPA. To date, little research has accurately captured MVPA bouts in real-world settings. In our recent analyses of the Capture-24 dataset, we observed that free-living walking bouts do not necessarily maintain a constant speed, in contrast to treadmill walking in laboratory settings¹⁷. There are accelerations, decelerations, and brief interruptions, and the intensity may intermittently fall slightly below the moderate threshold within an MVPA bout.

While the specific term “bout” is no longer emphasized in the updated PA guidelines^2,3, the underlying physiological concept remains a focus in PA research. Several metrics have been proposed to characterize the spectrum from sporadic to sustained activity and to elucidate the interplay between activity volume and pattern for health benefits. For example, the M60_Ratio (whether the most active 60 min of a day are consecutive or spread throughout the day) by Rowlands et al. and Schwendinger et al.^18,19, PA fragmentation (reciprocal of mean activity bout length) by Wanigatunga et al.²⁰, and PA variability (activity concentrated in a daily peak versus a stable, sporadic pattern) in our recent work²¹. However, none of these metrics can directly quantify the duration of sporadic and bouted MVPA.

Taken together, while a simple method of defining bouts is desirable for clarity, it may not be practical across diverse real-world situations. In contrast, machine-learning approaches can accommodate this complexity by leveraging contextual features and producing more nuanced classifications. In this study, we trained a random forest model to recognize patterns of accelerometer data associated with sporadic and bouted activity, as labeled by wearable camera data. This approach allowed us to accurately quantify the volume of each type of MVPA. We applied the algorithm to a large-scale, population-based cohort study of UK Biobank to investigate the associations of bouted MVPA and sporadic MVPA with all-cause mortality and CVD incidence. Here we show that the health benefits of sporadic MVPA follow an L-shaped pattern, plateauing at 150 min/week, while the benefits of bouted MVPA extend beyond 300 min/week.

Methods

Consent and ethical approval

Ethical approval of the Capture-24 dataset was obtained from the University of Oxford Inter-Divisional Research Ethics Committee (Reference number: SSD/CUREC1A/13-262). All participants provided written informed consent for data collection and research use. The publicly released dataset is de-identified and includes only text annotations derived from camera images²²; the images themselves are not shared.

This study is a secondary analysis and was conducted with approved access to de-identified data from the UK Biobank (Application ID: 85118). We did not require additional approval to use these data, as the approval from the National Health Service’s National Research Ethics Service (Reference number: 11/NW/0382) covers secondary use. All participants in the UK Biobank study provided written informed consent prior to their participation.

Study population

The UK Biobank is a prospective, population-based cohort study that recruited over 500,000 participants aged 40–79 years from across England, Scotland, and Wales²³. Baseline measures, such as sociodemographic questionnaires, bio-samples, and medical history, were collected in 22 assessment centers from 2006 to 2010. To enhance the measurement of PA, a subsample of more than 100,000 individuals wore accelerometers for 7 consecutive days. Accelerometry data were collected between 2013 and 2016, a median of 5.7 years after the initial enrollment. In this cohort study, we analyzed 96,054 participants (mean age 62.0 years, 56.4% women) who had valid accelerometry data and were followed for all-cause mortality through November 2022. In a subsample of 88,050 participants (mean age 61.4 years, 58.4% women) free of CVD at baseline, we analyzed CVD incidence outcomes through October 2022. The sample diagram is presented in Supplementary Fig. S13.

Main exposures

A tri-axial accelerometer (Axivity AX3, Newcastle, UK) was worn on the dominant wrist to record human movement signals over a period of 7 days, with a sampling frequency of 100 Hz. The UK Biobank expert team processed the raw accelerometer data by conducting gravity calibration, calculating Euclidean Norm Minus One (ENMO) of the three axes $\sqrt{x^2+y^2+z^2}-1$, removing non-wear time, and imputing missing data to account for bias resulting from non-wear time. Specifically, non-wear time was defined as any period of ≥60 min during which the standard deviation of each of the three axes was less than 13 mg^24,25. Non-wear periods were imputed by replacing missing values with the mean ENMO from the corresponding clock-time segments on the non-missing days, calculated at the minute-epoch level. Participant inclusion followed a three-step procedure: first, we excluded participants with device calibration failure; second, we excluded those with less than 72 h of wear time or lacking valid wear in at least one hourly bin on any of the 7 days (a criterion shown to yield an intra-class coefficient of approximately 0.90 relative to a complete 7-day wear protocol)²⁶; third, we excluded outliers whose average ENMO was more than 9 standard deviations from the sample mean²⁷ (Supplementary Fig. S13). The protocols and rationales for these procedures are described in detail elsewhere²⁶. The 5-s epoch ENMO data have been made available in the Biobank repository (Data-Field 90004). We converted the data to 60-s epochs, consistent with the NHANES protocol²⁸, to align with established accelerometer data processing procedures and facilitate comparison with existing knowledge. Participants were required to wear the device for ≥72 h, with data recorded for every hour of the day. This criterion resulted in an acceptable error margin of ≤10% compared to a full 7-day wear period²⁶. Metabolic Equivalents (METs) provide a standardized way to express the energy expenditure of different activities, with 1 MET representing the energy expenditure at rest. According to the Compendium of Physical Activities²⁹, a value of 3–5.9 METs is considered moderate intensity and ≥6 METs is considered vigorous-intensity PA. In this study, we used 125 milli-g and 400 milli-g as the accelerometer thresholds to correspond to 3 METs and 6 METs, respectively^27,30.

The main exposures were the time spent in either bouted MVPA (lasting longer than 10 min) or sporadic MVPA (shorter than 10 min), as estimated by accelerometry. We developed a random forest classifier, a well-recognized machine-learning technique in accelerometer-based human activity recognition³¹, to capture MVPA bouts lasting longer than 10 min. We did not use the traditional 80% criteria to define bouts, which identifies a 10-min bout of MVPA as consecutive accelerations above a moderate threshold, allowing for no more than 2-min breaks¹⁶. Despite its wide use, relevant evidence supporting the 80% criterion has been lacking.

The random forest classifier was trained using data from 151 participants (52 males and 99 females; age range: 18–91 years) in Oxfordshire County, United Kingdom. The ground truth for their free-living PA was obtained through wearable cameras (Vicon Autographer), which automatically captured images at intervals of 20–30 s, while concurrent acceleration data were recorded using wrist-worn accelerometers (Axivity AX3, Newcastle, UK). We chose this dataset to train the classifier because its data collection was geographically similar to that of the UK Biobank, and it used the same type of accelerometers. The image data provided informative annotations for a wide range of activities, such as

1. transportation walking at 3.0 METs;

2. transportation, waiting, and sitting at 1.3 METs;

3. leisure hiking or walking at 5.0 MET;

4. leisure standing and talking in person or using a phone at 1.8 METs.

The MET values of activities were determined by trained annotators using camera data and the Compendium of Physical Activities²⁹. The detailed annotations within this dataset make it particularly suitable for identifying bouts of activity because it allows for accurate determination of the duration of activities, such as hiking, as well as interruptions, such as sitting and using the phone. The validation of the annotations has been reported elsewhere²².

To form a dataset suitable for training the random forest classifier, we segmented the Capture-24 time-series data into 10-min intervals. Each interval contained 10 data points, representing the minute-by-minute ENMO over the 10 min. The ENMO values were calculated using the GGIR package in RStudio. The labeling of ground truth was based on walking MVPA (walking at the intensity of 3 METs or higher) because walking is the fundamental form of health-benefiting PA. Intervals with 10 min of walking MVPA were classified as “bouted MVPA.” Those with 1–9 min of walking MVPA were categorized as “sporadic MVPA.” Intervals with no MVPA were labeled as “non-MVPA.” The ENMO data were fed to the random forest classifier (100 trees, 10 features) to predict labels. Five-fold cross-validation revealed an overall accuracy of 0.78. The class-specific F1 score was 0.83 and recall was 0.85 for bouted MVPA. The confusion matrix is presented in Table 1. We further analyzed the domains of sporadic and bouted walking MVPA using the camera data. Table 2 provides a detailed breakdown of the domains in which sporadic and bouted walking MVPA occurred, offering insights into how these MVPA patterns are distributed across daily activities and contexts. We noticed that bouted MVPA primarily occurred in the transportation domains, while more than 50% of sporadic MVPA took place in household and occupational domains.

Table 1.

Confusion matrix of the random forest classifier in differentiating the three types of 10-min intervals

		Prediction; number of 10-min intervals
		Non-MVPA	Sporadic MVPA	Bouted MVPA
Ground truth	Non-MVPA	884	150	36
	Sporadic MVPA	147	535	152
	Bouted MVPA	21	132	850

Precision = 81.9%, indicating the proportion of the 10-min intervals predicted as bouted MVPA were actually bouted MVPA.

Recall = 84.7%, representing the proportion of true bouted MVPA intervals that were correctly identified as bouted.

Table 2.

Domain information of sporadic and bouted walking MVPA in the Capture-24 dataset

Patterns of walking MVPA	Domains (based on camera data)	Number of minutes	Percentage within each domain
Sporadic walking MVPA	Home Activity and Household Chores	284	32%
	Leisure Activity and Sports	158	19%
	Occupational Activity and Manual Work	194	22%
	Transportation Walking	237	27%
Bouted walking MVPA	Home Activity and Household Chores	377	11%
	Leisure Activity and Sports	722	21%
	Occupational Activity and Manual Work	509	15%
	Transportation Walking	1839	53%

In UK Biobank data, the MVPA accumulated in the bouted windows as predicted by the random forest classifier was defined as bouted MVPA. All other MVPA were defined as sporadic MVPA. Figure 1 illustrates the distinction between bouted and sporadic MVPA using a rolling mean visualization. This approach reduced minute-to-minute fluctuations, resulting in a clearer and more interpretable presentation of the overall pattern of intensity distribution over the 7-day wear period. Total duration of MVPA was calculated as the duration of moderate-intensity PA plus twice the duration of vigorous-intensity PA^2,3.

Fig. 1. Patterns of sporadic and bouted MVPA.

Presenting the 10-min rolling mean ENMO.

Primary outcome

The primary outcome in this study was all-cause mortality, ascertained via the national death registry. We excluded individuals who died within the first 2 years of follow-up. This was done to account for reverse causality, where poor health caused lower PA rather than lower PA caused poor health. The duration of follow-up was calculated from the date of accelerometer wear to the date of death, or the end of follow-up (November 30, 2022), whichever came first. The secondary outcome was CVD incidence, identified by inpatient hospital records, which included both fatal and non-fatal cases of ischemic heart disease (IHD) (I20-25), heart failure (HF) (I50), and cerebrovascular disease (CeVD) (I60-69), according to the International Classification of Diseases, Tenth Revision (ICD-10). The follow-up for inpatient hospital data ended on October 31, 2022, in England; May 31, 2022, in Wales; and August 31, 2022, in Scotland.

Covariates

Covariates were selected based on previous research in the UK Biobank cohort²⁷. Sociodemographic covariates included age at accelerometer wear, stratified into 5-year age groups for proportional hazards assumption, sex (male/female), race/ethnicity (White, non-White), assessment center (one of 22 locations), and Townsend Deprivation Index, divided into quintiles. Health-related covariates included smoking status (never, former, or current), alcohol consumption (never, former, or current), diet quality, assessed using a five-point scale based on dietary habits, body mass index (BMI), categorized as <18.5, 18.5-24.9, 25-29.9, or ≥30 kg/m². Additional covariates included systolic blood pressure (divided into quintiles), use of cholesterol-lowering medication (yes/no), self-reported sleep duration (<7 h, 7–9 h, and >9 h), accelerometer wear time (hours/day), and sedentary behavior as estimated by subtracting self-reported sleep duration from time spent at ≤30 mg (hours/day)³². In sensitivity analysis, we further adjusted for baseline CVD (yes/no, based on both questionnaires and hospital record), diabetes (yes/no, based on questionnaires, hospital record, hemoglobin A1c ≥ 48 mmol/mol, and medication usage), cancer (yes/no, based on both questionnaires and hospital record), and long-standing illness (yes/no, based on questionnaires), and general health status (excellent, good, fair, and poor, based on questionnaires).

Statistical analysis

We constructed Cox proportional hazards models to investigate the association of bouted MVPA and sporadic MVPA with outcome measures. We modeled bouted MVPA and sporadic MVPA as two continuous variables, respectively, using restricted cubic splines with three knots placed at the 25th, 50th, and 75th percentiles. Each exposure was included in the model separately without mutual adjustment to avoid multicollinearity issues. The model adjusted for age, sex, ethnicity, assessment centers, the Townsend Deprivation Index, smoking, alcohol consumption, diet quality, body mass index, systolic blood pressure, cholesterol-lowering medications, sleep duration, sedentary behavior, and accelerometer wear time. We did not adjust for the total volume of PA as measured by the overall acceleration average due to multicollinearity (variance inflation factor > 10). We excluded individuals who died within the first 2 years of follow-up to account for reverse causality. Importantly, PA occurs on a continuum from brief to prolonged bouts. Our binary classification of MVPA as either sporadic or in bouts cannot capture this continuous transition. To address this limitation, we applied multiple bout-duration thresholds (3, 5, 10, 15, and 20 min) to examine how bout threshold influences the dose-response association between MVPA patterns and health outcomes. To account for the potential influence of intensity on the observed associations, we calculated the vigorous physical activity (VPA) component (≥400 mg) within both sporadic and bouted MVPA and then examined the dose-response association between VPA and health outcomes.

To explore the combined impact of the two types of MVPA on health outcomes, we jointly classified participants into 16 mutually exclusive groups according to the quartiles of bouted MVPA and quartiles of sporadic MVPA. To investigate whether age and sex influenced the observed relationships, we performed stratified analyses by age group (<65 years vs. ≥65 years) and by sex. In sensitivity analysis, we further adjusted for baseline CVD (not in the analysis of CVD incidence), cancer, diabetes, long-standing illness, and self-reported general health status. We also conducted separate analyses for IHD, HF, and CeVD.

The proportional hazards assumption was verified through graphical examination of Schoenfeld residuals for continuous variables and log(−log) survival plots for categorical variables, with all variables meeting the assumption. The variance inflation factors were below 5 for all variables, indicating no severe multicollinearity. Missing data were managed using multivariate imputation by chained equations. All statistical analyses were performed using RStudio (RStudio Inc., Boston, MA, USA) in January 2025, utilizing the “survival”, “rms”, and “mice” packages. Statistical significance was determined using a two-tailed alpha value of 0.05.

Results

Testing the traditional bout algorithm

The widely used traditional bout algorithm is based on a simple rule of thumb: it identifies a 10-min bout of MVPA as consecutive accelerations above a moderate threshold, allowing for no more than 2-min breaks^15,16,33,34. Using Capture-24²², a camera data-validated dataset of PA in uncontrolled, free-living conditions, we find that nearly half of true MVPA bouts, such as 10 min of continuous transportation walking or 10 min of leisure hiking, do not meet the 80% criteria (Table 3). It is challenging to accurately capture bouted MVPA using the traditional percentage metric: A higher cutoff value (e.g., 80%) leads to low sensitivity, resulting in many bouted MVPA not being captured. Conversely, a lower cutoff value (e.g., 50%) reduces specificity, leading to misclassification of sporadic MVPA as bouted. In Supplementary Fig. S1, we visualize several patterns of real-world MVPA walking bouts and provide a direct comparison of the minutely microstructure between sporadic and bouted MVPA.

Table 3.

Performance of bout identification using the percentage of acceleration above the moderate threshold compared to the camera-based reference

Threshold	Specificity	Sensitivity
10%	0.33	0.97
20%	0.44	0.94
30%	0.58	0.93
40%	0.67	0.89
50%	0.78	0.82
60%	0.84	0.77
70%	0.90	0.69
80%	0.97	0.55
90%	1	0.38
100%	1	0.16

Sensitivity is the proportion of true bouted MVPA correctly identified as bouted.

Specificity is the proportion of true sporadic MVPA correctly identified as sporadic.

In the table, boldface highlights the conventional 80% bout-identification rule—requiring that within a 10‑min window at least 80% of minutes (≥8) exceed the moderate-intensity threshold with no more than 2 min below threshold—and its corresponding performance. At this threshold, the algorithm shows very high specificity (0.97) but only moderate sensitivity (0.55), indicating that while sporadic MVPA is rarely misclassified as bouted, a substantial proportion of true bouted MVPA fails to be detected under the traditional rule.

Descriptive analyses

Figure 1 depicts the differences between bouted MVPA and sporadic MVPA based on our machine-learning algorithm. The distribution of bouted MVPA exhibits a pronounced positive skewness, with a skewness coefficient of 2.0, indicating a significant departure from normality. The median of bouted MVPA is 201 min/week, with an interquartile range (IQR) of 95 to 363 min/week. In contrast, sporadic MVPA displays a mild positive skew, with a skewness coefficient of 0.6. The median of sporadic MVPA is 149 min/week, with an IQR of 102 to 202 min/week, indicating a more symmetric distribution (Supplementary Figs. S2 and S3). There is a moderate correlation between bouted and sporadic MVPA (Spearman correlation coefficient = 0.65, P < 0.001) (Fig. 2). Bouted MVPA accounts for 57% (IQR 43–68%) of total MVPA (total MVPA = bouted MVPA + sporadic MVPA).

Fig. 2.

Spearman correlation matrix of age, total PA, total MVPA, bouted MVPA, sporadic MVPA, light-intensity PA, and sedentary behavior.

The baseline characteristics of the study participants across the joint categories of the two MVPA patterns are presented in Table 4. A notable age gradient was observed, with age being inversely associated with both bouted MVPA and sporadic MVPA. Specifically, the most active group (high bouted MVPA and high sporadic MVPA) is, on average, 5 years younger than the least active group (low bouted MVPA and low sporadic MVPA). The correlation matrix of age, total PA, total MVPA, bouted MVPA, sporadic MVPA, light-intensity PA, and sedentary behavior is presented in Fig. 2. In terms of sex distribution, the group with high sporadic but low bouted MVPA has a significantly higher proportion of females (60.3%) compared to the group with high bouted but low sporadic MVPA (44.3% females).

Table 4.

The baseline characteristics by the medians of bouted and sporadic MVPA (n = 96,065)

	Four mutually exclusive groups by the median of bouted MVPA and the median of sporadic MVPA (min/week)
	Sporadic < 149 Bouted < 201	Sporadic ≥ 149 Bouted < 201	Sporadic < 149 Bouted ≥ 201	Sporadic ≥ 149 Bouted ≥ 201	P values
	n = 35440	n = 12593	n = 13008	n = 35013
Sociodemographic factors
Age (years) at Accelerometry	64.6 (7.2)	60.2 (7.8)	62.7 (7.3)	59.3 (7.7)	<2 × 10−16
Female (%)	20,450 (57.7)	7594 (60.3)	5768 (44.3)	20,380 (58.2)	<2.2 × 10−16
Race/Ethnicity
White (%)	32,916 (92.9)	11,592 (92.0)	12,050 (92.6)	31,851 (91.0)	<2.2 × 10−16
Non-White (%)	2390 (6.7)	961 (7.6)	910 (7.0)	3052 (8.7)
Missing (%)	134 (0.4)	40 (0.3)	48 (0.4)	110 (0.3)
Townsend Deprivation Indexa	−1.7 (2.8)	−1.6 (2.9)	−2 (2.7)	−1.7 (2.8)	<2 × 10−16
Modifiable risk factors
Tobacco use
Never (%)	19,096 (53.9)	7360 (58.5)	7446 (57.2)	20,913 (59.7)	<2.2 × 10−16
Previous (%)	13,341 (37.6)	4253 (33.8)	4835 (37.2)	11,960 (34.2)
Current (%)	2903 (8.2)	950 (7.5)	686 (5.3)	2055 (5.9)
Missing (%)	100 (0.3)	30 (0.2)	41 (0.3)	85 (0.2)
Alcohol intake
Never (%)	1224 (3.4)	401(3.2)	287 (2.2)	875 (2.5)	<2.2 × 10−16
Previous (%)	1164 (3.3)	348 (2.8)	306 (2.4)	814 (2.3)
Current (%)	33,020 (93.2)	11,830 (93.9)	12,404 (95.4)	33,291 (95.1)
Missing (%)	32 (0.1)	14 (0.1)	11 (0.1)	33 (0.1)
Healthy eating score (0–5 Scale)	3.2 (1.0)	3.2 (1.0)	3.2 (1.0)	3.3 (1.0)	<2 × 10−16
Body mass index (kg/m2)	28.1 (5.0)	26.9 (4.6)	26.2 (3.9)	25.5 (3.8)	<2 × 10−16

Data are mean (standard deviation) for continuous variables and n (percent) for categorical variables. One-way analysis of variance (ANOVA) is used to compare continuous variables across groups. The chi-squared test is used to compare categorical variables across groups. The Bonferroni test is used to account for multiple comparisons. All tests are two-sided.

^aA higher Townsend Index indicates a higher level of deprivation.

Main outcomes

During a follow-up period of 8.0 years, encompassing 765,990 person-years for all-cause mortality and 687,444 person-years for CVD incidence, we identify 3586 deaths among 96,054 participants and 4948 incident CVD events among 88,050 participants. For specific CVD types, there are 3017 cases of IHD, 1067 cases of HF, and 1615 cases of CeVD.

For sporadic MVPA, compared to the least active individuals (65 min/week of sporadic MVPA at 10th percentile), those accumulating 150 and 300 min/week of sporadic MVPA have 48% (95% confidence interval [CI]: 44–52%) and 50% (95% CI: 41–57%) lower all-cause mortality, respectively. For CVD outcome, those accumulating 150 and 300 min/week of sporadic MVPA have 33% (95% CI: 28–37%) and 34% (95% confidence interval [CI]: 25–61%) lower risk of CVD incidence, compared to the 10th percentile (Fig. 3), respectively. The associations between sporadic MVPA and individual CVD outcomes (IHD, HF, and CeVD) are consistent with the overall CVD findings (Supplementary Fig. S4).

Fig. 3. Associations of bouted MVPA and sporadic MVPA with outcomes.

Using the 10th percentile (rather than extreme reference points, e.g., 0 min/week) as the reference. The solid line represents estimates of hazard ratios, and the shaded area indicates 95% confidence intervals. Analyses adjusted for age, sex, ethnicity, locations of assessment centers, Townsend Deprivation Index, smoking status, alcohol intake, dietary intake, sleep duration, body mass index, systolic blood pressure, cholesterol-lowering medications, sedentary behavior, and accelerometer wear time. n = number of participants included in the analysis.

For bouted MVPA, compared to the least active individuals (38 min/week of bouted at 10th percentile), those accumulating 150 and 300 min/week of bouted MVPA have 33% (95% CI: 29–37%) and 49% (95% CI: 44–53%) lower all-cause mortality, respectively. For CVD outcome, those accumulating 150 and 300 min/week of bouted MVPA have 22% (95% CI: 18–26%) and 35% (95% CI: 29–40%) lower risk of CVD incidence, compared to the 10th percentile (Fig. 3). The associations between bouted MVPA and IHD and CeVD are consistent with the overall CVD findings. However, the association between bouted MVPA and HF was L-shaped, leveling off at 300 min/week (Supplementary Fig. S4).

In joint analysis, while higher levels of sporadic MVPA and bouted MVPA are both associated with lower all-cause mortality and CVD incidence, there is a plateau “effect” of sporadic MVPA at approximately 150 min/week. Further increases in sporadic MVPA beyond this point are not associated with additional benefits. On the other hand, the benefits of bouted MVPA extend well beyond 300 min/week. The lowest risks of all-cause mortality and CVD incidence are observed in the groups with the highest bouted MVPA, but not necessarily with the highest sporadic MVPA (Fig. 4).

Fig. 4. Joint association of bouted and sporadic MVPA with outcomes (quartiles × quartiles).

Analyses adjusted for age, sex, ethnicity, locations of assessment centers, Townsend Deprivation Index, smoking status, alcohol intake, dietary intake, sleep duration, body mass index, systolic blood pressure, cholesterol-lowering medications, sedentary behavior, and accelerometer wear time.

Further adjusting for chronic conditions slightly attenuates the associations between MVPA and health outcomes (Supplementary Fig. S5). In sex-stratified analysis, the optimal dose of 150 min/week of sporadic MVPA is consistently observed in males and females (Supplementary Figs. S6 and S7). In age-stratified analysis (Supplementary Figs. S8 and S9), the optimal dose of sporadic MVPA remains at 150 min/week. However, the association between sporadic MVPA and all-cause mortality becomes U-shaped in individuals aged 65 years or older, with an increased risk of all-cause mortality when sporadic MVPA exceeds 150 min/week. The U-shaped association among older adults persists in the healthy-sample analysis, which excludes individuals with CVD, cancer, chronic obstructive pulmonary disease (COPD), diabetes, or self-rated poor health (Supplementary Fig. S10).

In the analysis of VPA, we examine the duration accumulated in each pattern and find no substantial differences: median 1.00 (IQR 0.00–4.00) min/week for sporadic VPA and 1.00 (IQR 0.00–9.00) min/week for bouted VPA. We then examine the dose-response association between VPA and health outcomes. VPA accumulated in MVPA bouts demonstrates a linear inverse association with all-cause mortality and CVD incidence. In contrast, the benefits of sporadic VPA are curvilinear and most pronounced below 10 min/week, with a flatter association at higher doses (Supplementary Fig. S11). When testing different thresholds for bout duration, we find that the optimal dose of sporadic MVPA gradually increases with higher thresholds: 75 min/week for bouts under 3 min, 150 min/week for bouts under 10 min, and nearly 300 min/week for bouts under 20 min. We observe similar trajectories for CVD incidence (Supplementary Fig. S12).

Discussion

In this large-scale, population-based cohort study of the UK Biobank, we apply a machine-learning algorithm trained on “ground truth” camera data to categorize accelerometer-measured MVPA into two distinct patterns. The first pattern, termed “bouted MVPA,” is characterized by longer durations of continuous MVPA, presumably indicative of a more sustained up-regulation of aerobic metabolism, such as a 10-min walk from home to the grocery store. In contrast, the second pattern, known as “sporadic MVPA,” consists of brief episodes of MVPA that occur sporadically throughout the day, such as a 3-min walk from the office to the parking lot. The integration of wearable devices with machine learning data processing methods offers an evidence-based approach to detecting MVPA bouts in free-living environments, broadening the traditional bout concept. With these insights, our study offers an alternative method to assess the dose-response associations of sporadic MVPA with all-cause mortality and CVD incidence^15,19.

The main findings of this study highlight an L-shaped association of sporadic MVPA with all-cause mortality and CVD incidence, leveling off at 150 min/week. Higher levels of sporadic MVPA are associated with a rapid reduction in mortality and CVD incidence among the least active individuals. Achieving 150 min/week of sporadic MVPA is associated with a 48% reduction in all-cause mortality and a 33% reduction in CVD incidence, compared to the 10th percentile of sporadic MVPA (65 min/week). The magnitude of mortality risk reduction observed in our analysis is similar to findings from a meta-analysis of 10 studies using accelerometer-measured total MVPA, which reports a 60% reduction in all-cause mortality at 150 min/week³⁵. This magnitude of mortality reduction in our study is larger than that observed in questionnaire-assessed leisure-time MVPA (a 31% reduction in all-cause mortality at 150 min/week)⁴, which might be due to the overall strengthened MVPA-health association in accelerometer studies compared to questionnaire-based studies³⁶. Similarly, the magnitude of CVD risk reduction in our study is also larger than that reported in a meta-analysis of questionnaire-assessed PA (17% reduction in CVD incidence at 150 min/week)³⁷. Collectively, our findings indicate that the health benefits associated with brief, sporadic MVPA could be comparable to total MVPA or leisure-time MVPA at the guideline-recommended volume.

Our study adds to the current body of knowledge by revealing that the optimal dose of sporadic MVPA is at the guideline-recommended volume of 150 min/week. The biological plausibility of the health-enhancing effect of sporadic MVPA is supported by experimental studies. Several randomized controlled trials demonstrate that engaging in sporadic MVPA by interrupting sedentary time can modify a range of risk factors, including blood pressure, glucose levels, lipid profiles, etc.^12,13. Among older adults (≥ 65 years), we observe a U-shaped association between sporadic MVPA and all-cause mortality and CVD incidence. The U-shaped association should be interpreted with caution because of the small number of older adults who accumulated at least 150 min/week of sporadic MVPA, causing wider confidence intervals in the estimates. One possible explanation for the U-shaped association is reverse causality, where underlying diseases or reduced physical function have caused the sporadic PA pattern rather than the PA pattern causing higher mortality. This phenomenon, known as PA fragmentation¹⁹, has been associated with an increased risk of all-cause mortality in older adults²⁰. Because PA fragmentation could reflect underlying disease²⁰, we repeat the analysis in a healthy subsample, excluding participants with cancer, CVD, diabetes, COPD, or poor self-rated health. The U-shaped association persists in the healthy-sample analysis, implying that the health implications of high levels of sporadic MVPA in older adults warrant further investigation.

Our findings suggest a dose-response association between bouted MVPA and mortality with no appreciable threshold. The results agree well with existing evidence regarding the dose-response association between leisure-time MVPA and all-cause mortality⁴. Another insight from our joint analysis is that bouted MVPA is beneficial at any given level of sporadic MVPA, implying “add-on” health benefits of structured, long-duration MVPA beyond brief, sporadic MVPA. We speculate that bouted MVPA and sporadic MVPA might improve health via different mechanisms: it could be that sporadic MVPA enhances health primarily by improving traditional biomarkers, such as blood pressure and lipids^12,13, while bouted MVPA could be more effective in inducing additional physiological adaptations, such as improved cardiorespiratory fitness³⁸.

In the analysis of the VPA component, the VPA accumulated in MVPA bouts demonstrates a linear inverse association with all-cause mortality and CVD incidence. In contrast, the benefits of sporadic VPA are curvilinear and most pronounced below 10 min/week, with a flatter association at higher doses. This aligns with growing evidence that even a few minutes of vigorous intermittent lifestyle physical activity (VILPA) could be associated with up to 40% lower all-cause mortality and 45% lower major adverse cardiovascular events^39,40. Some of the proposed mechanisms include improvements in cardiorespiratory fitness, insulin sensitivity, and vascular function^40,41. In the analysis of different bout thresholds, the optimal dose of sporadic MVPA increases with higher thresholds: 75 min/week for bouts under 3 min, 150 min/week for bouts under 10 min, and nearly 300 min/week for bouts under 20 min. This observation aligns with the overload principle in exercise training⁴²: less demanding stimuli reach their maximal benefit sooner, and further improvements require progressively greater overload (e.g., higher intensity, longer duration, or increased frequency).

Our findings have important clinical implications. Given that fewer than 20% of middle-aged adults engage in regular exercise^43,44, accumulating 150 min of MVPA through short, occasional opportunities enables individuals to adopt an active lifestyle by seamlessly incorporating MVPA into their daily routines. Considering that many of the most sedentary individuals are unlikely to be capable of accomplishing longer-duration bouted MVPA, sporadic MVPA could serve as a valuable complement, or introduction, to traditional exercise strategies. Ideally, as sedentary individuals progressively increase their fitness levels, they should be encouraged to incorporate bouted MVPA, which is more akin to the traditional notion of “exercise,” to achieve further health benefits.

Our study has several limitations. In our sample, over 92% of participants accumulate 0–300 min/week of sporadic MVPA, leaving only 8% with more than 300 min/week, which increases statistical uncertainty. In contrast, 23% of participants achieved more than 300 min/week of bouted MVPA, allowing more robust estimation of hazard ratios in this group. These distinct distributions of sporadic and bouted MVPA may have contributed to the observed differences in dose-response relationships and associated health outcomes. Wrist-worn accelerometers, which are increasingly used in population-based studies to estimate PA levels, are associated with higher measurement errors compared to the well-established estimates from waist-worn devices⁴⁵. Wrist-worn devices may overestimate total MVPA due to hand movements that do not necessarily reflect whole-body physical effort. Also, accelerometers may have reduced accuracy in capturing certain types of activities, such as cycling or resistance training⁴⁶, and may be influenced by non-ambulatory movements. It has been shown that wrist accelerometers are particularly subject to artifacts such as hand gesturing or tapping, which may be misclassified as MVPA⁴⁷. Another limitation is that the UK Biobank sample is not representative of the general population, as it predominantly includes individuals from economically affluent regions and primarily white ethnic backgrounds. However, previous work indicates that this lack of representativeness of the UK Biobank participants does not materially influence the associations between PA and all-cause or CVD mortality⁴⁸. Future research is warranted to verify our findings across more diverse populations, including those with varied socioeconomic backgrounds.

In conclusion, achieving the guideline-recommended 150 min/week of MVPA through engaging in brief, sporadic MVPA episodes is associated with substantial health benefits. Incorporating bouted MVPA, rather than merely accumulating additional sporadic MVPA beyond the dose of 150 min/week, may confer further health benefits. These findings underscore the importance of promoting both sporadic and bouted MVPA to encourage an active and healthy lifestyle.

Author contributions

Y.C. and T.M. conceived the idea. Y.C. and T.M. wrote the first draft of the report with input from M.Y.C.P., J.S., Y.J.X., Q.Y., C.D.L., X.L., Y.L., and J.L. Y.C. and T.M. did the statistical analysis. X.W. contributed to the development of the machine learning algorithm. T.M. had full access to all the data in the study and takes responsibility for the integrity of the data and accuracy of the data analysis. All authors contributed to the critical revision and approved the final version.

Peer review

Peer review information

Communications Medicine Communications Medicine thanks Fabian Schwendinger and the other, anonymous, reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

UK Biobank data are available to eligible researchers via the formal application process described at https://www.ukbiobank.ac.uk/enable-your-research, and the datasets analyzed in this study can be requested through this route. The Capture-24 dataset can be freely accessed at https://ora.ox.ac.uk/objects/uuid:99d7c092-d865-4a19-b096-cc16440cd001⁴⁹. The numerical source data underlying the main-figure graphs and charts are provided as Supplementary Data and are labeled sequentially. Specifically, the source data for Fig. 1 is in Supplementary Data 1; the source data for Fig. 2 is in Supplementary Data 2; the source data for Fig. 3 is in Supplementary Data 3; and the source data for Fig. 4 is in Supplementary Data 4.

Code availability

The R code implementing the algorithm to capture bouted moderate-to-vigorous physical activity, along with all project scripts for preprocessing, analysis, and figure generation, is available at: https://figshare.com/articles/online_resource/Sporadic_bout_CommMed_code/30857627⁵⁰. The code is free to use for non‑commercial purposes; for other uses, please contact the corresponding authors.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s43856-026-01421-z.