Comparing Cadence vs. Machine Learning based Physical Activity Intensity Classifications: Variations in the Associations of Physical Activity with Mortality

Comparing Cadence vs. Machine Learning based Physical Activity Intensity Classifications: Variations in the Associations of Physical Activity with Mortality | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Comparing Cadence vs. Machine Learning based Physical Activity Intensity Classifications: Variations in the Associations of Physical Activity with Mortality View ORCID Profile Le Wei , Matthew N. Ahmadi , Raaj Kishore Biswas , Stewart G. Trost , Emmanuel Stamatakis doi: https://doi.org/10.1101/2024.03.23.24304766 Le Wei 1 Mackenzie Wearables Research Hub, Charles Perkins Centre, The University of Sydney , Sydney, Australia 2 School of Health Sciences, Faculty of Medicine and Health, The University of Sydney , Sydney, Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Le Wei Matthew N. Ahmadi 1 Mackenzie Wearables Research Hub, Charles Perkins Centre, The University of Sydney , Sydney, Australia 2 School of Health Sciences, Faculty of Medicine and Health, The University of Sydney , Sydney, Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Raaj Kishore Biswas 1 Mackenzie Wearables Research Hub, Charles Perkins Centre, The University of Sydney , Sydney, Australia 2 School of Health Sciences, Faculty of Medicine and Health, The University of Sydney , Sydney, Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Stewart G. Trost 3 School of Human Movement and Nutrition Sciences, The University of Queensland , St Lucia, Queensland, Australia 4 Children’s Health Queensland Hospital and Health Service. Centre for Children’s Health Research , South Brisbane, Queensland, Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site Emmanuel Stamatakis 1 Mackenzie Wearables Research Hub, Charles Perkins Centre, The University of Sydney , Sydney, Australia 2 School of Health Sciences, Faculty of Medicine and Health, The University of Sydney , Sydney, Australia Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: emmanuel.stamatakis{at}sydney.edu.au Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Background Step cadence-based and machine-learning (ML) methods have been used to classify physical activity (PA) intensity in health-related research. This study examined the association of intensity-specific PA daily duration with all-cause (ACM) and CVD mortality varied among cadence-based and ML methods in 68,561 UK Biobank participants. Methods The two-stage-ML method categorized activity type and then intensity. The one-level-cadence method (1LC) derived intensity duration using all detected steps and cadence thresholds of ≥100 steps/min (moderate intensity) and ≥130 steps/min (vigorous intensity). The two-level-cadence method (2LC) detected ambulatory activities and then steps with the same cadence thresholds. Results The 2LC exhibited the most pronounced association at the lower end of the duration, e.g., the 2LC showed the smallest minimum moderate-to-vigorous PA (MVPA) duration (amount associated with 50% of optimal risk reduction) (2LC vs 1LC vs ML, 2.8 minutes/day [95% CI: 2.6, 2.8] vs 11.1 [10.8, 11.4] vs 14.9 [14.6, 15.2]) while exhibiting similar corresponding ACM hazard ratio (HR) among methods (HR: 0.83 [95% CI: 0.78, 0.88] vs 0.80 [0.76, 0.85] vs 0.82 [0.76, 0.87]). The ML elicited the greatest mortality risk reduction, e.g., for VPA-ACM association, 2LC vs 1LC vs ML: median, 2.0 minutes/day [95% CI: 2.0, 2.0] vs 6.9 [6.9, 7.0] vs 3.2 [3.2, 3.2]; HR, 0.69 [0.61, 0.79] vs 0.68 [0.60, 0.77] vs 0.53 [0.44, 0.64]. After standardizing duration, the ML exhibited the most pronounced associations, e.g., for MPA-CVD mortality, 2LC vs 1LC vs ML, standardized minimum-duration: -0.77 vs - 0.85 vs -0.94; HR 0.82 [0.72, 0.92] vs 0.79 [0.69, 0.90] vs 0.77 [0.69, 0.85]. Conclusion The 2LC exhibited the most pronounced association with mortality at the lower end of the duration. The ML method provided the most pronounced association with mortality after standardizing the durations. 1. Introduction Current PA guidelines recommend weekly 150-300 moderate-to-vigorous intensity physical activity (MVPA) mins, or 75-150 vigorous intensity PA (VPA) mins and emphasize health benefits of all PA bouts regardless of their duration 1 , 2 . However, most of these guidelines were derived from self-reports which typically fail to recognize the health value of short PA bout (e.g., < 10-minute bout) 3 , thus affecting the accuracy of the estimates of the health impact of different PA intensities. Recent studies have shown that the short intermittent activity bursts were beneficially associated with long term health risks (e.g., all-cause 4 , 5 , CVD 4 and cancer mortality risk 4 , and CVD 5 and cancer 6 incidence). The advancements in wearables measurement technology and methods for processing and analyzing large volumes of accelerometer data provide researchers with multiple options for classifying PA intensity within the short bout. Besides the more recent machine learning (ML) classification models 7 , other methods include cut-points or thresholds based on proprietary activity counts 8 or gravitational units 9 , or step cadence thresholds 10 . The step cadence of ≥ 100 steps/min has been examined as a heuristic threshold for moderate-to-vigorous intensity for adults in both treadmill and overground walking 10 – 14 . Similarly, a cadence of ≥ 130 steps/min has been proposed as the heuristic threshold for vigorous intensity among young and middle-aged adults 10 , 11 . The step cadence methods are feasible with any accelerometer providing researchers with a relatively simple method to estimate the association of duration of PA across intensities with health outcomes in free-living samples. Prior studies have applied cadence thresholds to examine the association of PA intensities duration with various health outcomes 15 – 21 . For instance, a harmonized meta-analysis including 15 international accelerometry cohorts (13 waist, 1 thigh, 1 wrist) found that the association of the duration of moderate or higher intensity (defined as ≥ 100 steps/min) with mortality was attenuated after adjusting for total step counts, albeit remained significant 15 . Moreover, step is the fundamental unit of the ambulatory PA, step cadence-based thresholds may help people understand and monitor their required higher intensity PA minutes (e.g., at least 150 MPA minutes per week) 1 , 2 , 22 and support more people to be sufficiently active 23 . ML methods have gained considerable attention in PA research recent years due to the increasing application of accelerometers and accessibility to raw data (gravitational acceleration) 24 . By training a model with “features” of the accelerometer signal, such as time and frequency domain, that are extracted from the raw data, ML models classify patterns in the raw accelerometer signal corresponding to an activity type or intensity 24 . ML algorithm has the advantage of high accuracy in identifying different PAs and intensities in both lab and free-living environments 7 , 25 , 26 . Recent observational studies applying the ML method showed that approximately daily 2-3 short bouts or 3-4 minutes of vigorous intermittent lifestyle PA (VILPA) or 1-5 MV-ILPA minutes were associated with substantially reduced mortality risk, and cancer and CVD incidence 4 – 6 , amounts that are considerably lower than current major PA guidelines recommendations (at least 75 weekly VPA mins or 150 MPA mins) 1 , 2 , 22 primarily based on questionnaire-based data. Prior research indicated that the ML method demonstrated significantly higher agreement with ground-truth PA intensity than cut-point (counts) methods for preschool children in free-living environment 27 . A recent unpublished study of ours 28 indicated that the intensity-specific PA daily duration estimates based on step cadence thresholds were substantially different to those derived using the ML methods (mean absolute difference for MVPA: 24.2 mins/day). Such differences prompt the question of whether the PA intensity-mortality associations differ by intensity classification method in free-living conditions. To our knowledge, no previous study has examined the extent to which association between the duration of the PA intensity and mortality varies according to multiple methods derived by processing the accelerometer data to estimate PA intensity. Such information is important for guiding analytic and resource allocation decisions in observational research and for understanding the role of PA intensity classification in health research. The purpose of this study was to compare the association of intensity-specific daily PA duration with all-cause and CVD mortality, derived from two step cadence-based and one ML method, in a large UK cohort with wrist-worn accelerometer data. 2. Materials and Methods 2.1 Participants The UK Biobank is a large prospective cohort with 502,616 aged 40-69 years UK adults recruited between 2006 and 2010 29 . Participants completed baseline measurements and provided written consent to use their data. The ethical approval was provided by the UK National Health Service, National Research Ethics Service (Ref 11/NW/0382). From 2013 to 2015, 103,684 UK biobank participants were sent and wore the Axivity AX3 wrist-worn triaxial accelerometer (Newcastle upon Tyne, UK) on their dominant wrist for 7 days continuously. The AX3 was initialized to capture triaxial acceleration data at a sampling frequency of 100 Hz and a dynamic range of ±8 g. A monitoring day was considered valid if the wear time exceeded 16 hours. Participants returned the devices by mail and the data were calibrated 30 and non-wear periods were identified 31 . Participants needed at least four valid monitoring days including a minimum of one valid weekend day to be included in our study. As previous study 7 , we excluded participants with missing covariate data, CVD and cancer history (ascertained through hospital admission records), and outcome events occurring within one year after the PA assessment. In total, 68,561 participants were included in our study ( Supplementary figure 1 ). Download figure Open in new tab Supplemental fig 1 Flow diagram of participants 2.2 Physical activity intensity classification methods 2.2.1 Machine learning method This two-stage classifier first applied a validated random forest (RF) classifier trained on 48 time and frequency domain features to categorize each 10-second window into one of the four PA classes: sedentary (sitting still, reclined, etc.), standing utilitarian movements (standing stationary/active standing: washing dishes, ironing a shirt, etc.), walking activities (active commuting, ambulatory gardening, etc.), or running/energetic activity (chasing around with children, etc.) 7 . Time windows classified as walking were assigned to different PA intensity bands using ambulatory acceleration thresholds and those classified as running/energetic activities were classified as vigorous (≥ 6 METs). Time windows classified as walking with normalized gravitational units < 100 milli g’s were considered LPA (<3 METs) and those classified as walking with normalized gravitation units ≥100 milli g and < 400 milli g were considered MPA (3-6 METs). The time windows classified as walking with normalized gravitation units ≥ 400 milli g walking activities were considered vigorous PA (VPA) (≥6 METs) 32 , 33 . Windows classified as sedentary behaviors were considered sedentary (< 1.5 METs). We calculated the mean daily duration across intensities by averaging over valid monitoring days. 2.2.2 One-level cadence method We classified PA intensities using a validated one-level cadence method 10 (1LC). We implemented an open-source Windowed Peak Detection algorithm on the raw acceleration signal to detect steps 34 . This algorithm demonstrated high accuracy (Mean 89%, SD 12%) during treadmill and outdoor walking 34 . We calculated the step counts within each 60-second window and classified MPA, VPA and MVPA based on previously published step cadence thresholds (10–14). MPA was defined as a cadence ≥100 and <130 steps/min, VPA was defined as ≥ 130 steps/min, MVPA was defined as ≥ 100 steps/min. We considered a cadence of ≥ 20 and < 100 steps/min as light-PA (LPA). We used 20 steps/min as the lower cut-point for LPA to avoid misclassifying shuffling or pottering steps that occur in the 60-second window. 2.2.3 Two-level cadence method Accurate step detection is challenging in free-living environments. A study showed that routinely above 20% mean average percentage error (MAPE) existed in step counting accuracy in wrist-worn accelerometer cohort studies 35 ., likely due to the random wrist and upper limb movement (e.g., eating, brushing teeth, typing keyboard) misclassified as “steps”. To improve step counting accuracy, a hybrid two-level cadence method was developed (2LC). Firstly, we identified stepping activities (e.g., walking, running) by applying the RF classifier used in the ML method. Secondly, we summed up the step counts in 10-second windows classified as walking within each non-overlapping one minute interval, and classified that minute as LPA, MPA and VPA based on the same cadence thresholds applied in 1LC 10 . Then we added up the minutes corresponding to each intensity level, respectively, and calculated the mean daily duration across intensities by averaging over valid monitoring days. 2.3 Outcome ascertainment Participants were followed up until 30 November 2022. The deaths records were obtained through linkage with the National Health Service (NHS) Digital of England and Wales or the NHS Central Register and National Records of Scotland. Based on the ICD-10 (International Classification of Diseases, 10 th revision) from both primary and contributory death cause, we defined CVD mortality as death from diseases of the circulatory system (ICD-10 codes: I0, I11, I13, I20–I51, I60–I69), excluding hypertension and diseases of arteries and lymph. 2.4 Covariates We adjusted for covariates that are typical in studies examining intensity-specific PA and CVD risk based on previous peer-reviewed literature 36 . We included age, sex, ethnicity, accelerometer wear duration, LPA minutes, MPA minutes, VPA minutes, smoking status, alcohol consumption, sleep duration, diet, screen time, education level, parental history of CVD and cancer, and medication use of cholesterol, blood pressure, or diabetes. 2.5 Statistical analysis To minimize the influence of the sparse data, we excluded the MVPA, MPA, and VPA duration derived from the cadence-based and ML method above the 95th percentile 4 , 6 , 7 . We examined the time-to-event dose-response association (hazard ratios [HRs]) of duration across intensities classified from all three methods with ACM and CVD mortality using the Cox proportional hazards regression with the 5th percentile of the duration across intensities as the reference and age as timescale. We demonstrated the dose-response association using the restricted cubic spline with knots placed at the 6th, 34th, and 67th percentile to fit the right-skewed distribution 6 . We assessed the proportional hazards assumptions using Schoenfeld residuals and observed no violations. We presented overlapping plots to visually show and compare the dose-response associations of daily duration at each PA intensity with ACM and CVD mortality across the cadence-based and ML methods. To compare across methods, we assessed the minimum dose 4 , 6 , 7 , 25th and 75th percentile dose (indicated by the ED50, ED25, ED75 in figures and tables, which estimate the daily duration associated with 50%, 25% and 75% of the optimal risk reduction, respectively), and the optimal dose (the nadir of the dose-response curve, representing the optimal risk reduction) and median daily duration and their corresponding HRs (95% CI). We used bootstrapping with replacement (1000 iterations) to calculate the 95% confidence interval for above metrics. We also assessed the dose-response association of the standardized MVPA/MPA/VPA duration as computed by each method (defined as absolute duration divided by the corresponding standard deviation [SD]) with the two mortality outcomes. To assess the robustness of our results, we performed the sensitivity analysis on the 90th percentile of dataset derived from each method to further minimise the influence of sparse minutes, and on the participants based on the 95 th percentile of the 1LC method. We performed all analyses using R software (version 4.2.2) and fitted models using rms package (version 6.3.0). 3. Results 3.1. Description of the study sample Supplementary Table 1 shows the characteristics of the sample derived by the 1LC method in MVPA minutes, stratified by quartiles of daily MVPA minutes. The mean (SD) age was 61.8 (7.8), and 39,847 (58.1%) were women. Over a mean follow-up 8.0 (0.9) years, there were 2134 deaths, of which 575 were due to CVD related causes. Supplementary Fig. 1 shows the complete sample deprivation process. View this table: View inline View popup Supplementary Table 1 Characteristics of Participants, by quartile a of one-level cadence method-derived MVPA minutes 3.2. Associations of daily MVPA with all-cause and CVD mortality Fig. 1 shows that the L-shaped dose-response curve of MVPA duration with ACM derived from the ML and 1LC method were similar, which the association decreased in a near-linear fashion up to a similar nadir point (optimal duration: ML, 32.0 (30.3, 33.9), HR, 0.63 (0.53, 0.75); 1LC, 31.9 (30.0, 34.7), HR, 0.60 (0.53, 0.69)) and then increased in a near-linear fashion with 95% CI largely overlapped. The L-shaped association derived from the 2LC method exhibited a more pronounced ACM risk reduction than other methods till reaching its nadir point. The median, minimum, 25 th , 75 th and the optimal doses in 2LC were substantially smaller than those in the 1LC and ML method, with slightly higher corresponding ACM HRs, e.g., the minimum dose ( Supplementary Table 2 ) : ML vs 1LC vs 2LC, 14.9 (95% CI: 14.6, 15.2) vs 11.1 (10.8, 11.4) vs 2.8 (2.6, 2.8), HR: 0.82 (95% CI: 0.76, 0.87) vs 0.80 (0.76, 0.85) vs 0.83 (0.78, 0.88). Fig. 2 shows that the association derived from the 2LC method exhibited a more pronounced CVD risk reduction than other methods until its nadir point. ML method provided the lowest CVD HR across all the metrics among the methods, e.g., the ED75 dose, ML vs 2LC, 20.0 (19.0, 23.5) vs 5.6 (4.1, 9.9); HR: 0.60 (0.50, 0.71) vs 0.69 (0.58, 0.83) ( Supplementary Table 3 ) . However, there was less evidence (wider 95% CI) for CVD mortality than ACM due to fewer CVD mortality events. View this table: View inline View popup Supplementary table 2 The duration associated with the ED25, ED50, ED75, ED100 and the median of the intensities duration derived from cadence-based methods (1LC and 2LC)* and machine learning (ML) method with all-cause mortality View this table: View inline View popup Supplementary Table 3 The duration associated with the ED25, ED50, ED75, ED100 and the median of the intensities duration derived from cadence-based methods (1LC and 2LC)* and machine learning (ML) method with CVD mortality Download figure Open in new tab Fig. 1 Association of the MVPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with all-cause mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Cross, diamond, and triangle represent the daily MVPA minutes that associated with 25% (ED25), 50% (ED50), and 75% (ED75) of the optimal risk reduction, respectively, and circle represents the median daily MVPA minutes. Please see supplementary table 2 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, light physical activity duration, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Fig 2 Association of the MVPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with CVD mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Cross, diamond, and triangle represent the daily MVPA minutes that associated with 25% (ED25), 50% (ED50), and 75% (ED75) of the optimal risk reduction, respectively, and circle represents the median daily MVPA minutes. Please see supplementary table 2 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, Light PA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. 3.3. Associations of MPA duration with all-cause and CVD mortality Fig. 3 shows that the L-shaped dose-response association of MPA mins between the ML and 1LC method with ACM were similar with 95% CI largely overlapped, and both methods exhibited linear inverse association up to around the median duration (ML:32.2 [32.1, 32.3], HR, 0.60 [0.51, 0.70]; 1LC: 19.9 [19.9, 20.0], HR, 0.65 [0.58, 0.73]) ( Supplementary Table 2 ) where the curve flattened. Meanwhile, the 2LC method exhibited steeper risk reduction in a near-linear fashion up to its median (5.3 [5.3, 5.4], HR, 0.68 [0.60, 0.77]) and continued to show more risk reduction than other methods until 20 MPA minutes. The ML method provided substantially larger minimum, 25 th , and 75 th percentile durations than the 1LC and 2LC methods, with more ACM risk reduction, e.g., the minimum dose ( Supplementary Table 2 ) : ML vs 1LC vs 2LC, 14.9 (95% CI: 14.6, 15.2) vs 8.6 (8.3, 9.3) vs 2.3 (2.1, 2.8). HR: 0.77 (0.70, 0.83) vs 0.81 (0.76, 0.86) vs 0.82 (0.77, 0.88). Fig. 4 shows that L-shaped beneficial association with CVD mortality across methods, respectively, was similar to that in the MPA-ACM association in terms of the dose-response curve and other metrics ( Supplementary Table 3 ), except for the upward pattern of the 2LC method after the median MPA mins (roughly 6 mins). Download figure Open in new tab Fig. 3 Association of the MPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with all-cause mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Cross, diamond, and triangle represent the daily MVPA minutes that associated with 25% (ED25), 50% (ED50), and 75% (ED75) of the optimal risk reduction, respectively, and circle represents the median daily MVPA minutes. Please see supplementary table 2 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, light PA, VPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95 th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Fig 4 Association of the MPA minutes derived from the cadence-based methods (1LC and 2LC)* and machine learning (ML) method with CVD mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Cross, diamond, and triangle represent the daily MVPA minutes that associated with 25% (ED25), 50% (ED25), and 75% (ED25) of the optimal risk reduction, respectively, and circle represents the median daily MVPA minutes. Please see supplementary table 2 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, light PA, VPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. 3.4. Associations of VPA duration with all-cause and CVD mortality Fig. 5 shows that the 1LC method provided the least pronounced dose-response association of VPA mins with ACM, and the ML method provided the most pronounced one, e.g., the median ( Supplementary Table 2 ) : ML, 3.2 (95% CI: 3.2, 3.2), HR, 0.53 (95% CI: 0.44, 0.64); 1LC, 6.9 (6.9, 7.0), 0.68 (0.60, 0.77); 2LC, 2.0 (2.0, 2.0), 0.69 (0.61, 0.79). The dose-response association of the 2LC method was almost identical to that of the ML method up to roughly 1.5 mins, with both methods showing steep risk reduction. The ML method further exhibited steep ACM risk reduction up to around its median point and demonstrated substantially lower ACM risk reduction than that of the 2LC method. Fig. 6 shows that the dose-response association of VPA minutes derived by the 2LC method with CVD mortality exhibited more pronounced risk reduction compared to the ML method up to around 10 minutes where the curve flattened. The 2LC method provided smaller dose but higher corresponding CVD HRs for most point estimates, e.g, the minimum point ( Supplementary Table 3 ): ML: 9.0 (7.4, 9.3), HR, 0.70 (0.60, 0.81); 2LC, 0.6 (0.4, 0.9), HR, 0.75 (0.67, 0.85). However, there was less evidence (wider 95% CI) for CVD mortality than ACM due to fewer CVD mortality events. Download figure Open in new tab Fig. 5 Association of the VPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with all-cause mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Cross, diamond, and triangle represent the daily MVPA minutes that associated with 25%(ED25), 50% (ED50), and 75% (ED75) of the optimal risk reduction, respectively, and circle represents the median daily MVPA minutes. Please see supplementary table 2 for the list of metrics (duration and corresponding HR). Adjusted for age, sex, wear time, light PA, MPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Fig 6 Association of the VPA minutes derived from the cadence-based methods (1LC and 2LC)* and machine learning (ML) method with CVD mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Cross, diamond, and triangle represent the daily MVPA minutes that associated with 25% (ED25), 50% (ED50), and 75% (ED75) of the optimal risk reduction, respectively, and circle represents the median daily MVPA minutes. Please see supplementary table 2 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, light PA, MPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. 3.5. Associations of the standardised duration with all-cause and CVD mortality Supplementary Fig 2 - 4 show that the dose-response associations of standardized duration with ACM across intensities, respectively, was similar between the 1LC and 2LC method, with 95% CI largely overlapped. The ML method provided the most pronounced association of standardized duration with ACM across intensities, especially for the standardized VPA minutes-ACM association ( Supplementary Fig 4 ), e.g., for the median of the standardized VPA minutes/day and the corresponding HR (ML: -0.31, HR, 0.53 (0.44, 0.64); 1LC: -0.35, HR, 0.68 (0.60, 0.77); 2LC: -0.39, HR, 0.71, (0.63, 0.81)) ( Supplementary Table 4 ) . Supplementary Table 4 shows that almost all the metrics related to MPA and VPA between the 2LC and the ML method were different. Supplementary Fig 5 and 7 show that the 95% CI of the L-shaped dose-response association of the standardized MVPA and VPA duration with CVD mortality were largely overlapped. Supplementary Fig 6 shows that the ML method provided the most pronounced association of standardized MPA duration with CVD mortality compared to other methods, e.g., for the median of the standardized MPA minutes/day and the corresponding HR (ML: -0.20, HR, 0.54 (0.43,0.68); 1LC: -0.27, HR, 0.59 (0.43,0.81); 2LC: -0.35, HR, 0.64 (0.48, 0.86)) ( Supplementary Table 5 ) . We observed less evidence (wider 95% CI) for association of standardized duration with CVD mortality due to fewer CVD mortality events than ACM. Download figure Open in new tab Supplementary fig 2 Association of the standardized MVPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with all-cause mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Diamond represents the minimum dose (ED50), which estimates the daily standardised MVPA minutes associated with 50% of optimal risk reduction across methods. Circle represents the standardised daily median MVPA minutes across methods. Please see supplementary table 3 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, LPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Supplementary fig 3 Association of the standardized MPA minutes derived from the cadence-based methods (1LC and 2LC)* and machine learning (ML) method with All-cause mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Diamond represents the minimum dose (ED50), which estimates the daily standardised MPA minutes associated with 50% of optimal risk reduction across methods. Circle represents the standardised median MPA minutes across methods. Please see supplementary table 3 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, LPA, VPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95 th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Supplementary fig 4 Association of the standardized VPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with all-cause mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Diamond represents the minimum dose (ED50), which estimates the daily standardised VPA minutes associated with 50% of optimal risk reduction across methods. Circle represents the standardised median VPA minutes across methods. Please see supplementary table 3 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, LPA, MPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Supplementary fig 5 Association of the standardized MVPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with CVD mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Diamond represents the minimum dose (ED50), which estimates the daily standardised MVPA minutes associated with 50% of optimal risk reduction across methods. Circle represents the standardised median MVPA minutes across methods. Please see supplementary table 4 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, LPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Supplementary fig 6 Association of the standardized MPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with CVD mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Diamond represents the minimum dose (ED50), which estimates the daily standardised MPA minutes associated with 50% of optimal risk reduction across methods. Circle represents the standardised median MPA minutes across methods. Please see supplementary table 4 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, LPA, VPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. Download figure Open in new tab Supplementary fig 7 Association of the standardized VPA minutes derived from the cadence-based methods (1LC and 2LC) and machine learning (ML) method with CVD mortality. 1LC indicates the one-level cadence method. 2LC indicates the two-level cadence method. Diamond represents the minimum dose (ED50), which estimates the daily standardised VPA minutes associated with 50% of optimal risk reduction across methods. Circle represents the standardised median VPA minutes across methods. Please see supplementary table 4 for the list of metrics (duration and corresponding HR). The association was adjusted for age, sex, wear time, LPA, MPA, smoking status, alcohol consumption, sleep duration, diet, screen-time, education, self-reported parental history of CVD and cancer, and self-reported medication use (cholesterol, blood pressure, and diabetes). The range was capped at the 95th percentile to minimize the influence of sparse data. All methods used their corresponding 5th percentile as the reference level to calculate each HR. View this table: View inline View popup Supplementary table 4 The standardized duration associated with ED25, ED50, ED75, ED100 and the median of the intensities duration derived from cadence-based methods (1LC and 2LC)* and machine learning (ML) method with all-cause mortality View this table: View inline View popup Supplementary table 5 The standardized duration associated with the ED25, ED50, ED75, ED100 and the median of the intensities duration derived from cadence-based methods (1LC and 2LC)* and machine learning (ML) method with CVD mortality 4. Discussion To our knowledge, this is the first study to compare the dose-response associations of different PA intensity classification methods (cadence-based methods and ML method) with standard health outcomes used routinely in observational studies. The ML and 1LC methods provided similar dose-response associations. The 2LC method exhibited the most pronounced rate of decrease in the association with mortality across intensities at the lower end of the duration. The ML elicited more risk reduction in mortality than cadence-based methods, particularly for the VPA-ACM association. The ML method provided the most pronounced dose-response associations of the standardized duration with mortality outcomes across different intensity levels. Future research could focus on deriving the cadence-thresholds for short windows, e.g., 10-second window for each PA intensity level, to inform the step-based public health guideline and interventions, etc. A potential explanation for the variation in dose-response associations is the different window length applied in the 2LC and ML method. Our ML method classified PA intensity using 10-second window, enabling the detection of finer pattern of MVPA derived by very brief bouts, thus leading to different dose-response association relative to the 2LC method which applied 60-second window. The shorter window is more likely to detect MVPA burst 37 . Prior studies applying various window lengths (15 seconds, 30 seconds, 60 seconds) demonstrated that shorter windows detect significantly more MVPA minutes 27 , 38 . Applying a longer epoch such as 60-seconds window might result in a smoothing effect 39 , e.g., a 30-second jogging to catch bus (MVPA burst) might be degraded to lower intensity using 2LC method. A recently published large cohort study has indicated that the MVPA accumulated through bouts that less than 1 min (with VPA being at least 15% proportion of the MVPA time) would elicit beneficial change on ACM risk 5 . Such smoothing effect cause higher intensity PA to decrease to a lower level, potentially misleading researchers to unreliable intensity time – morality association and therefore overlooked the health effect of the valuable granularity of intensity time. Another potential explanation is that the 2LC method, although be able to detect stepping activities and thus counting steps accurately, might overlook the variation in step intensity (e.g., steps involving upper-body movements). The ML method, in conjunction with the wrist accelerometer, could detect steps associated with upper-body movements 40 (e.g., ambulatory gardening), thus capturing wider range of PAs and providing more representative duration of PA intensities than 2LC method. Prior studies have indicated that PAs involving upper-body movements such as occupational and household activities account for a large proportion of MVPA time (e.g., the household PAs was the most prevalent domain for those who reported under weekly 150 MVPA mins) 41 . The association of the VPA duration derived by the ML method with ACM was similar to that of 2LC method at the lower end of distribution (up to 2 mins), suggesting that these VPA duration might be the result of incidental brisk walking or running (e.g., quickly moving for 10 seconds). Prior study has shown that a large portion (92.3%) of VILPA bouts in non-exercisers were less than a minute 4 . The 2LC using 60-seconds window may overlook many such bursts and thus unlikely to derive the dose-response association close to ML method in free-living environment. The ML-derived VPA dose-response association appeared the steepest HR reduction for ACM with the lowest HR associated in all point estimates, suggesting that the VPA duration including granularity duration might demonstrate more beneficial change on ACM than that of the VPA duration calculated based on step counts per min. This enhanced the value of the short VPA burst for achieving beneficial health change. After standardization of the MVPA/MPA/VPA minutes, the dose-response association with mortality between the 2LC and the ML method showed variation in terms of the curve and most point estimates. This was especially evident in the association of standardized VPA duration with the ACM. This finding implied that the propensity of the estimation methods not only provide different estimates of the duration, but also the additional duration of the MVPA/MPA/VPA bursts detected by the ML method elicited higher degree of mortality risk reduction than other methods. This enhanced the role of the short bout higher intensity PA in potentially beneficial changes on health. Wrist, as the most active part of body, produces many extraneous wrist movements (e.g., eating) which might be recorded as ambulatory by wrist-worn accelerometers (“error steps”) 42 . Although the 1LC method was based on the same cadence-thresholds as 2LC, it counted steps without identifying the stepping activities first, which leads to substantial higher “error steps” counts than the 2LC method. Therefore, the 1LC method inflated MVPA/MPA durations and thus derived similar association to the ML method. A study comparing the step detection accuracy of 13 wearables found that wrist-worn accelerometers detected roughly 450 more steps per day than waist-worn in free-living environment 43 . The inflated MVPA/MPA duration might lead to associations with mortality close to that of ML method. Nonetheless, the inherent randomness in wrist movements might introduce variations in error steps count, influenced by factors such as the variation of wrist-movements, sample characteristics or step-detection methods making 1LC-derived dose-response association with mortality unstable. To our knowledge, this is the first study comparing the PA intensity classification methods (cadence-based and ML method) in terms of the association with standard health outcomes used routinely in observational studies. Our study had large sample with long follow-up years. The ML method could detect the micro-pattern of PA and upper-body movements in free-living environment. The 2LC method used advanced step detection algorithm and provide better accuracy. We have conducted multiple sensitivity analyses to show robustness of our findings. Our study had several limitations. Our study categorized confounders (e.g., sleep duration, LPA) to avoid statistical assumption violation. The ML method is not the ground-truth PA intensity time estimation method, although showed high accuracy for MPA (precision: 92.7) and VPA (94.6) comparing to the criterion data 7 . The measurement error in the ML method for PA intensity time might affect the comparison of the intensity time - mortality association among methods. The UK biobank has a low response-rate (5.5%) and thus not representative of the population 44 , although a study has demonstrated that poor representativeness in UK biobank didn’t materially affect the PA-mortalities associations 45 . 5. Perspective The 1LC method produced similar dose-response MVPA/MPA-mortality associations to the ML method. The 2LC method exhibited the most pronounced associations with mortality across intensities than the ML method at lower end of the duration. The ML method elicited more risk reduction in mortality than cadence-based methods, particularly for the VPA-ACM association. After standardization of the duration, the ML method generally provided most pronounced associations across intensities. These suggested that the MPA/VPA bursts detected by the shorter detection window (10-second) in the ML method can capture the associations between activities and health outcomes with greater granularity. The method that based on the threshold of walking steps in a minute long window might underestimate both the duration and potential health benefits of the MPA and VPA bouts under the free-living conditions. Future studies could focus on the cadence-based thresholds for windows less than a minute for detecting the MPA and VPA bursts. Funding This study was funded by the National Health and Medical Research Council (NHMRC) through a Leadership level 2 Fellowship to Emmanuel Stamatakis (APP1194510). Authors’ contributions E.S., M.N.A, L.W., R.K.B, and S.G.T contributed to the study conception and design, and interpretation of data. Data preparation was performed by M.N.A. The draft of the manuscript and data analysis were performed by L.W. All authors revised it critically and gave final approval and agreed to be accountable for all aspects of the work, ensuring integrity and accuracy. Conflict of Interest Statement The authors declare that they have no conflict of interest. Informed Consent Informed consent was obtained from all individual participants included in the study. Data Availability Statement This research has been conducted using the UK Biobank Resource under Application Number 25813. Bona fide researchers can register and apply to use the UK Biobank dataset at http://ukbiobank.ac.uk/register-apply/ . Acknowledgements This research has been conducted using the UK Biobank Resource under Application Number 25813. The authors would like to thank all the participants and professionals contributing to the UK Biobank. References 1. ↵ Bull FC , Al-Ansari SS , Biddle S , et al. World Health Organization 2020 guidelines on physical activity and sedentary behaviour . Br J Sports Med . 2020 ; 54 ( 24 ): 1451 – 1462 . doi: 10.1136/bjsports-2020-102955 OpenUrl Abstract / FREE Full Text 2. ↵ Piercy KL , Troiano RP , Ballard RM , et al. The physical activity guidelines for Americans . JAMA - Journal of the American Medical Association . 2018 ; 320 ( 19 ): 2020 – 2028 . doi: 10.1001/jama.2018.14854 OpenUrl CrossRef PubMed 3. ↵ Sallis JF , Saelens BE. Assessment of physical activity by self-report: Status, limitations, and future directions . Res Q Exerc Sport . 2000 ; 71 : 1 – 14 . doi: 10.1080/02701367.2000.11082780 OpenUrl CrossRef PubMed Web of Science 4. ↵ Stamatakis E , Ahmadi MN , Gill JMR , et al. Association of wearable device-measured vigorous intermittent lifestyle physical activity with mortality . Nat Med . 2022 ; 28 ( 12 ): 2521 – 2529 . doi: 10.1038/s41591-022-02100-x OpenUrl CrossRef 5. ↵ Ahmadi MN , Hamer M , Gill JMR , et al. Brief bouts of device-measured intermittent lifestyle physical activity and its association with major adverse cardiovascular events and mortality in people who do not exercise: a prospective cohort study . Lancet Public Health . 2023 ; 8 ( 10 ): e800 – e810 . doi: 10.1016/S2468-2667(23)00183-4 OpenUrl CrossRef 6. ↵ Stamatakis E , Ahmadi MN , Friedenreich CM , et al. Vigorous Intermittent Lifestyle Physical Activity and Cancer Incidence Among Nonexercising Adults: The UK Biobank Accelerometry Study . JAMA Oncol. Published online July 27 , 2023 . doi: 10.1001/jamaoncol.2023.1830 OpenUrl CrossRef 7. ↵ Ahmadi MN , Clare PJ , Katzmarzyk PT, del Pozo Cruz B, Lee IM, Stamatakis E. Vigorous physical activity, incident heart disease, and cancer: how little is enough? Eur Heart J. Published online December 7 , 2022 . doi: 10.1093/eurheartj/ehac572 OpenUrl CrossRef 8. ↵ Trost SG , Mciver KL , Pate RR. Conducting accelerometer-based activity assessments in field-based research . In: Medicine and Science in Sports and Exercise . Vol 37 . ; 2005 . doi: 10.1249/01.mss.0000185657.86065.98 OpenUrl CrossRef PubMed Web of Science 9. ↵ Stemland I , Ingebrigtsen J , Christiansen CS , et al. Validity of the Acti4 method for detection of physical activity types in free-living settings: comparison with video analysis . Ergonomics . 2015 ; 58 ( 6 ): 953 – 965 . doi: 10.1080/00140139.2014.998724 OpenUrl CrossRef PubMed 10. ↵ Tudor-Locke C , Ducharme SW , Aguiar EJ , et al. Walking cadence (steps/min) and intensity in 41 to 60-year-old adults: the CADENCE-adults study . International Journal of Behavioral Nutrition and Physical Activity . 2020 ; 17 ( 1 ). doi: 10.1186/s12966-020-01045-z OpenUrl CrossRef 11. ↵ Tudor-Locke C , Aguiar EJ , Han H , et al. Walking cadence (steps/min) and intensity in 21-40 year olds: CADENCE-adults . International Journal of Behavioral Nutrition and Physical Activity . 2019 ; 16 ( 1 ). doi: 10.1186/s12966-019-0769-6 OpenUrl CrossRef PubMed 12. Tudor-Locke C , Mora-Gonzalez J , Ducharme SW , et al. Walking cadence (steps/min) and intensity in 61–85-year-old adults: the CADENCE-Adults study . International Journal of Behavioral Nutrition and Physical Activity . 2021 ; 18 ( 1 ). doi: 10.1186/s12966-021-01199-4 OpenUrl CrossRef 13. Aguiar EJ , Mora-Gonzalez J , Ducharme SW , et al. Cadence-based classification of moderate-intensity overground walking in 41- to 85-year-old adults . Scand J Med Sci Sports. Published online 2022 . doi: 10.1111/SMS.14274 OpenUrl CrossRef 14. ↵ Aguiar EJ , Gould ZR , Ducharme SW , Moore CC , McCullough AK , Tudor-Locke C. Cadence-based Classification of Minimally Moderate Intensity during Overground Walking in 21-To 40-Year-Old Adults . J Phys Act Health . 2019 ; 16 ( 12 ): 1092 – 1097 . doi: 10.1123/jpah.2019-0261 OpenUrl CrossRef 15. ↵ Paluch AE , Bajpai S , Bassett DR , et al. Daily steps and all-cause mortality: a meta-analysis of 15 international cohorts . Lancet Public Health . 2022 ; 7 ( 3 ): e219 – e228 . doi: 10.1016/S2468-2667(21)00302-9 OpenUrl CrossRef PubMed 16. Paluch AE , Bajpai S , Ballin M , et al. Prospective Association of Daily Steps With Cardiovascular Disease: A Harmonized Meta-Analysis . Circulation. Published online December 20 , 2022 . doi: 10.1161/CIRCULATIONAHA.122.061288 OpenUrl CrossRef 17. Blodgett JM , Ahmadi MN , Atkin AJ , et al. Device-measured physical activity and cardiometabolic health: the Prospective Physical Activity , Sitting, and Sleep (ProPASS) consortium. Eur Heart J. Published online November 10 , 2023 . doi: 10.1093/eurheartj/ehad717 OpenUrl CrossRef 18. Senaratne N , Stubbs B , Werneck AO , Stamatakis E , Hamer M. Device-measured physical activity and sedentary behaviour in relation to mental wellbeing: An analysis of the 1970 British cohort study . Prev Med (Baltim ). 2021 ; 145 . doi: 10.1016/j.ypmed.2021.106434 OpenUrl CrossRef 19. Blodgett JM , Mitchell JJ , Stamatakis E , Chastin S , Hamer M. Associations between the composition of daily time spent in physical activity, sedentary behaviour and sleep and risk of depression: Compositional data analyses of the 1970 British cohort Study . J Affect Disord . 2023 ; 320 : 616 – 620 . doi: 10.1016/j.jad.2022.09.110 OpenUrl CrossRef 20. Cooper R , Stamatakis E , Hamer M. Associations of sitting and physical activity with grip strength and balance in mid-life: 1970 British Cohort Study . Scand J Med Sci Sports . 2020 ; 30 ( 12 ): 2371 – 2381 . doi: 10.1111/sms.13793 OpenUrl CrossRef 21. ↵ Huang BH , Hamer M , Chastin S , Pearson N , Koster A , Stamatakis E. Cross-sectional associations of device-measured sedentary behaviour and physical activity with cardio-metabolic health in the 1970 British Cohort Study . Diabetic Medicine . 2021 ; 38 ( 2 ). doi: 10.1111/dme.14392 OpenUrl CrossRef 22. ↵ Ross R , Chaput JP , Giangregorio LM , et al. Canadian 24-Hour Movement Guidelines for Adults aged 18-64 years and Adults aged 65 years or older: an integration of physical activity, sedentary behaviour, and sleep. In: Applied Physiology, Nutrition, and Metabolism = Physiologie Appliquee , Nutrition et Metabolisme . Vol 45 . NLM (Medline); 2020 : S57 – S102 . doi: 10.1139/apnm-2020-0467 OpenUrl CrossRef 23. ↵ Stamatakis E , Ahmadi M , Murphy MH , et al. Journey of a thousand miles: From ‘Manpo-Kei’ to the first steps-based physical activity recommendations . Br J Sports Med. Published online 2023 . doi: 10.1136/bjsports-2023-106869 OpenUrl FREE Full Text 24. ↵ Narayanan A , Desai F , Stewart T , Duncan S , MacKay L. Application of raw accelerometer data and machine-learning techniques to characterize human movement behavior: A systematic scoping review . J Phys Act Health . 2020 ; 17 ( 3 ): 360 – 383 . doi: 10.1123/jpah.2019-0088 OpenUrl CrossRef PubMed 25. ↵ Willetts M , Hollowell S , Aslett L , Holmes C , Doherty A. Statistical machine learning of sleep and physical activity phenotypes from sensor data in 96,220 UK Biobank participants . Sci Rep . 2018 ; 8 ( 1 ). doi: 10.1038/s41598-018-26174-1 OpenUrl CrossRef PubMed 26. ↵ Pavey TG , Gilson ND , Gomersall SR , Clark B , Trost SG. Field evaluation of a random forest activity classifier for wrist-worn accelerometer data . J Sci Med Sport . 2017 ; 20 ( 1 ): 75 – 80 . doi: 10.1016/J.JSAMS.2016.06.003 OpenUrl CrossRef 27. ↵ Ahmadi MN , Pavey TG , Trost SG. Machine learning models for classifying physical activity in free-living preschool children . Sensors (Switzerland ). 2020 ; 20 ( 16 ): 1 – 14 . doi: 10.3390/s20164364 OpenUrl CrossRef 28. ↵ Le Wei , Matthew N. Ahmadi , Mark Hamer , et al. Comparing Cadence and ML based Estimates for PA Intensities classification . Unpublished results 29. ↵ Sudlow C , Gallacher J , Allen N , et al. UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age . PLoS Med . 2015 ; 12 ( 3 ). doi: 10.1371/journal.pmed.1001779 OpenUrl CrossRef PubMed 30. ↵ Sipoš M , Pačes P , Roháč J , Nováček P. Analyses of Triaxial Accelerometer Calibration Algorithms . IEEE Sens J . 2012 ; 12 ( 5 ). doi: 10.1109/JSEN.2011.2167319 OpenUrl CrossRef 31. ↵ Ahmadi MN , Nathan N , Sutherland R , Wolfenden L , Trost SG. Non-wear or sleep? Evaluation of five non-wear detection algorithms for raw accelerometer data . Published online 2019 . doi: 10.1080/02640414.2019.1703301 OpenUrl CrossRef 32. ↵ Hildebrand M , Van Hees VT , Hansen BH , Ekelund U. Age group comparability of raw accelerometer output from wrist-and hip-worn monitors . Med Sci Sports Exerc . 2014 ; 46 ( 9 ): 1816 – 1824 . doi: 10.1249/MSS.0000000000000289 OpenUrl CrossRef PubMed Web of Science 33. ↵ Ramakrishnan R , Doherty A , Smith-Byrne K , et al. Accelerometer measured physical activity and the incidence of cardiovascular disease: Evidence from the UK Biobank Cohort study . PLoS Med . 2021 ; 18 ( 1 ). doi: 10.1371/journal.pmed.1003487 OpenUrl CrossRef 34. ↵ Brondin A , Nordström M , Olsson CM , Salvi D. Open source step counter algorithm for wearable devices . In: ACM International Conference Proceeding Series. Association for Computing Machinery ; 2020 . doi: 10.1145/3423423.3423431 OpenUrl CrossRef 35. ↵ Toth LP , Park S , Springer CM , Feyerabend MD , Steeves JA , Bassett DR. Video-Recorded Validation of Wearable Step Counters under Free-living Conditions . Med Sci Sports Exerc . 2018 ; 50 ( 6 ): 1315 – 1322 . doi: 10.1249/MSS.0000000000001569 OpenUrl CrossRef PubMed 36. ↵ Ahmadi MN , Clare PJ , Katzmarzyk PT, del Pozo Cruz B, Lee IM, Stamatakis E. Vigorous physical activity, incident heart disease, and cancer: how little is enough? Eur Heart J. Published online October 27 , 2022 . doi: 10.1093/eurheartj/ehac572 OpenUrl CrossRef 37. ↵ Arvidsson D , Fridolfsson J , Börjesson M. Measurement of physical activity in clinical practice using accelerometers . J Intern Med . 2019 ; 286 ( 2 ): 137 – 153 . doi: 10.1111/joim.12908 OpenUrl CrossRef PubMed 38. ↵ Banda JA , Haydel KF , Davila T , et al. Effects of varying epoch lengths, wear time algorithms, and activity cut-points on estimates of child sedentary behavior and physical activity from accelerometer data . PLoS One . 2016 ; 11 ( 3 ). doi: 10.1371/journal.pone.0150534 OpenUrl CrossRef 39. ↵ Aadland E , Andersen LB , Anderssen SA , Resaland GK , Kvalheim OM. Associations of volumes and patterns of physical activity with metabolic health in children: A multivariate pattern analysis approach . Prev Med (Baltim ). 2018 ; 115 : 12 – 18 . doi: 10.1016/j.ypmed.2018.08.001 OpenUrl CrossRef 40. ↵ Liu F , Wanigatunga AA , Schrack JA. Assessment of Physical Activity in Adults Using Wrist Accelerometers . Epidemiol Rev . 2021 ; 43 ( 1 ): 65 – 93 . doi: 10.1093/epirev/mxab004 OpenUrl CrossRef 41. ↵ Strain T , Fitzsimons C , Foster C , Mutrie N , Townsend N , Kelly P. Age-related comparisons by sex in the domains of aerobic physical activity for adults in Scotland . Prev Med Rep . 2016 ; 3 : 90 – 97 . doi: 10.1016/j.pmedr.2015.12.013 OpenUrl CrossRef PubMed 42. ↵ Bassett DR , Toth LP , LaMunion SR , Crouter SE. Step Counting: A Review of Measurement Considerations and Health-Related Applications . Sports Medicine . 2017 ; 47 ( 7 ): 1303 – 1315 . doi: 10.1007/s40279-016-0663-1 OpenUrl CrossRef 43. ↵ Nakagata T , Murakami H , Kawakami R , et al. Step-count outcomes of 13 different activity trackers: Results from laboratory and free-living experiments . Gait Posture . 2022 ; 98 : 24 – 33 . doi: 10.1016/j.gaitpost.2022.08.004 OpenUrl CrossRef 44. ↵ Fry A , Littlejohns TJ , Sudlow C , et al. Comparison of Sociodemographic and Health-Related Characteristics of UK Biobank Participants with Those of the General Population . Am J Epidemiol . 2017 ; 186 ( 9 ): 1026 – 1034 . doi: 10.1093/aje/kwx246 OpenUrl CrossRef PubMed 45. ↵ Stamatakis E , Owen KB , Shepherd L , Drayton B , Hamer M , Bauman AE. Is Cohort Representativeness Passél]? Poststratified Associations of Lifestyle Risk Factors with Mortality in the UK Biobank . Epidemiology . 2021 ; 32 ( 2 ): 179 – 188 . doi: 10.1097/EDE.0000000000001316 OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted March 24, 2024. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Comparing Cadence vs. Machine Learning based Physical Activity Intensity Classifications: Variations in the Associations of Physical Activity with Mortality Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Comparing Cadence vs. Machine Learning based Physical Activity Intensity Classifications: Variations in the Associations of Physical Activity with Mortality Le Wei , Matthew N. Ahmadi , Raaj Kishore Biswas , Stewart G. Trost , Emmanuel Stamatakis medRxiv 2024.03.23.24304766; doi: https://doi.org/10.1101/2024.03.23.24304766 Share This Article: Copy Citation Tools Comparing Cadence vs. Machine Learning based Physical Activity Intensity Classifications: Variations in the Associations of Physical Activity with Mortality Le Wei , Matthew N. Ahmadi , Raaj Kishore Biswas , Stewart G. Trost , Emmanuel Stamatakis medRxiv 2024.03.23.24304766; doi: https://doi.org/10.1101/2024.03.23.24304766 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Epidemiology Subject Areas All Articles Addiction Medicine (573) Allergy and Immunology (865) Anesthesia (303) Cardiovascular Medicine (4457) Dentistry and Oral Medicine (445) Dermatology (383) Emergency Medicine (610) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1517) Epidemiology (15244) Forensic Medicine (30) Gastroenterology (1132) Genetic and Genomic Medicine (6620) Geriatric Medicine (669) Health Economics (1002) Health Informatics (4557) Health Policy (1372) Health Systems and Quality Improvement (1615) Hematology (543) HIV/AIDS (1272) Infectious Diseases (except HIV/AIDS) (15936) Intensive Care and Critical Care Medicine (1106) Medical Education (624) Medical Ethics (147) Nephrology (670) Neurology (6634) Nursing (346) Nutrition (999) Obstetrics and Gynecology (1148) Occupational and Environmental Health (957) Oncology (3348) Ophthalmology (980) Orthopedics (369) Otolaryngology (421) Pain Medicine (436) Palliative Medicine (130) Pathology (665) Pediatrics (1696) Pharmacology and Therapeutics (693) Primary Care Research (714) Psychiatry and Clinical Psychology (5463) Public and Global Health (9257) Radiology and Imaging (2210) Rehabilitation Medicine and Physical Therapy (1371) Respiratory Medicine (1198) Rheumatology (598) Sexual and Reproductive Health (716) Sports Medicine (532) Surgery (714) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a031cefcc9d2c13d',t:'MTc4MDAxOTAxMA=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();