A machine learning approach to measure and monitor physical activity in children to help fight overweight and obesity

Physical Activity is important for maintaining healthy lifestyles. Recommendations for physical activity levels are issued by most governments as part of public health measures. As such, reliable measurement of physical activity for regulatory purposes is vital. This has lead research to explore standards for achieving this using wearable technology and artificial neural networks that produce classifications for specific physical activity events. Applied from a very early age, the ubiquitous capture of physical activity data using mobile and wearable technology may help us to understand how we can combat childhood obesity and the impact that this has in later life. A supervised machine learning approach is adopted in this paper that utilizes data obtained from accelerometer sensors worn by children in free-living environments. The paper presents a set of activities and features suitable for measuring physical activity and evaluates the use of a Multilayer Perceptron neural network to classify physical activities by activity type. A rigorous reproducible data science methodology is presented for subsequent use in physical activity research. Our results show that it was possible to obtain an overall accuracy of 96 % with 95 % for sensitivity, 99 % for specificity and a kappa value of 94 % when three and four feature combinations were used.

Validation of a computerized use of time recall for activity measurement in advanced-aged adults

BACKGROUND: The Multimedia Activity Recall for Children and Adults (MARCA) is a computerized recall instrument that records use of time during 24 hr the previous day and has been developed to address limitations of current self-report physical activity measures for those in advanced age. METHODS: Test-retest reliability and convergent validity of the adult MARCA were assessed in a sample of 45 advanced-age adults (age 84.9 SD ± 1.62 yr) as a subsample of the Life and Living in Advanced-Age Cohort Study New Zealand (LiLACS NZ). Test-retest methods required participants to recall the previous day's activity using the MARCA twice within the same day. Convergent validity was assessed against accelerometry. RESULTS: Test-retest reliability was high, with ICCs greater than .99 for moderate to vigorous physical activity (MVPA) and physical activity level (PAL). Compared with accelerometry, the MARCA demonstrated validity comparable to other self-report instruments with Spearman's coefficients of .34 and .59 for time spent in nonsedentary physical activity and PAL. CONCLUSION: The MARCA is a valid and reliable self-report tool for physical activity behaviors in advanced-age adults.

Measuring physical activity in a cardiac rehabilitation population using a smartphone-based questionnaire

BACKGROUND: Questionnaires are commonly used to assess physical activity in large population-based studies because of their low cost and convenience. Many self-report physical activity questionnaires have been shown to be valid and reliable measures, but they are subject to measurement errors and misreporting, often due to lengthy recall periods. Mobile phones offer a novel approach to measure self-reported physical activity on a daily basis and offer real-time data collection with the potential to enhance recall.

OBJECTIVE: The aims of this study were to determine the convergent validity of a mobile phone physical activity (MobilePAL) questionnaire against accelerometry in people with cardiovascular disease (CVD), and to compare how the MobilePAL questionnaire performed compared with the commonly used self-recall International Physical Activity Questionnaire (IPAQ).

METHODS: Thirty adults aged 49 to 85 years with CVD were recruited from a local exercise-based cardiac rehabilitation clinic in Auckland, New Zealand. All participants completed a demographics questionnaire and underwent a 6-minute walk test at the first visit. Subsequently, participants were temporarily provided a smartphone (with the MobilePAL questionnaire preloaded that asked 2 questions daily) and an accelerometer, which was to be worn for 7 days. After 1 week, a follow-up visit was completed during which the smartphone and accelerometer were returned, and participants completed the IPAQ.

RESULTS: Average daily physical activity level measured using the MobilePAL questionnaire showed moderate correlation (r=.45; P=.01) with daily activity counts per minute (Acc_CPM) and estimated metabolic equivalents (MET) (r=.45; P=.01) measured using the accelerometer. Both MobilePAL (beta=.42; P=.008) and age (beta=-.48, P=.002) were significantly associated with Acc_CPM (adjusted R(2)=.40). When IPAQ-derived energy expenditure, measured in MET-minutes per week (IPAQ_met), was considered in the predicted model, both IPAQ_met (beta=.51; P=.001) and age (beta=-.36; P=.016) made unique contributions (adjusted R(2)=.47, F2,27=13.58; P<.001).There was also a significant association between the MobilePAL and IPAQ measures (r=.49, beta=.51; P=.007).

CONCLUSIONS: A mobile phone-delivered questionnaire is a relatively reliable and valid measure of physical activity in a CVD cohort. Reliability and validity measures in the present study are comparable to existing self-report measures. Given their ubiquitous use, mobile phones may be an effective method for physical activity surveillance data collection.

Adolescent physical activity: does implementation intention have a role?

BACKGROUND: The current study tested the utility of an integrated social cognitive model to predict physical activity (PA) intentions and behavior in New Zealand adolescents. METHOD: Seventy-two adolescents (mean age = 16.92, SD = 0.66) completed measures consistent with the integrated model (attitude, subjective norm, perceived behavioral control [PBC], goal intention, task-efficacy, barrier efficacy, and implementation intention). Pedometer data (Yamax SW200 pedometer) were collected for 7 days, and a self-report 7-day recall questionnaire was administered at the end of this week. A series of hierarchical regression analyses were conducted to examine the contribution of the model to PA goal intention, implementation intention, self-reported and objective PA. RESULTS: The integrated model accounted for 41% of goal intention, 33% of implementation intention, and 41% and 18% of subjectively and objectively measured PA, respectively. PBC had the strongest association with goal intention whereas attitude had the strongest association with implementation intention. Task-efficacy made the greatest contribution to objectively measured PA, whereas implementation intention had the strongest association with subjectively measured PA. CONCLUSION: These findings have implications regarding PA measurement in adolescent populations, and suggest that social cognitive variables play an important role in adolescent PA. Recommendations for future research are discussed.

Correlates of agreement between accelerometry and self-reported physical activity

PURPOSE: Understanding factors that influence accurate assessment of physical activity (PA) and sedentary behavior (SB) is important to measurement development, epidemiologic studies, and interventions. This study examined agreement between self-reported (International Physical Activity Questionnaire-Long Form [IPAQ-LF]) and accelerometry-based estimates of PA and SB across six countries and identified correlates of between-method agreement. METHODS: Self-report and objective (accelerometry-based) PA and SB data were collected in 2002-2011 from 3865 adult participants in eight cities from six countries (Belgium, Czech Republic, Denmark, Spain, United Kingdom, and United States). Between-method relative agreement (correlation) and absolute disagreement (mean difference between conceptually and intensity-matched IPAQ-LF and accelerometry-based PA and SB variables) were estimated. Also, sociodemographic characteristics and PA patterns were examined as correlates of between-method agreement. RESULTS: Observed relative agreement (relationships of IPAQ-LF with accelerometry-based PA and SB variables) was small to moderate (r = 0.05-0.37) and was moderated by sociodemographic (age, sex, weight status, and education) and behavioral (PA-type) factors. The absolute disagreement was large, with participants self-reporting higher PA intensity and total time in moderate-to-vigorous-intensity PA than accelerometry. Also, self-reported sitting time was lower than accelerometry-based sedentary behavior. After adjusting for sociodemographic and behavioral factors, the absolute disagreement between pairs of IPAQ-LF and accelerometry-based PA variables remained significantly different across cities/countries. CONCLUSIONS: Present findings suggest systematic cultural and/or linguistic and sociodemographic differences in absolute agreement between the IPAQ-LF and the accelerometry-based PA and SB variables. These results have implications for the interpretation of international PA and SB data and correlate/determinant studies. They call for further efforts to improve such measures.

Exploring the relationship between patterns of physical activity, sedentary behaviour and cardiometabolic health in middle-aged Irish adults

Background and Study Rationale Being physically active is a major contributor to both physical and mental health. More specifically, being physically active lowers risk of coronary heart disease, high blood pressure, stroke, metabolic syndrome (MetS), diabetes, certain cancers and depression, and increases cognitive function and wellbeing. The physiological mechanisms that occur in response to physical activity and the impact of total physical activity and sedentary behaviour on cardiometabolic health have been extensively studied. In contrast, limited data evaluating the specific effects of daily and weekly patterns of physical behaviour on cardiometabolic health exist. Additionally, no other study has examined interrelated patterns and minute-by-minute accumulation of physical behaviour throughout the day across week days in middle-aged adults. Study Aims The overarching aims of this thesis are firstly to describe patterns of behaviour throughout the day and week, and secondly to explore associations between these patterns and cardiometabolic health in a middle-aged population. The specific objectives are to: 1 Compare agreement between the International Physical Activity Questionnaire-Short Form (IPAQ-SF) and GENEActiv accelerometer-derived moderate-to-vigorous (MVPA) activity and secondly to compare their associations with a range of cardiometabolic and inflammatory markers in middle-aged adults. 2 Determine a suitable monitoring frame needed to reliably capture weekly, accelerometer-measured, activity in our population. 3 Identify groups of participants who have similar weekly patterns of physical behaviour, and determine if underlying patterns of cardiometabolic profiles exist among these groups. 4 Explore the variation of physical behaviour throughout the day to identify whether daily patterns of physical behaviour vary by cardiometabolic health. Methods All results in this thesis are based on data from a subsample of the Mitchelstown Cohort; 475 (46.1% males; mean aged 59.7±5.5 years) middle-aged Irish adults. Subjective physical activity levels were assessed using the IPAQ-SF. Participants wore the wrist GENEActiv accelerometer for 7 consecutive days. Data was collected at 100Hz and summarised into a signal magnitude vector using 60s epochs. Each time interval was categorised based on validated cut-offs. Data on cardiometabolic and inflammatory markers was collected according to standard protocol. Cardiometabolic outcomes (obesity, diabetes, hypertension and MetS) were defined according to internationally recognised definitions by World Health Organisation (WHO) and Irish Diabetes Federation (IDF). Results The results of the first chapter suggest that the IPAQ-SF lacks the sensitivity to assess patterning of activity and guideline adherence and assessing the relationship with cardiometabolic and inflammatory markers. Furthermore, GENEActiv accelerometer-derived MVPA appears to be better at detecting relationships with cardiometabolic and inflammatory markers. The second chapter examined variations in day-to-day physical behaviour levels between- and within-subjects. The main findings were that Sunday differed from all other days in the week for sedentary behaviour and light activity and that a large within-subject variation across days of the week for vigorous activity exists. Our data indicate that six days of monitoring, four weekdays plus Saturday and Sunday, are required to reliably estimate weekly habitual activity in all activity intensities. In the next chapter, latent profile analysis of weekly, interrelated patterns of physical behaviour identified four distinct physical behaviour patterns; Sedentary Group (15.9%), Sedentary; Lower Activity Group (28%), Sedentary; Higher Activity Group (44.2%) and a Physically Active Group (11.9%). Overall the Sedentary Group had poorer outcomes, characterised by unfavourable cardiometabolic and inflammatory profiles. The remaining classes were characterised by healthier cardiometabolic profiles with lower sedentary behaviour levels. The final chapter, which aimed to compare daily cumulative patterns of minute-by-minute physical behaviour intensities across those with and without MetS, revealed significant differences in weekday and weekend day MVPA. In particular, those with MetS start accumulating MVPA later in the day and for a shorted day period. Conclusion In conclusion, the results of this thesis add to the evidence base regards an optimal monitoring period for physical behaviour measurement to accurately capture weekly physical behaviour patterns. In addition, the results highlight whether weekly and daily distribution of activity is associated with cardiometabolic health and inflammatory profiles. The key findings of this thesis demonstrate the importance of daily and weekly physical behaviour patterning of activity intensity in the context of cardiometabolic health risk. In addition, these findings highlight the importance of using physical behaviour patterns of free-living adults observed in a population-based study to inform and aid health promotion activity programmes and primary care prevention and treatment strategies and development of future tailored physical activity based interventions.

Workplace pedometer interventions for increasing physical activity

Background
The World Health Organization and the World Economic Forum have recommended further research to strengthen current knowledge of workplace health programmes, particularly on effectiveness and using simple instruments. A pedometer is one such simple instrument that can be incorporated in workplace interventions.

Objectives
To assess the effectiveness of pedometer interventions in the workplace for increasing physical activity and improving subsequent health outcomes.

Search methods
Electronic searches of the Cochrane Central Register of Controlled Trials (671 potential papers), MEDLINE (1001), Embase (965), CINAHL (1262), OSH UPDATE databases (75) and Web of Science (1154) from the earliest record to between 30th January and 6th February 2012 yielded 3248 unique records. Reference lists of articles yielded an additional 34 papers. Contact with individuals and organisations did not produce any further records.

Selection criteria
We included individual and cluster-randomised controlled trials of workplace health promotion interventions with a pedometer component in employed adults. The primary outcome was physical activity and was part of the eligibility criteria. We considered subsequent health outcomes, including adverse effects, as secondary outcomes.

Data collection and analysis
Two review authors undertook the screening of titles and abstracts and the full-text papers independently. Two review authors (RFP and MC) independently completed data extraction and risk of bias assessment. We contacted authors to obtain additional data and clarification.

Main results
We found four relevant studies providing data for 1809 employees, 60% of whom were allocated to the intervention group. All studies assessed outcomes immediately after the intervention had finished and the intervention duration varied between three to six months. All studies had usual treatment control conditions; however one study’s usual treatment was an alternative physical activity programme while the other three had minimally active controls. In general, there was high risk of bias mainly due to lack of blinding, self reported outcome measurement, incomplete outcome data due to attrition, and most of the studies had not published protocols, which increases the likelihood of selective reporting.

Three studies compared the pedometer programme to a minimally active control group, but the results for physical activity could not be combined because each study used a different measure of activity. One study observed an increase in physical activity under a pedometer programme, but the other two did not find a significant difference. For secondary outcomes we found improvements in body mass index, waist circumference, fasting plasma glucose, the quality of life mental component and worksite injury associated with the pedometer programmes, but these results were based on limited data from one or two small studies. There were no differences between the pedometer programme and the control group for blood pressure, a number of biochemical outcomes and the quality of life physical component. Sedentary behaviour and disease risk scores were not measured by any of the included studies.

One study compared a pedometer programme and an alternative physical activity programme, but baseline imbalances made it difficult to distinguish the true improvements associated with either programme.

Overall, there was insufficient evidence to assess the effectiveness of pedometer interventions in the workplace.

There is a need for more high quality randomised controlled trials to assess the effectiveness of pedometer interventions in the workplace for increasing physical activity and improving subsequent health outcomes. To improve the quality of the evidence available, future studies should be registered in an online trials register, publish a protocol, allocate time and financial support to reducing attrition, and try to blind personnel (especially those who undertake measurement). To better identify the effects of pedometer interventions, future studies should report a core set of outcomes (total physical activity in METs, total time sitting in hours and minutes, objectively measured cardiovascular disease and type II diabetes risk factors, quality of life and injury), assess outcomes in the long term and undertake subgroup analyses based upon demographic subgroups (e.g. age, gender, educational status). Future studies should also compare different types of active intervention to test specific intervention components (eligibility, duration, step goal, step diary, settings), and settings (occupation, intervention provider).

Authors’ conclusions
There was limited and low quality data providing insufficient evidence to assess the effectiveness of pedometer interventions in the workplace for increasing physical activity and improving subsequent health outcomes.

P L A I N  L A N G U A G E  S U M M A R Y

Do workplace pedometer interventions increase physical activity?
The World Health Organization recommends that most people should undertake at least 30 minutes of moderate-intensity physical activity on most days, as it reduces the risk of cardiovascular disease, diabetes and some cancers. However, less than 40% of the world’s population are undertaking adequate amounts of physical activity and rates have been declining. Here we assess whether pedometer workplace interventions increase physical activity and thereby lead to subsequent health benefits.

To assess this, we searched for randomised controlled trials of workplace health promotion interventions that involved the use of a pedometer undertaken in employed adults. Between 30th January and 6th February 2012 we searched a range of electronic libraries and references of relevant papers, retrieving 3282 potential papers.

We eventually included four studies in the review. One study compared pedometer programmes with an alternative physical activity programme, but there were important baseline differences between the intervention and control groups that made it difficult to distinguish the true effect. The three remaining studies compared pedometer programmes with minimally active control groups. One study observed an improvement in physical activity in the pedometer programme, but two other studies found no significant difference between the pedometer group and the control group. We could not combine these results together, as each study used a different measure for physical activity, so it is not clear what the overall effect is. Single studies found beneficial changes in body mass index, fasting plasma glucose, the mental component of quality of life and worksite injury associated with the pedometer programmes as opposed to the control group. However, none of the studies identified consistent differences between the pedometer programme and the control group for waist circumference, blood pressure and quality of life outcomes. In addition, we judged the majority of included studies to have a high risk of bias, mainly due to participants and staff knowing who was in the intervention and who was in the control group, attrition of participants and not having published a protocol prior to running the study.

We conclude that there was insufficient evidence to assess whether workplace pedometer interventions are of benefit. There is a need for further high quality randomised controlled trials to be undertaken with a range of health outcomes and assessment in the long term.