Sensing human activity to improve sedentary lifestyle

Research output: ThesisPhD Thesis - Research external, graduation UTAcademic

148 Downloads (Pure)

Abstract

Recent public health campaigns often communicate the alarming phrase: “Sitting is the new smoking”. Sitting is related to all-cause mortality, cardiovascular disease, type 2 diabetes, and metabolic syndrome. Sedentary behavior is generally understood as “sitting or reclining while expending ≤1.5 metabolic equivalents” and the interesting aspect of sedentary behavior is that it is a modifiable health risk. The health risk can be reduced if a person changes his or her behavior towards a healthier one; to sit less and to become more physically active.
Research focusing on patterns of sedentary behavior has taken off since the rise of both wearable technologies and activity sensors. They provide opportunities for uncovering sedentary patterns within the context of daily life. As a consequence, the sedentary research field moved forward towards fine-grained, objective monitoring of sedentary behavior in free-living conditions for substantial time frames. Current wearable activity sensors are, however, not flawless in measuring sedentary behavior. It is therefore important to understand the effects of possible measurement bias, in order to deal with it in the best way.
People are often unaware of their sedentary behavior, making it difficult to change the behavior. mHealth interventions can improve awareness and trigger behavior change by tailoring the intervention to the user’s needs by providing direct feedback and coaching on physical activity and sedentary behavior together with real-time information on the context. Context information can be gathered by integrating relevant data sources or by posing questions about the here-and-now. Further increase of acceptance of mHealth interventions can be achieved by tailoring to individual values, and barriers and facilitators to these values.
The aim of this thesis is to determine how wearable activity sensors can be applied successfully in health interventions focused on sedentary behavior.
This thesis follows an expanding scope: starting from the level of the activity sensor up to the level of public health. The first part of this thesis focuses on the measurement of sedentary behavior and its patterns by means of wearable activity sensors (Chapter 2, 3 and 4). The second part of this thesis focuses on the development and evaluation of mHealth interventions that utilize these wearable activity sensors (Chapter 5 and 6).
The pattern of sedentary behavior during the day is an independent health risk. Prolonged sedentary time affects cardio-metabolic and inflammatory biomarkers, independent of the total sedentary time. Since the rise of both wearable technologies and activity sensors, there is however, no consensus among researchers on the best outcome measures for representing the sedentary pattern during the day, based on wearable activity sensors. Chapter 2 provides an overview of current pattern measures of sedentary behavior in adults, by means of a literature review. Simple measures of sedentary behavior were most often reported, like the number of bouts, the medium or median bout length. More complex pattern measures, such as the GINI index or the W50 were reported sparsely. Due to the differences among measurement devices, data analysis protocols and a lack of basic outcome parameters such as total wear-time and total sedentary time, the sedentary patterns, reported in the various studies, were difficult to compare. The simple and complex measures of sedentary time accumulation serve different goals, varying from a quick overview to in-depth analysis and prediction of behavior. The answer to which measures are most suitable to report, is therefore strongly dependent on the research question. From this overview in Chapter 2 we conclude that the reported measures were dependent on 1) the sensing method, 2) the classification method, 3) the experimental and data cleaning protocol, and 4) the applied definitions of bouts and breaks. Based on these findings, we recommend to always report total wear-time, total sedentary time, number of bouts and at least one measure describing the diversity of bout lengths in the sedentary behavior such as the W50. Additionally, we recommend to report the measurement conditions and data processing steps.
One of the factors influencing the output of activity sensors mentioned above is the experimental protocol (Chapter 2). This was studied in more depth in Chapter 3 in which we focused on optimal sensor placement for measuring physical activity. Subjects walked at various speeds on a treadmill, performed a deskwork protocol, and walked on level ground, while simultaneously wearing five activity sensors with a snug fit on an elastic waist belt. We found that sensor location, type of activity, and their interaction-effect affected sensor output. The most lateral positions on the waist belt were the least sensitive for interference. Additionally, the effect of mounting was explored by repeating the experimental protocol with sensors more loosely fitted to the elastic belt. The loose fit resulted in lower sensor output, except for the deskwork protocol, where output was higher. We conclude that, in order to increase the reliability and to reduce the variability of sensor output, researchers should place activity sensors on the most lateral position of a participant‘s waist belt. If the sensor hampers free movement, it may be positioned slightly more forward on the belt. Finally, we recommend to wear sensors tightly fitted to the body.
Another factor influencing the output of activity sensors mentioned above (Chapter 2) is the classification method. Currently, the most applied method to distinguish sedentary from active time is by applying a cut-point to accelerometry-based data. This means that the intensity of the measured behavior is classified as being sedentary when below this cut-point. The effect of the classification method on sedentary pattern measures was studied in detail in laboratory and free-living conditions with office workers (Chapter 4). In this study we found that the outcome measures are robust – meaning that the outcome measures do not change –, when cut-points for classifying sedentary behavior are within the boundaries of ±10-20% of the optimal cut-point. This conclusion implies that results from studies analyzing sedentary patterns based on different cut-points, can only be compared if the cut-points are within these boundaries.
In Chapter 5, we combined the knowledge on sensor use, data processing steps and outcome measures with context-aware technology in an intervention for older office workers towards sitting less and breaking up sitting time. Office workers spend a high percentage of their time sitting, often in long periods of time. Research suggests that it is healthier to break these long bouts into shorter periods by being physically active. In order to promote breaking up long sedentary bouts, we developed an innovative, context-aware activity coach for older office workers. This coach provides activity suggestions, based on a physical activity prediction model, consisting of past and current physical activity (measured by a wearable activity sensor) and digital agendas. The total sedentary time in the intervention week, was not reduced compared to the baseline week. However, the pattern of the sedentary behaviour did change – the office workers reduced their total time spent in long sitting bouts (≥45 minutes). Additionally, the office workers indicated that the main added value of the intervention resided in creating awareness about their personal sedentary behaviour pattern. Finally, the participants were compliant to 53% of the suggestions; a number that could be increased by improving the timing of suggestions. We conclude that the mobile intervention (using an activity sensor, smartphone application and context information) has the potential to improve the sedentary behaviour of older office workers. The gain can especially be found in breaking up long sedentary periods by being physically active. Older office workers value that it makes them aware of their sedentary behaviour. We also found that focusing on total sedentary time as an outcome of a public health intervention, aimed at reducing sedentary behaviour, is too simplistic. Rather, one should take into account both the duration and the number of bouts when determining the effect of the intervention. We conclude this article by summarizing our design recommendations for eHealth interventions that aim to improve sedentary behaviour.
In Chapter 6 we focused on a design approach to further increase acceptance of mHealth interventions – by tailoring to individual values, and barriers and facilitators to these values. In this study, we demonstrated how value-based design can contribute to the design of a product or service that addresses real needs and thus, lead to high acceptance. We described the methods and application of value-based design. We elicited values, facilitators and barriers of their reduced mobility – meaning difficulty with walking, biking, and/or activities of daily living – of older adults via in-depth interviews. These interviews resulted in a myriad of key values, such as ‘independence from family’ and ‘doing their own groceries’. Co-creation design sessions resulted in innovative mobility aids from which three designs for a wheeled walker were selected for evaluation on acceptance, again via in-depth interviews. Their acceptance was rather low. Current mobility device users were more eager to accept the designs than non-users. The value-based approach offered designers a close look into the lives of the elderly, thereby opening up a wide range of innovation possibilities that better fit the actual needs. However, mobility is related to physical capacity and not being sedentary. In-depth understanding of the values of life to be mobile, can therefore directly inspire designers focused on mobility aids. Nevertheless, this understanding can as well tap into the context and personal goals needed to tailor health interventions on sedentary behavior.
In Chapter 7, the general discussion, we discuss the rapid development of sensors and analysis methods, as well as gathering rich context information by means Experience Sampling. Cut-point-based analyses make only very limited use of the wealth of data that can be measured by activity sensors and has various challenges which hinders generalization of the current body of knowledge (Chapter 2, 3 and 4). It seems to make more sense to express the measured behavior in terms of the actual behavior, such as bicycling and climbing stairs, rather than expressing physical activity in counts or metric units representing its intensity. Machine learning techniques are very capable of this with higher accuracy, and their application seems to be a logical step forwards. And when expressed as behavior, machine learning output will be easier to adopt in interventions, as total sitting time and active breaks are better understandable than for example counts – matching the real-world knowledge of users. The Experience Sampling Method (ESM) incorporated in the developed intervention (Chapter 5) provided in-depth understanding of the context of sedentary behavior – the where, when, why, with whom and experienced emotions. Future interventions should incorporate this type of context-awareness in real-time tailored coaching strategies to increase awareness and behavior change.
The studies in this thesis have shown that activity sensors can provide valuable information on the pattern of sedentary behavior, and that these can be useful for health interventions. We have shown the strengths of current practice and opportunities for improvement, by applying a broad scope with a focus on measuring sedentary behavior and developing and evaluating mHealth interventions. Based on the study we conclude that:
The combination of the bottom-up approach (from sensor, to data to information levels) and the top-down approach (from user values related to public health to interventions and its specific feedback and coaching strategies), contribute to diverse, strongly connected and interdependent domains of applying wearable activity sensors in health interventions focused on sedentary behavior.
Original languageEnglish
Awarding Institution
  • University of Twente
Supervisors/Advisors
  • Hermens, Hermanus J., Supervisor
  • van Velsen, Lex Stefan, Co-Supervisor
Thesis sponsors
Award date21 Sep 2018
Place of PublicationEnschede, The Netherlands
Print ISBNs978-90-365-4604-1
DOIs
Publication statusPublished - 21 Sep 2018

Fingerprint

Sedentary Lifestyle
Human Activities
Telemedicine
Health
Public Health
Outcome Assessment (Health Care)
Social Conditions
Interviews

Keywords

  • Sedentary behavior
  • Sitting behavior
  • Activity coaching systems
  • Sedentary behaviour
  • Activity sensor
  • Ecological momentary assessment
  • Experience sampling method
  • Classification
  • Design

Cite this

@phdthesis{447a18f22c6d464e9f32e16944a742b7,
title = "Sensing human activity to improve sedentary lifestyle",
abstract = "Recent public health campaigns often communicate the alarming phrase: “Sitting is the new smoking”. Sitting is related to all-cause mortality, cardiovascular disease, type 2 diabetes, and metabolic syndrome. Sedentary behavior is generally understood as “sitting or reclining while expending ≤1.5 metabolic equivalents” and the interesting aspect of sedentary behavior is that it is a modifiable health risk. The health risk can be reduced if a person changes his or her behavior towards a healthier one; to sit less and to become more physically active.Research focusing on patterns of sedentary behavior has taken off since the rise of both wearable technologies and activity sensors. They provide opportunities for uncovering sedentary patterns within the context of daily life. As a consequence, the sedentary research field moved forward towards fine-grained, objective monitoring of sedentary behavior in free-living conditions for substantial time frames. Current wearable activity sensors are, however, not flawless in measuring sedentary behavior. It is therefore important to understand the effects of possible measurement bias, in order to deal with it in the best way. People are often unaware of their sedentary behavior, making it difficult to change the behavior. mHealth interventions can improve awareness and trigger behavior change by tailoring the intervention to the user’s needs by providing direct feedback and coaching on physical activity and sedentary behavior together with real-time information on the context. Context information can be gathered by integrating relevant data sources or by posing questions about the here-and-now. Further increase of acceptance of mHealth interventions can be achieved by tailoring to individual values, and barriers and facilitators to these values. The aim of this thesis is to determine how wearable activity sensors can be applied successfully in health interventions focused on sedentary behavior. This thesis follows an expanding scope: starting from the level of the activity sensor up to the level of public health. The first part of this thesis focuses on the measurement of sedentary behavior and its patterns by means of wearable activity sensors (Chapter 2, 3 and 4). The second part of this thesis focuses on the development and evaluation of mHealth interventions that utilize these wearable activity sensors (Chapter 5 and 6). The pattern of sedentary behavior during the day is an independent health risk. Prolonged sedentary time affects cardio-metabolic and inflammatory biomarkers, independent of the total sedentary time. Since the rise of both wearable technologies and activity sensors, there is however, no consensus among researchers on the best outcome measures for representing the sedentary pattern during the day, based on wearable activity sensors. Chapter 2 provides an overview of current pattern measures of sedentary behavior in adults, by means of a literature review. Simple measures of sedentary behavior were most often reported, like the number of bouts, the medium or median bout length. More complex pattern measures, such as the GINI index or the W50 were reported sparsely. Due to the differences among measurement devices, data analysis protocols and a lack of basic outcome parameters such as total wear-time and total sedentary time, the sedentary patterns, reported in the various studies, were difficult to compare. The simple and complex measures of sedentary time accumulation serve different goals, varying from a quick overview to in-depth analysis and prediction of behavior. The answer to which measures are most suitable to report, is therefore strongly dependent on the research question. From this overview in Chapter 2 we conclude that the reported measures were dependent on 1) the sensing method, 2) the classification method, 3) the experimental and data cleaning protocol, and 4) the applied definitions of bouts and breaks. Based on these findings, we recommend to always report total wear-time, total sedentary time, number of bouts and at least one measure describing the diversity of bout lengths in the sedentary behavior such as the W50. Additionally, we recommend to report the measurement conditions and data processing steps.One of the factors influencing the output of activity sensors mentioned above is the experimental protocol (Chapter 2). This was studied in more depth in Chapter 3 in which we focused on optimal sensor placement for measuring physical activity. Subjects walked at various speeds on a treadmill, performed a deskwork protocol, and walked on level ground, while simultaneously wearing five activity sensors with a snug fit on an elastic waist belt. We found that sensor location, type of activity, and their interaction-effect affected sensor output. The most lateral positions on the waist belt were the least sensitive for interference. Additionally, the effect of mounting was explored by repeating the experimental protocol with sensors more loosely fitted to the elastic belt. The loose fit resulted in lower sensor output, except for the deskwork protocol, where output was higher. We conclude that, in order to increase the reliability and to reduce the variability of sensor output, researchers should place activity sensors on the most lateral position of a participant‘s waist belt. If the sensor hampers free movement, it may be positioned slightly more forward on the belt. Finally, we recommend to wear sensors tightly fitted to the body.Another factor influencing the output of activity sensors mentioned above (Chapter 2) is the classification method. Currently, the most applied method to distinguish sedentary from active time is by applying a cut-point to accelerometry-based data. This means that the intensity of the measured behavior is classified as being sedentary when below this cut-point. The effect of the classification method on sedentary pattern measures was studied in detail in laboratory and free-living conditions with office workers (Chapter 4). In this study we found that the outcome measures are robust – meaning that the outcome measures do not change –, when cut-points for classifying sedentary behavior are within the boundaries of ±10-20{\%} of the optimal cut-point. This conclusion implies that results from studies analyzing sedentary patterns based on different cut-points, can only be compared if the cut-points are within these boundaries. In Chapter 5, we combined the knowledge on sensor use, data processing steps and outcome measures with context-aware technology in an intervention for older office workers towards sitting less and breaking up sitting time. Office workers spend a high percentage of their time sitting, often in long periods of time. Research suggests that it is healthier to break these long bouts into shorter periods by being physically active. In order to promote breaking up long sedentary bouts, we developed an innovative, context-aware activity coach for older office workers. This coach provides activity suggestions, based on a physical activity prediction model, consisting of past and current physical activity (measured by a wearable activity sensor) and digital agendas. The total sedentary time in the intervention week, was not reduced compared to the baseline week. However, the pattern of the sedentary behaviour did change – the office workers reduced their total time spent in long sitting bouts (≥45 minutes). Additionally, the office workers indicated that the main added value of the intervention resided in creating awareness about their personal sedentary behaviour pattern. Finally, the participants were compliant to 53{\%} of the suggestions; a number that could be increased by improving the timing of suggestions. We conclude that the mobile intervention (using an activity sensor, smartphone application and context information) has the potential to improve the sedentary behaviour of older office workers. The gain can especially be found in breaking up long sedentary periods by being physically active. Older office workers value that it makes them aware of their sedentary behaviour. We also found that focusing on total sedentary time as an outcome of a public health intervention, aimed at reducing sedentary behaviour, is too simplistic. Rather, one should take into account both the duration and the number of bouts when determining the effect of the intervention. We conclude this article by summarizing our design recommendations for eHealth interventions that aim to improve sedentary behaviour.In Chapter 6 we focused on a design approach to further increase acceptance of mHealth interventions – by tailoring to individual values, and barriers and facilitators to these values. In this study, we demonstrated how value-based design can contribute to the design of a product or service that addresses real needs and thus, lead to high acceptance. We described the methods and application of value-based design. We elicited values, facilitators and barriers of their reduced mobility – meaning difficulty with walking, biking, and/or activities of daily living – of older adults via in-depth interviews. These interviews resulted in a myriad of key values, such as ‘independence from family’ and ‘doing their own groceries’. Co-creation design sessions resulted in innovative mobility aids from which three designs for a wheeled walker were selected for evaluation on acceptance, again via in-depth interviews. Their acceptance was rather low. Current mobility device users were more eager to accept the designs than non-users. The value-based approach offered designers a close look into the lives of the elderly, thereby opening up a wide range of innovation possibilities that better fit the actual needs. However, mobility is related to physical capacity and not being sedentary. In-depth understanding of the values of life to be mobile, can therefore directly inspire designers focused on mobility aids. Nevertheless, this understanding can as well tap into the context and personal goals needed to tailor health interventions on sedentary behavior. In Chapter 7, the general discussion, we discuss the rapid development of sensors and analysis methods, as well as gathering rich context information by means Experience Sampling. Cut-point-based analyses make only very limited use of the wealth of data that can be measured by activity sensors and has various challenges which hinders generalization of the current body of knowledge (Chapter 2, 3 and 4). It seems to make more sense to express the measured behavior in terms of the actual behavior, such as bicycling and climbing stairs, rather than expressing physical activity in counts or metric units representing its intensity. Machine learning techniques are very capable of this with higher accuracy, and their application seems to be a logical step forwards. And when expressed as behavior, machine learning output will be easier to adopt in interventions, as total sitting time and active breaks are better understandable than for example counts – matching the real-world knowledge of users. The Experience Sampling Method (ESM) incorporated in the developed intervention (Chapter 5) provided in-depth understanding of the context of sedentary behavior – the where, when, why, with whom and experienced emotions. Future interventions should incorporate this type of context-awareness in real-time tailored coaching strategies to increase awareness and behavior change.The studies in this thesis have shown that activity sensors can provide valuable information on the pattern of sedentary behavior, and that these can be useful for health interventions. We have shown the strengths of current practice and opportunities for improvement, by applying a broad scope with a focus on measuring sedentary behavior and developing and evaluating mHealth interventions. Based on the study we conclude that: The combination of the bottom-up approach (from sensor, to data to information levels) and the top-down approach (from user values related to public health to interventions and its specific feedback and coaching strategies), contribute to diverse, strongly connected and interdependent domains of applying wearable activity sensors in health interventions focused on sedentary behavior.",
keywords = "Sedentary behavior, Sitting behavior, Activity coaching systems, Sedentary behaviour, Activity sensor, Ecological momentary assessment, Experience sampling method, Classification, Design",
author = "Boerema, {Simone Theresa}",
year = "2018",
month = "9",
day = "21",
doi = "10.3990/1.9789036546041",
language = "English",
isbn = "978-90-365-4604-1",
series = "DSI Ph.D. Thesis Series",
publisher = "University of Twente",
number = "18-011",
school = "University of Twente",

}

Sensing human activity to improve sedentary lifestyle. / Boerema, Simone Theresa.

Enschede, The Netherlands, 2018. 204 p.

Research output: ThesisPhD Thesis - Research external, graduation UTAcademic

TY - THES

T1 - Sensing human activity to improve sedentary lifestyle

AU - Boerema, Simone Theresa

PY - 2018/9/21

Y1 - 2018/9/21

N2 - Recent public health campaigns often communicate the alarming phrase: “Sitting is the new smoking”. Sitting is related to all-cause mortality, cardiovascular disease, type 2 diabetes, and metabolic syndrome. Sedentary behavior is generally understood as “sitting or reclining while expending ≤1.5 metabolic equivalents” and the interesting aspect of sedentary behavior is that it is a modifiable health risk. The health risk can be reduced if a person changes his or her behavior towards a healthier one; to sit less and to become more physically active.Research focusing on patterns of sedentary behavior has taken off since the rise of both wearable technologies and activity sensors. They provide opportunities for uncovering sedentary patterns within the context of daily life. As a consequence, the sedentary research field moved forward towards fine-grained, objective monitoring of sedentary behavior in free-living conditions for substantial time frames. Current wearable activity sensors are, however, not flawless in measuring sedentary behavior. It is therefore important to understand the effects of possible measurement bias, in order to deal with it in the best way. People are often unaware of their sedentary behavior, making it difficult to change the behavior. mHealth interventions can improve awareness and trigger behavior change by tailoring the intervention to the user’s needs by providing direct feedback and coaching on physical activity and sedentary behavior together with real-time information on the context. Context information can be gathered by integrating relevant data sources or by posing questions about the here-and-now. Further increase of acceptance of mHealth interventions can be achieved by tailoring to individual values, and barriers and facilitators to these values. The aim of this thesis is to determine how wearable activity sensors can be applied successfully in health interventions focused on sedentary behavior. This thesis follows an expanding scope: starting from the level of the activity sensor up to the level of public health. The first part of this thesis focuses on the measurement of sedentary behavior and its patterns by means of wearable activity sensors (Chapter 2, 3 and 4). The second part of this thesis focuses on the development and evaluation of mHealth interventions that utilize these wearable activity sensors (Chapter 5 and 6). The pattern of sedentary behavior during the day is an independent health risk. Prolonged sedentary time affects cardio-metabolic and inflammatory biomarkers, independent of the total sedentary time. Since the rise of both wearable technologies and activity sensors, there is however, no consensus among researchers on the best outcome measures for representing the sedentary pattern during the day, based on wearable activity sensors. Chapter 2 provides an overview of current pattern measures of sedentary behavior in adults, by means of a literature review. Simple measures of sedentary behavior were most often reported, like the number of bouts, the medium or median bout length. More complex pattern measures, such as the GINI index or the W50 were reported sparsely. Due to the differences among measurement devices, data analysis protocols and a lack of basic outcome parameters such as total wear-time and total sedentary time, the sedentary patterns, reported in the various studies, were difficult to compare. The simple and complex measures of sedentary time accumulation serve different goals, varying from a quick overview to in-depth analysis and prediction of behavior. The answer to which measures are most suitable to report, is therefore strongly dependent on the research question. From this overview in Chapter 2 we conclude that the reported measures were dependent on 1) the sensing method, 2) the classification method, 3) the experimental and data cleaning protocol, and 4) the applied definitions of bouts and breaks. Based on these findings, we recommend to always report total wear-time, total sedentary time, number of bouts and at least one measure describing the diversity of bout lengths in the sedentary behavior such as the W50. Additionally, we recommend to report the measurement conditions and data processing steps.One of the factors influencing the output of activity sensors mentioned above is the experimental protocol (Chapter 2). This was studied in more depth in Chapter 3 in which we focused on optimal sensor placement for measuring physical activity. Subjects walked at various speeds on a treadmill, performed a deskwork protocol, and walked on level ground, while simultaneously wearing five activity sensors with a snug fit on an elastic waist belt. We found that sensor location, type of activity, and their interaction-effect affected sensor output. The most lateral positions on the waist belt were the least sensitive for interference. Additionally, the effect of mounting was explored by repeating the experimental protocol with sensors more loosely fitted to the elastic belt. The loose fit resulted in lower sensor output, except for the deskwork protocol, where output was higher. We conclude that, in order to increase the reliability and to reduce the variability of sensor output, researchers should place activity sensors on the most lateral position of a participant‘s waist belt. If the sensor hampers free movement, it may be positioned slightly more forward on the belt. Finally, we recommend to wear sensors tightly fitted to the body.Another factor influencing the output of activity sensors mentioned above (Chapter 2) is the classification method. Currently, the most applied method to distinguish sedentary from active time is by applying a cut-point to accelerometry-based data. This means that the intensity of the measured behavior is classified as being sedentary when below this cut-point. The effect of the classification method on sedentary pattern measures was studied in detail in laboratory and free-living conditions with office workers (Chapter 4). In this study we found that the outcome measures are robust – meaning that the outcome measures do not change –, when cut-points for classifying sedentary behavior are within the boundaries of ±10-20% of the optimal cut-point. This conclusion implies that results from studies analyzing sedentary patterns based on different cut-points, can only be compared if the cut-points are within these boundaries. In Chapter 5, we combined the knowledge on sensor use, data processing steps and outcome measures with context-aware technology in an intervention for older office workers towards sitting less and breaking up sitting time. Office workers spend a high percentage of their time sitting, often in long periods of time. Research suggests that it is healthier to break these long bouts into shorter periods by being physically active. In order to promote breaking up long sedentary bouts, we developed an innovative, context-aware activity coach for older office workers. This coach provides activity suggestions, based on a physical activity prediction model, consisting of past and current physical activity (measured by a wearable activity sensor) and digital agendas. The total sedentary time in the intervention week, was not reduced compared to the baseline week. However, the pattern of the sedentary behaviour did change – the office workers reduced their total time spent in long sitting bouts (≥45 minutes). Additionally, the office workers indicated that the main added value of the intervention resided in creating awareness about their personal sedentary behaviour pattern. Finally, the participants were compliant to 53% of the suggestions; a number that could be increased by improving the timing of suggestions. We conclude that the mobile intervention (using an activity sensor, smartphone application and context information) has the potential to improve the sedentary behaviour of older office workers. The gain can especially be found in breaking up long sedentary periods by being physically active. Older office workers value that it makes them aware of their sedentary behaviour. We also found that focusing on total sedentary time as an outcome of a public health intervention, aimed at reducing sedentary behaviour, is too simplistic. Rather, one should take into account both the duration and the number of bouts when determining the effect of the intervention. We conclude this article by summarizing our design recommendations for eHealth interventions that aim to improve sedentary behaviour.In Chapter 6 we focused on a design approach to further increase acceptance of mHealth interventions – by tailoring to individual values, and barriers and facilitators to these values. In this study, we demonstrated how value-based design can contribute to the design of a product or service that addresses real needs and thus, lead to high acceptance. We described the methods and application of value-based design. We elicited values, facilitators and barriers of their reduced mobility – meaning difficulty with walking, biking, and/or activities of daily living – of older adults via in-depth interviews. These interviews resulted in a myriad of key values, such as ‘independence from family’ and ‘doing their own groceries’. Co-creation design sessions resulted in innovative mobility aids from which three designs for a wheeled walker were selected for evaluation on acceptance, again via in-depth interviews. Their acceptance was rather low. Current mobility device users were more eager to accept the designs than non-users. The value-based approach offered designers a close look into the lives of the elderly, thereby opening up a wide range of innovation possibilities that better fit the actual needs. However, mobility is related to physical capacity and not being sedentary. In-depth understanding of the values of life to be mobile, can therefore directly inspire designers focused on mobility aids. Nevertheless, this understanding can as well tap into the context and personal goals needed to tailor health interventions on sedentary behavior. In Chapter 7, the general discussion, we discuss the rapid development of sensors and analysis methods, as well as gathering rich context information by means Experience Sampling. Cut-point-based analyses make only very limited use of the wealth of data that can be measured by activity sensors and has various challenges which hinders generalization of the current body of knowledge (Chapter 2, 3 and 4). It seems to make more sense to express the measured behavior in terms of the actual behavior, such as bicycling and climbing stairs, rather than expressing physical activity in counts or metric units representing its intensity. Machine learning techniques are very capable of this with higher accuracy, and their application seems to be a logical step forwards. And when expressed as behavior, machine learning output will be easier to adopt in interventions, as total sitting time and active breaks are better understandable than for example counts – matching the real-world knowledge of users. The Experience Sampling Method (ESM) incorporated in the developed intervention (Chapter 5) provided in-depth understanding of the context of sedentary behavior – the where, when, why, with whom and experienced emotions. Future interventions should incorporate this type of context-awareness in real-time tailored coaching strategies to increase awareness and behavior change.The studies in this thesis have shown that activity sensors can provide valuable information on the pattern of sedentary behavior, and that these can be useful for health interventions. We have shown the strengths of current practice and opportunities for improvement, by applying a broad scope with a focus on measuring sedentary behavior and developing and evaluating mHealth interventions. Based on the study we conclude that: The combination of the bottom-up approach (from sensor, to data to information levels) and the top-down approach (from user values related to public health to interventions and its specific feedback and coaching strategies), contribute to diverse, strongly connected and interdependent domains of applying wearable activity sensors in health interventions focused on sedentary behavior.

AB - Recent public health campaigns often communicate the alarming phrase: “Sitting is the new smoking”. Sitting is related to all-cause mortality, cardiovascular disease, type 2 diabetes, and metabolic syndrome. Sedentary behavior is generally understood as “sitting or reclining while expending ≤1.5 metabolic equivalents” and the interesting aspect of sedentary behavior is that it is a modifiable health risk. The health risk can be reduced if a person changes his or her behavior towards a healthier one; to sit less and to become more physically active.Research focusing on patterns of sedentary behavior has taken off since the rise of both wearable technologies and activity sensors. They provide opportunities for uncovering sedentary patterns within the context of daily life. As a consequence, the sedentary research field moved forward towards fine-grained, objective monitoring of sedentary behavior in free-living conditions for substantial time frames. Current wearable activity sensors are, however, not flawless in measuring sedentary behavior. It is therefore important to understand the effects of possible measurement bias, in order to deal with it in the best way. People are often unaware of their sedentary behavior, making it difficult to change the behavior. mHealth interventions can improve awareness and trigger behavior change by tailoring the intervention to the user’s needs by providing direct feedback and coaching on physical activity and sedentary behavior together with real-time information on the context. Context information can be gathered by integrating relevant data sources or by posing questions about the here-and-now. Further increase of acceptance of mHealth interventions can be achieved by tailoring to individual values, and barriers and facilitators to these values. The aim of this thesis is to determine how wearable activity sensors can be applied successfully in health interventions focused on sedentary behavior. This thesis follows an expanding scope: starting from the level of the activity sensor up to the level of public health. The first part of this thesis focuses on the measurement of sedentary behavior and its patterns by means of wearable activity sensors (Chapter 2, 3 and 4). The second part of this thesis focuses on the development and evaluation of mHealth interventions that utilize these wearable activity sensors (Chapter 5 and 6). The pattern of sedentary behavior during the day is an independent health risk. Prolonged sedentary time affects cardio-metabolic and inflammatory biomarkers, independent of the total sedentary time. Since the rise of both wearable technologies and activity sensors, there is however, no consensus among researchers on the best outcome measures for representing the sedentary pattern during the day, based on wearable activity sensors. Chapter 2 provides an overview of current pattern measures of sedentary behavior in adults, by means of a literature review. Simple measures of sedentary behavior were most often reported, like the number of bouts, the medium or median bout length. More complex pattern measures, such as the GINI index or the W50 were reported sparsely. Due to the differences among measurement devices, data analysis protocols and a lack of basic outcome parameters such as total wear-time and total sedentary time, the sedentary patterns, reported in the various studies, were difficult to compare. The simple and complex measures of sedentary time accumulation serve different goals, varying from a quick overview to in-depth analysis and prediction of behavior. The answer to which measures are most suitable to report, is therefore strongly dependent on the research question. From this overview in Chapter 2 we conclude that the reported measures were dependent on 1) the sensing method, 2) the classification method, 3) the experimental and data cleaning protocol, and 4) the applied definitions of bouts and breaks. Based on these findings, we recommend to always report total wear-time, total sedentary time, number of bouts and at least one measure describing the diversity of bout lengths in the sedentary behavior such as the W50. Additionally, we recommend to report the measurement conditions and data processing steps.One of the factors influencing the output of activity sensors mentioned above is the experimental protocol (Chapter 2). This was studied in more depth in Chapter 3 in which we focused on optimal sensor placement for measuring physical activity. Subjects walked at various speeds on a treadmill, performed a deskwork protocol, and walked on level ground, while simultaneously wearing five activity sensors with a snug fit on an elastic waist belt. We found that sensor location, type of activity, and their interaction-effect affected sensor output. The most lateral positions on the waist belt were the least sensitive for interference. Additionally, the effect of mounting was explored by repeating the experimental protocol with sensors more loosely fitted to the elastic belt. The loose fit resulted in lower sensor output, except for the deskwork protocol, where output was higher. We conclude that, in order to increase the reliability and to reduce the variability of sensor output, researchers should place activity sensors on the most lateral position of a participant‘s waist belt. If the sensor hampers free movement, it may be positioned slightly more forward on the belt. Finally, we recommend to wear sensors tightly fitted to the body.Another factor influencing the output of activity sensors mentioned above (Chapter 2) is the classification method. Currently, the most applied method to distinguish sedentary from active time is by applying a cut-point to accelerometry-based data. This means that the intensity of the measured behavior is classified as being sedentary when below this cut-point. The effect of the classification method on sedentary pattern measures was studied in detail in laboratory and free-living conditions with office workers (Chapter 4). In this study we found that the outcome measures are robust – meaning that the outcome measures do not change –, when cut-points for classifying sedentary behavior are within the boundaries of ±10-20% of the optimal cut-point. This conclusion implies that results from studies analyzing sedentary patterns based on different cut-points, can only be compared if the cut-points are within these boundaries. In Chapter 5, we combined the knowledge on sensor use, data processing steps and outcome measures with context-aware technology in an intervention for older office workers towards sitting less and breaking up sitting time. Office workers spend a high percentage of their time sitting, often in long periods of time. Research suggests that it is healthier to break these long bouts into shorter periods by being physically active. In order to promote breaking up long sedentary bouts, we developed an innovative, context-aware activity coach for older office workers. This coach provides activity suggestions, based on a physical activity prediction model, consisting of past and current physical activity (measured by a wearable activity sensor) and digital agendas. The total sedentary time in the intervention week, was not reduced compared to the baseline week. However, the pattern of the sedentary behaviour did change – the office workers reduced their total time spent in long sitting bouts (≥45 minutes). Additionally, the office workers indicated that the main added value of the intervention resided in creating awareness about their personal sedentary behaviour pattern. Finally, the participants were compliant to 53% of the suggestions; a number that could be increased by improving the timing of suggestions. We conclude that the mobile intervention (using an activity sensor, smartphone application and context information) has the potential to improve the sedentary behaviour of older office workers. The gain can especially be found in breaking up long sedentary periods by being physically active. Older office workers value that it makes them aware of their sedentary behaviour. We also found that focusing on total sedentary time as an outcome of a public health intervention, aimed at reducing sedentary behaviour, is too simplistic. Rather, one should take into account both the duration and the number of bouts when determining the effect of the intervention. We conclude this article by summarizing our design recommendations for eHealth interventions that aim to improve sedentary behaviour.In Chapter 6 we focused on a design approach to further increase acceptance of mHealth interventions – by tailoring to individual values, and barriers and facilitators to these values. In this study, we demonstrated how value-based design can contribute to the design of a product or service that addresses real needs and thus, lead to high acceptance. We described the methods and application of value-based design. We elicited values, facilitators and barriers of their reduced mobility – meaning difficulty with walking, biking, and/or activities of daily living – of older adults via in-depth interviews. These interviews resulted in a myriad of key values, such as ‘independence from family’ and ‘doing their own groceries’. Co-creation design sessions resulted in innovative mobility aids from which three designs for a wheeled walker were selected for evaluation on acceptance, again via in-depth interviews. Their acceptance was rather low. Current mobility device users were more eager to accept the designs than non-users. The value-based approach offered designers a close look into the lives of the elderly, thereby opening up a wide range of innovation possibilities that better fit the actual needs. However, mobility is related to physical capacity and not being sedentary. In-depth understanding of the values of life to be mobile, can therefore directly inspire designers focused on mobility aids. Nevertheless, this understanding can as well tap into the context and personal goals needed to tailor health interventions on sedentary behavior. In Chapter 7, the general discussion, we discuss the rapid development of sensors and analysis methods, as well as gathering rich context information by means Experience Sampling. Cut-point-based analyses make only very limited use of the wealth of data that can be measured by activity sensors and has various challenges which hinders generalization of the current body of knowledge (Chapter 2, 3 and 4). It seems to make more sense to express the measured behavior in terms of the actual behavior, such as bicycling and climbing stairs, rather than expressing physical activity in counts or metric units representing its intensity. Machine learning techniques are very capable of this with higher accuracy, and their application seems to be a logical step forwards. And when expressed as behavior, machine learning output will be easier to adopt in interventions, as total sitting time and active breaks are better understandable than for example counts – matching the real-world knowledge of users. The Experience Sampling Method (ESM) incorporated in the developed intervention (Chapter 5) provided in-depth understanding of the context of sedentary behavior – the where, when, why, with whom and experienced emotions. Future interventions should incorporate this type of context-awareness in real-time tailored coaching strategies to increase awareness and behavior change.The studies in this thesis have shown that activity sensors can provide valuable information on the pattern of sedentary behavior, and that these can be useful for health interventions. We have shown the strengths of current practice and opportunities for improvement, by applying a broad scope with a focus on measuring sedentary behavior and developing and evaluating mHealth interventions. Based on the study we conclude that: The combination of the bottom-up approach (from sensor, to data to information levels) and the top-down approach (from user values related to public health to interventions and its specific feedback and coaching strategies), contribute to diverse, strongly connected and interdependent domains of applying wearable activity sensors in health interventions focused on sedentary behavior.

KW - Sedentary behavior

KW - Sitting behavior

KW - Activity coaching systems

KW - Sedentary behaviour

KW - Activity sensor

KW - Ecological momentary assessment

KW - Experience sampling method

KW - Classification

KW - Design

U2 - 10.3990/1.9789036546041

DO - 10.3990/1.9789036546041

M3 - PhD Thesis - Research external, graduation UT

SN - 978-90-365-4604-1

T3 - DSI Ph.D. Thesis Series

CY - Enschede, The Netherlands

ER -

Boerema ST. Sensing human activity to improve sedentary lifestyle. Enschede, The Netherlands, 2018. 204 p. (DSI Ph.D. Thesis Series; 18-011). https://doi.org/10.3990/1.9789036546041