• What is assessed?
• How is the measurement conducted?
• When is this method used?
• How are estimates of physical activity derived?
• Strengths and limitations
• Populations
• Further considerations
• Resources required
• Instrument library
• References

There is strong evidence indicating that measuring multiple biological signals using different types of sensors can considerably improve the accuracy of estimates of physical activity parameters as opposed to measuring only a single signal (Strath et al., 2005). Here, we define multi-sensor monitoring as methods relying on three or more types of sensors, e.g. skin temperature, near-body ambient temperature, heat flux, galvanic skin response, accelerometer, gyroscope, magnetometer, pressure sensor, respiration, etc. Activity parameters that multi-sensor monitoring may provide include energy expenditure, intensity, frequency, sleep time, step counts, distance and speed. Some features and activity parameters that can be assessed using multi-sensor monitors are summarised in Table P.3.29.

Several studies have demonstrated the high validity of multi-sensor monitoring in relation to doubly-labelled water techniques in diverse populations, including children (Calabro et al., 2013), adults (Johannsen et al., 2010) and older adults (Calabro et al., 2015). Moreover, given that multi-sensor monitoring typically detect multiple physiological responses to activities performed, they can assess energy expenditure of some specific activities (such as weight lifting, cycling) which may not be fully captured by measurements of only one signal (e.g. accelerometer worn at hip). Another advantage is that multi-sensor monitoring can better discriminate wear time from non-wear time as this inference is also informed by the multi-channel information available.

Table P.3.29 Physical activity dimensions which can be assessed by multi-sensor monitors.

Multi–sensor arrays

Multi-sensor systems measure three or more phenomena in order to estimate physical activity. The number and array of sensors varies by system, and can include any combination of the following:

1. Electrocardiograph (ECG): Electrodes usually positioned on the chest or embedded within textile connecting to a central data-logger. ECG data can be analysed in order to derive heart rate (HR) and heart rate variability (HRV) parameters. Some systems store the raw ECG signal but others may only store summary data from the signal (ie. QRS detection, inter-beat-intervals). Respiration rate may be inferred from R-wave magnitude and/or variability in inter-beat intervals.
2. Photoplethysmograph (PPG): An optical sensor usually placed at the index finger/ear lobe/wrist can detect the fluctuation in blood perfusion of the skin during the cardiac cycle. This signal can be used to infer heart rate (via peak detection, inter-beat intervals). Respiration rate may be inferred from variability in inter-beat intervals.
3. Strain gauge sensors: Sensors which detect strain are usually embedded in a textile or a belt positioned around the chest. Respiratory rate can be inferred from the strain gauge time-series which reflects chest wall movements (i.e. thoracic expansion).
4. Blood pressure sensor: Directly measured in ambulatory systems as peripheral pressure using a cuff which inflates at regular intervals. In some systems, a non-invasive cuff-less method is used: ECG and PPG waveforms can be analysed to derive systolic and diastolic blood pressure. This is considered less invasive than the conventional cuff-based method.
5. Galvanic skin response: Electrodes are connected to a central circuit of the monitor that acts as a galvanic skin response sensor. An assessment of the skin’s conductivity between the two electrodes is galvanic skin response. Galvanic skin response reflects evaporative heat loss through detecting differences in sweat rates.
6. Bioimpedance: A measure of the resistance to a small electrical current as it travels through the body tissue(s). Used in some monitors to infer heart rate, respiration rate, hydration.
7. Skin temperature sensor: A thermistor-based sensor (a resistor) measures the surface temperature of the skin. The resistor measures electrical resistance that varies with (skin) temperature.
8. Near-body ambient temperature sensor: This sensor measures the near-body ambient temperature, which is defined as temperature on the device facing away from the body. Near-body ambient temperature varies according to heat coming off the skin around the monitor and its surrounding environmental conditions. Heat flux can be derived as the difference between near-body ambient temperature and skin temperature.
9. Heat flux sensor: A heat flux sensor measures the rate of heat dissipating from the body, e.g. heat convection from the skin to the air.
10. Pressure sensors: Sensors measuring compression, typically incorporated into a specific body garment or piece of equipment (such as insole of shoe) can monitor pressure of key body weight support points. This can help discriminate between static postures and weight/non-weight bearing activities.
11. Accelerometers: Accelerometers in multi-sensor systems are similar to standard accelerometry-based activity monitors, typically utilising a micro-electro-mechanical sensor (MEMS) device. Some multi-sensor monitors use a bi-axial accelerometer while others use a tri-axial accelerometer. Postures and steps are inferred from acceleration data in some devices.
12. Gyroscopes: Gyroscopes measure angular (rotational) velocity, so can detect movements is not detected by accelerometer; however the combination of accelerometer and gyroscopes (6 degrees of freedom) can be used to measure body movements and express these in coordinates relative to gravity (as opposed to local coordinates).

1. Magnetometers: Magnetometers sense the Earth’s magnetic field and can be used to get a compass heading (indicating absolute orientation in the North-South-East-West plane). Local sources of magnetic fields, e.g. electric appliances, may also be transiently apparent in the signal. The combination of magnetometer with gyroscope and accelerometer (9 degrees of freedom) is also known as an inertial measurement unit (IMU) and allows expressing body movements in global coordinates, eg quaternions.
2. GNSS receiver: A sensor which receives signals from global navigation satellite system(s) to determine geographic location.

Configurations and wear locations

Some or all of the above sensors can be combined in different configurations and wear locations. Examples of how sensors have been combined in single instruments are described in Table P.3.30.

The number, size, and location of sensors will impact on the acceptability to participants; pilot work is essential if intending to use untried technologies in populations where these have not yet been applied. One multi-sensor monitor is designed to be worn on the middle of wearers’ triceps; a population-based study using this approach has reported high compliance rate of wearing multi-sensor monitors to be nearly 99% (Welk et al., 2014).

Table P.3.30 Wear locations and sensor arrays of multi-sensor monitors.

1. Observational studies
2. Interventions and randomised controlled trials to examine intervention or treatment efficacy
3. Studies undertaking association analyses between exposure(s) and outcome(s)
4. Describing temporal patterns of the intensity of physical activity throughout the day

Data from multi-sensor monitors are downloaded and typically processed using the manufacturers’ software or alternatively using user-written programs. Commercial software packages typically use proprietary algorithms to derive estimates of activity parameters. Derived estimates (inference) from commercial multi-sensor monitors may incorporate all or just some of the data from its sensors  as well as participants’ demographic data (e.g. age, sex, height, weight, smoking status, and handedness); however, the exact algorithms through which collected data produce activity parameters are often unknown and manufacturers may periodically update their proprietary algorithms to improve accuracy or expand the list of activity parameters being estimated (Backlund et al., 2010; Calabro et al., 2009; Leet et al., 2014) but this also presents challenges for comparisons between new and previously reported results.

In addition to proprietary models, there are published models developed by independent researchers. A recent study comparing the inference potential of multiple signals showed that the accelerometer information was most useful in classifying activity intensity of laboratory activities in a small sample; significant improvements in precision were achieved by adding information from heart rate, followed by smaller improvements using near-body temperature and skin temperature, whilst galvanic skin response did not add any further value to the model (Cvetkovic et al., 2016).

Another study using two proprietary algorithms to estimate intensity based on acceleration, galvanic skin response, skin and near-body temperature of the upper arm underestimated (absolute) activity energy expenditure of free-living older adults by 18.5 and 26.8%, when compared against criterion measures using the doubly-labelled water technique; however correlations were high (r>0.75) (Mackey et al., 2011).

Characteristics of multi-sensor monitor methods are described in Table P.3.31.

Strengths

1. Prospective data collection, so no recall bias
2. No social desirability bias (except possibly as reactivity or differential non-wear time)
3. No requirement for literacy and numeracy; data quality should not differ by educational attainment, ethnicity or socio-economic status
4. Continuously captures movement from second-to-second in much finer detail than even the most detailed PA diary
5. Greater sensitivity to changes in behaviour over time, so useful for evaluating interventions
6. Generally more options for making more robust inferences about activity as more information is collection
7. Ability to take into account the additional work required to move against resistance, such as lifting weights or cycling against a strong wind or incline
8. Better discrimination between wear time and non-wear time if the monitor senses contact with the skin and/or presence of other physiological signals known to have a narrower range when worn by a human.
9. Easy to administer (possibility to collect via post so no need for direct contact with participant)
10. No need to have high analytical skills for processing data
11. Provides time-stamped data, showing durations of PA and number of transitions from one PA level to another
12. Useful for self-monitoring and providing real-time feedback to wearer in situations where behaviour change is a desirable objective - e.g. RCTs, community health interventions
13. Regular updates of proprietary algorithms to improve accuracy
14. High compliance with monitoring protocols in observation studies (e.g. 99% Welk et al., 2014)

Limitations

Missing data / non-wear time / non-compliance:

1. Contextual data: objective methods provide no information on domain or behaviour type (e.g. driving, working, TV/computer use)
2. Representativeness: devices tend to be worn only for short-periods of a few days so may not capture infrequent, yet significant, physical activities
3. Negative aesthetic effects: unsightly bulges below close-fitting clothes may decrease wear time and compliance
4. Adverse effects: Individuals with metal allergies may experience skin irritations
5. Water resistance: Some multi-sensor monitors may not be water-proof but more recent versions tend to be water-resistant.

Cost and resources:

1. Purchase costs: Multi-sensor monitors are considerably more expensive than self-reported methods, and somewhat more expensive than typical accelerometer-based monitors.
2. Losses: The cost of lost or damaged devices may be substantial. Although in an ideal situation a study should lose no more than 2-3% of their devices, studies of more challenging populations (e.g. young children) may lose >25%.
3. Instruction sheet for participants: Particularly for more complex and cumbersome systems

Bias:

1. Prone to reactivity, whereby knowledge that one is being monitored causes changes in typical behaviour

Reproducibility / Transparency:

1. The use of proprietary algorithms results in uncertainty about the precise methodology employed and also prevents comparisons with studies using other devices.

Table P.3.31 Strengths and limitations of multi–sensor monitoring.

Considerations relating to the use of multi-sensor monitoring for assessing physical activity are summarised by population in Table P.3.32.

Table P.3.32 Physical activity assessment by multi-sensor monitors in different populations.

1. It is essential to check if participants are allergic to metal since some individuals may experience some skin irritation if device has metal electrodes designed to be in contact with the skin
2. Data should be processed using methodology that is fit for purpose; this may involve decisions on which versions of manufacturer software and algorithms to use to obtain estimates of activity parameters
3. Participants should be instructed to clean any electrodes using wet wipes or alcohol swaps when there is too much dirt on them

1. Multi-sensor monitors
2. Sufficient computers or laptops are required to initialise and download the monitors
3. Additional docking stations may be needed to recharge the monitors
4. In the field, attention must be paid to the logistics of distributing and collecting and re-distributing the monitors
5. Data storage

A list of specific multi-sensor instruments is being developed for this section. In the meantime, please refer to the overall instrument library page by clicking here to open in a new page.

1. Bäcklund C, Sundelin G, Larsson C. Validity of armband measuring energy expenditure in overweight and obese children. Medicine and Science in Sports and Exercise. 2009;42:1154-61
2. Banos O, Villalonga C, Damas M, Gloesekoetter P, Pomares H, Rojas I. PhysioDroid: combining wearable health sensors and mobile devices for a ubiquitous, continuous, and personal monitoring. TheScientificWorldJournal. 2014;2014:490824
3. Calabro MA, Kim Y, Franke WD, Stewart JM, Welk GJ. Objective and subjective measurement of energy expenditure in older adults: a doubly labeled water study. European Journal of Clinical Nutrition. 2014;69:850-5
4. Calabró MA, Stewart JM, Welk GJ. Validation of pattern-recognition monitors in children using doubly labeled water. Medicine and Science in Sports and Exercise. 2013;45:1313-22
5. Calabró MA, Welk GJ, Eisenmann JC. Validation of the SenseWear Pro Armband algorithms in children. Medicine and Science in Sports and Exercise. 2009;41:1714-20
6. Chowdhury EA, Western MJ, Nightingale TE, Peacock OJ, Thompson D. Assessment of laboratory and daily energy expenditure estimates from consumer multi-sensor physical activity monitors. PLOS ONE. 2016;12:e0171720
7. Cvetkovic B, Milic R, Lustrek M. Estimating Energy Expenditure With Multiple Models Using Different Wearable Sensors. IEEE Journal of Biomedical and Health Informatics. 2015;20:1081-7
8. D'Angelo LT, Weber S, Honda Y, Thiel T, Narbonneau F, Luth TC. A system for respiratory motion detection using optical fibers embedded into textiles. Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference. 2009;2008:3694-7
9. Grossman P. The LifeShirt: a multi-function ambulatory system monitoring health, disease, and medical intervention in the real world. Studies in Health Technology and Informatics. 2005;108:133-41
10. Johannsen DL, Calabro MA, Stewart J, Franke W, Rood JC, Welk GJ. Accuracy of armband monitors for measuring daily energy expenditure in healthy adults. Medicine and Science in Sports and Exercise. 2010;42:2134-40
11. Lee JM, Kim Y, Bai Y, Gaesser GA, Welk GJ. Validation of the SenseWear mini armband in children during semi-structure activity settings. Journal of Science and Medicine in Sport. 2014;19:41-5
12. Mackey DC, Manini TM, Schoeller DA, Koster A, Glynn NW, Goodpaster BH, Satterfield S, Newman AB, Harris TB, Cummings SR, et al. Validation of an armband to measure daily energy expenditure in older adults. The Journals of Gerontology. Series A, Biological sciences and medical sciences.2011;66:1108-13
13. Pandian PS, Mohanavelu K, Safeer KP, Kotresh TM, Shakunthala DT, Gopal P, Padaki VC. Smart Vest: wearable multi-parameter remote physiological monitoring system. Medical Engineering & Physics. 2006;30:466-77
14. Sazonov E, Hegde N, Browning RC, Melanson EL, Sazonova NA. Posture and activity recognition and energy expenditure estimation in a wearable platform. IEEE Journal of Biomedical and Health Informatics. 2015;19:1339-46
15. Sazonova N, Browning RC, Sazonov E. Accurate prediction of energy expenditure using a shoe-based activity monitor. Medicine and Science in Sports and Exercise. 2010;43:1312-21
16. Strath SJ, Brage S, Ekelund U. Integration of physiological and accelerometer data to improve physical activity assessment. Medicine and Science in Sports and Exercise. 2005;37:S563-71
17. Welk GJ, Kim Y, Stanfill B, Osthus DA, Calabro MA, Nusser SM, Carriquiry A. Validity of 24-h physical activity recall: physical activity measurement survey. Medicine and Science in Sports and Exercise. 2014;46:2014-24