Musculoskeletal health outcomes of the low back and neck/shoulder are among the most prevalent and burdensome of all occupational injuries. Estimates from the Global Burden of Disease 2010 study suggest that low back pain causes more global disability than any other condition, accounting for 83 million disability-adjusted life years (DALYs) in 2010 alone (Hoy et al., 2012; Hoy et al., 2014a). Low back pain arising from occupational exposure to physical risk factors caused 21.7 million DALYs; an increase of 22% between 1990 and 2010 (Driscoll et al., 2014; Murray et al., 2013). Neck/shoulder pain was estimated to account for 33.6 million DALYs in 2010, ranking it the 4^th highest in terms of disability and 21^st in terms of overall burden (Hoy et al., 2014b).

Occupational exposure to non-neutral postures has been associated with musculoskeletal health outcomes of the low back and neck/shoulder (da Costa and Vieira, 2010 ; van Rijn et al., 2010; Vieira and Kumar, 2004). Methods for assessing exposure to non-neutral working postures in a field setting include self-report, observational, and direct measurement (Burdorf and Van Der Beek, 1999; David, 2005). The standard approach for directly measuring occupational exposure to non-neutral postures of the low back and shoulder is with piezoresistive accelerometers or, more recently, inertial measurement units (IMUs) secured to the trunk and/or upper arms (David, 2005; Li and Buckle, 1999; Teschke et al., 2009).

An IMU is a solid-state device that measures and reports an object’s spatial orientation and motion characteristics using multiple electromechanical sensors (i.e., accelerometers, gyroscopes, and/or magnetometers). They are considered advantageous to accelerometers for posture assessment as fusion of the components of an IMU may address limitations of each individual sensor component (Luinge and Veltink, 2005; Roetenberg et al., 2007). For example, gyroscope measurements can be used to compensate for accelerometer-based measurements that are known to be negatively affected by dynamic and complex motions (Amasay et al., 2009; Godwin et al., 2009). Several recent studies have indicated that IMUs are reasonably stable and accurate when estimating trunk and upper arm postures in comparison to “gold-standard” optoelectric motion capture systems (Cuesta-Vargas et al., 2010; El-Gohary and McNames, 2012; Kim and Nussbaum, 2013; Schall Jr et al., 2015a) and a field-capable reference device (Schall Jr et al., 2015b). Additionally, the small size and increasing affordability of IMUs make them practical instruments for measuring posture in field-based research.

In addition to posture, an IMU’s accelerometer may be used to measure other aspects of worker health that are typically difficult to observe, such as intensity of physical activity (PA). Although leisure-time PA is widely considered beneficial to overall health (Haskell et al., 2007; Warburton et al., 2006), and some studies have observed favorable effects of PA on musculoskeletal pain among worker populations (Hildebrandt et al., 2000), high intensity occupational PA has been associated with increased risk of several chronic health conditions and may be deleterious to health (Harari et al., 2015; Heneweer et al., 2011; Holtermann et al., 2012a; Holtermann et al., 2012b; Holtermann et al., 2010; Sitthipornvorakul et al., 2011). Moreover, some investigators have suggested a U-shaped relationship between intensity of PA and undesirable health outcomes (i.e. both inactivity and excessive activity may increase risk; Heneweer et al., 2009; Sesso et al., 2000). Recent work has thus emphasized the need for more accurate ascertainment of exposure to occupational and leisure-time PA using objective and validated field measurement instruments to better establish future public and occupational health recommendations (Holtermann, 2015; Prince et al., 2008).

Modern PA monitors are portable, electronic devices used to measure and track fitness-related metrics such as energy expenditure, caloric consumption, sleep patterns, and total activity. PA monitors are commonly used in epidemiological research because of their small size, freedom from biases associated with self-report, and accuracy and precision in estimating intensity of activity (Freedson et al., 2012; Matthews et al., 2012). Most PA monitors are worn at the waist or wrist and use single or multi-axis accelerometers to provide summary metrics of the quantity and intensity of movements during set time intervals (Taraldsen et al., 2012; Troiano et al., 2008; Van Remoortel et al., 2012). PA monitors have been used extensively to promote workplace health, with promising results (Bravata et al., 2007; Freak-Poli et al., 2011; Pal et al., 2009; van Sluijs et al., 2006).

Because IMUs and commercially available PA monitors incorporate fundamentally identical hardware (accelerometers), the use of small numbers of sensors in anatomic locations to capture both biomechanically meaningful posture information and PA information may be possible. Commercially available PA monitors positioned on the upper arms and on the upper back provide valid inclination data for direct long-term field measurements of upper arm and trunk inclinations in comparison to a magnetic tracking device (Korshøj et al., 2014) and a universal goniometer (Hirschhorn, 2015). However, the extent to which PA measurements obtained from sensors located on the upper arms or on the trunk agree with PA measurements obtained from PA monitor worn at the waist is not known. The objective of this study was, therefore, to estimate the agreement between measures of PA obtained with IMUs attached to the upper arms and to the trunk and measures of PA obtained with a commercially available PA monitor worn at the waist.

Complete ActiGraph data were obtained on all 36 participants. Instrumentation failure (IMUs) led to the loss of one participant’s trunk data, three participants’ left upper arm data, and three participants’ right upper arm data. All but three participants worked a 12 hour shift (two worked for eight hours and one worked for 11 hours). Twenty-one participants worked day shifts (starting at 7 am) and 15 participants worked night shifts (starting at 7pm). Overall, a small percentage of work time was spent performing moderate PA across all participants regardless of sensor location. On average, participants spent less than 0.25% of their work day performing any vigorous activity (>5725 counts/min or >6 METS). Due to the small percentage of work time spent performing vigorous activity, the percentage of work time spent performing vigorous or very vigorous activity was combined with the percentage of work time performing moderate activity for all analyses.

Statistically significant differences were observed between all sensor locations for all PA summary metrics except for between the left and right arm for percentages of work time in the light and moderate counts/min categories (Table 1). In general, the IMU worn on the dominant (right) arm displayed the greatest range and mean percentage of work time performing moderate intensity PA, followed by the non-dominant (left) arm, the waist, and the trunk (e.g., percentage of work time METs moderate; Figure 2).

Bland-Altman evaluations used to assess agreement between the PA measurements from each IMU and the wGT3X-BT further suggested that PA measurements from the trunk and each upper arm were substantially different (Table 2). Mean bias estimates indicated that the IMUs worn on the upper arms both estimated about 200 counts/min more, on average, than the wGT3X-BT (195 counts/min for the left arm and 208 for the right arm). In addition, estimates for the limits of agreement indicated potentially large differences between measurements. The IMU worn on the right (dominant) upper arm had a greater range for the limits of agreement (2229 counts/min) in comparison to that of the left upper arm (1743 counts/min). The IMU worn on the trunk had a smaller mean bias than the upper arm IMUs, estimating about 8 counts/min fewer, on average, than the wGT3X-BT. Despite the smaller mean bias, however, the IMU worn on the trunk displayed the greatest range of the limits of agreement (2581 counts/min). A similar pattern of mean bias and limits of agreement results was observed for METs, although the greatest range of the limits of agreement occurred with the right (dominant) upper arm (1.33 METs) rather than the trunk.

Qualitative assessment of the summarized Bland-Altman plots (Figure 3) indicated that, in general, as the mean of the average of two measurement approaches increased, the difference between the measurement approaches also increased. An exception was an observed decrease in the difference in METs between the IMU at the trunk and the wGT3X-BT with an increase in the average METs of these two sensors. Within-subject variability of both the averages of the measurement approaches and the differences between them also generally increased as the average increased. The existence of proportional bias, especially for the upper arms, suggests that PA measurements (i.e., counts/min and METs) from the IMUs at the trunk and upper arms did not agree equally through the full range of measurements with PA measurements from the wGT3X-BT at the waist. This observation was most pronounced when comparing the IMUs on the upper arms with the wGT3X-BT.

While participation in moderate- or vigorous-intensity leisure-time PA is considered an important aspect of good overall health (Haskell et al., 2007; Warburton et al., 2006), large amounts of both low intensity (i.e., sedentary) and high intensity occupational PA may be detrimental to health (Harari et al., 2015; Heneweer et al., 2011; Heneweer et al., 2009; Holtermann et al., 2012a; Holtermann et al., 2010; Sitthipornvorakul et al., 2011). Further development of objective, field capable measurement devices are needed to more accurately ascertain exposure to occupational and leisure-time PA and understand their relationship with health outcomes (Holtermann, 2015). Several recent studies and systematic reviews have assessed the validity and/or reliability of PA monitors among participants in controlled laboratory environments and/or during free-living conditions (Aguilar-Farías et al., 2014; Brazeau et al., 2014; Grydeland et al., 2014; Jarrett et al., 2014; John et al., 2010,2013; Kelly et al., 2013; Kowalski et al., 2012; Van Remoortel et al., 2012; Thiese, 2014; Vanhelst et al., 2010; Welk et al., 2012; Yang and Hsu, 2010). Results have generally suggested a high level of heterogeneity (Van Remoortel et al., 2012), indicating a need for researchers to cautiously select monitors and summary metrics that are appropriate for their research questions (Kowalski et al., 2012). Recommendations and protocols for evaluating and using wearable monitors for researchers, developers, and end users have thus been developed in an effort to control for differences among monitors and summary metrics (Brazendale et al., 2015; Freedson et al., 2012; Welk et al., 2012).

IMUs are reasonably stable and accurate instruments capable of simultaneously measuring multiple components of worker health such as exposure to high intensity PA and non-neutral working postures. Despite the large number of studies comparing various PA monitors and activity outcomes, the current study represents the first effort we are aware of to estimate the agreement between measures of PA obtained with IMUs attached to the upper arms and to the trunk to measures of PA obtained with an ActiGraph wGT3X-BT worn at the waist. Overall, results suggested that PA measurements obtained from the four sensor locations evaluated in this study (the waist, trunk, and each upper arm) varied substantially. The IMU worn on the dominant (right) arm displayed the greatest amount of moderate intensity PA for all summary metrics evaluated, followed by the non-dominant (left) arm, the waist, and the trunk.

Accelerometers worn on the waist may underestimate the energy expenditure of free-living individuals because of an inability to detect arm movements and upper body work (Bassett Jr et al., 2000; Hendelman et al., 2000; Matthews, 2005; Matthews et al., 2012). Given this limitation, it is worth considering if PA measurements for individuals performing complex, high intensity upper body activities should be assessed using PA monitors worn on the upper arm instead of the waist. Patient care workers, for example, have been observed to report a substantially greater duration of moderate to vigorous PA during a work week than what is directly measured by PA monitors worn at the waist (Umukoro et al., 2013). Although nurses may perceive their work as physically-demanding, PA monitors worn at the waist may not be sensitive to the tasks potentially culpable for the sentiment (e.g., extremes of posture during patient transfers). The results of the current study imply that an IMU worn on the upper arm may provide more representative estimates of PA than a waist-worn PA monitor for field-based occupational research. However, while we have observed a greater range and higher percentage of work time performing moderate PA for IMUs located on the upper arms in comparison to the waist-worn PA monitor, it is unknown what specific tasks participants were doing that led to this observation.

An additional incentive to wearing PA monitors on the upper arm is the ability to capture information regarding exposure to non-neutral shoulder postures, a risk factor for the development of work-related musculoskeletal health outcomes (da Costa and Vieira, 2010; Miranda et al., 2001; Ohlsson et al., 1994; NIOSH, 1997; Silverstein et al., 2008; Silverstein et al., 2006; Svendsen et al., 2004a; Svendsen et al., 2004b; van Rijn et al., 2010; van der Windt et al., 2000). Estimates of arm and trunk inclination obtained with PA monitors have been observed to compare favorably to estimates from a magnetic tracking device (Korshøj et al., 2014) and universal goniometer (Hirschhorn, 2015). While PA monitors have been used to assess PA levels in the context of worker health promotion (Malik et al., 2014) and have strong potential for posture measurement, few studies have simultaneously measured occupational exposure to working postures and PA. This study provides a critical step towards understanding the potential limitations of using IMUs worn on the trunk and upper arms to simultaneously measure PA and exposure to non-neutral working postures.

Limitations of the study include the lack of a wrist-worn PA monitor and the collection of PA data during a single work shift. Several studies have recommended multiple measurement days for reliable accelerometer estimates of habitual daily PA (Berlin et al., 2006; Trost et al., 2000; Welk et al., 2012). Agreement between sensor locations may have improved had measurements been obtained over multiple days. Additionally, while we used raw acceleration to measure PA and summarized our estimates of PA using two approaches, the use of traditional PA count cut points is becoming less common (John et al., 2013). Finally, it is important to note that while the IMU worn on the dominant (right) arm displayed greater levels of moderate intensity PA for all summary metrics evaluated, on average, than the non-dominant (left) arm, it is unknown if this result was a function of worker handedness or if the work itself required more right arm movement. Further research on the effects of worker handedness and the design of the work environment on PA estimation is needed.

Efficient estimation of the physical demands of work remains somewhat limited by the need for multiple sensors to capture different aspects of physical demands (e.g., ActiGraph or other commercially available monitors to capture PA and a separate accelerometer or IMU to measure posture). The results of this study indicate that the location of the sensor used to obtain acceleration information may substantially affect PA estimates when compared to estimates from a waist-worn PA monitor. Additional research is needed to better understand how PA measurements obtained from the upper arm and trunk relates to measurements obtained from waist and wrist-worn PA monitors. Specifically, validation studies conducted in controlled laboratory environments or using systematic observation of field-based activities are needed to develop conversion algorithms intended to relate or synthesize PA estimates from various studies (e.g., Brazendale et al., 2015). Standardized, non-proprietary metrics and procedures for using ambulatory sensors such as IMUs in field-based research must also continue to be developed and evaluated (Freedson et al., 2012).

While PA measurements obtained from the upper arm and trunk may not agree with measurements obtained from the waist, they may provide more representative PA estimates for demanding work tasks than a waist-worn PA monitor. The existence of a potential interaction between intensity of PA and non-neutral working postures should, therefore, also be further investigated. Such research may provide valuable insight into the association between physical risk factors in the workplace and musculoskeletal health outcomes.