Construct validity of the Actiwatch‐2 for assessing movement in people with profound intellectual and multiple disabilities

Abstract Background Valid measures to assess either small or assisted performed movements of people with profound intellectual and multiple disabilities (PIMD) are required. We analysed the construct validity of the Actiwatch‐2 to assess movement in people with PIMD. Method Twenty‐two persons with PIMD were video recorded while wearing an Actiwatch‐2. We used 15s‐partial‐interval recording to record upper body movement, body position and activity situation. Multilevel analyses were used to evaluate if the Actiwatch‐2, based on produced counts, could detect changes in these factors. Results The presence versus absence of upper body movement and an activity situation in which participants were involved versus not involved resulted in significantly higher counts, with a large variety in predicted counts between participants. No relationship between body position and counts was found. Conclusions The Actiwatch‐2 seems able to assess obvious upper body movement in people with PIMD, and whether there is involvement in an activity situation.

van aLPHEn Et aL. Vlaskamp, Reynders, & Nakken, 2005). In addition, recently, also new technologies such as an interactive ball are introduced to increase body movement in persons with PIMD Van Delden et al., 2020).
These movement activities can be used for a wide range of goals and encourage different domains of human functioning, such as the motor domain, but also beyond the motor domain, for example in social and cognitive functioning Jones et al., 2007;Van Alphen et al., 2019;Van der Putten et al., 2014). As a result, several studies recommend movement activities to be directed towards individual and specific measurable goals integrated within the overall support provided for people with PIMD Van Alphen et al., 2019;Van der Putten et al., 2005;Wessels, Bossink, & van der Putten, 2017). To identify whether goals are achieved and to what extent improvement of movement has contributed to outcomes on different domains, researchers and practitioners should be able to accurately assess the amount of movement of people with PIMD.
In general, movement (or physical activity) is assessed based on energy expenditure or the execution of movements in daily life (e.g. steps per day: Hilgenkamp, Reis, Van Wijck, & Evenhuis, 2012).

Several studies have been performed into the validation of a wide
range of devices to assess movement in ambulatory people, but hardly in non-ambulatory people such as people with PIMD (Berlin, Storti, & Brach, 2006;Warms & Belza, 2004). In addition, algorithms that predict the activity energy expenditure of people with PIMD are lacking (Waninge et al., 2013). Moreover, measurement evidence (e.g. validity and reliability) among subgroups of people with intellectual disability, such as people with PIMD, are lacking in this field (Pitchford, Dixon-Ibarra, & Hauck, 2018). Therefore, there is an urgent need for research into instruments measuring movement in people with PIMD. Most movements of people with PIMD are either small and assisted or passively performed. Therefore, we suggest that instruments measuring movement in people with PIMD should capture actively as well as assisted and passively performed movements. In addition, even small movements of the limbs performed from different body postures (i.e. lying, sitting and standing) as well as changes in body position are important to identify in people with PIMD, because these are not self-evident.
To date, a few subjective and objective measures are used to assess the movement behaviour of people with PIMD (Van Delden & Reidsma, 2018; Van der Putten, Bossink, Frans, Houwen, & Vlaskamp, 2017;Waninge et al., 2013). A previous study investigated the degree and type of strategies offered to facilitate movement in people with PIMD by the use of a diary ( Van der Putten et al., 2017). This study did provide a valuable insight into the number of transfers, relocations and motor activities offered in the support of people with PIMD ( Van der Putten et al., 2017). However, it did not focus on the actual amount of movement of people with PIMD. In addition, diaries in general are susceptible to inaccurate recall and in comparison with objective measures less accurate to assess the amount of movement performed. Objective measures such as heart rate monitors have been used to provide an insight into the daily activity patterns of persons with PIMD (Waninge et al., 2013). Heart rate monitors maybe useful to roughly evaluate initiatives directed at the facilitation of movement, but it is unclear if those monitors based on heart rate patterns also could identify passively and assisted performed movements of people with PIMD.
In addition, heart rate patterns are influenced by differences in physiological responses and with time of day, age, and probably also other personal and psychosocial factors (Waninge et al., 2013;Warms, 2006). As a result, the influence of movement on heart rate in people with PIMD is not fully clear. Automatic measurements of movement based on video recordings have also been used in people with PIMD . In the simplified motion energy analysis, for instance, the amount of pixels that changed beyond a certain threshold is measured. Although, the use of persuasive technological measurements is highly valued, the outcomes can become difficult due to unforeseen side-effects and incorrect values (e.g. influence of auto-focus, shaking camera, moving material and other persons who entered the view of the camera) . All in all, movement can be measured in different ways, but specific instruments with clear psychometric properties are needed to assess the amount of movement of people with PIMD.
Accelerometers can provide objective and continuous information about the duration, frequency and intensity of movements and are relatively easy to wear (Ainsworth, Cahalin, Buman, & Ross, 2015;Berlin et al., 2006). An Actiwatch, a wrist-worn accelerometer, is originally developed to measure rest-activity patterns based on body movement and is previously used in the support of people with PIMD to investigate sleep problems (Drenth, Poppes, & Vlaskamp, 2007;Van de Wouw, Evenhuis, & Echteld, 2013;Van Dijk, Hilgenkamp, Evenhuis, & Echteld, 2012). Because an Actiwatch records wrist accelerations which are directly related to the amount of movement performed, this instrument may be useful to distinct between facilitated movements and small involuntary movements in people with PIMD. In addition, an Actiwatch may be able to distinct activities performed from different body postures and possibly also different activity situations and ways of stimulation. Moreover, actively performed as well as passively and assisted performed movements will be identified by an Actiwatch which is important particularly in people with PIMD because of their severe motor disabilities.
Only a few studies have investigated (in other populations than people with PIMD) whether the Actiwatch-2 (Philips, Respironics) can be used as a measure of movement behaviour (Lambiase, Gabriel, Chang, Kuller, & Matthews, 2014;Lee & Suen, 2017;Neil-Sztramko, Rafn, Gotay, & Campbell, 2017). These studies suggest that an Actiwatch-2 is able to discriminate different intensities of movement activity (Neil-Sztramko et al., 2017), although may better capture low-intensity activities instead of higher intensity activities (Lambiase et al., 2014;Warms, 2006). This may be particularly pertinent for people with PIMD. In addition, an instrument such as an Actiwatch-2 that could measure both sleep and movement behaviour simultaneously will reduce the burden on participants with PIMD. Moreover, the Actiwatch-2 is already used to measure sleep of persons with PIMD on a regular basis. Hence, the research could benefit from the fact that participants as well as their support professionals are already acquainted with the use of this instrument.
The validity of the Actiwatch-2 to assess movement in people with PIMD has not been previously investigated. The purpose of this study was to investigate the construct validity of the Actiwatch-2 to assess movement in people with PIMD. We evaluated if the Actiwatch-2 could detect observed changes in upper body movement, body position and activity situation. We have added activity situation to this study, because movement in persons with PIMD largely depend on stimulation by the environment and could result from different activity situations and ways of stimulation, even when not directly aimed at movement. Future research on the effectiveness of movement interventions may benefit from the results if for instance passive and active participation in movement can be distinguished.

| Participants
In the present study, the participants were enrolled in an intervention study registered at the Netherlands Trial Register (number 6627), which was approved by the Ethics Committee for Pedagogical Sciences and Educational Science of the University of Groningen.
Based on funding cooperating parties, participants were recruited by physical therapists of three different residential facilities offering 24-hr support to people with intellectual and visual disabilities, including people with PIMD. For 26 participants, written informed consent was obtained from parents or legal representatives. Inclusion criteria were (a) severe or profound intellectual disability (intelligence quotient (IQ) under 35 points or a developmental age up to 36 months), (b) severe or profound motor disability (classified as Gross Motor Function Classification System (GMFCS) IV or V: Palisano et al., 2000), and (c) a continuous need for support for all activities in daily life (Nakken & Vlaskamp, 2007;WHO, 2001). In addition, all participants had moderate to profound visual impairment or blindness (a visual acuity of less than 0.3 (WHO, 2016)), because they were recruited from the cooperating residential facility for people with visual impairment and intellectual disability. Participants of the above-mentioned study were included in the present study if they had available Actiwatch and video data collected within the same time frames. Three participants were excluded because of missing Actiwatch data due to oversensitivity or reluctance to wear the device on their wrist. In addition, one participant was excluded because of missing video data. Therefore, the current study is based on 22 participants with PIMD (11 males and 11 females) with a mean age of 35.1 ± 13.6 years. Table 1 shows the characteristics of the participants in terms of mobility and health problems.

| Procedures
The data were retrieved from the above-mentioned intervention study including three measurement periods that lasted two weeks each measurement period. Per measurement period, wrist accelerations of the dominant wrist (their most mobile arm/hand) were measured with an Actiwatch-2 for at least seven consecutive days, 24 hr per day. In addition, participants were video recorded during their regular program (without prescription of any activity by the researchers and except for caring activities where clothes were taken off) each meas-

| Actiwatch measurements
Actiwatch-2 (Philips, Respironics) data were collected with an epoch duration of 15 s. The Actiwatch-2 contains an acceleration-responsive piezoelectric sensor and is set up to record the intensity, frequency and duration of movements which is converted into voltage.
This means that an increase in speed and motion produces an increase in voltage (sampling rate 32 Hz), which was integrated and stored as an activity count in the Actiwatch memory reflecting the peak acceleration per 15 s. Actiwatch data were transferred offline to a computer and automatically stored in activity counts by date and time using the Philips Actiware 6.0.9. software.
4. Standing/moving: Participants were engaged in walking activity with physical assistance or use of a body support walker, for example.
Activity situation was coded in four categories indicating a different involvement of people with PIMD due to a different aim of stimulation in relation to movement activity (adapted from Special Heroes, 2013). The four categories are as follows: 1. Being present: Participants were present, but not actively engaged or involved in the activity situation. For example, audio-visual activities, activities focusing on other participants in the same environment, or even no activities were provided and resulting in, for instance, movements arising from behavioural states.
2. Being part of: Participants were part of the activity situation, but not directly stimulated by their environment to move actively. For example, activities like massage, grooming or moments of social interaction were offered.
3. Passive participation: Participants were involved in activities with the help of support aimed at an active an engaged movement experience. For instance, the limbs of the participants were moved by powered exercise machines or participants experienced the wind while swinging and being moved in a hammock.
4. Active participation: Participants were actively involved and engaged in the activity situation and had a motorically active participation with little support. For instance, participants were eating independently (e.g. holding a cup or picking up a piece of bread), splashing the water while swimming, initiated bouncing movements on a bouncy castle, or were walking with physical assistance or use of a body support walker when positioned.

| Reliability
To ensure reliable coding, 12 video recordings (two randomly chosen video recordings of six participants) were coded by two independent observers. The interrater-reliability was calculated by using Cohen's kappa (Cohen, 1960

| Statistical analysis
To determine the construct validity of the Actiwatch-2, it was ana-

| Data from similar periods of time
The participants had available Actiwatch data for at least one video recording (of about 15 min) with a maximum of 11 video recordings.
As we used 15s-partial-interval recording, each video record-

| Relationship between movement and activity counts
The mean number of counts was 12.5 times higher for the presence of movement (M = 90.1, SD = 139.3) compared with the absence of movement (M = 7.2, SD = 29.0) (See Table 2). As shown in Figure 1, there is a wide variety in the count range between participants, but the mean number of counts for all participants except one (participant 3) was higher for the presence of movement versus the absence of movement for each of the body positions and activity situations (see Figure 1 and Table 2). The results of the multilevel models are presented in   Table 3). This suggests that the Actiwatch-2 is able to detect changes in the occurrence of upper body movement. Table 3 shows the influence of body position (Model 4) and activity situation (Model 5) on the count level. Sitting and standing/moving versus lying had no significant influence on the count level, while standing still yielded significantly lower counts than lying. However, standing still is based on only four minutes of observation (seven and nine video-based observations related to the absence and presence of movement, respectively (see Table 2 in comparison with a situation in which people with PIMD were not involved in the activity (being present). The logit model confirmed a relationship between activity situation and counts (see Table 3). The probability of higher category counts increased with involvement in an activity situation. The risk (in terms of odds) of higher category counts for being part, passive participation and active participation were 1.6, 2.8, and 1.4 times the risk of being present, respectively (Logit model, Table 3).

| Relationship with body position and activity situation
In summary, the Actiwatch gives significantly higher counts for the presence versus absence of movement and significantly higher counts for three of the activity situations versus being present when adding activity situation in addition to the occurrence of movement (Model 5 and Logit model). There is, however, a large variance between participants when it comes to the counts that are associated with the occurrence of movement and activity situation. The level-3 variance (between participants) of the intercept and slope is 457.1 and 475.0 (Model 5, Table 3).

| Main findings
This study investigated the construct validity of the Actiwatch-2 to assess the occurrence of upper body movement in people with PIMD. The major finding is that the Actiwatch-2 is able to distinguish the presence of upper body movement from the absence of upper body movement in people with PIMD. This study did not find a significant effect of body position, in particular of a lying and sitting position, on the count level, In addition, the Actiwatch-2 gave significantly higher counts in situations in which a person with PIMD is involved in the activity from situations in which a person with PIMD is present but not involved. The Actiwatch-2 is, however, not able to distinct different types of activity at which people with PIMD are involved. For instance, the Actiwatch is not able to distinguish if the presence of movement is derived from massage (being part of), swinging in a hammock (passive participation) or initiated bouncing movements (active participation). In addition, the results showed a wide variety in the count range between participants. Therefore, cut-off values should be defined person-to-person.

| Theoretical reflection and implications
The validity evidence with regard to the measurement of physical activity in people with intellectual disability, and in particular in people with PIMD, is limited (Pitchford et al., 2018). This study provides evidence with regard to the construct validity of the Actiwatch-2 as a measurement of movement in people with PIMD. The results can be used to evaluate interventions directed at the facilitation of upper body movement of people with PIMD. The Actiwatch-2 may be suitable to determine whether movement activity results from facilitation and whether obvious movements instead of none or small involuntary movements (scored as the absence of movement) were seen in people with PIMD. In general, an Actiwatch is sensitive for small movements and will register involuntary movements (Warms, 2006).
However, based on the current study and those of Warms and Belza (2004) & Vlaskamp, 2009). In addition, with the ability of the Actiwatch-2 to distinguish the presence and absence of movement, the Actiwatch-2 may be useful to evaluate whether inactivity has decreased in people with PIMD. This is important, because these people are at risk for being physically inactive (Bjornson, Belza, Kartin, Logsdon, & McMaughlin, 2007;Hilgenkamp et al., 2012;Van der Putten et al., 2017) and even small improvements in physical activity can be very beneficial for these persons (Jones et al., 2007;Levine, 2007;Woodcock, Franco, Orsini, & Roberts, 2011).
Based on the results with regard to activity situation, the outcomes of the Actiwatch-2 can be best explained with both the occurrence of movement and activity situation added to the model.
In addition, the Actiwatch-2 give significantly higher counts in situations in which a person with PIMD is involved in the activity in comparison with situations in which a person with PIMD is present but not involved. Therefore, we suggest that activities including the stimulation of social interaction, tactile stimulation or stimulation of the motor domain could possibly be distinguished from none and audio-visual activities, such as watching television or listening to music. This finding may contribute to future research emerged at improvement of the quality of support of people with PIMD (Van der Putten & Vlaskamp, 2011). The current study, however, showed, that the Actiwatch is not able to distinct activity situations with the involvement of a person with PIMD and thus between movement activities and activities such as massage. An explanation might be that the functional use of the arms of people with PIMD is limited (Nakken & Vlaskamp, 2007) resulting in movements with a low frequency, intensity and duration independent of the type of stimulation. Based on the variety in count range for the occurrence of movement between participants, it is possible that the ability in handling objects differed between participants. A less severe limitation in motor functioning may increase the active participation including the speed of performed movements and therefore may increase the accuracy of measurement in people with PIMD. For future studies, it is, therefore, recommended that an individual approach is used or that the manual ability of the participants is included as factor in the analysis.
With regard to the relationship between body position and counts, the Actiwatch did show similar outcomes for different body postures and is, similar to the ability of accelerometers in general, limited in accurately measuring body postures (Ainsworth et al., 2015).
One explanation might be the place where the Actiwatch is worn. An accelerometer device worn on the leg has been shown to accurately measure reductions in sitting time (Kozey-Keadle et al., 2011). It may be that just wrist derived data, as collected in the current study, are inappropriate to distinct between body postures.
This study concludes a difference in counts for the absence versus presence of upper body movement as well as for an activity situation in which participants were involved versus not involved.
Nevertheless, thresholds to summarize the counts into specific activity categories for persons with PIMD have to be further calibrated. This is important in order to be able to predict the time spent prediction equations should also be validated (Plasqui & Westerterp, 2007). Moreover, as the group of people with PIMD is heterogeneous (Nakken & Vlaskamp, 2007) and the relationship between the occurrence of movement and outcomes of the Actiwatch significantly differs between participants, cut-off values should be individually defined. To be able to predict the outcomes of the Actiwatch-2 that relate to activity and inactivity for an individual with PIMD, an individual count pattern could be investigated by the use of video observation in combination with Actiwatch-2 measurements.
Although group-based comparative intervention-based research in people with PIMD seems inappropriate, we want to emphasize with a rough indication for inactivity that the outcomes for this target group should be viewed differently. A count range 0-100 (or even  Table 3, which should be fine-tuned and tailored on an individual level.

| Methodological reflection and implications
When interpreting the results of this study a few remarks need to bear in mind. As this study was a validity study, outliers were not excluded from the analysis. As stated in the result section, the mean number of counts for all participants except participant 3 was higher for the presence versus absence of movement. The data of this participant seem to be influenced by one video recording presenting a mean count of 218.8 while no actual performance of movement was observed. Based on the count pattern, it seems that the Actiwatch had been stuck in the value during the performance of movement earlier in the video. Although further analysis showed that exclusion of this video remains the same study conclusion, it needs to be taken in mind that such deviations may occur by using technical devices such as an Actiwatch. Despite that, it has also been showed that the correlation between measured and true exposure was higher for accelerometers compared with questionnaire measurements (Ferrari, Friedenreich, & Matthews, 2007). However, it should be taken in mind that the Actiwatch may not suitable for every person with PIMD. In the current study, 11.5% of the participants had to be excluded due to oversensitivity or reluctance to wear the device on their wrist.
The data used in current study were retrieved from three measurement periods at which Actiwatch and video data were collected.
As IV are able to walk with physical assistance, unsurprisingly, standing and walking/moving were seen very infrequent in the current study.
Therefore, the influence of body position on the outcomes of the Actiwatch-2 should be further investigated in persons with PIMD.
The distribution of counts in the current study was right skewed.
Therefore, as a sensitivity analysis, a logistic model for binomial re- We suggest that in case of subtle head movements, for instance, techniques such as motion history (Iwabuchi et al., 2014)  of people with PIMD. However, it can actually be discussed if those kind of movements should be seen as movement when there is no active participation of the person with PIMD. This type of activity includes sensory stimulation (e.g. experiencing the wind while swinging) as well as vestibular stimulation which may evoke reflex responses in the muscles (Mittal & Narkeesh, 2012). As these type of activities are used in current practice to activate people with PIMD (Van Alphen et al., 2019) and may evoke a motor response, professional consensus about what movement should actually consist of for people with PIMD is needed. In addition, this study did not identify movements of the legs. Although this can be seen as a limitation, we expect other instruments than the Actiwatch to be needed to identify leg movements. Most movements of people with PIMD are performed from a lying and sitting position and we do not expect leg movements from these body postures to influence wrist movements. An accelerometer device worn on the leg could possibly offer a solution here (Kozey-Keadle et al., 2011).
Based on the current study, it can be suggested that the Actiwatch-2 is able to distinguish obvious movement activity from small involuntary movements. Spasticity and stereotypical behaviour, however, could also manifest as obvious limb movements, identified as the presence of movement. Although these movements are a form of activity (Warms & Belza, 2004), those are usually not aimed to improve by interventions (although they could be in some cases an expression of enthusiasm). Moreover, to our opinion, frequently seen stereotypical behaviour hamper the opportunity to explore the environment and the development of functional skills. However, identifying those involuntary movements with an Actiwatch is difficult, because the acceleration signal related to movement needed for manipulating material could in fact be the same as the acceleration signal related to movement shown during stereotypical behaviour. Therefore, for future research, it is recommended that participants with significant involuntary movements be analysed separately in a way that the movement elicited can be distinguished from involuntary movements. In addition, it is important to maximize the benefits of movement in persons with PIMD by integrating individual tailored movement activities into their support.

| CON CLUS ION
The Actiwatch-2 may be useful to assess the occurrence of movement of people with PIMD, and whether there is involvement in an activity situation. However, further studies are needed to calibrate cut-off points to define the counts and patterns of change for individuals with PIMD.

ACK N OWLED G M ENTS
The authors kindly acknowledge and thank the involved practitioners as well as the participants (and their representatives for giving permission) for study participation. Furthermore, we kindly acknowledge and thank Lisanne Werkhoven and Annemiek Huurdeman for their contributions to the data analysis.

CO N FLI C T O F I NTE R E S T
There are no conflicts of interest.