How Many Days Are Necessary to Represent an Infant's Typical Daily Leg Movement Behavior Using Wearable Sensors?

Abstract

Background

Characteristics of movement can differentiate infants with typical development and infants with or at risk of developmental disabilities. We used wearable sensors to measure infants’ typical movement patterns in the natural environment.

Objective

Our objectives were to determine (1) how many days were sufficient to represent an infant's typical daily performance, and (2) if there was a difference in performance between weekdays and weekend days.

Design

This was a prospective, observational study.

Methods

We used wearable sensors to collect 7 consecutive days of data for leg movement activity, from 10 infants with typical development (1–5 months old). We identified each leg movement, and its average acceleration, peak acceleration, and duration. Bland-Altman plots were used to compare the standard (average of 7 days) with 6 options (1 day, the average of days 1 and 2, through the average of days 1 through 6). Additionally, the average of the first 2 weekdays was compared with the average of 2 weekend days.

Results

The absolute difference between the average of the first 2 days and the standards fell below 10% of the standards (movement rate = 8.5%; duration = 3.7%; average acceleration = 2.8%; peak acceleration = 3.8%, respectively). The mean absolute difference between weekdays and weekends for leg movement rate, duration, average acceleration, and peak acceleration was 11.6%, 3.7%, 7.2%, and 7.3% of the corresponding standard.

Limitations

The small sample size and age range limit extrapolation of the results.

Conclusions

Our results suggest the best option is to collect data for 2 consecutive days and that movement did not differ between weekdays and weekend days. Our results will inform the clinical measurement of full-day infant leg movement for neuromotor assessment and outcome purposes.

Spontaneous kicking movements in infancy have been proposed as foundational and necessary practice to support the motor skill of walking, a skill that emerges later in life.^1,2 Current theories of infant motor learning support the idea that infants learn motor skills through repeated cycles of exploratory movements, perceiving and acting on errors and successes repeatedly.^1,3,4 It is currently unknown, however, what types of exploratory movements are most beneficial for learning specific motor skills and how many repetitions of movement practice are required for an infant to achieve a new motor skill.

Further, infants with atypical neuromotor systems produce different leg movement characteristics than infants with typical development. Differences in leg movement characteristics have been observed in infants with cerebral palsy,⁵ Down syndrome,^6,7 and myelomeningocele.^8,9 Considering that they move differently than infants with typical development, infants with atypical development could require more practice, or different types of practice, to learn motor skills.

In order to relate the amount and type of leg movement practice to motor skill development we are using wearable sensors to measure infant leg movements across a full day in the natural environment. We can identify each leg movement produced across a full day, as well as its duration, average acceleration, and peak acceleration. Here we assessed the reliability of movement characteristics across days.

Previously, researchers have used video and motion analysis systems to quantify infants’ motor behavior. However, these approaches are limited to a few minutes of data collection and cannot capture comprehensive and detailed characteristics of infants’ spontaneous leg movements across a day in the natural environment. Moreover, the frame-by-frame analysis of these types of data requires considerable resources (including trained personnel and time). Although the amount of time clearly depends on the type of analysis being performed, our experience in the Infant Neuromotor Control Laboratory, for example, is that it can take 5-8 hours to code the type of leg movements an infant makes in a 5-minute video (using 12 different codes for type of leg movement based on Thelen et al^10,11). Wearable sensors, in contrast, can capture comprehensive and detailed information about highly variable “real-world” infant behavior across days in the natural environment. And although validating the data to ensure the analysis of sensor data accurately represents the construct being measured can be time consuming, once a measure is established it can be calculated almost immediately. For example, once a company like Fitbit supplies software to calculate steps taken from their sensor, the calculation is displayed almost immediately.

We use wearable sensors to identify each infant leg movement and its duration, average acceleration, and peak acceleration. This is in contrast to activity monitor sensors, which are used to measure physical activity levels in the natural environment. Instead of identifying each distinct movement as we do, activity monitor sensors use area under the acceleration curve or acceleration threshold crossing counts per epoch of time to classify physical activity as high intensity versus low intensity versus sedentary activity. To date, activity monitors (eg, ActiGraph, ActiGraph LLC, Pensacola, FL, USA) have been validated in children aged 2 years and older to measure physical activity, but their use has not been validated in infants or children aged 2 years and younger.^12–17 To do so would require comparison of sensor measurement of infant physical activity with a gold-standard measurement, for example, observer rating of high intensity versus low intensity versus sedentary.

At the Infant Neuromotor Control Laboratory we use wearable sensors to collect more detailed quantitative data about infant leg movement than typical activity monitors allow. We are uniquely able to capture the quantity of infant leg movements, along with the type (unilateral or bilateral), duration, and peak and average acceleration of each movement across full days in the natural environment.^18,19 Due to known variability in infant performance and behavior, our goal here was to determine how many days are needed to accurately capture an infant's typical daily performance. The same issue has been raised about using activity monitors to measure physical activity levels.²⁰ Our objectives here were to determine: (1) whether 1 day is sufficient to represent an infant's typical daily performance, or whether more days are needed; and (2) whether there is a difference between weekdays and weekend days.

Methods

Participants

Data from 10 infants (male = 7, female = 3) with typical development, aged 1 to 5 months (mean [SD] age = 97.5 [41.2] days), were included. We chose to investigate the age range before sitting onset because the leg movement patterns will likely change when infants shift to using their legs to stabilize sitting. The infants were recruited by word of mouth and with fliers placed around the University of Southern California campuses and public spaces, such as local libraries. Fliers were also distributed electronically via email and social media sites. Infants from singleton, full-term births were included. Infants from multiple births, infants with a history of gestation less than 38 weeks, or infants experiencing complications during birth were excluded. Infants with any known visual, orthopedic, or neurological impairment at the time of testing were excluded. Motor development was assessed using the Alberta Infant Motor Scale.²¹ If an infant scored below the 10th percentile for their age, they were excluded. No existing data describe day-to-day variability in infant movement in a quantified way, therefore we were unable to perform a power analysis based on intrasubject variability, and the data from this study will be used to determine this information.

Procedure

This research study was approved by the Institutional Review Board of the University of Southern California (HS-16-00170). A parent or legal guardian signed an informed consent form prior to their infant's participation. Data were collected in participants’ homes using wearable sensors that collected triaxial accelerometer, gyroscope, and magnetometer data at 20 Hz (Opal sensor; APDM Inc, Portland, Oregon). Each sensor weighed 22 g and measured 48.4 × 36.1 × 13.4 mm (L × W × H) (eFig. 1, available at https://academic.oup.com/ptj). Participants wore the sensors for 7 consecutive days of full-day leg movement activity, for 7 to 13 hours per day (mean [SD] wear time = 9.46 [1.71] hours). A researcher placed the sensors on the participant's ankles on the first morning (1 sensor on each leg). A caregiver placed the sensors on the participant's ankles each morning thereafter. The caregiver removed the sensors when putting the participant to bed for the night, and plugged the sensors in to charge overnight. Caregivers were also instructed to remove the sensors if the participant was in water. If the participant had a bath before bed, the sensors were removed for the day. Each sensor was labeled for the right or left leg directly on the sensor. The family was encouraged to perform their typical daily activities. The caregiver filled out a daily survey describing whether it was or was not a typical day for their infant. All days included in our analysis were considered to be typical for the infant per caregiver report.

Data were stored in the internal memory of each sensor and were downloaded from the sensors after the seventh day using Motion Studio software (APDM Inc, Portland, OR, USA). We used Matlab programs (The MathWorks Inc, Natick, MA, USA) to calculate leg movement characteristics from the wearable sensor data. First, the algorithm determined thresholds for each leg based on the resultant acceleration and angular velocity signals across the full day. Next, the algorithm identified the beginning and end of each leg movement based on threshold crossings. A new movement was identified each time the infant paused or changed the direction of the leg movement. Once we had identified the start and end of each leg movement we were able to calculate its duration(s) and average and peak acceleration (m/s²).¹⁸ The definition and validation of this threshold-based approach is described fully in our previous publication.¹⁹ Leg movement rate was calculated as the number of leg movements per hour of awake time of sensor wear, in order to normalize across infants and days.¹⁹ We classified the infant as asleep when there were fewer than 3 leg movements in 5 minutes. In summary, we previously validated that infant leg movements can be accurately measured using wearable sensors,¹⁹ and we were addressing reliability of the wearable sensor measures in this study.

Statistics

All days were described as typical by the caregivers. No days were excluded from our analysis. We included data from when the sensors were placed on the participant in the morning until they were removed at the end of the day. One participant did not wear the sensors for 2 hours in the middle of 1 day. In this case, the algorithm counted it as sleep time. For each participant for each day, we calculated the daily average of leg movement rate per hour of awake time, duration, peak acceleration, and average acceleration of leg movements. We chose to use the average of 7 days as the standard here. We compared the standard with 6 sequential options: the average value from the first day, the average of the first 2 days, through the average of the first 6 days. There were 6 pairs of comparison in total. Consecutive days starting at the first day were chosen to reflect the most likely scenarios of home data collection. We also calculated the average of the first 2 weekdays to compare with the average of 2 weekend days. Because we wanted to see the consistency of the data from a single sensor, and the right and left legs were highly correlated across all variables (Spearman correlation coefficient (ρ): leg movement rate = 0.86; duration = 0.88; average acceleration = 0.88; peak acceleration = 0.89), we only present the right leg data here.

Bland-Altman plots were used to determine the amount of difference between each measurement option (6 sequential options: the average value from the first day, the average of the first 2 days, through the average of the first 6 days) and the standard (the average of all 7 days).^22,23 Limits of agreement were calculated and are shown in the plots. We also determined whether there was proportional bias by examining whether the magnitude of the difference was related to the value of each variable. Due to the small sample size, we used the median instead of the mean as our measure of central tendency. The absolute differences between each measurement option were compared with the standards to determine how consistency changed with adding more days to the analysis. Spearman correlation coefficients were calculated to measure the strength of the associations between the 6 sequential options and the standard. The same analysis was used to compare weekdays and weekends; the average of the first 2 weekdays was compared with the average of the 2 weekend days. Statistical tests were performed using SPSS software (Version 22; IBM Corporation, Armonk, NY, USA). Bland-Altman plots were generated using MedCalc statistical software 18.11.3 (MedCalc Software, Ostend, Belgium).

Role of Funding Source

The funding sources had no role in the design or execution of the study, nor the analyses, interpretation of the data, or decision to submit results.

Results

Leg Movement Rate

Leg movement rates from the first day to the seventh day for each participant are shown in Figure, part a. The leg movement rate ranged from 601 to 2962 movements per hour of awake time. The absolute difference between each day and its standard (average of 7 days) ranged from 0 to 453 movements per hour of awake time. Figure, part b, shows that there was no proportional bias because the magnitude of the difference is not related to the amount of leg movement. The difference between the average of the first 2 days and the standard was: median(min, max) = −56(−326, 201). All of the differences were within the limits of agreement (−348, 274). The median of the absolute difference decreased from 16.8% to 8.5% of the standard when increasing from 1 day of data to the average of the first 2 days. The Spearman correlation coefficient increased from 0.39 to 0.58 when increasing from 1 day of data to the average of the first 2 days (Tab. 1).

Figure.

a, Average leg movement rate (right leg movements rate per hour awake time) per day from the first day of data collection to the seventh day. Each line is a different participant, and lines are connected between days for visualization. b–g, Bland-Altman plots comparing the average of the first n (n = 1–6) days and the average of 7 days for right leg movement rate.

Table 1.

Average Values for Right Leg Movement Rate

Parameter	Median (Range)	Median of Absolute Difference (Range) Compared With Average for 7 d	Median of Difference (Range) Compared With Average for 7 d	Spearman Correlation for Each Method Compared With Average for 7 d
First day	1150 (601–2363)	213 (79–453)	−136 (−453 to 379)	0.39
Average for first 2 d	1210 (793–2485)	107 (34–326)	−56 (−326 to 201)	0.58
Average for first 3 d	1277 (852–2239)	48 (19–338)	−20 (−338 to 178)	0.71
Average for first 4 d	1259 (865–2262)	52 (6–289)	−19 (−289 to 84)	0.86
Average for first 5 d	1239 (862–2284)	25 (0–269)	1 (−269 to 73)	0.87e
Average for first 6 d	1256 (854–2307)	19 (2–244)	−6 (−244 to 86)	0.87
Average for 7 d	1266 (871–2284)

Duration

Average daily right leg movement durations from the first day to the seventh day for each participant are shown in eFigure 2 (available at https://academic.oup.com/ptj). The average duration of right leg movements ranged from 0.23 to 0.36 s. The absolute difference between each day and its standard ranged from 0 to 0.03 s. eFigure 3A-F (available at https://academic.oup.com/ptj) shows that there was no proportional bias because the magnitude of the difference is not related to the length of duration. The difference between the average of the first 2 days and the standard was: median(min, max) = 0.01(−0.02, 0.03). The limit of agreement was (−0.023, 0.030), and only 1 data point was outside the limits of agreement. The median of absolute difference was 3.7% of the standard when increasing the data from the average of the first 2 days. The Spearman correlation coefficient increased from 0.78 to 0.92 when increasing from 1 day of data to the average of the first 2 days (Tab. 2).

Table 2.

Average Values for Duration of Right Leg Movement

Parameter	Median (Range; s)	Median of Absolute Difference (Range; s) Compared With Average for 7 d	Median of Difference (Range; s) Compared With Average for 7 d	Spearman Correlation for Each Method Compared With Average for 7 d
First day	0.27 (0.23–0.36)	0.01 (0–0.03)	<−0.01 (−0.03 to 0.02)	0.78
Average for first 2 d	0.27 (0.25–0.36)	0.01 (0–0.03)	<0.01 (−0.02 to 0.03)	0.92
Average for first 3 d	0.27 (0.25–0.36)	<0.01 (0–0.01)	<0.01 (−0.01 to 0.01)	0.90
Average for first 4 d	0.27 (0.25–0.35)	<0.01 (0–0.01)	<0.01 (−0.01 to 0.01)	0.84
Average for first 5 d	0.27 (0.25–0.35)	<0.01 (0–0.01)	<0.01 (−0.01 to 0.01)	0.89
Average for first 6 d	0.27 (0.25–0.35)	<0.01 (0–0.01)	<0.01 (−0.01 to 0)	0.92
Average for 7 d	0.27 (0.25–0.36)

Average Acceleration

The daily averages for right leg movement average acceleration from the first day to the seventh day for each participant are shown in eFigure 4 (available at https://academic.oup.com/ptj). The average acceleration of right leg movements for each day ranged from 1.570 to 2.820 m/s². The absolute difference between each day and its standard ranged from 0.003 to 0.275 m/s². eFigure 5A–F (available at https://academic.oup.com/ptj) shows that there was no proportional bias because the magnitude of the difference is not related to the value of acceleration. The difference between the average of the first 2 days and the standard was: median(min, max) = −0.008(−0.272, 0.059). The limits of agreement were (−0.237, 0.179), and only 1 data point was outside the limits of agreement. The median of the absolute difference decreased from 5.7% to 2.8% of the standard when increasing from 1 day of data to the average of the first 2 days. The Spearman correlation coefficient was 0.77 when using the average of the first 2 days (Tab. 3).

Table 3.

Average Acceleration of Right Leg Movement

Parameter	Median (Range; m/s2)	Median of Absolute Difference (Range) Compared With Average for 7 d (m/s2)	Median of Difference (Range) Compared With Average for 7 d (m/s2)	Spearman Correlation for Each Method Compared With Average for 7 d
First day	1.888 (1.570–2.820)	0.111 (0.032–0.190)	−0.082 (−0.190 to 0.115)	0.77
Average for first 2 d	1.907 (1.58–2.760)	0.055 (0.005–0.272)	−0.008 (−0.272 to 0.059)	0.77
Average for first 3 d	1.918 (1.586–2.730)	0.050 (0.021–0.275)	−0.023 (−0.275 to 0.202)	0.81
Average for first 4 d	1.941 (1.593–2.717)	0.039 (0.012–0.160)	0.014 (−0.160 to 0.101)	0.90
Average for first 5 d	2.006 (1.624–2.694)	0.030 (0.003–0.101)	−0.003 (−0.064 to 0.101)	0.95
Average for first 6 d	1.961 (1.643–2.740)	0.028 (0.004–0.093)	0.014 (−0.093 to 0.060)	0.95
Average for 7 d	1.964 (1.627–2.705)

Peak Acceleration

The daily averages of the right leg movement peak acceleration values from the first day to the seventh day for each participant are shown in eFigure 6 (available at https://academic.oup.com/ptj). The average peak acceleration of right leg movements ranged from 2.793 to 5.635 m/s². The absolute difference between each day and its standard ranged from 0.002 to 0.606 m/s². eFigure 7A–F (available at https://academic.oup.com/ptj) shows that there was no proportional bias because the magnitude of the difference is not related to the value of acceleration. The difference between the average of the first 2 days and the standard was: median (min, max) = 0.054 (−0.606, 0.144). The limits of agreement were (−0.550, 0.449), and only 1 data point was outside the limits of agreement. The mean absolute difference decreased from 6.7% to 3.8% of the standard when increasing from 1 day of data to the average of the first 2 days. The Spearman correlation coefficient was 0.67 when using the average of the first 2 days (Tab. 4).

Table 4.

Average Peak Acceleration of Right Leg Movement

Parameter	Median (Range; m/s2)	Median of Absolute Difference (Range) Compared With Average for 7 d (m/s2)	Median of Difference (Range) Compared With Average for 7 d (m/s2)	Spearman Correlation for Each Method Compared With Average for 7 d
First day	3.490 (2.821–5.635)	0.251 (0.011–0.438)	−0.248 (−0.438 to 0.390)	0.79
Average for first 2 d	3.576 (2.793–5.390)	0.143 (0.009–0.606)	0.054 (−0.606 to 0.144)	0.67
Average for first 3 d	3.619 (2.941–5.326)	0.145 (0.081–0.588)	0.088 (−0.588 to 0.444)	0.69
Average for first 4 d	3.627 (2.990–5.251)	0.097 (0.006–0.319)	−0.002 (−0.319 to 0.221)	0.90
Average for first 5 d	3.784 (3.072–5.195)	0.058 (0.003–0.221)	−0.011 (−0.089 to 0.221)	0.96
Average for first 6 d	3.729 (3.108–5.325)	0.054 (0.002–0.127)	0.033 (−0.127 to 0.118)	0.96
Average for 7 d	3.769 (3.070–5.245)

Weekdays Versus Weekends

Table 5 shows the comparison of the average of the first 2 weekdays with the average of the 2 weekend days. The median absolute difference between weekdays and weekends for leg movement rate, duration, average acceleration, and peak acceleration was 11.6%, 3.7%, 7.2%, and 7.3% of the corresponding standard. The Spearman correlation coefficients between the average of first 2 weekdays and the average of the 2 weekend days showed weak to moderate strength of correlation (0.36–0.44). Bland-Altman plots for each variable showed that 90% to 100% of the data points were within the limits of agreement (eFigs 8–11, available at https://academic.oup.com/ptj).

Table 5.

Movement Rate, Duration, and Acceleration (Average and Peak) of Right Leg Movements on Weekdays vs Weekends

Parameter	Median (Minimum to Maximum) on:		Median Absolute Difference Between Weekend Days and Weekdays (Range)	Median Difference Between Weekend Days and Weekdays (Range)	Spearman Correlation for Leg Movement Rate, Duration, Average Acceleration and Peak Acceleration Between Weekend Days and Weekdays
	Weekend Days	Weekdays
Leg movement rate	1238 (601–2607)	1281 (781–2471)	147 (0–607)	40 (−329 to 607)	0.44
Duration: second	0.27 (0.23–0.36)	0.27 (0.23–0.36)	0.01 (0–0.05)	0.01 (−0.02 to 0.05)	0.36
Average acceleration: m/s2	1.893 (1.570–2.820)	2.013 (1.503–2.968)	0.141 (0–0.523)	0.060 (−0.353 to 0.523)	0.38
Peak acceleration: m/s2	3.566 (2.821–5.635)	3.853 (2.765–5.973)	0.276 (0.024–1.326)	0.048 (−0.773 to 1.326)	0.36

Discussion

Our goal here was to determine the reliability of our measures of infant leg movement characteristics across days. To the best of our knowledge, our results also provide the first report of average daily infant leg movement behavior across 7 successive typical days. Given the lack of research quantifying variability in infant behavior across full days, it is difficult to define a cutoff standard for reliability that is “good enough.” Not surprisingly, our results revealed that the reliability of wearable sensor values increased with each additional day of data collected. However, this needs to be balanced against the burden to the family and increased use of resources. Our assessment is that 2 days is the “best bet.” After 2 days, the absolute difference between the average of the first 2 days and the standards of leg movement rate, duration, average acceleration, and peak acceleration all fell below 10% (8.5%, 3.7%, 2.8%, and 3.8%, respectively). The Spearman correlation coefficients comparing the average of the first 2 days and the standards are all above 0.6, and 2 days is the shortest time needed to avoid a significant difference between the 2 methods. Using more than 2 days did not greatly improve the results and would increase burden to the family and use of resources.

Previous researchers have tested the reliability of accelerometers in quantifying levels of physical activity across different age groups. For example, Pedersen et al found that 4.7 days for work and 5.5 days for leisure time were needed to reliably estimate adults’ activity patterns using ActiGraph.²⁴ Another study found that 4 days were needed to achieve desirable reliability estimates using accelerometers to measure physical activity levels in 9- to 11-year-old children.²⁵ For studies measuring preschooler's physical activity and sedentary time, the minimum time to achieve an acceptable measurement using accelerometers was 3 days per week.^26,27 One study from Pitchford et al investigated infant physical activity level using an ActiGraph. They found that 2 days and 12 hours of data collection at the ankle, and 3 days and 15 hours per day at the wrist gave reliable estimates for infants’ physical activity level.²⁰ Notably, these studies all measured intensity of physical activity as the dependent variable, which is different than our dependent variables. Despite the different variables being assessed, our study is consistent with the finding of the Pitchford et al study, because we both found that ∼2 days of data provide reliable measurements of infant leg movement activity across days. Together, these findings suggest that overall measures of infant activity are more similar from day to day than measures in older children and adults. These findings do not; however, explore specific behaviors within a day or fluctuations in behavior across minutes and hours, they only generalize very broadly across a full day.

We did not find differences between weekend days and weekdays in infant movement characteristics; median absolute differences between weekdays and weekends for leg movement rate, duration, average acceleration, and peak acceleration were 11.6%, 3.7%, 7.2%, and 7.3% of the corresponding standard. Although infants do not have workday or school-day schedules themselves, their behavior could possibly be influenced by their caregivers’ and siblings’ differing schedules. Adults were shown to have greater habitual physical activity during weekdays compared with weekend days using pedometer- or accelerometer-determined physical activity.^28,29 However, conflicting results were found in preschoolers. Byun et al found that preschool children spent more time in sedentary behavior on weekdays than at weekends, whereas Bingham et al found no difference between them.^30,31 These different results could be caused by different participant ages and preschool activities (mean age = 2.9 years in the Bingham et al study and no sitting study time vs mean age = 4.1 years in the Byun et al study and sitting study time occurred). There was also a difference in mean wear time between the 2 studies (14 hours in the Bingham et al study vs 8.4 hours in the Byun et al study). Of note, the reliability between the average of 2 weekend days and the average of the first 2 weekdays is lower than when comparing the average of the first 2 days with the average of 7 days. This is not surprising, because in the former case the average of 2 days is being compared with the average of 2 days, whereas in the latter case the average of 2 days is being compared with the average of 7 days. An estimate from 7 days is likely to be closer to the ground truth than an estimate of 2 days, and this finding is reflected in our reliability measures. Our recommendation is that it is appropriate to include weekend days when the goal of the analysis is to describe an infant's typical day.

Wearable sensors now allow the collection and analysis of infant movement behavior across days and weeks. With the progressively falling cost of wearable sensors, they could become as ubiquitous as mobile phones, providing the opportunity for mobile health platforms and widespread use. By creating a platform for collecting and analyzing information about quantity and type of infant movements provided by wearable sensors, the opportunity exists to learn more about the relationship between movement experience and skill development. Further, there is an opportunity to promote, monitor, and improve skill learning and physical functioning outside formal training times, such as therapy sessions.³²

Using 4 characteristics (leg movement rate, duration, acceleration, and peak acceleration) from infant leg movements across days, we suggest that the best balance between minimal wear time and increasing reliability of measurements with more data is 2 consecutive days when the goal is to describe typical daily activity. Weekdays and weekends were not different, and in all cases the caregiver described the day as typical for their infant. How large a deviation is necessary to label a day as “atypical” is currently unknown.

Limitations

The small sample size and different ages among the participants limit the generalizability of the study. We limited the study to infants aged 1 to 5 months not yet sitting independently. At sitting onset, we expect infants’ leg movement behavior to change as they use their legs for stability. A 1-month-old infant can still differ from a 5-month-old infant, however, and our sample size was not large enough to determine this. Moreover, we did not assess whether performance of developmental skills changed during this week-long study. Regarding sensor measurement, a common concern is error caused by the presence of external motion (caregiver handling or mechanical movement).^24,33 In this study, there was no reason to suspect that these errors would be different across typical days; however, this is worth investigating in future studies. It would also be interesting to measure how changes in variability across days and weeks relate to changes in performance of developmental skills.

Conclusions

To represent infants’ typical daily leg movement behavior, 2 consecutive days of wear time is the best option to balance increased reliability/sufficient data with resource use/participant burden. The difference between the average of the first 2 days and the standard (average of 7 days) is consistently smaller across measures than using only the first day, and using more than the first 2 days does not appear to reduce differences much more. These results will inform the rapidly expanding use of wearable sensors to measure full-day infant behavior for infant neuromotor assessment and outcome measurement. Future research should assess different age groups or functional levels during infancy. Another important question is what is the minimum length of data acquisition within a day that is necessary to represent the leg movement patterns of that whole day.

Author Contributions and Acknowledgments

Concept/idea/research design: B.A. Smith

Writing: W. Deng, B.A. Smith

Data collection: B.A. Smith

Data analysis: W. Deng, I.A. Trujillo-Priego, B.A. Smith

Project management: B.A. Smith

Fund procurement: B.A. Smith

Providing participants: B.A. Smith

Providing facilities/equipment: B.A. Smith

Consultation (including review of manuscript before submitting): I.A. Trujillo-Priego, B.A. Smith

We thank all the infants and their families who participated in this study.

Ethics Approval

This research study was approved by the Institutional Review Board of the University of Southern California (HS-16-00170). A parent or legal guardian signed an informed consent form prior to each infant's participation.

Funding

This research was supported by a grant from the CAL-PT FUND (PI: B.A. Smith). Additionally, B.A.S. was supported in part by NIH K12-HD055929 (PI: K. Ottenbacher). Statistical consultation was supported in part by funding from the National Institutes of Health from the National Center for Advancing Translational Science UL1TR001855 and UL1TR000130 (PI: T. Buchanan). I.A.T-P. is partially supported by Consejo Nacional de Ciencia y Tecnología (CONACYT, Mexico). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Disclosure and Presentations

The authors completed the ICJME Form for Disclosure of Potential Conflicts of Interest and reported no conflicts of interest.

Preliminary data from this paper were presented at the 2017 California Physical Therapy Association Annual Conference, San Diego, California; and internally at the Dentistry Research Day at the University of Southern California, 2018, Los Angeles, California.