Abstract
Purpose:
To assess the concurrent validity of 2 wearable sensors, GENEActiv (GA) and MonBaby (MB), for tracking tummy time in full-term and preterm infants at home.
Methods:
Nineteen full-term infants and 13 infants born preterm, aged 3 to 6 months, wore GA and MB sensors during caregivervideo-recorded active play at home over 3 days. Prone (tummy time) durations detected by each sensor were compared to video for validation.
Results:
Both GA and MB sensors demonstrated excellent (κ = 0.86) and substantial (κ = 0.78) second-by-second agreement with video, respectively, for tracking tummy time. For cumulative tummy time, the GA showed higher accuracy with video (GA =60 minutes, video =58 minutes, difference =2 minutes) compared to MB (MB =43 minutes, video =47 minutes, difference =4 minutes). No differences in accuracy were found between the 2 sensors.
Conclusion:
Wearable sensors can accurately track tummy time at home and can support adherence to movement guidelines in infants. Establishing caregiver feasibility is crucial for broader use.
Keywords: tummy time, wearable sensors, preterm infant, infant motor development, movement guidelines
INTRODUCTION
Movement experiences in diverse play positions like supine, prone, and sitting are foundational for infant development. These experiences build antigravity muscle strength, optimize muscle length,1 and improve visual access to objects and people.2 Together, these benefits support environmental exploration and cognitive learning, laying the groundwork for essential developmental skills.2 In the first months of life, infants rely on their caregivers to provide these positioning experiences. This is particularly relevant for infants born preterm (<37 weeks of gestation), who often experience hypotonia and muscular weakness that may limit their ability to actively engage in positioning.3 Moreover, parents of preterm infants may be more cautious and may avoid engaging their babies in physical activities due to concerns about their infant’s perceived fragility or a lack of confidence in handling.4 Such cautious behavior may inadvertently reduce opportunities for optimal positioning and may exacerbate the risk of delays in preterm infants.
National organizations, such as the American Academy of Pediatrics (AAP), guide caregivers in optimizing infant positioning. The AAP advocates that 30 to 60 minutes of “tummy time” or supervised and awake “prone play” is the primary recommended movement experience for infants under 6 months of age.5 While tummy time is usually assessed through caregiver-recalls, this approach can be challenging and labor-intensive for caregivers due to the sporadic nature of tummy time throughout the day. A recent study validating caregiver recalls against direct observation found only a moderate correlation, with caregivers of preterm infants overestimating tummy time by approximately 22 minutes per day.6 These findings indicate a significant discrepancy and highlight the need for objective tummy time tracking measures for young infants.
Wearable sensors present a promising solution to quantify tummy time. Sensors that include accelerometers and gyroscopes provide objective data and can estimate body positions during functional activities in infants.7–10 For instance, Greenspan et al utilized pitch angle cut-points to classify infant positions, including prone, using a single DIY Arduino sensor during a researcher-supervised free play assessment at home.7 Franchak et al explored the feasibility of classifying infant positions using 4 sensors and a machine learning approach during unsupervised play sessions at home with 2 older infants.8 Hewitt et al employed 2 commercially available sensors, GENEActiv (GA) and MonBaby (MB), to specifically track the prone position in controlled laboratory settings by analyzing 360° body rotation and elevation angles from scatter plots of the x-, y-, and z-axes.9
Hewitt et al’s study, which focused specifically on tummy time, serves as a key reference for our research.9 We followed the sensor validation framework by Keadle et al., which involves 4 phases. Phase 0 and Phase I are conducted in controlled settings to test the sensor signal and develop processing methods like cut-points or algorithms. Phase II involves semi-structured evaluation with researcher control. Phase III comprises rigorous free-living validation in natural settings, and Phase IV involves the adoption of validated sensors in applied studies.11 Hewitt et al completed Phase 0 testing for the GA (Activinsights Ltd) and MB (MonDevices, Inc) sensors with the manufacturers and performed Phase I testing with infants in a lab setting. In the Hewitt study, infants wore the 2 sensors, while their caregivers placed them in 12 different positions, including prone, for 3 minutes each. These positions were recorded on video, axis thresholds were established, and algorithms to identify the prone position were developed. The study found that the GA sensor’s algorithm was more accurate (95%) in detecting tummy time compared to the MB sensor (79%).9
This raises several important questions. Can laboratory-derived sensor algorithms accurately capture natural tummy time transitions, such as pivoting and crawling, which occur in real-life settings? How might the absence of researcher supervision at home affect the accuracy of these algorithms? Finally, can findings from full-term infants be generalized to at-risk preterm infants?
Our study aimed to evaluate the validity of 2 wearable sensors, GA and MB, for measuring tummy time in home settings through Phase II (researcher-supervised) and Phase III (caregiver-driven) testing. Our secondary aim was to determine if there are differences in accuracy between the 2 sensors and if accuracy varies based on birth status.
METHODS
This observational cross-sectional study included a convenience sample of 32 infants (19 born full term and 13 born preterm), recruited from the Virginia Commonwealth University Health systems and the surrounding community. Eligibility criteria included infants aged between 3 and 6 months (corrected age for preterm infants), tolerance to tummy time as reported by parents—defined as crying for no more than 30 seconds when placed on the tummy, and if the caregivers were at least 18 years of age and spoke English. Infants with medical conditions that prevented them from lying on their tummies were excluded. Written informed consent was obtained from 1 parent or legal guardian, and the study was approved by the Institutional Review Board.
Measures
Two laboratory-validated wearable sensors, the GA (Activinsights Ltd, Cambridgeshire, UK) and the MB (MonDevices, Inc, New York, NY, USA), were used (see Figure 1 for sensor features). The GA device is a smartwatch that operates without a display, collects data at a frequency of 30 Hz, and does not require charging throughout the 3-day period at the given frequency. In contrast, the MB sensor is a baby monitor that interfaces with a smartphone application (iOS and Android) to display the infant’s position in real time, collects data at a frequency of 6.25 Hz, and requires daily recharging via a portable charger. A Panasonic HC-770 camcorder (1080p at 30 frames per second) mounted on a tripod was used as the gold standard for direct observation.
Fig. 1. Placement and technical features of the tummy time sensors. MonBaby is attached to the center of the chest using a snap-on ring, and GENEActiv is attached to the anterolateral right hip using an elastic strap.*Parental consent obtained for image publication.
Study procedures
This 2-phase (Phases II and III) study was conducted in participants’ homes over 3 days.
Phase II: researcher-guided semi-structured home validation.
Phase II began on Day 0 or Day 1, following the provision of study materials, which included the 2 sensors, a camcorder, and an illustrated manual. Due to COVID-19 restrictions (April 2021-March 2022), caregivers could choose between in-person setup and doorstep equipment drop-off with remote guidance, utilizing Zoom Video Communications (Version 5.13.11). Caregivers were instructed on sensor activation, camera placement, and synchronization. They synchronized sensor-camera timestamps preceding and post each play session by shaking both sensors in front of the camera for 3 to 5 seconds before attaching them to their infant’s clothing. A motor skills assessment using the Alberta Motor Infant Scale (AIMS) was recorded for validation purposes. As the AIMS is an observational tool, it facilitated the simulation of natural infant play in 4 positions: supine, prone, sitting, and standing.12 The recorded AIMS session was used to compare the duration of “prone/tummy time” captured on video with that recorded by the sensors. The entire session lasted 15 to 30 minutes, after which the researcher left the participant’s home or ended the Zoom session.
Phase III: caregiver-driven, unstructured, home validation.
In Phase III, the caregivers managed synchronization events, sensor attachment, and recorded their infant’s play sessions over 3 consecutive days. Play sessions were defined as times when the infant was awake and actively playing with the caregiver and could occur anywhere in the home (eg, bed, crib, or floor). The duration of each play session was left to the caregiver’s discretion, with instructions to replicate their daily play routines as closely as possible. For privacy, sensors and video were deactivated during clothing changes, diaper changes, bathing, and sleep periods. To minimize attachment errors, caregivers received printed checklists, twice-daily text reminders, and round-the-clock troubleshooting support via phone calls or Zoom sessions. On Day 3, the researcher conducted doorstep pickup.
Data coding and processing
Video data coding.
Infant video data were labeled using Datavyu software, categorizing position variables into “prone” (such as prone with head down, on forearms or hands, various forms of prone mobility, and quadruped), “prone-supported” (lying on the stomach with supportive devices or inclined on a parent’s chest), and “non-prone positions” (supine, side lying, sitting upright with or without support, and supported standing).9 Two trained coders completed the coding, with 20% of videos secondarily coded for reliability (Inter-rater ICC = 0.93; Intra-rater ICC = 0.97). Daily durations of prone and prone-supported positions were calculated using a customized Excel position calculator and summed to obtain the total 3-day prone duration for video analysis.
Sensor data processing.
Raw data from both sensors were exported as an Excel sheet, containing timestamps and acceleration values for the x-, y-, and z-axes (measured in g, where 1 g= 9.8 m/s2). For Phase II, second-by-second agreement was established to assess the sensors’ ability to distinguish between prone and non-prone positions during natural play transitions. Timestamp drift between sensors and video was assessed using synchronization events in a subset of participants, ie, 20% of randomly selected infants from both phases.
For both phases, cumulative agreement between video and sensor data for the prone position was evaluated by comparing total prone and prone-supported times from video and sensors using customized MATLAB scripts (see Supplemental Digital Content Data 1, available at: http://links.lww.com/PPT/A651 and Supplemental Digital Content Data 2, available at: http://links.lww.com/PPT/A652) based on the laboratory-validated algorithm.9 For Phase III, daily prone durations were summed to obtain cumulative 3-day prone times.
The GA sensor’s algorithm analyzes a scatter plot with 2 key angles: (1) rotation (360° angle) on the x-axis representing the body’s orientation in the horizontal plane and (2) elevation (up/down angle) on the y-axis representing the body’s tilt or incline in the vertical plane. These angles classified positions into prone, prone-supported, and non-prone categories. For the MB sensor, a 360° angle was calculated using the following formula in Excel: =ACOS(E3/SQRT(C2*C2 + D2*D2 + E2*E2)) * 180/PI(), where the numbers represent the row on the Excel spreadsheet, and C, D, and E are the columns containing x-, y-, and z-axis data. Angles less than 134° were categorized as non-prone positions. Prone and prone-supported positions were identified using a z-axis threshold of less than −0.10 g.9
Data analysis
Descriptive statistics summarized participant characteristics. Spearman’s correlation coefficients (ρ) assessed associations between sensors and video, categorized as weak (0.1–0.4), moderate (0.4–0.7), or strong (0.7–1.0).13 For Phase II, second-by-second agreement was evaluated using weighted kappa, interpreted as excellent (k = 0.81–1.0), substantial (k = 0.61–0.80), moderate (k = 0.41–0.60), fair (k < 0.21–0.40), or poor (k < 0.20).14 During Phase III, significant data loss (~45%) in the MB sensor due to Bluetooth disconnections resulted in shorter video durations for comparison. To account for this, separate linear mixed models (LMMs) were constructed to evaluate absolute differences in cumulative tummy time between each sensor and their time-matched videos, adjusting for birth status (full-term vs preterm). Another LMM tested differences in tummy time accuracy between GA and MB. Effect sizes (Cohen’s d) in LMM were calculated and interpreted as small (0.2), medium (0.5), or large (0.8).15 Sensor accuracy was clinically interpreted using a predefined acceptable limit of ±10% difference (absolute difference/direct observation value × 100), based on previous physical activity studies.16,17 Bland-Altman plots and limits of agreement visualized systematic bias between sensors and video.18 Statistical significance was set at ≤0.05 (JMP® Pro v15.1.0, MedCalc® v20.218, SPSS v29).
RESULTS
The final sample comprised 32 infants (n =19 full term, mean age = 5.35 months, SD = 1.17; n =13 preterm, mean adjusted age = 4.60 months, SD = 1.02). Descriptive statistics are presented in Table 1. Across both validation phases, no infants were placed in prone-supported positions, so this category was merged into the prone category for analysis. Sensor and video-recorded prone durations for each infant across the 2 phases are presented in a box-plot provided in Supplemental Digital Content Data 3, available at: http://links.lww.com/PPT/A653. Drift calculations, across the 2 phases, revealed that the GA sensor had a drift of 4 seconds (SD = 3 seconds) over an average 31.3-minute recording, while the MB sensor had a drift of 1 minute 17 seconds (SD = 1 minute 30 seconds) over an average 29.4-minute recording.
TABLE 1.
Participant Characteristics (N = 32)
| Term Infants (n = 19) | Preterm Infants (n = 13) | |
|---|---|---|
|
| ||
| Infant demographics | ||
| Age in monthsa | 5.35 (1.2) | 4.60 (1.0) |
| Birth weight in lbs | 6.9 (0.9) | 3.96 (1.7) |
| Current weight in lbs | 14.5 (1.5) | 14.5 (11.3, 16.1)b |
| Head circumference in cm | 40.8 (39.1,43.9)b | 41.5 (2.8) |
| Body length in cm | 63.3 (3.6) | 62.5 (3.9) |
| Gestational age in weeks | ||
| >37 | 19 (100%) | 0 (0%) |
| 32–37 | 0 (0%) | 6 (46.2%) |
| 28–32 | 0 (0%) | 5 (38.4%) |
| <28 | 0 (0%) | 2 (15.4%) |
| AIMS scores | ||
| Prone domain | 7.47 (1.4) | 6.69 (2.5) |
| Total AIMS score | 19.84 (4.2) | 16.54 (4.5) |
| Percentile (range, min to max) | 10th to 90th | <5th to 50th |
| Ethnicity | ||
| Hispanic | 0 (0%) | 0 (0%) |
| Non-Hispanic | 19 (100%) | 11 (84.6%) |
| Not reported | 0 (0%) | 2 (15.%) |
| Race | ||
| White | 15 (78.9%) | 4 (30.8%) |
| Black | 0 (0%) | 4 (30.8%) |
| Asian | 0 (0%) | 1 (7.7%) |
| Multi-racial | 4 (21.1%) | 1 (7.7%) |
| Not reported | 0 (0%) | 3 (23.0%) |
| Caregiver demographics | ||
| Age in years | ||
| 18–35 | 16 (84.2%) | 9 (69.2%) |
| 36–55 | 1 (5.3%) | 3 (23.1%) |
| 56+. | 1 (5.3%) | 0 (0%) |
| Not reported | 1 (5.3%) | 1 (7.7%) |
| Ethnicity | ||
| Hispanic | 0 (0%) | 11 (84.6%) |
| Non-Hispanic | 19 (100%) | 1 (7.7%) |
| Not reported | 0 (0%) | 1 (7.7%) |
| Race | ||
| White | 17 (89.5%) | 5 (38.5%) |
| Black | 0 (0%) | 4 (30.8%) |
| Asian | 1 (5.3%) | 1 (7.7%) |
| Multi-racial | 1 (5.3%) | 0 (0%) |
| Not reported | 0 (0%) | 3 (23.1%) |
Abbreviation: AIMS, Alberta Infant Motor Scale.
Adjusted age is reported for infants born preterm, and means (standard deviations) are reported for normally distributed continuous variables.
Medians and interquartile range are reported for not normally distributed continuous variables, and frequencies (percentages) are reported for categorical variables.
Phase II: researcher-guided semi-structured home validation
Three caregivers chose remote testing via video conferencing, while the remaining 29 selected in-person testing. Data from 3 infants were missing due to fussiness during prone play and poor video quality. An additional 3 datasets were missing for the MB: 2 due to Bluetooth errors and 1 data download error (N: GA = 29, MB = 26).
Both the GA (ρ = 0.97, P < .0001) and the MB sensors (ρ = 0.88, P < .0001) exhibited significant, strong positive correlations with video for measuring prone time (Table 2). For the second-by-second agreement between the sensors and video, the GA showed “excellent” agreement (k = 0.858, 95% CI = 0.813–0.904) and the MB showed “substantial” agreement (k = 0.789, 95% CI = 0.705–0.873).
TABLE 2.
Spearman’s Correlations Between Sensors and Video Stratified by Birth Status
| Full-term Infants |
Preterm Infants |
|||
|---|---|---|---|---|
| n | ρ (P-value) | n | ρ (P-value) | |
|
| ||||
| Phase II: researcher-guided | ||||
| GENEActiv vs video | 17 | 0.98 (<.0001) | 12 | 0.99 (<.0001) |
| MonBaby vs video | 15 | 0.96 (<.0001) | 11 | 0.72 (.01) |
| Phase III: caregiver-driven | ||||
| GENEActiv vs video | 17 | 0.98 (<.0001) | 13 | 0.99 (<.0001) |
| MonBaby vs video | 19 | 0.97 (<.0001) | 13 | 0.99 (<.0001) |
ρ = Spearman’s rho, interpreted as weak (0.1–0.4), moderate (0.4–0.7), or strong (0.7–1.0), P < .05 is statistically significant.
For cumulative prone durations Table 3, compared to video, the GA overreported tummy time by 0.04 minutes (P −= .83, d = 0.04) in a 6.5-minute recording, while the MB sensor underreported tummy time by 0.4 minutes (P = .59, d = 0.10) in a 4.5-minute recording. These differences did not vary by birth status (P > .05). The percentage differences between sensor and video were within acceptable limits (±10%) for both sensors (GA = +0.61%, MB = −6.5%).
TABLE 3.
Least Square Mean Estimates for Sensor vs Video Comparisons for Tummy Time Duration (in Minutes)
| Phase II: Researcher-Guided |
Phase III: Caregiver-Driven |
|||
|---|---|---|---|---|
| Estimate (SE) | 95% CI | Estimate (SE) | 95% CI | |
|
| ||||
| GENEActiv vs video | ||||
| GENEActiv prone time | 6.53 (0.97) | (4.54, 8.52) | 60.39 (8.80) | (42.39, 78.39) |
| Video prone time | 6.49 (0.97) | (4.50, 8.48) | 58.63 (8.80) | (40.63, 76.64) |
| MonBaby vs video | ||||
| MonBaby prone time | 4.06 (0.53) | (2.97, 5.15) | 43.50 (6.20) | (30.85, 56.15) |
| Video prone time | 4.46 (0.53) | (3.36, 5.55) | 46.98 (6.20) | (34.33, 59.63) |
Abbreviations: CI, confidence interval; SE, standard error.
Note: GENEActiv overestimated tummy time compared to the video, and MonBaby underestimated tummy time compared to the video.
Phase III: caregiver-driven, unstructured, home validation
Both sensors encountered instances of data loss, either complete (resulting in no sensor recordings across the 3 days) or partial (involving bouts of data loss during 1 or more tummy time sessions across the 3 days). These partial sections were subtracted from time-matched video sections, and the remaining data was included in the analyses. For the GA, 2 infants (6.3%) had complete data loss due to caregivers accidentally switching off the sensor (N = 30). Additionally, 5 infants (15.6%) had partial data loss due to similar issues and sensor displacement altering accelerometer axes. No complete data loss occurred for the MB (N = 32), but partial data loss was observed in 15 infants (47%), mainly due to Bluetooth disconnection from the caregiver’s phone and failure to charge the sensor overnight.
The GA (ρ = 0.99, P ≤ .0001) and the MB sensors (ρ = 0.97, P ≤ .0001) demonstrated significant, strong positive correlations with time-matched video (Table 2). No significant difference in sensor accuracy were found across the 3 testing days for either the GA (P = .61) or MB (P = .13) sensors and data from the 3 days were combined for the final analysis. Additionally, no significant differences in sensor accuracy by birth status were observed (GA, P = .48; MB, P = .52).
The GA overestimated tummy time by an average of 1.76 minutes (P = .08, d = 0.3) in a 58.6-minute recording. The MB underestimated tummy time by 3.5 minutes in a 46.9-minute recording, and this underestimation was statistically significant (P = .007, d = 0.5) (Table 3). Despite larger absolute differences compared to Phase II, these remained within clinically acceptable limits (GA = +3% and MB = −7.5%). While directly comparing the sensors, no significant difference in accuracy was found (difference = 1.4 minutes, SE = 0.87, P = .25, d = 0.2). This indicates that, while GA had greater precision with the video individually, the overall difference in performance between the 2 sensors was not statistically significant.
The Bland-Altman analysis mirrored the LMM analyses. As seen in Figure 2, the MB exhibited a significant (P = .007) systematic bias compared to video. This bias increased as tummy time durations increased (evident from the dispersion of points moving away from the zero line for durations exceeding 30 minutes on the x-axis; Figure 2b). In contrast, the GA had a non-significant bias and narrower limits of agreement, indicating better precision in its measurements. However, both sensors displayed some outliers beyond these limits of agreement, suggesting occasional deviations from the expected accuracy.
Fig. 2. Bland-Altman plots with 95% limits of agreement between the sensors and video for tummy time durations (Phase III; GENEActiv, N = 30; MonBaby, N = 32). Differences between the sensor and video (y-axis), average of sensor and video (x-axis), the solid-orange line represents the zero line or the point where the sensor and video have no differences, the solid-blue line represents the bias, the solid brown lines show the upper (mean +1.96 SD) and lower (mean −1.96 SD) 95% limits of agreement, individual data points for full term infants are seen as orange circles, and purple squares for preterm infants. Tummy time duration is reported in minutes. Abbreviations: FT, full-term infants; PT, preterm infants; TT, tummy time.
DISCUSSION
This study established the validity of 2 commercially available sensors, GA and MB, for tracking tummy time in infants across 3 days in home settings. Our findings build on previous research on wearables for infants by extending validation beyond controlled laboratory settings9,10 and researcher supervision.7 Both sensors were strongly correlated with video for tracking tummy time in home environment. However, the GA sensor demonstrated greater individual accuracy with video compared to MB sensor, which showed limitations in long-duration recordings. Sensor accuracy was consistent across full-term and preterm infants.
During Phase II (researcher-guided) testing, both sensors demonstrated excellent to substantial second-by-second agreement with video, highlighting their ability to precisely identify bouts of prone positioning during natural play transitions at home. These findings are consistent with those of Greenspan et al7 and Franchak et al,8 who reported similar levels of agreement using DIY and commercial sensors, respectively, for tracking infant positions including prone, during supervised play sessions in infants. While the second-by-second agreement was comparable between the 2 sensors, drift analysis revealed key differences that likely influenced their performance over time. The GA exhibited minimal drift (~4 seconds), whereas the MB displayed larger drift (>1 minute) over an average 30-minute recording. This discrepancy may stem from differences in data transmission technologies, with MB relying on Bluetooth Low Energy (BLE), which was susceptible to timestamp inaccuracies due to connection disruptions. Over time, this drift likely compounded, resulting in a misalignment between MB and video data that may have reduced second-by-second accuracy, particularly in longer recordings. Previous studies that evaluated wearable sensors for tracking infant motor activity have often focused on short-duration recordings or laboratory-based settings, where drift is less likely to occur.9,10 Our results emphasize the need for future studies to evaluate sensor accuracy over prolonged periods and to account for drift as a potential source of error, particularly for BLE-based devices.
In addition to the drift, we found that the MB misclassified certain seated positions like prop sitting as prone. This misclassification likely reflects differences in sensor placement, with GA positioned laterally to the hip and MB placed at the center of the chest. For younger infants, prop sitting often corresponded to the infant’s chest leaning forward toward the floor, leading MB to detect these bouts as prone. These results underscore the importance of selecting appropriate sensor placement for the target activity. Based on current literature and our findings, when using a single sensor for position/posture tracking in infants, lateral placements at the hip or ankle are preferred.7,19
Despite the second-by-second discrepancies, both the GA and MB sensors demonstrated comparable accuracy for tracking cumulative tummy time durations in home settings in Phase III (caregiver-driven), achieving an accuracy of >90%. This cumulative accuracy was similar to that observed in Hewitt et al (Phase I),9 hinting that the cumulative tummy time accuracy was less affected by drift. As with the second-by-second agreement, the GA had a greater cumulative accuracy with video for tracking tummy time compared to the MB. The MB also exhibited systematic bias, with its accuracy diminishing as tummy time durations exceeded 30 minutes. This indicates a potential limitation of the MB in accurately capturing prolonged periods of tummy time, which should be considered when utilizing the device.
Apart from algorithm-related issues, we encountered several feasibility challenges including a substantial proportion of partial data loss for the MB due to loss of Bluetooth connection and failure to charge. For the GA, placement and sensor activation errors led to complete or partial data loss across the 3 days. The robustness of wearable data depends on the motion-sensing platform and transmission method employed.19 Our results suggest that beyond assessing validity, it is crucial to evaluate feasibility in terms of sensor placement, activation, charging, transmission, and data download for unsupervised sensor use at home. These findings will be further detailed in a separate paper.
Implications for practice and research
Our findings have several implications for clinical and research practice: (1) Tummy time monitoring—Both the GA and MB sensor algorithms can be used to accurately identify tummy time bouts at home without researcher supervision. Confidence in these findings will be further strengthened after discussing the feasibility barriers and their potential solutions for home use. Objective monitoring of tummy time can assist clinicians, therapists, and public health researchers in identifying the percentage of at-risk infants who adhere to AAP-recommended tummy time guidelines.5 This could facilitate early detection of atypical tummy time skills in preterm infants and guide their referral to early intervention services. (2) Dose-response relationships—These wearables can be used to track dose-response relationships between tummy time and health outcomes in infants. For instance, results from caregiver-recall of tummy time show that full-term infants engaging in over 1 hour and 21 minutes of tummy time daily tend to achieve gross motor milestones earlier compared to those with less tummy time.20 Those who surpass 12 minutes of tummy time daily at 2 months of age tend to have a lower body mass index at 4 months.21 Similarly, very preterm infants (<32 weeks gestation) who average 46 minutes of tummy time daily exhibit higher scores on motor assessments compared to very preterm infants who predominantly play in the supine position by 8 months corrected age.22 Given the low accuracy of caregiver-recalls, these important findings can be confirmed using the GA or MB sensors and can guide the development of precise interventions to target specific health outcomes in infancy. (3) Intervention studies—After establishing the feasibility of these devices, they can be implemented in intervention studies both as outcomes and intervention delivery modes. For instance, a recent study used the GA sensor to assess the feasibility and effectiveness of a tummy time intervention at home.23 Physical activity studies in adults often use wearables like Fitbit to generate biofeedback reports for modifying behaviors.24 Similar reports can be used to design targeted tummy time goals for infants and could encourage the involvement of caregivers in goal-setting and intervention delivery.
Limitations
Although positioning devices like Boppy Pillows are often used for prone play, none of the caregivers in our study used them, so we could not assess their impact on sensor accuracy. Due to the video-recording requirement, caregivers preferred floor-based prone play, and no prone-supported positions on the caregiver’s chest or lap were documented.
Our study is the first to report concurrent validity between wearable sensors and video to identify prone positioning in infants born preterm. However, the small sample size and variability in gestational ages among preterm infants limit generalizability. The sample size may also have hindered detecting accuracy differences between sensors and identifying differences between preterm and full-term infants. A larger, more diverse preterm cohort could enhance robustness of our conclusions. Additionally, our focus on prone-tolerant infants restricts applicability to prone-intolerant infants, who may benefit most from objective tummy time-tracking. Some participants in our study provided more tummy time data than others. This is a common variation in infant positioning practices, and could have introduced bias. The LMM analyses may have partially mitigated this bias by weighting participant observations. A large drift for the MB, particularly during longer recording sessions, highlights the need for drift correction in sensors relying on BLE technology. Subsequently, addressing feasibility challenges for both sensors for home use is crucial before their implementation in Phase IV.
CONCLUSIONS
Our study underscores the potential of commercially available wearable sensors as robust tools for objectively tracking tummy time in both full-term and preterm infants within home settings. The GA sensor demonstrates high accuracy in measuring tummy time across multiple days, making it particularly valuable for clinical studies that aim to assess changes in daily prone exposure over time or pre- to post-intervention. Conversely, for those who prioritize real-time visualization of infant position over shorter periods, the MB sensor offers a viable alternative. Early identification of reduced tummy time holds significant implications for developmental surveillance and targeted management strategies.
Supplementary Material
Supplemental digital content is available for this article. Direct URL citation appears in the printed text and is provided in the HTML and PDF versions of this article on the journal’s Web site (www.pedpt.com).
WHAT THIS ADDS TO THE EVIDENCE.
Current evidence:
Adhering to the daily recommended tummy time for infants is associated with positive health outcomes. However, traditional methods of tracking tummy time like parental recall are often unreliable, particularly for at-risk infants. The adoption of wearable sensors for monitoring infant posture and movement is on the rise, offering potential solutions to the subjectivity challenges inherent in conventional tracking methods.
Gap in the evidence:
A previous study developed wearable sensor algorithms for the commercially available GENEActiv (GA) and MonBaby (MB) sensors to automate tummy time tracking in full-term infants, testing them in controlled laboratory settings. However, questions persist regarding the suitability of these sensors for preterm infants, their effectiveness in natural home settings where tummy time involves various movement transitions, and their reliance on caregivers for sensor application.
How does this study fill this evidence gap?
To address the evidence gap, our study conducted a 2-phase validation of the GA and MB sensors in both full-term and preterm infants within home environments. First, we evaluated the sensors’ accuracy in identifying tummy time during natural play transitions at home, under researcher supervision. Subsequently, we assessed their validity over 3 days of video-recorded play, where caregivers applied the sensors independently, without supervision. Our findings demonstrated that both sensors accurately tracked tummy time in all infants, with GA showing higher precision compared to MB. Potential feasibility challenges for unsupervised sensor use were identified.
Implication of all the evidence to clinicians:
This study establishes the validity of 2 commercially available wearable sensors for tracking tummy time at home. Clinicians can leverage these sensors to monitor adherence to tummy time guidelines and develop tailored tummy time interventions. The objective data provided by these sensors can also enhance parental engagement and education, leading to better adherence to tummy time practices. For researchers, these sensors offer a scalable tool for collecting large-scale data in naturalistic settings, facilitating studies on dose-response relationships between tummy time and health outcomes.
ACKNOWLEDGMENTS
We extend our heartfelt gratitude to the infants and their families for their participation in this study. We are especially thankful to Dhriti Thakur and Kayla Bowler for their time and effort in data collection. We also wish to acknowledge the dedicated staff and students of the Motor Development Lab (MDL) and the Sensorimotor Processing and Rehabilitation Engineering (SPRE) labs at VCU for their valuable contributions throughout various stages of this project.
This work was supported by funding from the Center for Smart Use of Technology to Assess Real-world Outcomes (C-STAR) pilot project grant (National Institute of Health sub-award, #P2C HD101899) and Virginia Commonwealth University graduate school dissertation assistantship awarded to Ketaki Inamdar in the final year of her doctoral studies.
Footnotes
At the time this work was completed, Ketaki Inamdar was a doctoral candidate at Virginia Commonwealth University, Richmond, VA, USA.
The authors declare no conflict of interest.
REFERENCES
- 1.Jones MW. Supine and prone infant positioning: a winning combination. J Perinat Educ. 2005;13(1):10–20. doi: 10.1624/105812404826388 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Franchak JM. Changing opportunities for learning in everyday life: infant body position over the first year. Infancy. 2019;24(2):187–209. doi: 10.1111/infa.12272 [DOI] [PubMed] [Google Scholar]
- 3.Majnemer A Caregiver practices that influence motor development: what are the next moves? Dev Med Child Neurol. 2007;49(11):804. doi: 10.1111/j.1469-8749.2007.00804.x [DOI] [PubMed] [Google Scholar]
- 4.Stern M, Karraker KH, Sopko AM, Norman S. The prematurity stereotype revisited: impact on mothers’ interactions with premature and full-term infants. Infant Ment Health J. 2000;21(6):495–509. doi: 10.1002/1097-0355(200011/12)21:6<495::aid-imhj7>3.0.co;2-f [DOI] [Google Scholar]
- 5.Moon RY, Darnall RA, Feldman-Winter L, Goodstein MH, Hauck FR. SIDS and other sleep-related infant deaths: evidence base for 2016 updated recommendations for a safe infant sleeping environment. Pediatrics. 2016;138(5). doi: 10.1542/PEDS.2016-2940/60296 [DOI] [PubMed] [Google Scholar]
- 6.Inamdar K, Dusing SC, Thacker L, Pidcoe PE, Finucane S, Chu V. Tummy time tracking: examining agreement between parent recall and direct observation in infants. Am J Occup Ther. 2024;78(4):7804185020. doi: 10.5014/ajot.2024.050521 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Greenspan B, Cunha AB, Lobo MA. Design and validation of a smart garment to measure positioning practices of parents with young infants. Infant Behav Dev. 2021;62(1–2):101530. doi: 10.1016/j.infbeh.2021.101530 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Franchak JM, Scott V, Luo C. A contactless method for measuring full-day, naturalistic motor behavior using wearable inertial sensors. Front Psychol. 2021;12:701343. doi: 10.3389/fpsyg.2021.701343 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Hewitt L, Stanley RM, Cliff D, Okely AD. Objective measurement of tummy time in infants (0–6 months): a validation study. PLoS One. 2019;14(2):e0210977. doi: 10.1371/journal.pone.0210977 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Airaksinen M, Räsänen O, Ilén E, et al. Automatic posture and movement tracking of infants with wearable movement sensors. Sci Rep. 2020;10(1):1–13. doi: 10.1038/s41598-019-56862-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Keadle SK, Lyden KA, Strath SJ, Staudenmayer JW, Freedson PS. A framework to evaluate devices that assess physical behavior. Exerc Sport Sci Rev. 2019;47(4):206–214. doi: 10.1249/JES.0000000000000206 [DOI] [PubMed] [Google Scholar]
- 12.Piper MC, Pinnell LE, Darrah J, Maguire T, Byrne PJ. Construction and validation of the Alberta Infant Motor Scale (AIMS). Can J Public Heal. 1992;83(SUPPL. 2):S46–50. [PubMed] [Google Scholar]
- 13.Akoglu H User’s guide to correlation coefficients. Turkish J Emerg Med. 2018;18(3):91–93. doi: 10.1016/J.TJEM.2018.08.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.McHugh ML. Interrater reliability: the kappa statistic. Biochem Medica. 2012;22(3):276–282. doi: 10.11613/bm.2012.031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Sullivan GM, Feinn R. Using effect size—or why the P value is not enough. J Grad Med Educ. 2012;4(3):279–282. doi: 10.4300/jgme-d-12-00156.1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Feehan LM, Geldman J, Sayre EC, et al. Accuracy of fitbit devices: systematic review and narrative syntheses of quantitative data. JMIR mHealth uHealth. 2018;6(8):e10527. doi: 10.2196/10527 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Martinko A, Karuc J, Jurić P, Podnar H, Sorić M. Accuracy and precision of consumer-grade wearable activity monitors for assessing time spent in sedentary behavior in children and adolescents: systematic review. JMIR mHealth uHealth. 2022;10(8):e37547. doi: 10.2196/37547 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Giavarina D Understanding Bland Altman analysis. Biochem Medica. 2015;25(2):141–151. doi: 10.11613/BM.2015.015 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Lobo MA, Hall ML, Greenspan B, Rohloff P, Prosser LA, Smith BA. Wearables for pediatric rehabilitation: how to optimally design and use products to meet the needs of users. Phys Ther. 2019;99(6):647–657. doi: 10.1093/ptj/pzz024 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Dudek-Shriber L, Zelazny S. The effects of prone positioning on the quality and acquisition of developmental milestones in four-month-old infants. Pediatr Phys Ther. 2007;19(1):48–55. doi: 10.1097/01.pep.0000234963.72945.b1 [DOI] [PubMed] [Google Scholar]
- 21.Koren A, Kahn-d’angelo L, Reece SM, Gore R. Examining childhood obesity from infancy: the relationship between tummy time, infant BMI-z, weight gain, and motor development—an exploratory study. J Pediatr Heal Care. 2019;33(1). doi: 10.1016/j.pedhc.2018.06.006 [DOI] [PubMed] [Google Scholar]
- 22.Bartlett DJ, Fanning JEK. Relationships of equipment use and play positions to motor development at eight months corrected age of infants born preterm. Pediatr Phys Ther. 2003;15(1):8–15. doi: 10.1097/01.PEP.0000051693.49199.41 [DOI] [PubMed] [Google Scholar]
- 23.Hewitt L, Stephens S, Spencer A, Stanley RM, Okely AD. Weekly group tummy time classes are feasible and acceptable to mothers with infants: a pilot cluster randomized controlled trial. Pilot Feasibility Stud. 2020;6 (1). doi: 10.1186/s40814-020-00695-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Schmidt LI, Jansen CP, Depenbusch J, Gabrian M, Sieverding M, Wahl HW. Using wearables to promote physical activity in old age: feasibility, benefits, and user friendliness. Z Gerontol Geriatr. 2022;55 (5):388–393. doi: 10.1007/s00391-022-02083-x [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.