Mandibular Movement Monitoring in Children With Neurodisability

ABSTRACT

Background

Obstructive sleep apnoea (OSA) in children is diagnosed using polysomnography (PSG), but children with neurodisability are >3 times more likely to not tolerate leads and sensors. The diagnostic accuracy of Sunrise, a small sensor applied to the chin, has not been assessed in this special population.

Objective

Compare the diagnostic accuracy of Sunrise to simultaneously recorded in‐laboratory PSG.

Methods

We conducted a prospective, cross‐sectional study of children aged 3–18 years with neurodisability, who attended the Queensland Children's Hospital sleep laboratory for diagnostic PSG between February 2024 and September 2025. Sunrise‐ and PSG‐derived measures of sleep disordered breathing were compared, and questionnaire data examining interest in novel sensor technology, and portable, home‐based sleep studies were also collected.

Results

There were 48 children with a neurodisability (mean [SD] age = 11.2 [3.6]; 39% female). For 43 children with complete Sunrise data, sensitivity and specificity for identifying moderate‐severe obstructive sleep apnoea (OSA) using calibrated MM‐ORDI were 0.88 (95% CI: 0.45–1.00) and 0.63 (95% CI: 0.45–0.79), respectively. The concordance correlation coefficient of the obstructive respiratory disturbance index was 0.65 (95% CI: 0.5–0.8) and the median absolute error was 5.7 (IQR: 2.7–13.1) events per hour. Most carers in attendance preferred in‐laboratory sleep studies (56%) to home (33%), but the majority preferred Sunrise (50%) to limited‐channel (17%) or full PSG (10%).

Conclusion

Sunrise shows excellent ability to exclude moderate‐severe OSA in children with neurodisability with moderate specificity. There is substantial appetite among parents and carers for alternative sleep testing modalities.

1. Introduction

Obstructive sleep apnoea (OSA) is a pervasive disorder that affects up to 6% of children internationally [1], contributing to increased health service utilisation [2] attention/behavioural deficits and cardiometabolic disease extending into adulthood [3, 4]. Specialist paediatric sleep services are key to timely diagnosis, yet services are suboptimal for children with neurodisability, for whom OSA prevalence may be as high as 80% [5]. Polysomnography (PSG) is used to diagnose and monitor OSA but has been shown to be poorly tolerated by children with neurodisability (greater than three times more frequently than in neurotypical children) [6], impacting data quality, diagnosis and subsequent management. Unlike for adults, home PSG is neither recommended or considered equivalent to Level 1 in‐laboratory studies by international guidelines [7, 8], further complicating sleep monitoring in a cohort of children who have often had previously challenging experiences in a hospital setting [9, 10]. Poor sleep in children with neurodisability has profound impacts on the family functioning, including for siblings and parents of affected children [11, 12].

PSG is difficult to undertake in children with neurodisability as they will often become distressed and anxious in the hospital environment, with resultant disrupted sleep that does not reflect their typical pattern. Additionally, these children are often poorly tolerant of PSG sensors, particularly those applied to sensitive areas such as the midface [6]. However, this group of children are the ones in whom PSG is often most required; their OSA is more complex, and PSG assists in assessing disease severity and their risk for surgical therapeutic interventions such as adenotonsillectomy. Recent surveys of parents and health care professionals show that sleep is a prioritised health outcome for children with neurodisability [13, 14, 15]. Utilisation of portable and non‐invasive sleep monitoring technologies could improve data quality, enable sleep monitoring within the home, improve outcomes, and meaningfully improve the patient journey for children with neurodisability and their parents or carers.

One such technology is Sunrise, a small device worn on the chin that uses an inertial measurement unit (IMU) to record movement and rotation of the mandible during sleep. Analysis of IMU data is automated using a suite of machine learning algorithms to estimate sleep time, arousals from sleep, and respiratory disturbances [16]. Sunrise has been previously validated in adults [16, 17, 18, 19] and in children without significant co‐morbidities [20], but not in children with neurodisability. OSA prevalence in children with neurodisability may be as high as 80% and is complex to diagnose and manage due to a confluence of factors such as muscle weakness, orofacial and/or airway abnormalities, and anatomical and physiological changes to the central nervous system and its pathways [5]. Validation of monitoring technologies in this inherently more challenging population is therefore essential.

The primary aim of this study was to evaluate the diagnostic performance of calibrated mandibular movement‐derived obstructive respiratory disturbance index (MM‐ORDI) for detecting moderate paediatric OSA, defined as PSG‐obstructive apnoea–hypopnoea index (PSG‐OAHI) ≥ 5 events per hour. Secondary aims were to assess agreement between the MM‐ORDI and PSG‐ORDI and to describe parent and carer perspectives on the Sunrise device and on portable and home‐based sleep monitoring. The American Academy of Sleep Medicine (AASM) Scoring Manual (Version 3, February 2023) provided the PSG scoring framework for this study, and PSG‐OAHI was utilised as the reference metric because this metric is recommended for use in defining OSA in children by the International Classification of Sleep Disorders Manual version 3 (ICSD‐3), which notes that paediatric OSA is characterised by intermittent partial or complete obstructive episodes i.e., obstructive apnoea and/or hypopnea events.

2. Methods

2.1. Study Design, Setting, and Participants

This was a prospective, cross‐sectional study comparing Sunrise (Sunrise, Belgium) and PSG‐derived sleep parameters obtained from simultaneously acquired Sunrise and nocturnal Level 1 PSG recordings. Children aged 3–18 years undergoing diagnostic PSG at the Queensland Children's Hospital were opportunistically recruited between February 2024 and September 2025. Children were excluded if they (i) did not have a neurodisability, or (ii) received treatment including CPAP, non‐invasive ventilation, or supplemental oxygen during their PSG. Informed written consent was obtained from parents or carers in attendance, and ethical approval for this study was obtained by the Children's Health Queensland Human Research Ethics Committee (review reference HREC/2023/QCHQ/102838).

2.2. Polysomnography

A Level 1 Natus PSG system (Natus Medical Inc, Pleasanton CA, United States) was used to acquire nocturnal PSG data. Sleep staging and scoring of respiratory and arousal events, including respiratory effort‐related arousals (RERAs), were performed by trained paediatric sleep professionals in accordance with the most recent AASM Scoring Manual (Version 3, February 2023). Where sensor drop‐out occurred, investigators used AASM‐recommended alternatives (e.g., oronasal thermistor and respiratory inductance plethysmography Sum in place of airflow measured through nasal cannulae). PSG was marked as invalid for the purpose of this study where sensor drop‐out occurred with no AASM‐recommended alternative available, for a period of ≥ 10% total sleep time. PSG scoring and interpretation followed usual clinical processes and were performed independently of Sunrise analysis. Sunrise disconnection periods were therefore not considered in the scoring of PSG.

2.3. Sunrise

Sunrise is a small sensor that adheres to the chin and records mandibular movement and rotation using an IMU, as well as other data depending on the version of hardware used. Version 1.1 of the hardware, that includes an IMU only, was used for this study. Sunrise sensors were synchronised to an Apple iPad (Apple, CA, United States) via Bluetooth using the Sunrise application available through the App Store. Sunrise records were commenced as close to PSG “lights out,” and ceased as close to “lights on,” as much as possible while ensuring the clinical PSG workflow was undisturbed. Recordings were automatically transferred to the Sunrise clinical portal for automatic analysis using a set of machine learning algorithms known as gradient boosting decision trees. The results of these analyses are made available through the clinical portal for clinicians to review. The contents of these reports were transcribed by the research team into a study database. Sunrise analyses are automated and occurred independently of clinical PSG scoring.

2.4. Questionnaire

Parent and carer perspectives on sleep monitoring within the home versus in‐hospital, and the utilisation of portable sensor technologies were assessed using a short questionnaire that consisted of multiple choice and free‐text responses. Similar free‐text responses were grouped to aid in analysis.

2.5. Statistical Methods

Distribution‐appropriate methods were used for descriptive statistics; mean and standard deviation (SD) were reported for normally distributed variables or median and inter‐quartile range (IQR) were reported for non‐normally distributed variables.

The clinical utility of Sunrise was assessed using a variety of metrics, including sensitivity, specificity, positive and negative predictive values. Binomial exact (Clopper‐Pearson) confidence intervals were computed for each metric. These were calculated using a threshold of five events per hour for the MM‐ORDI [20]. Calibrated MM‐ORDI (9.61 events per hour) was compared to a PSG‐OAHI threshold of five events per hour [21]. Mean and median absolute errors, and concordance correlation coefficients (CCCs) [22, 23] were calculated for the MM‐ORDI and calibrated MM‐ORDI [20]. CCCs allow us to assess agreement between Sunrise and PSG, and to assess equality of MM‐ and PSG‐derived parameters.

A Kruskal‐Wallis test was used to compare the absolute values of the ORDI residuals across groups, and Chi‐square tests of independence were used to compare survey data of children with and without previous PSG. Statistical analyses were performed using Stata 17 (StataCorp LLC, TX, United States), and data visualisation using Stata 17. Calculation of CCCs were performed using the CONCORD module for Stata [24].

3. Results

Forty‐nine children were approached to participate, and informed written consent was obtained from 48 children and their guardian in attendance. Questionnaire data were obtained for all 48 children who participated, and demographic, anthropometric, PSG, and neurodisability diagnosis data for this cohort are shown in Table 1.

Table 1.

Demographic, anthropometric, polysomnographic, and neurodisability diagnosis data for 48 children included in this study.

	n (%), Median (IQR), or Mean (SD)
Age (Years)	11.2 (SD 3.6)
Female	19 (39%)
Height (cm)	146 (SD: 22.7)
Weight (kg)	51.7 (IQR: 36.3–83.5)
BMI (kg/m2)	23.4 (IQR: 18.4–29.9)
Neck Circumference (cm)	37 (SD: 7.0)
PSG‐OAHI	0.9 (IQR: 0.2–2.9)
PSG‐ORDI	2.6 (IQR: 1.5–5.6)
PSG‐AHI	2.4 (1.0–5.4)
Neurodisability	ADHD (n = 12)
	Trisomy 21 (n = 10)
	Autism spectrum disorder and ADHD (n = 10)
	ASD (n = 7)
	Chromosomal abnormality* (n = 6)
	Prader‐Willi syndrome (n = 2)
	Chiari malformation (n = 1)

Abbreviations: ADHD, Attention‐deficit/hyperactivity disorder; AHI, Apnoea‐hypopnea index; ASD, Autism spectrum disorder; OAHI, Obstructive AHI; ORDI, Obstructive respiratory disturbance index; PSG, Polysomnography.

^*Chromosomal abnormalities not inclusive of Trisomy 21 and Prader–Willi syndrome.

Valid PSG and Sunrise sleep monitoring data were obtained for 43 children. For five children without Sunrise data, four did not tolerate the sensor, and one recording was a technical failure (recording upload failure). Twenty‐three of 43 (53%) children had no OSA, 12/43 (28%) had mild OSA (PSG‐OAHI: 1–5) 4/43 (9%) had moderate OSA (PSG‐OAHI: 5–10), and 4/43 (9%) had severe OSA (PSG‐OAHI ≥ 10). OSA prevalence by neurodisability is plotted in Figure 1.

Figure 1.

Frequency of OSA severity by neurodisability in 43 children with paired PSG and Sunrise data. [Color figure can be viewed at wileyonlinelibrary.com]

3.1. Agreement

Accelerometer and gyroscopic data are used in combination with machine learning to produce an MM‐ORDI. Agreement of calibrated MM‐ORDI and uncalibrated MM‐ORDI with equivalent PSG‐derived sleep parameters is shown in Table 2.

Table 2.

Mandibular‐movement (MM) derived sleep parameter agreement with PSG‐derived sleep parameters for 43 children with sensor data.

	Calibrated MM‐ORDI[20]	MM‐ORDI
Mean absolute error (SD)	N/A	8.6 (8.6)
Median absolute error (IQR)	N/A	5.7 (2.7–13.1)
CCC (95% CI)	N/A	0.65 (0.50–0.80)
Sensitivity* (95% CI)	0.88 (0.45–1.00)	1.00 (0.63–1.00)
Specificity* (95% CI)	0.63 (0.45–0.79)	0.37 (0.17–0.49)
PPV* (95% CI)	0.35 (0.15–0.65)	0.21 (0.11–0.43)
NPV* (95% CI)	0.96 (0.78–1.00)	1.00 (0.72–1.00)

Abbreviations: CCC, Concordance correlation coefficient; NPV, Negative predictive value; ORDI, Obstructive respiratory disturbance index; PPV, Positive predictive value.

^*Binary classification using a cut‐off of 5. *Calibrated MM‐ORDI (9.61 events/hour) to predict moderate paediatric PSA (PSG‐OAHI ≥ 5).

Performance metrics for Sunrise‐derived measurements are compared to PSG‐derived in Table 2, and a Bland‐Altman plot comparing the MM‐ORDI and PSG‐ORDI is shown in Figure 2. Calibrated MM‐ORDI of 9.61 events per hour was compared to an PSG‐OAHI of five events per hour. The absolute values of the ORDI residuals were not statistically significantly different across neurodisability groups (p = 0.22).

Figure 2.

Bland–Altman plot comparing the Sunrise‐ and PSG‐derived ORDI. [Color figure can be viewed at wileyonlinelibrary.com]

3.2. Questionnaire

We obtained questionnaire data for 48/48 (100%) of children for whom consent was obtained, and a summary of these data collated from parent or carer responses are shown in Table 3. Twenty‐two of 48 (46%) of children had previously had PSG.

Table 3.

Summary of questionnaire data for 48 participants.

	Had PSG previously (n = 22)	Did not have PSG previously (n = 26)	Total (n = 48)
Preferred sleep study location
Home	10/22 (45%)	6/26 (23%)	16/48 (33%)
Hospital	12/22 (55%)	15/26 (57%)	27/48 (56%)
Not sure	0 (0%)	5/26 (19%)	5/48 (10%)
Preferred sleep study type
Full PSG (all leads and sensors)	1/22 (5%)	4/26 (15%)	5/48 (10%)
Limited‐channel (nasal cannula, bands, pulse oximetry)	2/22 (9%)	6/26 (23%)	8/48 (17%)
Sunrise	12/22 (55%)	12/26 (46%)	24/48 (50%)
Not sure	0 (0%)	2/26 (8%)	2/48 (8%)
No response	7/22 (32%)	2/26 (8%)	9/48 (19%)

There was a statistically significant association between preferred sleep study location (home, hospital, or not sure) and whether the child had previously undergone PSG (p = 0.049). There was no evidence to suggest the sleep study preference was associated with having previously completed in‐laboratory PSG (p = 0.27).

The most frequently cited concerns regarding sleep studies in the home by parents or carers in attendance were anticipated issues with compliance (n = 13, 27%), concerns relating to setup of equipment and the application of leads and sensors (n = 10, 21%), and environmental distractions (n = 2, 4%). When asked to provide other comments, key themes were concerns relating to accuracy of home sleep studies compared to in‐laboratory studies (n = 8, 19%), preference for smaller devices (n = 3, 6%) and those easier to use (n = 2, 4%).

4. Discussion

In this study of children with neurodisability, we compared simultaneously acquired PSG and Sunrise recordings from a cohort requiring evaluation for SDB and gathered parent and carer perspectives on home sleep study monitoring. Using the calibrated MM‐ORDI as the outcome measure, Sunrise was highly sensitive and reasonably specific, with excellent NPV in this cohort. Thus, Sunrise demonstrated excellent ability to exclude OSA in children with neurodisability but is less well suited to “ruling in” OSA with a high proportion of false positives. This may limit the usefulness of Sunrise for surgical decision‐making, where false positives may result in a decision to undertake surgery for a child with neurodisability, a cohort in whom perioperative risk is known to be higher [25, 26]. Overall, our analysis calibrated MM‐ORDI supports previous findings in neurotypical children, and we observed similar sensitivity, and slightly reduced specificity [20]. In this study, four children had moderate OSA, and four had severe OSA. Further analysis in a larger sample of children is needed to improve our understanding of device performance across the wide range of diagnoses that fall within the umbrella term of neurodisability. While previous validation work in a paediatric population showed generally better performance than what we have presented in this work [20], our work focuses on children with more complex diagnoses in whom additional challenges are faced when undertaking monitoring. This is an increasing population encountered by tertiary sleep laboratories worldwide. Surveys completed by guardians of children suggest there may be substantial appetite for home sleep testing, particularly among those with previous experience with in‐laboratory PSG. However, concerns relating to compliance and accuracy compared to in‐laboratory testing were raised and require further exploration.

Sunrise offers several key advantages over traditional in‐laboratory PSG, and other forms of limited channel monitoring in children with neurodisability. It is a small device attached to the chin, allowing acquisition of data without the discomfort of more invasive monitoring that is typically poorly tolerated by children with neurodisability [6]. Parents and carers overwhelmingly preferred Sunrise monitoring to in‐laboratory and limited‐channel sleep monitoring. Owing to sophisticated machine learning analysis of recording, Sunrise offers insight into sleep architecture, obstructive sleep disordered breathing, and restlessness. Compared to other devices evaluated in the paediatric population to date [27], automated analysis and reduces the burden on clinical teams to manually analyse recorded data and the associated training. A diagnosis of neurodisability is not contraindicated for Sunrise testing, and our study demonstrates that Sunrise is highly effective at excluding OSA in this population of children. Furthermore, utilisation of Sunrise in the home has the advantage of enabling clinical teams to effectively characterise typical sleep patterns over multiple nights of monitoring, with reduced technical support burden compared to other forms of monitoring.

We identified several important considerations associated with Sunrise testing in this population. Adenotonsillectomy is the first‐line treatment for paediatric OSA [28], so high rates of false positives may result in unnecessary exposure of children with neurodisability to anaesthesia and surgical intervention. Clinical teams may prefer technologies that are comparatively better at “ruling in” OSA in this population. The calibrated MM‐ORDI showed better correlation with PSG measurements, definitions used in frequently cited published works on paediatric OSA that have historically informed clinical management use AHI‐based definitions [23, 29]. This is an important consideration for clinical services that care for children with neurodisability.

There are some important limitations to our work. While every effort was made to ensure Sunrise analysis periods were as close as possible to PSG, some discordance may nevertheless be attributable to differences in analysable recording. We were unable to assess the accuracy of the sleep staging algorithms used by Sunrise, which, to the best of our knowledge, were trained only on adult data, and have yet to be validated in any paediatric population. Thus, the reliability of these aspects of Sunrise is uncertain. During analysis we reviewed children for whom the absolute ORDI residual was large, and in one case noted substantial discordance in sleep and wake classification compared to PSG.

Another important limitation of this work is that this was a study of only a small number of children which included only four children with moderate OSA, and four with severe OSA, limiting the precision and stability of our statistical analysis. Our sampling strategy was opportunistic, and no a priori sample size calculation was performed. Therefore, our findings should be considered exploratory and be interpreted with caution. This is particularly true for children with neurodisability, as this overarching term encompasses many heterogenous conditions that have varied impacts on sleep and breathing. It follows that the distribution of different neurodisability diagnoses in any assessed population may impact both distribution of OSA severity, and performance of Sunrise. In a previous 6‐month long cross‐sectional study at our centre, we found 37% of children with neurodisability had at least mild disease [6], compared to 47% in the present study. Furthermore, our data do not include children with significant behavioural intolerance; in this study, 8% of children were unable to tolerate Sunrise, and investigators did not recruit children for whom PSG was unsuccessful or was unlikely to succeed if Sunrise was applied as an extra device. Thus, we cannot be certain of the generalisability of our findings to children with moderate‐severe OSA, and together with the small sample size, we cannot be certain about generalizability to children with neurodisability more broadly. Finally, our questionnaire is not a validated instrument, and so these data should be interpreted with caution. A strength of our research is the incorporation of consumer perspectives, from which we observed both an appetite for at‐home testing using smaller, less invasive technologies, coupled with salient concerns around accuracy compared to in‐laboratory testing.

In this limited evaluation of the Sunrise sleep test, we found it showed excellent ability to identify moderate‐severe OSA in children with neurodisability when using the calibrated MM‐ORDI. Overall, our data is concordant with previous findings in neurotypical children for calibrated MM‐ORDI [20]. Half of surveyed parents and carers indicated they would prefer home sleep monitoring and overwhelmingly favoured the Sunrise compared to more traditional devices, though concerns were frequently raised around the diagnostic accuracy of limited home monitoring relative to in‐laboratory PSG. Further data are needed to confirm our findings, to better understand Sunrise performance across the spectrum of neurodisability, and to assess its use in the home.

Author Contributions

Andrew J. Collaro: conceptualization, funding acquisition, writing – original draft, investigation, methodology, visualization, writing – review and editing, formal analysis, data curation, project administration, supervision. Kiara D. Sclip: data curation, writing – review and editing, investigation, project administration. Lydia R. Penny: investigation, writing – review and editing, data curation, project administration. Laura R. Mead: investigation, writing – review and editing, data curation, project administration. Ajay Kevat: investigation, writing – review and editing, formal analysis. Jasneek K. Chawla: conceptualization, funding acquisition, writing – review and editing, supervision, formal analysis.

Consent

Informed written consent was obtained from parents or carers in attendance, and ethical approval for this study was obtained by the Children's Health Queensland Human Research Ethics Committee (review reference HREC/2023/QCHQ/102838).

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.