How often do caregivers attend their child when a seizure detection device alerts for a nocturnal major motor seizure?

Abstract

Objective

Development in seizure detection devices has focused mainly on detection performance. Yet, in order to serve their function in preventing harmful situations and even sudden unexpected death in epilepsy (SUDEP), caregivers need to respond to seizure alarms. This aspect has not been studied so far. We therefore determined caregiver attendance in response to an alarm from a seizure detection device in a real‐world family home setting and evaluated the determinants affecting this rate.

Methods

We examined caregiver attendance using video recordings from the PROMISE trial, a home‐based study designed to determine the performance of the NightWatch seizure detection device. Attendance was recorded when a caregiver approached the child within 15 min after the alarm. We evaluated attendance to each true alarm, and we randomly selected one false alarm of the same subject, if available, to evaluate attendance as well. We also collected several child‐ and alarm‐related determinants, which we analyzed for effect on attendance using a generalized estimated equation (GEE).

Results

We included 461 true positive alarms for 31 children and matched them to 311 false alarms. The overall attendance rate for true positive alarms was 64%, with a median individual attendance rate of 100% per child. The individual attendance rate to false alarms (median 50%) was significantly lower when comparing the response to true positive alarms (p < .001). Nine caregivers always responded to alarms regardless of their nature, whereas two never attended to any alarm. The presence of seizure‐related sounds (odds ratio [OR] 7.73, 95% confidence interval [CI] 3.74–15.96) and having a lower seizure frequency (OR 0.37, 95% CI 0.19–0.75) were associated with higher attendance rates.

Significance

We found that rates of attendance to nocturnal major motor seizure alarms were generally high, although variation existed among caregivers. These findings highlight the need for counseling when implementing seizure detection devices in epilepsy care.

Key points.

• We evaluated caregiver attendance to alarms for nocturnal major motor seizures of 31 children wearing the NightWatch seizure detection device and examined the potential determinants.

• We found that rates of attendance to nocturnal major motor seizure alarms were generally high, although variation existed among caregivers.

• Nine caregivers always responded to alarms regardless of their nature, whereas two never attended to any alarm.

• Some variance in attendance was explained by two factors: we found higher rates of attendance to seizures with sounds and those with a lower seizure frequency.

• Optimal implementation of seizure detection devices requires targeted education to guide caregivers in accurate attendance to alarms.

1. INTRODUCTION

Sudden unexpected death in epilepsy (SUDEP) is one of the most important causes of premature death in people with epilepsy, with an average incidence of 1 per 1000 person‐years. ¹ SUDEP is mostly seizure related, in particular with generalized or focal to bilateral convulsive seizures. SUDEP risk increases with the number of convulsive seizures. ¹ Three key situational factors appear to be strongly associated with SUDEP occurrence: (1) nocturnality; (2) prone position; (3) being unwitnessed. ² , ³ , ⁴ These factors interact closely: SUDEP is more frequent among individuals with nocturnal convulsive seizures who sleep alone compared to those who have a bed partner. ³ SUDEP rates are lower with increasing levels of nocturnal supervision, which strengthens the previous statement. ⁵ , ⁶ An additional association has been observed between SUDEP and nocturnality ² , ⁴ and SUDEP and the prone position. ⁴ , ⁷ , ⁸ One possible explanation is that individuals may experience impaired arousal and an inability to reposition during the postictal phase, potentially leading to apnea through airway obstruction and CO₂ rebreathing. ⁷ , ⁸ The presence of a caregiver might lower SUDEP risk through repositioning or arousing the subject. Seizure detection devices may help to alert to seizures that were unwitnessed. Picard et al. describe a 20‐year‐old man, who experienced a nocturnal seizure that triggered the alarm of a seizure detection device, which resulted in SUDEP. ⁹ No caregiver responded to the alarm for a period of 15 min, illustrating, although in a single example, the importance of caregiver attendance. Caregiver attention to alarms is essential, as it determines the effectiveness of these devices in preventing potentially serious adverse outcomes. ¹⁰ , ¹¹ Previous studies have focused on improving the accuracy and usability of nocturnal seizure detection devices, ¹² but to our knowledge they have not reported on the frequency and determinants of caregiver response to alarms. ¹³ We therefore determined caregiver attendance to seizures detected by a seizure detection device in a real‐world family home setting, and examined which factors influenced this rate.

2. METHODS

2.1. Recruitment and subject selection

We reviewed the video recordings from the PROMoting Implementation of Seizure detection devices in Epilepsy care (PROMISE) trial (NCT03909984) to evaluate attendance rates. ¹⁴ This long‐term, prospective, home‐based study aimed to assess the performance of the NightWatch device in detecting nocturnal major motor seizures in 51 children. The participants, who were between 4 and 16 years of age, were recruited from three tertiary epilepsy centers in The Netherlands (SEIN, Kempenhaeghe, or University Medical Centre Utrecht). All children were living at home and had at least one nocturnal major motor seizure a week. Participants wore the NightWatch seizure detection device during the night on their upper arm, preferably on the side where the motor seizures typically begin, for a period of 2 months between 2018 and 2020. The NightWatch transmits a wireless signal to a separate station, which alerts parents or guardians when a major motor seizure is suspected. Detection is based on movement (three‐dimensional [3D] accelerometry) and changes in heart rate (photoplethysmography). ¹⁴ , ¹⁵ Performance of NightWatch was evaluated through video recordings taken during the trial nights and by comparing them with other detection methods (i.e., video detection and caregiver journal). The video recordings, captured in the child's bedroom, focused on the bed and its immediate surroundings, including walls and occasionally the floor. All alarms and annotations (either generated by NightWatch, video, or caregiver's journal) were blindly reviewed by trial nurses. In addition, 5% of all nights without alarms or annotations were screened for missed motor seizures. All annotated events were labeled as major motor seizures or false positive alarms. Major motor seizures were defined as generalized or focal to bilateral tonic–clonic, generalized tonic lasting longer than 30 s, hyperkinetic, and other major seizures (focal onset clonic, generalized onset, and “tonic–clonic” like seizures, i.e., bilateral seizures without a classical generalized or focal to bilateral tonic–clonic pattern). All other seizures were categorized as minor seizures. Alarms that did not coincide with a seizure or with a minor seizure were labeled as false positive alarms, as the NightWatch was developed to detect major motor seizures.

2.2. Methods and procedures

2.2.1. Aim I: Caregiver attendance to alarms for nocturnal major motor seizures

Attendance was scored using the video tracings. We included all children with video tracings of true positive NightWatch alarms, that is, alarms signaling major motor seizures, to determine the caregiver attendance rate.

We classified a response as attendance only if a caregiver—parent, guardian, or other family member—responded to an alarm by coming in proximity to the child, that is, touching or bending over the child or their immediate surroundings (e.g., the bedside, bedding, or items in the bed). All other situations, for example, a caregiver responding but not entering the room or entering the room but without approaching the child, were not considered as attendance.

Attendance was scored during the full time period of the seizure, as well as for the 15 min after the alarm, referred to collectively as the “period of interest.” We selected this timeframe based on the average latency between the end of a major motor seizure and the occurrence of SUDEP in the postictal phase. ¹⁶ For seizures lasting longer than 15 min, the recording was reviewed until the seizure ended. In case a major motor seizure or alarm occurred within 3 min of the last alarm or if a second alarm was triggered for the same seizure, we labeled these events as a cluster and counted them as a single alarm.

Alarms were excluded from analysis if the video footage did not cover the entire period of interest (e.g., due to early termination of the recording) or if the child was not sufficiently visible (e.g., if the child's head and/or upper body was out of view or if the child left the bed before the caregiver attended them). An alarm was also excluded when a caregiver was already in proximity to the child in response to a previous seizure or other event. An exception was made for caregivers sleeping next to the child and already in contact with them or their bedding. In these cases, attendance was recorded if there was a clear reaction to the alarm, such as touching or stroking the child.

2.2.2. Aim II: Caregiver attendance to false positive alarms (i.e., alarms for minor or no seizures)

We randomly selected for each true positive alarm one false positive alarm (if available), that is, an alert to a minor seizure or no seizure for the same child. Attendance to these alarms was evaluated in the same manner as for the true positive alarms. If a false alarm did not meet our selection criteria (e.g., due to early termination of the recording), we randomly selected another false positive alarm (if available).

2.2.3. Aim III: Identify determinants of caregiver attendance

To determine which factors might further influence caregiver attendance, we collected the following fixed clinical‐related determinants: age at the start of inclusion of the PROMISE trial, the use of a tent bed, the Caregiver Strain Index (CSI; the highest score was selected if a child had two caregivers) at start of the trial and presence of a learning disability. To evaluate alarm‐related determinants, we collected the time of the alarm (12 p.m. to 12 a.m., 12 a.m. to 6 a.m., or 6 a.m. to 12 p.m.); whether the alarm was true or false positive; if the alarm coincided with a seizure (i.e., a major motor or minor seizure); the number of preceding NightWatch alarms during that particular night (categorized as 0, 1, or >1 alarms); if the caregiver was sleeping in the same room; if the alarm was clustering (i.e., if another alarm occurred in the 15 min before or after); the number of total, false, and true positive alarms the child had so far; the seizure frequency during the trial (the number of major seizures the child had, including the ones missed by NightWatch, divided by the number of days in the trial); the positive predictive value (PPV; number of true positive alarms up to the event/total alarms up to the event); and F1‐score [2*PPV*sensitivity/(PPV + sensitivity)] determined at the time of the alarm; the number of nights with NightWatch use up to the event; and audible seizure‐ or non–seizure‐related sounds. Sounds were recorded in the period 3 min before and 3 min after the alarm and, if present, during the seizure. For seizure‐related sounds, we included all vocal productions during a seizure, as well as postictal snoring and external sounds caused by the seizure (e.g., rhythmic sounds due to the child hitting their surroundings caused by motor symptoms). All other sounds were classified as non–seizure‐related sounds, including, for example, calling for a parent, playing with toys, and so on.

Sound production was only considered as a determinant when sounds were recognizable (louder than whispering or sighing) and when sounds preceded caregiver arrival by at least 5 s, since caregivers need time to respond. This did not apply when a caregiver sleeps next to a child, since response time could have been less.

2.2.4. Body position and postictal recovery

For all included major motor seizures, we evaluated if the child was in the prone position. A prone position was defined as the child lying on their stomach, with the body's transverse plane parallel to the sternum and <45 degrees from the horizontal mattress. In addition, the upper body was lifted <45 degrees from the mattress. We also evaluated the seizure semiology duration by noting the start and end of visible symptoms (i.e., seizure sound or motor symptoms, not including postictal jerks). To evaluate the state of consciousness for the child, the presence and timing of the first spontaneous movement during the immediate postictal period was noted.

2.3. Statistical analysis

Attendance rates were calculated per child for true positive alarms and false positive alarms separately. We used the Wilcoxon signed‐rank test to evaluate the difference in median individual attendance rate between these alarms.

To analyze the influence of the clinical and alarm‐related determinants on caregiver attendance while correcting for multiple observations per child, we performed generalized estimating equations (GEEs). All significant (p < .05) determinants of the univariate GEE model were checked for multicollinearity and included in the multivariate analysis while applying the Bonferroni correction to correct for multiple testing.

3. RESULTS

We selected 31 children from the PROMISE database. Nineteen of 51 children were not included in our study because they did not have any major motor seizures (n = 18) and one was not included because they had no true positive alarms meeting our criteria (Figure 1). For these 31 children, we reviewed a total of 490 true alarms, of which 29 were excluded due to early termination of the recording, insufficient view of the child, or prior caregiver presence. This resulted in a total of 461 true positive alarms. About one third (36%) of the true positive alarms signaled generalized or focal to bilateral tonic–clonic (TC) seizures, 4% generalized tonic (T > 30s) (seizures, 10% for hyperkinetic motor (HM) seizures, and 50% for other major seizures. The number of true positive alarms per child varied from 1 to 114 (median 5 alarms, interquartile range [IQR] 2–11).

FIGURE 1.

Study flowchart of inclusion of children and alarms from the PROMISE trial. ¹⁰ For the 461 true positive NightWatch alarms that met the inclusion criteria, we randomly matched (1:1) false positive NightWatch alarms per subject, if possible. We identified 311 false positive alarms that met our inclusion criteria. *Multiple reasons for exclusion possible.

We selected 1044 false positive alarms for the 461 true positive alarms involving 31 children. From these, a maximum of 348 were randomly selected to pursue a 1:1 ratio. Eventually, 311 of the available false alarms met the selection criteria. For one child, there were no false positive alarms that met the selection criteria (Figure 1). The number of false positive alarms varied from 0 to 84 per child (median 5 alarms, IQR 2–10), with 54% of the alarms for minor seizures and 46% for no seizures. Clinical characteristics of the study population are summarized in Table 1.

TABLE 1.

Clinical characteristics of the study population.

Study population	Participants (n = 31)	Missing data (no. of participants)
Age (years)	Median 10 (IQR 8–14)	0
Sex (% female)	n 14 (45%)	0
Etiology
Genetic	n 15 (48%)	0
Structural	n 11 (35%)
Unknown	n 5 (16%)
Epilepsy with generalized or focal to bilateral tonic‐clonic seizures	n 28 (90%)	0
No. of anti‐seizure medications	Median 2 (IQR 2–3)	0
Presence of learning disability	n 23 (74%)	0
Use of a tent bed	n 5 (16%)	0
Usage of a listening device	n 3 (10%)	0
NightWatch usage (no. of nights)	Median 64 (IQR 45–75)	0
Caregiver Strain Index score at start of trial a	n 10 (IQR 9–11)	n 4 (13%)

Abbreviations: IQR, interquartile range; No., number.

^a Scores range from 0 to 13, with 13 indicating maximum strain.

3.1. Aim I: Caregiver attendance to alarms for nocturnal major motor seizures

Attendance was observed for 294 of the 461 true positive alarms (64%), including 19 of 31 children for whom all true alarms were attended. The median individual attendance rate was 100% (IQR 70–100%) (Figure 2).

FIGURE 2.

Caregiver attendance rates in response to NightWatch alarms for nocturnal major motor seizures. The size of the blue circle is proportional to the number of true alarms for each child. The red line represents the median individual attendance rate per child.

Median time of caregiver arrival for true positive alarms was 27 (IQR 0–54) Seconds (Figure 3). Caregivers attended to the child before the NightWatch alarm in 24% of all true alarms.

FIGURE 3.

Time (in min) of caregiver attendance to NightWatch alarms for nocturnal major motor seizures relative to the seizure alarm. The 0 value on the x‐axis indicates the alarm going off. The red line represents the median time of arrival relative to alarm onset.

3.2. Aim II: Attendance to false positive alarms (i.e., minor or no seizures)

Caregivers attended 125 false positive alarms (40%), including nine attending all false alarms for their child. Median attendance rate per child to false alarms was 0.49 (IQR 0.2–1). The Wilcoxon signed‐rank test revealed significantly lower individual attendance rates for false positive alarms (p < .001) when compared with attendance to true positive alarms, with 11 caregivers having either attended all alarms or none at all (Figure 4).

FIGURE 4.

Caregiver attendance rates to the true (blue) and false positive NightWatch alarms (red) per child. Cases are presented according to the attendance rate for true positive alarms (Figure 2).

3.3. Aim III: Identify determinants of caregiver attendance

Attendance for each determinant is summarized in Table 2. With the univariate GEE analysis, we found significantly lower attendance rates for a higher PPV of the NightWatch system (p = .017, OR 0.97, 95% CI 0.95–1.00), for a higher seizure frequency (p = .007, OR 0.31, 95% CI 0.13–0.72), and when more than one alarms occurred before that current night (p = .036, OR 0.57, 95% CI 0.34–0.96). Attendance rates were significantly higher when the child made seizure‐related sounds (p = .001, OR 3.66, 95% CI 1.70–7.85). For the clinical, child, or caregiver‐related factors, we found lower attendance with increasing age of the child (p = .033, OR 0.77, 95% CI 0.61–0.98) and if the child slept in a tent bed (p = .012, OR 0.18, 95% CI 0.05–0.69).

TABLE 2.

Determinants of caregiver attendance to the NightWatch alarms (n = 772), including 419 attended and 353 unattended alarms.

	Attended NightWatch alarms, no. (%) or median (IQR), n = 419	Not attended NightWatch alarms, no. (%) or median (IQR), n = 353	Univariate GEE analysis, OR (95% CI)	Multivariate GEE analysis, OR (95% CI)
Age of participants (years)	10 (9–13)	14 (9–14)	0.77 (0.61–0.98) a	0.80 (0.65–0.98)
Presence of a learning disability	384 (54)	324 (46)	0.74 (0.13–4.41)	–
Use of a tent bed	52 (25)	155 (75)	0.18 (0.05–0.69) a	– b
Caregiver Strain Index (CSI) score at start trial (scored 0–13)	9 (9–11)	10 (9–11)	0.99 (0.96–1.03)	–
Caregiver sleeping in same room as the child	134 (71)	54 (29)	2.37 (0.64–8.75)	–
Nature of the alarm
False positive (i.e. minor or no seizure)	125 (40)	186 (60)	Ref.	–
True positive (i.e. major motor seizure)	294 (64)	167 (36)	2.62 (0.66–10.33)
Presence of a seizure
No	63 (44)	81 (56)	Ref.	–
Yes (i.e. major motor or minor seizure)	356 (57)	272 (43)	1.68 (0.71–4.02)
Timing of the alarm
<12 AM	120 (50)	119 (50)	Ref.	–
12–6 AM	226 (56)	178 (44)	1.26 (0.61–2.58)
>6 AM	73 (57)	56 (43)	1.29 (0.49–3.43)
Sounds coinciding with the alarm
Seizure‐related	270 (70)	117 (30)	3.66 (1.70–7.85) a	7.73 (3.74–15.96) a
Non‐seizure‐related	74 (50)	75 (50)	0.80 (0.41–1.53)	–
Positive Predictive Value (PPV) up to the event	46 (33–53)	74 (50–78)	0.97 (0.95–1.00) a	0.98 (0.97–1.00)
F1‐score (2*PPV*sensitivity /PPV + sensitivity)	0.51 (0.40–0.63)	0.73 (0.42–0.76)	0.06 (0.00–1.48)	–
Clustering of alarms (i.e. coinciding with another alarm 15 min before or after the event)	101 (70)	43 (30)	2.29 (0.39–13.62)	–
Number of prior alarms that night
0	203 (60)	134 (40)	Ref.	–
1	81 (56)	63 (44)	0.85 (0.60–1.21)	–
>1	135 (46)	157 (54)	0.57 (0.34–0.96) a	0.94 (0.68–1.31)
Number of alarms before the event
True positive alarms so far	12 (3–38)	55 (11–78)	1.00 (0.99–1.01)	–
False positive alarms so far	17 (17–31)	17 (6–34)	1.00 (1.00–1.01)
All alarms so far	31 (11–63)	55 (21–114)	1.00 (1.00–1.00)
Seizure frequency (number of seizures during the trial/trial duration)	11 (4–33)	24 (6–70)	0.31 (0.13–0.72) a	0.37 (0.19–0.75) a
Number of nights of NW usage up to the event	24 (11–44)	29 (12–46)	0.50 (0.98–1.01)	–

Note: The right column presents the odds ratio (OR) and confidence interval (95% CI) for the univariate Generalized Estimating Equation (GEE) analysis.

Abbreviations: AM, ante meridiem; CSI, Caregiver Strain Index; IQR, interquartile range; No, number of; NW, NightWatch; PPV, positive predictive value; Ref., Reference value.

^a Indicates significant results (threshold p = 0.05 univariate GEE or threshold p = 0.05/5 = 0.01 multivariate GEE).

^b This variable was not included in the multivariate analysis due to multicollinearity.

The variable tent bed was not included in the multivariate analysis, due to multicollinearity with age. In the multivariate analysis only seizure sounds (p = .000, OR 7.73, 95% CI 3.74–15.96) and seizure frequency (p = .005, OR .37, 95% CI .19–.75) remained significant.

3.4. Body position and postictal recovery

We found that 17% (n = 80) of the alarmed nocturnal major motor seizures ended in a prone position. For 19 of these seizures, the child was not lying in a stomach position at the beginning of the seizure (24%). Of the 80 seizures ending in the prone position, 39% were unattended. We found that the median time of first spontaneous movement was 23 s (IQR 2–54) with respect to the alarm. Caregivers arrived on median 1 s before the first spontaneous movement (IQR –37–50). Seven children did not demonstrate first spontaneous movement within the full period of interest. The median seizure duration was 51 s, with a median of 51 s until first spontaneous movement since the start of the seizure. For attended alarms, the median seizure duration was 72 s, with 58 s until first spontaneous movement since seizure start. For unattended alarms, median duration was 26 s and time to movement was 39 s.

4. DISCUSSION

Our study shows that attendance rates to nocturnal major motor seizure alarms at home are generally high, although variation exists among caregivers. Most caregivers respond within 1 min, arriving in time to reduce SUDEP risk. Although attendance was lower for false positive alarms, some caregivers showed equal attendance to both true and false positive alarms. This suggests that differences in attendance are due primarily to individual caregiver behavior. Some variance was explained by specific factors, as we found higher attendance when the child made seizure‐related sounds and lower attendance in children with a higher seizure frequency. There was also a trend for lower attendance among children of older age and when the NightWatch system yielded a higher PPV.

There are some limitations to our study. We only noted attendance if the caregiver was in proximity to the child due to the limited view in the recordings. This meant that we did not record all forms of caregiver response; for example, when caregiver was standing in the door opening but not entering the room. Not all forms of attendance are clinically meaningful. To prevent harmful situations, including SUDEP, it is crucial that caregivers approach the child. We therefore considered our definition of attendance more relevant, while we acknowledge that it does not cover the full spectrum of caregiver responses. We could not determine whether caregiver attendance was triggered by the NightWatch alarm or by other cues such as seizure‐related sounds or movements. For some attended seizures, caregivers arrived before the alarm, making it unclear whether they would have responded to the alarm had other cues not been present. Even when attendance followed the alarm, it remains uncertain whether the alarm was the actual reason for their response. Our study could therefore not demonstrate the added value of the NightWatch on caregiver responsiveness. Our study included a small number of children which, combined with significant variance in the number of alarms per child, resulted in limited power for our analyses. This may have prevented us from detecting some effects, although the study was capable of detecting heterogeneity among participants and several determinants. Participants were selected based on a high seizure burden, making them not generalizable for the whole epilepsy population. This group does probably resemble the majority of the key users, since we expect that, given the performance of the current Seizure Detection Devices (SDDs), these devices are best suited for people with a relatively high seizure frequency. ¹² , ¹⁷ Another limitation is that we had limited information about the caregivers themselves, such as their caregiver training, burden of care, baseline alertness, fatigue, or prior experiences with seizures. These factors could have influenced their response patterns but were not accounted for in our analyses. Future research should address this gap by incorporating caregiver questionnaires to explore the reasons underlying attendance patterns in response to alarms.

Our study provides novel insights into caregiver responses to nocturnal seizure alarms in a home setting, a topic that, to our knowledge, has not been explored extensively in the existing literature. Previous research has primarily focused on response times to seizures in epilepsy monitoring units (EMUs). For instance, one study reported a median response time of 22 s during the day, with a significantly longer response time at night (49 s) for EMU nurses, with 16% of seizures going unattended. ¹⁸ Another EMU study compared response times during both day and night, finding that caregivers responded more quickly than staff (11 s vs 22 s), with only 2% of seizures going unattended, one of which occurred when a caregiver was present. ¹⁹ In contrast, our study observed a larger proportion of unattended seizures and slightly longer response times. This difference may be due to the fact that in the EMU, attendance also served a diagnostic function, requiring prompt responses to perform necessary tests during and after seizures. In addition, caregivers in the EMU were constantly present in the EMU room, while in a home setting, they may have been further away when the alarm sounded. Our data therefore provide a unique insight into the natural behavior of caregivers. It should be mentioned that not all attendance can be attributed to NightWatch use. For example, a seizure might be noted by caregivers before detection (average latency for detection is estimated at 13 s). ²⁰ This is reflected in the attendance of some seizures before the alarm went off.

Several factors could play a role in the variation in attendance between caregivers. Previous literature shows that there is variation in the wishes of SDD users regarding the acceptance of the number of false positive alarms, which might also vary with the number of seizures occurring. ²¹ This could explain the differences in attendance, as some alarms might be unwanted; caregivers may assess from a distance and choose not to approach if they judge the alarm to be unnecessary.

Variation in caregiver behavior could also be attributed to differences in training and education on seizure hazards, as we know from previous studies that knowledge gaps exist for SUDEP and first aid during seizures. ²² , ²³ Another explanation could be varying levels of fear and anxiety, with some caregivers reacting more consistently due to heightened concern about the risks associated with seizures.

Some variance was explained by specific factors. The strongest factor was the presence of seizure‐related sounds. Notably, in some of the alarms, the caregiver arrived before the alarm went off, suggesting that not all attendance related to NightWatch alarms and that seizure sounds themselves may act as an effective and independent alert. In instances where the NightWatch alarm coincided with seizure‐related sounds, the sounds may have served as an additional marker convincing the caregiver to attend the event. The relationship between attendance and seizure frequency highlights how a high seizure burden can impact caregiver responsiveness. It is understandable that frequent seizures may lead to fatigue or desensitization, potentially reducing attendance. We initially expected that SDD performance would impact caregiver attendance, for example, by reducing attendance in those with high false alarm rates or high alarm burden, causing “alarm fatigue”. Although we did find that attendance rates were higher for true positive alarms than for false ones when comparing individual attendance rates, we did not observe any influence of alarm nature in our multivariate model, which considered all alarms while adjusting for clustering within individual caregivers. In addition, the number of prior alarms that night was associated with lower attendance rates but did not remain significant in the multivariate model. Instead, a higher PPV showed a trend toward a lower odds of caregiver attendance, suggesting that poorer performance might actually encourage attendance. The F1 score, another metric for SDD performance, did not show any significant impact on attendance. Furthermore, our data suggest that caregiver characteristics rather than SDD performance are the main determinants of attendance. This parallels previous findings that caregivers' overall perception of the device was not solely shaped by its performance, but also by parent‐related factors, such as their flexibility or perceived burden of care. ²⁴ We found that some caregivers consistently did attend or did not attend in response to the alarms, regardless of whether the alarm was a true or false positive. This likely contributed to the unexpected patterns we observed in our analysis. These consistent response behaviors might have overshadowed the potential impact of SDD performance on attendance. Although we considered the CSI, we had no other data on caregiver characteristics to explain the observed variability in attendance. We observed a trend for lower caregiver attendance in older children, possibly due to an assumed self‐sufficiency of older children. This is also reflected in previous literature, which indicates that caregivers are more likely to use monitoring devices for seizure detection when the child is younger. ²⁵

Our study underscores the dynamics of body position of nocturnal motor seizures. One study found that 9% of observed seizures ended with the person with epilepsy in a prone (stomach) position, with 69% of these seizures not beginning in that position. ²⁶ Although the proportion of seizures ending prone was lower than in our study, this particular study was conducted in an EMU setting and only studied subjects older than age 15 with convulsive seizures. Our study suggests that seizures ending prone are even more frequent in a more realistic, home‐based environment, highlighting once again the critical importance of seizure detection in everyday settings. We found that the median seizure semiology duration and the time until the first spontaneous movement were shorter in unattended seizures compared to attended seizures. Although this result is unexpected, it does not contradict the hypothesis that attendance supports SUDEP prevention. A possible explanation is that attendance is influenced by seizure severity—longer seizures with extended recovery times are more likely to be attended.

This research emphasizes the importance of encouraging caregivers of people with epilepsy using an SDD to respond promptly to alarms signaling nocturnal seizures. Although our findings indicate generally good attendance, it is not optimal in all cases. Notably, more than one third of seizures resulting in a prone position went unattended, which could lead to potentially dangerous situations. We observed that some caregivers never respond to the alarms. This implies that an SDD, even if it functions accurately, does not guarantee adequate attendance. Optimal SDD implementation would therefore require targeted education to guide caregivers on the importance of attending nocturnal major motor seizures and appropriate first aid response. On the other hand, we also found that some caregivers consistently respond to all alarms, including the false positive ones. Our work thus underscores the importance of counseling on the background and purpose of an SDD system, as not all alarms warrant attendance.

AUTHOR CONTRIBUTIONS

Study Conception and Design: The initial idea and framework for the study were conceptualized and designed collaboratively by A.T.B., F.S.L., and R.D.T., who contributed to refining the research objectives, hypotheses, and overall approach. Data Collection: A.T.B. was responsible for gathering the data required for the study, including designing and implementing the data collection process, ensuring accuracy, and maintaining the integrity of the dataset. Analysis and Interpretation of Results: The analysis and interpretation of the data were jointly carried out by A.T.B., F.S.L., and R.D.T. This included statistical evaluations, synthesis of findings, and drawing meaningful conclusions in the context of the study's aims. Draft Manuscript Preparation: A.T.B. and R.D.T. took the lead in drafting the manuscript, organizing the findings, and presenting them in a coherent narrative that aligned with the study objectives and methods. Review and Approval: All authors (A.T.B., F.S.L., and R.D.T.) contributed to the critical review of the manuscript, ensuring the accuracy and completeness of the results and interpretations. Each author provided feedback and revisions, culminating in approval of the final version for publication.

FUNDING INFORMATION

This work was supported by EpilepsieNL (2023–04) and the Anna Teding van Berkhout Stichting. The funders had no role in the study design, data collection, analysis, interpretation, or decision to submit for publication.

CONFLICT OF INTEREST STATEMENT

Roland D. Thijs (R.D.T.): Reports lecture and consultancy fees from Medtronic, UCB, Angelini Pharma, Theravarance, Zogenix, Novartis, LivAssured, and Arvelle, as well as grants from Medtronic and NewLife Wearables. None of the other authors has any conflict of interest to disclose. The authors are part of the Dutch TeleEpilepsy Consortium; none of the authors have shares in LivAssured, the company developing the NightWatch device, nor have they received or will they receive any compensation referring to future sales of the NightWatch. LivAssured had no role in the study design, analysis, or decision to submit the manuscript. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

ETHICS APPROVAL STATEMENT

The PROMISE trial was approved by the Medical Research Ethics Committee of the University Medical Center Utrecht.

PATIENT CONSENT STATEMENT

Written informed consent was obtained from the participants or assent from parents or legal guardians.

CLINICAL TRIAL REGISTRATION

The PROMISE trial was registered at the Dutch Trial Registry (NCT03909984).

Bosch AT, Leijten FSS, Thijs RD. How often do caregivers attend their child when a seizure detection device alerts for a nocturnal major motor seizure? Epilepsia. 2025;66:3810–3821. 10.1111/epi.18534

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.