Sleep in Dementia Caregivers and the Effect of a Nighttime Monitoring System


Dr. Meredeth Rowe, P.O. Box 100187, Gainesville, FL. E-mail:


Purpose: The purpose of this study was to determine if a nighttime home monitoring system, designed to track the movements of a care recipient with dementia, would relieve worry and improve sleep in caregivers of persons with dementia.

Design and Methods: In this controlled clinical trial, 49 dementia caregivers were followed for up to 1 year. Sleep was measured for 7-day intervals at nine points in time using actigraphy and a sleep diary.

Findings: Although the experimental caregivers generally reported that the system was “of great help” in relieving worry about nighttime activity, no significant group differences were found using multilevel modeling analyses. With regard to total sleep time, time awake after sleep onset, and sleep quality, multilevel models did not demonstrate any changes in sleep between groups, either averaged over time or for the interaction of group and time.

Conclusions: Since previous analysis of our qualitative data suggested improvements in caregiver worry and sleep, problems other than night awakenings may be perpetuating the sleep problem. Future studies should include testing of multimodal sleep interventions.

Clinical Relevance: Caregivers have high amounts of unwanted wake time during the night and additional research is needed to identify effective interventions to improve their sleep.

Informal caregivers are a crucial component of the system for providing care to persons with chronic health conditions. Arno, Levine, and Memmott (1999) estimated that caregivers provide 22 to 26 billion hours of care each year in the United States. The economic value of that care in 1997 dollars was estimated at $196 billion. In a study using 1998 data that estimated the cost of informal caregiving to a single individual with dementia, the cost was estimated to be $18,385 overall and $6,295 for direct caregiving time (at $8.20/hr; Moore, Zhu, & Clipp, 2001). Clearly, neither the cost in human hours nor the cost in dollars could be sustained by any private or government program, especially with the increased cost of care since these data were analyzed.

Recent research, however, has demonstrated that caregiving exacts a toll, worsening the overall physical and psychological health of informal caregivers, particularly when caregiving is perceived as stressful (Lee, Colditz, Berkman, & Kawachi, 2003; Schulz & Beach, 1999; Schulz, O'Brien, Bookwala, & Fleissner, 1995). While caregiver stress is composed of a number of factors, caregivers of persons with dementia (PWDs) cite inadequate sleep as a predominant problem (Creese, Bedard, Brazil, & Chambers, 2008; Willette-Murphy, Todero, & Yeaworth, 2006). A key factor impacting sleep is the nighttime activity common in many PWDs. Of all changes in PWD sleep patterns (e.g., longer sleep intervals, early morning awakenings, nighttime awakenings), caregivers rated nighttime awakenings the most stressful (McCurry et al., 1999). Estimates of nighttime activity in PWDs range from 19% to 54%, with most studies finding rates in the 30% to 40% range (McCurry, Logsdon, Teri, & Vitiello, 2007).

Caregiver Sleep Changes

Approximately 60% of caregivers of PWDs report nighttime awakenings in their care recipients (Creese et al., 2008; Wilcox & King, 1999). Creese et al. studied spousal caregivers of PWDs and found that 58% reported sleep quality as “poor” or “fair,” with sleep disruptions occurring due to PWDs wandering, using the bathroom, or exhibiting restlessness. Lee, Morgan, and Lindesay (2007) studied the sleep of these caregivers using both sleep quality questionnaires and objective sleep measures before, during, and after a planned respite stay for the PWD. Of the 39 caregivers, 77% exceeded the cutoff score for clinically disturbed sleep at baseline. They demonstrated prolonged wake after sleep onset times and poor sleep efficiency. However, during the time that the PWD was in respite care (2-week interval), caregivers had significant improvement in total minutes asleep and sleep quality. These improvements were not sustained once the PWD returned from respite care, suggesting a direct association between caregiving and sleep disruption (Lee et al., 2007).

Sleep complaints of caregivers have not always been corroborated by subjective or objective measures of sleep, nor is there a close match between actual sleep disruptions related to providing supervision and reported sleep deficits (Hoekert, der Lek, Swaab, Kaufer, & Van Someren, 2006; McCurry, Vitiello, Gibbons, Logsdon, & Teri, 2006). In a recent review, McCurry and colleagues (2007) proposed that changes to caregiver sleep result from a combination of several factors. These include disrupted routines, caregiver burden, and declining health of caregivers, who tend to be older adults.

We propose that one of the important perpetuating factors is caregivers’ reported worry that they will not awaken when their assistance is needed. Worry has been hypothesized to be a component of insomnia (Harvey, Sharpley, Ree, Stinson, & Clark, 2007; Jansson & Linton, 2006), and caregiver worry regarding getting enough sleep and being alerted to awakenings may perpetuate a cascade of events leading to chronically poor sleep patterns. In our clinical experience, caregivers report that they “sleep with an open ear” or “try to sleep lightly” in order to make sure they will be awakened. In this respect, they exhibit similar high levels of fatigue documented in mothers caring for infants sent home with apnea monitors, where increased surveillance is required during mothers’ sleeping hours (Gill, Robison, Williams, & Tinetti, 1999).

When the PWD is up at night, serious consequences can occur, including falls and unattended home exits (Rowe & Fehrenbach, 2004; Rowe, Feinglass, & Wiss, 2004). PWDs are at particular risk for falls, and nursing home placement may be required if injuries result (Horikawa et al., 2005; Rowe & Fehrenbach, 2004). Approximately 40% of PWDs who died after becoming lost in the community exited the home during the night (Rowe & Bennett, 2003). Thus, it is often critical for caregivers to awaken to provide supervision when the PWD has left the bed at night.

The intervention tested in this study was a home monitoring system designed to reliably awaken a caregiver when the PWD left the bed at night. We hypothesized that providing reliable information regarding the nighttime location of the PWD would improve caregiver sleep by eliminating unnecessary awakenings and by reducing the need to lighten sleep patterns in order to ensure awakening if supervision was needed. The home monitoring system was developed for use in homes of PWDs, and details of that system have been previously reported (Rowe, Lane, & Phipps, 2007). Briefly, this system uses components of a home security system, as well as a bed occupancy sensor, to provide information regarding movement of the PWD out of bed and through the home. Information comes in the form of text, voice, and alarm alerts on a keypad at the caregiver's bedside. A bed occupancy sensor is used to determine when the PWD leaves the bed. The caregiver is notified via text and voice messages when this happens, as well as when the PWD enters each room of the home. An emergency alert is provided in the event of an unsafe situation, such as an outside door being opened. Provided with this accurate information, the caregiver can decide what level of supervision to provide. This system is operated completely by the caregiver and does not require any external monitoring by a security company.

As part of the project to develop and test the system, a controlled clinical trial was conducted to test the following hypotheses: (a) Caregivers of PWDs who use the night monitoring system (NMS) will have less worry about unsafe nighttime activity than caregivers not using the system; (b) Caregivers who use the NMS will have longer subjective and objective total sleep times, shorter subjective and objective wake after sleep onset, and better sleep quality than caregivers who do not use the NMS; and (c) Caregivers who use the NMS will have less night-to-night variability in total sleep time and wake after sleep onset than caregivers who do not use the system.



The study was conducted using a pretest-posttest control group design with repeated measures at baseline and posttest months 2, 3, 4, 5, 6, 8, 10, and 12. Although both caregivers and PWDs were recruited, only caregiver data were included in hypothesis testing. Data on sleep were collected for 7 consecutive nights at each data collection point (month) using both subjective and objective measures. Descriptive data about the PWDs are reported.


Fifty-three participants were originally recruited for this study, with 26 caregivers in the experimental group and 27 in the control group. The study inclusion/exclusion criteria for the caregiver were: primary caregiver in the home without provisions for professional care at night; providing care for an individual with a medical diagnosis of Alzheimer's disease or other dementia as reported by the caregiver; 21 years of age or older; expressed concern about or report of nighttime activity in the PWD; have no physical impairments that would prevent the caregiver from providing rapid assistance when alarms sound (e.g., able to walk through the home without assistance); not undergoing active treatment for sleep disorders (e.g., using prescription sleeping medications on a nightly basis or therapies for sleep apnea); able to speak and read English; no cognitive impairments (Mini-Mental State Exam [MMSE] score>27).

Recruitment of subjects was done by placing study notices in caregiver newsletters, the local paper, local memory disorder and geriatric clinics, and during presentations at caregiver meetings and support groups. Clinicians who had provided care to clients with dementia bolstered recruitment efforts by notifying caregivers of these clients about the study. Those interested in participating were instructed to phone the researchers for further information. An initial phone screening for inclusion/exclusion criteria was conducted, and a home visit was scheduled with potential participants. During that visit, full disclosure of the study was provided through a signed informed consent document and discussion. MMSE screening of the caregiver was conducted as well as assignment to group. Two caregivers who had undergone an initial home visit did not participate in the study, one due to a score of less than 27 on the MMSE and the other due to family concerns about participation.

Forty-five caregivers were randomly assigned to the treatment condition (NMS installed in the home) or the control condition. Caregivers in the control group were paid $15 at each data collection point and provided with some educational material unrelated to any study goals (e.g., coping with the holidays or understanding the diagnosis of dementia). All visits were conducted in participants’ homes.

Four caregivers in the experimental condition had participated in a study on system reliability that directly preceded this study. Two caregivers requested the control condition after randomization to the experimental condition; we chose to keep them as control participants due to difficulty in recruiting this population for a long intervention study. Two participants initially assigned to the experimental group had a home/sleeping configuration not amenable to monitoring with the NMS (multiple sleep locations and not sleeping in a bed). Both subjects were asked and agreed to participate as control subjects. Thus, 45 participants were randomly assigned, and 8 of them (4 to each group) were assigned by preference or participation in a previous reliability study. There were no significant differences in the study variables between the group that was randomized and the unrandomized subjects. Of the 53 participants, 4 were excluded from the analyses reported in this paper (1 began sleep apnea treatment, 2 dropped from trial with baseline data only, and 1 had uninterpretable sleep data). There was approximately equal loss of participants over time between the experimental and control groups.

Dependent Measures

Both objective and subjective sleep data were collected simultaneously for 7 consecutive days at each data collection point, consistent with American Association of Sleep Medicine guidelines (Littner et al., 2003). For both objective and subjective measures, two common sleep variables were calculated: total sleep time and wake after sleep onset. Variables from objective sleep measurement are indicated with “o” and with “s” for subjective sleep measurement.

Objective Sleep

Actigraphy, a noninvasive method of monitoring sleep-wake cycles by analyzing wrist activity, was used to collect objective sleep data. The ActiWatch L by Respironics, Inc. was used in this study. This device collects activity on one channel that employs an omnidirectional accelerometer (sensitivity of=0.01 g-force) and on another channel that incorporates ambient light levels (Mini Mitter Co., 2001).

The use of actigraphy in this project has been described elsewhere (Rowe, McCrae, Campbell, Benito, & Cheng, 2008). In summary, bedtime and out-of-bed were generally set by using the times listed in the caregiver's sleep diary. However, if the light or activity data indicated that time did not seem to be accurate (i.e., too much or too little activity or light), these data were also used in setting this interval. Actiware established sleep start as the beginning of the first 10-minute period where no more than one epoch is scored as wake. Sleep end was the last 10-minute period where no more than one epoch was scored as sleep, or the out-of-bed time if the period was shorter than 10 minutes.

Subjective Sleep

Sleep diary data were collected on bedtime, sleep start, number of awakenings, minutes awake during night, wake time, out-of-bed time, and minutes spent napping the previous day, as well as a rating of sleep quality for that night's sleep (Lichstein, Riedel, & Means, 1999). These values were used to create subjective values for total sleep time and wake after sleep onset. For the sleep quality rating, participants were asked to rate quality of the previous night's sleep on a 5-point scale ranging from 1 (very poor) to 5 (excellent). In a recent review of sleep diary validity studies, it was concluded that sleep diaries provide a valid and relatively reliable index of sleep, and strongly recommend the inclusion of both subjective and objective measures in evaluation of sleep (Lichstein et al., 1999).

Caregiver Distress About Nighttime Activity

An instrument generated for this study was used to measure amount of worry-distress about nighttime activity affecting caregivers’ lives. The scale consisted of three items: How much does worrying about your relative (a) being up during the night without you knowing, (b) leaving the home, or (c) being injured affect your life? These items were measured on a Likert-type scale with a range of scores from 0 to 10; the 0 anchor was “not at all distressed.” The 5 anchor was “moderately,” and the 10 anchor was “very distressed.” The worry score is a mean of the above three items.

Construct validity was partially supported in this sample by correlations between the worry score and the negative subscale of the Positive and Negative Affect Scale (r=0.29, p=.056), and with the caregiver rating of whether nighttime activity was the hardest aspect for caregivers to manage (r=0.39, p=.007). The test-retest correlation of the worry score for the control group demonstrated stability of the measure between baseline and month 1 with a correlation coefficient of 0.78 (p<.001).

Experimental subjects were also asked how much the NMS improved distress on each of the three items using a 10-point Likert-type scale, and how helpful the NMS was in improving their sleep and reducing their worry.


For subjects in the experimental group, the NMS was installed in the home after baseline data collection was completed, usually 1 week later when the sleep diary and ActiWatchL were collected. After installation, a reliability period was conducted in which data streamed from the NMS were examined to ensure there were no false negative bed exits or home exits (PWD exited but system did not activate). This period, typically lasting around 3 weeks, was also used to educate the caregiver about system operation and achieve a 100% score demonstrating this knowledge. Caregiver reports and examination by the researcher of the embedded system event log at each visit demonstrated that all caregivers were able to easily learn operation of the NMS and used it nightly during the study. Details of the system's reliability throughout the study period were published elsewhere (Rowe et al., 2007). Briefly, there were no unattended home exits the NMS did not catch, nor were there any caregiver reports indicating that the PWD arose during the night without an NMS alert.

Statistical Analysis

SAS version 9.1.3 (SAS Institute, Cary, NC) was used for data analysis. Univariate and bivariate analyses were conducted for all demographic, clinical, and model variables. Repeated measurement analyses were performed using linear mixed effects models through SAS PROC MIXED with restricted maximum likelihood (REML) estimation. The mixed models can efficiently handle unbalanced longitudinal data and simultaneously account for both among- and within-subject sources of variation. Baseline measure for the variable and quadratic time trends were included in all models. Since most subjects were randomly assigned and there were no differences in baseline values of demographic and clinical variables, no other control variables were included in the models. Employing the “full model” (see Appendix), which allows for between-subject random variation of intercept and slope, we analyzed the four main outcomes of interest: worry score, objective and subjective total sleep time, objective and subjective wake after sleep onset, and standard deviation of each of these four variables. In order to induce normality, square root transformation was performed on the objective and subjective wake after sleep onset variables. The full process for the multilevel models can be found in the Appendix. Due to the number of outcomes examined in this work, an alpha level of 0.01 was used for statistical determination of the main hypotheses of group differences.

With regard to power, computations detailed in our grant application found that by using mixed-effects models for the computation of effects, and incorporating one measure on each participant, 60 subjects would yield a power of approximately 0.70. In this study, we ended up with 49 participants with usable longitudinal data, with 320 month-level time points (∼6.5 per independent subject) and should allow for the study to be adequately powered.


Descriptive Analyses

Caregivers ranged in age from 38 to 86 years with a mean of 62 (SD=11.90; Table 1). Most were female (82%), and the predominant race was white, non-Hispanic (78%). The targeted sample size of African Americans was achieved (18%), but due to the requirement that caregivers must speak English, only four participants were Hispanic. Most caregivers were spouses (51%), followed by adult daughters (38%). Thirty-nine percent of caregivers slept in the same room with the PWD with no significant differences between groups. All had at least a high school diploma, and 86% had at least some college or technical school; there were no significant differences in level of education by group (X2=3.04, p=.38). Thirty-three percent were currently employed. Number of prescribed medications was used as a proxy variable for health, with 30% of caregivers taking no medications, 30% taking 1 to 2 medications, 20% taking 3 to 4 medications, and 20% taking 5 to 10 medications. Over the 12-month study, caregivers’ depression scores averaged 13.22 on the Center for Epidemiologic Studies Depression Scale. This is close to the cutoff for significant psychological distress (16.0).

Table 1.  Descriptive Statistics of Demographic and Clinical Variables
 Experimental groupControl groupTest of difference
  1. Note. MMSE, Mini-Mental State Exam; PWD, person with dementia. at-test. bChi-square test.

Mean age in years (SD)61.52 (13.53)62.81 (10.50)0.37a, p=.71
Gender (female)74%88%1.72b, p=.19
Race (white, not Hispanic)70%85%1.58b, p=.21
Mean depression score11.914.41.00a, p=.32
Relationship to person with dementia  6.50b, p=.16
 Husband13% 8% 
 Adult daughter22%53% 
 Adult son13% 4% 
Prescribed medications2.12.50.64a, p=.52
Work status (% currently employed)30%35%0.10b, p=.75
PWD mean age in years (SD)78.39 (7.68)80.85 (9.29)1.00a, p=.32
PWD gender (female)44%56% 0.21b, p=.65
PWD mean MMSE (SD)13.6714.000.12a, p=.91

Statistics of the demographic and clinical variables by group assignment are displayed in Table 1. There were no significant group differences in these variables. The PWD mean age was 80 years (SD=8.58) with a range of 62 to 97, and 53% were male. Alzheimer's disease (81%) was the most common cause of dementia, followed by vascular dementia (4%), Lewy body dementia (2%), and dementia with unknown cause (13%). The average MMSE score was 13.83 (SD=7.42). There were no group differences on these variables.

At baseline, caregivers were moderately worried about nighttime activity of the PWD (Table 2) and reported moderate disruptions in sleep due to nighttime activity. There was not a significant group difference for these variables at baseline. Experimental caregivers generally reported that the sleep was improved and worry was reduced with use of the NMS. At month 12, 81% (n=22) reported that the NMS was of “great benefit” in improving their sleep or reducing their worry. Thirteen percent of caregivers (n=2) reported that the NMS provided little benefit; these two subjects were wives who slept in the same room as the PWD.

Table 2.  Descriptive Statistics of Baseline Sleep Variables
 Mean (SD)Test of difference t value, p value
Experimental groupControl group
Worry score5.06 (2.63)5.02 (2.67)0.05, .96
Caregiver sleep disruption from nighttime activity5.00 (2.7) 4.72 (2.84)0.33, .74
Total sleep time (minutes, actigraphy)386.77 (54.25) 390.74 (50.36) 0.25, .80
Wake after sleep onset (minutes, actigraphy)44.93 (23.80)50.73 (18.88)0.88, .38
Sleep onset latency (minutes, actigraphy)23.58 (22.58)25.34 (16.49)0.29, .77
Total sleep time (minutes, sleep diary)388.27 (53.34) 370.46 (72.45) 0.95, .34
Wake after sleep onset (minutes, sleep diary)32.12 (26.39)58.50 (37.62) 2.76, .008
Sleep onset latency (minutes, sleep diary)26.86 (20.11)29.17 (18.67)0.73, .68
Sleep quality (sleep diary)3.15 (0.62)3.08 (0.56)0.40, .69

Descriptive statistics for sleep variables are presented in Table 2. In general, values for the sleep diary were worse than those of actigraphy. Subjectively, caregivers reported on average 6.5 hours of sleep, and had prolonged wake after sleep onset times. Objectively, total sleep time values were somewhat greater at 7.45 hours of sleep, and wake after sleep onset values somewhat lower, although still representing a significant amount of unwanted wakefulness. Experimental caregivers had significantly lower subjective wake after sleep onset times than control group caregivers despite no significant difference in objective wake after sleep onset times. A high amount of total wake time during the night was found for all caregivers, in both subjective and objective measures of sleep.

Multilevel Models: Main Study Variables

Worry score Using the model selection and testing procedure described above and in Appendix, no significant effect was found for an interaction or group effect (Table 3). Additionally, using the proposed time effect test (H0: ß2=ß3=0), no significant effect for time was found (p=.33).

Table 3.  Final Model Results for Sleep Variables
 Fixed effectsWorry scoreTSToTSTsSqRt(WASOo)SqRt(WASOs)SQ
  1. Note. Table includes parameter estimates and standard errors. TST, total sleep time; WASO, wake after sleep onset; SQ, sleep quality; o, objective measurement; s, subjective measurement; SqRt, variable transformed using square root.

  2. Asterisks are used to indicate the p value as follows: *p≤.05; **p≤.01; ***p≤.001. Bold rows for the first two models indicate parameter test of interest for model. For the third model, the parameter test of interest is H0: β23=0 while controlling for baseline. For the reduced models, only the parameter tests of interest are presented.

Full modelIntercept (β0)1.54173.16148.482.941.311.27
Baseline (β1)0.510.600.630.500.550.62
Time (days; β2)−0.005−−0.0050.0007
Group (β4)0.72−3.91−16.890.320.950.05
Group * Time (β5)−0.004−0.04−0.04−0.001−0.002−0.0002
Reduced model 1Group (β4)0.12−11.80−21.620.130.690.04
Reduced model 2Time23)0.00001−0.00002−0.0003−2×10−60.00001−2×10−6

Sleep variables Using the model selection and testing procedure described above and in the Appendix, no statistically significant effects were found for any sleep variable for both the interaction and group terms. There were no significant effects for time: objective total sleep time (p=.20); subjective total sleep time (p=.16); objective wake after sleep onset (p=.67); or subjective wake after sleep onset (p=.13).

Sleep quality Using the model selection and testing procedure described above, there were no significant group findings in the interaction test or the main effects test. The test of overall effect of time was not significant (p=.62).

Multilevel Models: Standard Deviation Analysis of Sleep Variables

An additional area of investigation in this study was to determine if the NMS might reduce the high levels of night-to-night variability typically found in caregiver sleep studies. Using the standard deviation from each subject's 7-night sleep measurements, we used the procedures stated above to test for interaction, group, and time effects. There were no significant interaction or group effects for any sleep variable, and no time effects were found for any sleep variable: objective total sleep time (p=.57); subjective total sleep time (p=.94); objective wake after sleep onset (p=.70); subjective wake after sleep onset (p=.83).


Using sleep measured by actigraphy and a sleep diary, there was no significant improvement when the NMS was used by caregivers of PWDs over a 1-year period. Since the tested version of the NMS awoke the caregiver with all PWD nighttime activity, there may have been a varying impact on caregiver sleep. Caregivers who previously slept through some or all of the PWDs’ nighttime activity may have actually been awakened more, while caregivers with frequent awakening from high levels of vigilance may have had improved sleep. The qualitative interviews in a substudy provide some evidence of this effect (Spring, Rowe, & Kelly, 2009). In future development of the NMS, it will be important to add “smart” features; this would allow patterns of safe nighttime activity to be programmed or learned. For instance, in PWDs for whom falls were not a great concern, awakenings to use the bathroom and return safely to bed may not require a caregiver alert, allowing the caregiver to continue sleeping.

However, inability to measure sleep architecture may have influenced the finding of no group differences. In the qualitative study, experimental caregivers reported sleeping more deeply or more settled (Spring et al., 2009). A possible reflection of improved sleep architecture, this differentiation would not be evident with actigraphic measures of sleep. Because caregivers need to rapidly respond to the PWD during the night, it may be difficult to use full polysomnography measurements in the home. Less intrusive measurements of sleep architecture should therefore be considered in future studies testing the effectiveness of the NMS.

Interviewed caregivers also indicated that aspects of caregiving, other than nighttime awakenings of the PWD, caused sleep disruptions (Spring et al., 2009), and other researchers have indicated that sleep changes in caregivers are multifactorial (McCurry et al., 2007). This may explain prolonged sleep onset latency times, with caregivers taking on average 20 to 30 minutes to fall asleep. Presumably, caregivers would not retire when the PWD was in an unsafe situation, so this finding supports the multifactorial nature of caregiver sleep changes. Since sleep onset latency can likely be improved without impacting the safety of the PWD, interventions to improve sleep onset latency should be tested in this group.

Another interesting difference between quantitative and qualitative findings occurred in the analysis of worry about nighttime activity. While a strong qualitative finding was globally improved “peace of mind” and “less worry” (Spring et al., 2009), no quantifiable evidence was found of reduced worry about nighttime activity. Since an instrument was constructed specifically to measure worry, it is possible it did not correctly tap the intended construct. For instance, caregivers may continue to worry about unsafe nighttime activity despite confidence in the NMS to reduce injury or unattended exit risk. They may have developed a habitual pattern of learned/conditioned worry about the care recipient's nighttime behaviors. Although chronic cognitive patterns can sometimes respond to interventions that do not specifically target those cognitions (e.g., worries), it is not uncommon for such cognitive patterns to remain unchanged unless specifically targeted using cognitive restructuring and/or other cognitive techniques. While it makes “intuitive” sense that the NMS would alleviate caregivers’ worries about nighttime activity and also improve their sleep, it is very possible that worry and sleep disturbance have “taken on lives of their own.” Thus, treating caregiver sleep difficulties may require a multifactorial approach in which the original source of nighttime worry is reduced or eliminated with introduction of the NMS and learned cognitive patterns are treated directly.

The idea that the factors contributing to development of a problem are not the same factors as the ones that maintain that problem is not a new one in the sleep literature. According to Spielman's 3Ps Model of the development of chronic insomnia, conditions that precipitate sleep disturbance are frequently not the conditions that perpetuate it (Spielman & Glovinsky, 1991). This may help explain why there were no improvements in either subjective or objective sleep in the present study. Although we targeted one of the factors that may have precipitated poor nighttime sleep in caregivers (worry about the PWD's nighttime activity), factors that may be perpetuating those sleep difficulties (learned-conditioned cognitions and behaviors) were not targeted. Given the multifactorial nature of caregiver sleep problems, it is also not surprising that targeting only a single factor is not sufficient to produce dramatic sleep improvements. Thus, while providing caregivers with the NMS may be a necessary part of any treatment to improve their nighttime sleep, our study suggests it is not likely to be sufficient as a standalone treatment. It is probably best viewed as an important component of a multi-component treatment.


Although significant improvements in sleep were not demonstrated quantitatively, the NMS offered an array of important benefits in homes of PWDs. Analysis of other data collected for this study indicated that not only was nightly use of the NMS associated with a significant reduction in unsafe episodes (injuries and unattended exits; Rowe et al., 2009) but with improved “peace of mind,” reduced worry, and better perceived sleep quality. All experimental caregivers chose to keep and continued to use the NMS at the conclusion of the study (Rowe et al., 2007).

This study highlighted the poor sleep of many caregivers of PWDs, with low levels of total sleep time and high levels of wake time during the night. While the NMS did not significantly improve these sleep variables, other benefits of the NMS emerged in the overall study.


Appendix: Multilevel Models

Full Model


where yi is a (ni×1) response vector for subject i (I=1 to 49) at each time point j (j=1 to ni). Note that ni ranges from 7 to 56 with a mean of 46.

Xi is the (ni×6) fixed effects design matrix for subject i (1 to 49):


0β1β2β3β4β5)’ as the 6×1 vector of fixed effect parameters with interpretations:


Intercept parameter


Effect of baseline value on predicted response


Linear time effect (per day from baseline) on predicted response


Quadratic time effect (per day from baseline) on predicted response


Increment on response for being in treatment group


Increment to group effect for each additional day (interaction effect)

Zi is the (ni×2) random effects design matrix for subject i:


bi=(bi1 bi2)’ is the 2×1 random vector of random effects for subject i corresponding to random intercept and linear increments.

ei is the (ni×1) vector of within-unit error terms for the ni time points within subject i.

Distributional Assumptions

We assume that bi∼N2(0,D) where D is the 2×2 common covariance matrix of the random effects. We assume that D is unstructured for this analysis since the model formation already takes the time factor into account, and there is no reason to believe that the subject-specific intercept and linear values about the population values would have a defined structure:




Variance of the subject specific intercepts about the population intercept


Variance of the subject-specific linear slope effect about the population effect


D21: Covariance between subject-specific intercept and linear effect

We also assume that ei∼Nni(0, σ2Ini) independently from bi. Here σ2 is the common variance parameter of within-unit errors and σ2Ini is the covariance matrix of the random deviations about the ith subject's random regression line. As a result, we assume yi∼Nni(Xiβ, Σi) where:


The primary hypotheses involved testing for fixed effect treatment group differences in rate of change over time (interaction test, H0: β5=0) for the primary outcome variables. If the interaction terms were not significant, they were followed by tests for group differences averaged over time (main effects test, H0: β4=0) in a reduced model without the interaction term. Finally, if no group differences were found, a test of time effect was performed within a model containing only baseline values and the two time terms to see if any trend over time was present in the data (time effect test, H0: β23=0). For all tests, we used the widely employed Kenward and Roger (1997) denominator degrees of freedom adjustment to account for small sample uncertainty of the unadjusted approach.


The authors wish to acknowledge the National Institute of Nursing Research, National Institutes of Health, for providing funding for this study (2R42NR004952, 5R42NR004952). We thank our valuable team members: Stephen Lane, PhD; Annette Kelly, RN, PhD; Claydell Horne, RN, PhD; Judy Campbell, RN, PhD; Brandy Lehman, RN, PhD; Chad Phipps, BSN; Meredith Keller, BSN, and Andrea Pe Benito, BSN.