## Introduction

Sleep disturbances are prevalent in patients with Alzheimer’s disease (AD). Approximately 45% of AD patients have sleep problems1. Sleep disturbances are characterized by symptoms of fragmented sleep, night–day reversal, disrupted sleep–wake rhythms, and so on. This has been known to increase caregiver burden and the risk of institutionalization2,3.

It has been reported that sleep disturbance is associated with cognitive dysfunction of AD patients4. Besides, AD patients who are experiencing sleep disturbances are likely to suffer from mood and/or behavioral symptoms concurrently. A study on patients with mild and moderate AD has reported the association of sleep disturbance with behavioral symptoms, notably aggressiveness5. There are some evidences showing a significant loss in suprachiasmatic nucleus (SCN) of AD patients that is correlated with diminished sleep–wake rhythmicity6. To some extent, underlying circadian disturbances in AD patients could be responsible for their sleep and behavioral disturbances. In addition, insufficient light exposure which might be a cause of sleep and behavior disturbances can be problematic, particularly in institutionalized patients7.

As the light is a central modulator of circadian rhythms, therapeutic exposure of AD patients to bright light is expected to improve their sleep mainly by stabilizing their sleep–wake rhythms8,9,10,11. Besides, bright light treatment (bright LT) can improve the mood and cognitive function of AD patients by stabilizing their circadian rhythm or leading to neural stimulation of brain regions involved in alertness, emotion, and cognition12,13. It has been generally shown that both morning and evening bright LTs have beneficial effects on sleep and behavior of AD patients only when bright LT with enough duration and intensity is given to them14,15. However, many studies, even those with a randomized controlled design, have found no definite effect of bright LT on patients with dementia, including those with AD9,10.

These conflicting findings might result from different timing of applied bright LT, implicating that some patients might have received the light exposure during a sensitive region of the phase response curve while others have not9,10,11,16. Thus, it is important to determine the timing of LT based on individual circadian phase for each patient in order to elucidate the therapeutic effect of LT on AD patients. Another reason for these conflicting findings could be that study subjects are less likely to have greater responses to light since most studies have been conducted on severely demented patients whose SCNs were to be more degenerated8,17.

Our circadian system is known to be more sensitive to short wavelength light. A blue-enriched white light has been reported to have saturation effects even with an intensity of 750 lx. Its therapeutic effects are comparable to those of standard LT for patients with seasonal affective disorder18. However, it remains unclear whether the administered blue-enriched white light can enhance sleep quality of AD patients, although some studies have reported improvement of subjective sleep quality after its intervention in the patients with AD and/or related dementia2,16.

Phase delay in circadian rhythms has been found in AD patients. It is apparently discrete from what is observed in normal aging7,15,19. In a perspective of normalizing circadian rhythms, phase advance caused by LT would be reasonable to investigate its therapeutic effects on AD patients. Meanwhile, in a previous study on healthy young subjects, advance portions of blue light phase response curve have been presented from 09:00 to 20:00 after dim light melatonin onset (DLMO)20, a well-known marker for estimating the circadian phase21. Sletten et al.22 have also reported similar findings in healthy elderly subjects. However, to the best of our knowledge, no study has reported the effectiveness of LT in dementia patients based on their endogenous circadian rhythm.

Thus, the objective of this study was to investigate the effectiveness of timed blue-enriched white light treatment (timed BLT) on sleep, cognition, mood, and behavior of patients with mild and moderate AD based on individual circadian phase using DLMO compared to the effectiveness of timed blue-attenuating LT as control. Our hypothesis was that timed BLT applied based on individual DLMO would have more effects on outcome measures including subjective and objective sleep quality, mood and behavior symptoms, and cognitive function than the control (blue—attenuating LT) immediately post-treatment and at 4-week follow-up.

## Methods

### Participants

Patients with mild and moderate Alzheimer’s disease (AD) who had a sleep disturbance and their caregivers were recruited from the Dementia Clinic at Kangwon National University Hospital, physician referrals, and advertisements between March 2017 and April 2018. The diagnosis of AD and its severity were determined by a psychiatrist or a neurologist using the Diagnostic and Statistical Manual of Mental Disorder, 5th Edition (DSM-5)23 and the Clinical Dementia Rating Scale (CDR)24, respectively. Patient eligibility criteria included a diagnosis of probable or possible AD for major neurocognitive disorder, a CDR score ranging from 0.5 to 2, a sleep disturbance verified by a score of 5 or greater on the Pittsburgh Sleep Quality Index (PSQI)25 and/or by one or more complaints among difficulty initiating sleep (DIS), difficulty maintaining sleep (DMS), and early morning awakening (EMA) for 3 or more days per week.

Patients were excluded if they met the following criteria: (1) current substance related disorders, depressive disorders, or other psychiatric disorders by DSM-5; (2) current medical illness including liver cirrhosis, chronic pulmonary disease, cancer, uncontrolled diabetes, and uncontrolled hypertension; (3) being suspected or diagnosed with primary sleep disorders except for insomnia disorder (i.e., restless legs syndrome, sleep-disordered breathing disorder; hypersomnia, or narcolepsy); (4) a prior history of cerebrovascular disease or central nervous system (CNS) disease, or evidence of CNS injury; (5) current use of any medication affecting their circadian rhythms (i.e., CircadinR); (6) changes in type or dose of taking hypnotics or CNS active drugs for the study period; (7) significant impairment of hearing ability, visual acuity, or language ability which hindered the completion of neurocognitive tests; (8) abnormalities in complete blood cell count, liver function test, urine analysis, electrocardiogram (ECG), or chest X-ray; and (9) pupillary abnormalities on a neuro-ophthalmological examination.

In a single-blind, enrolled patients were assigned into one of the treatment group (TG) and control group (CG) by turns, based on a constant allocation ratio of 1:1 if possible. Fourteen patients (77.36 ± 5.79 years; M:F = 2:12) in the TG and 11 patients (78.55 ± 7.71 years; M:F = 5:6) in the CG completed our study protocol.

The study protocol was approved by the Institutional Review Board of Kangwon National University Hospital following all relevant guidelines and regulations (Protocol ID: KNUH-2016-10-007-001). The study was registered with the clinical studies database Clinical Research Information Service(CRiS), Republic of Korea (KCT0005282, registration date: 08/04/2020; http://cris.nih.go.kr/cris/inc/login.jsp). Written informed consent was obtained from each participant and his/her legally authorized representatives prior to commencement of this study. All procedures were carried out in accordance with principles of the Declaration of Helsinki.

### Procedures

In the laboratory, Dim Light Melatonin Onset (DLMO) was determined from seven hourly saliva samples obtained starting from 6 h prior to the mean sleep onset measured by actigraphy (Actiwatch 2; Philips Respironics, Murrysville, PA, USA) recording for 5 days at baseline. Home-based 1-h timed BLT was applied between 9 and 10 h after DLMO for 2 weeks. The study protocol consisted of assessments of outcome and actigraphical measures at baseline (T0), immediate post-treatment (T1), and 4 weeks after the end of the 2 weeks of LT (T2).

#### Clinical assessment

Enrolled patients were further evaluated for insomnia including its type (i.e., DIS only, DMS only, or DIS and DMS), its duration, and so on. Self-report questionnaires including the Korean Version of Epworth Sleepiness Scale (KESS)26, Korean Version of the Geriatric Depression Scale (GDS-K)27, the Korean Version of Blessed Dementia Scale-Activity of Daily Living (BDS-ADL-K)28, and Clinical Dementia Rating Scale (CDR)24 were then administered to each patient at baseline. Patients with KESS score ≥ 12 were further evaluated to determine whether they would have any primary sleep disorder except insomnia disorder by a psychiatrist through a clinical interview. When these questionnaires were compared between the two groups, there was no significant difference in scores of KESS or PSQI, the distribution of insomnia type, or the duration of insomnia.

#### Outcome measures

Assessments of outcome measures were conducted at T0, T1, and T2. Assessments for AD patients included measures of subjective sleep quality using PSQI, cognitive functions using the MMSE in the Korean Version of CERAD Packet (MMSE-KC), Trail Making Test (TMT-A), Digit Span Test Forward (DSF), and Digit Span Test Backward (DSB), mood using the Korean Version of the Cornell Scale for Depression in Dementia (CSDD-K), Visual Analogue Scale (VAS)-Global Vgor (GV) and VAS-Global Affect (GA), and behavior symptoms using the Korean Version of the Neuropsychiatric Inventory Questionnaire (severity) (KNPI-Qs). Objective sleep was also assessed using sleep parameters from actigraphy (Actiwatch 2; Philips Respironics, Murrysville PA, USA) recording along with sleep diary. Caregiver burden was measured using the Korean Version of Zarit Burden Interview(ZBI-K)29 and Korean Version of the Neuropsychiatric Inventory Questionnaire (distress) (KNPI-Qd)30.

#### Actigraphical measures

The enrolled patients were made to wear the wrist actigraphy for 5 consecutive days from the start of T0, T1 and T2, respectively. Actiwatch recording was done at one-min epochs using wake-threshold value of 40 activity counts that would give the best compromise between detecting sleep and wake in terms of sensitivity and specificity compared to polysomnography31. Actigraphy data were derived with Actiware-Sleep Software (version 6.0.2, Philips Respironics, Murrysville, PA, USA). Quality assessment was performed for actigraphy data prior to analysis. Invalid actigraphy data were excluded due to problems such as technical errors of actiwatch software and fairly deviated sleep–wake patterns caused by special events or alcohol drinking. Data without movement and/or light signal for 1 h or more were treated as missing data. When compliance with wearing the actiwatch was problematic (i.e., no movement or light signal for 4 h/day or more), data for those days were discarded. Objective nocturnal sleep parameters were calculated based on the sleep period from light off to light on according to their sleep diaries. Sleep parameters included time in bed (TIB), total sleep time (TST), sleep onset (SO), sleep onset latency (SOL), wake time after sleep onset (WASO), sleep efficiency (SE), and fragmentation index.

#### Saliva melatonin assay

Participants were not allowed to take chocolate, bananas, aspirin, or analgesics on the day of the saliva melatonin assay. They were asked to visit our research laboratory, 1 h before the starting time of the saliva melatonin assay. During the saliva melatonin assay, they were made to remain where light intensity remains in dim light (~ 15 lx). If necessary, they were allowed to stay with their caregiver during the saliva melatonin assay. Seven hourly saliva samples were obtained starting from 6 h prior to sleep onset measured by actigraphy at baseline. Salivary samples were collected by the passive drool method, in which the participant allows saliva to pool in their mouth and then drools (rather than spits) through a straw into the collection tube. Each time a quantity of saliva was collected less than 2 ml. Samples were then stored in a − 20° freezer until shipped for analysis32. A commercially available Direct Saliva Melatonin ELISA Assay Kit (Bühlmann Laboratories AG, Switzerland) was used for the salivary melatonin assay process, assay process, according to the procedure based in the manual33. The assay procedure follows the basic principle of a competitive ELISA whereby there is competition between a biotinylated and a non-biotinylated antigen for a fixed number of biotinylated antigens bound to the antibody is inversely proportional to the analyte concentration of the sample32. Analytical sensitivity was 0.3 pg/mL32. The DLMOs were determined as a threshold calculated at 2 SD above the average baseline samples. There was one case as a low secretor. For its melatonin values were ranged from 1.6 to 2.3 pg/ml, we could not determine the DLMO. Thus, we assumed the DLMO as the half hour after the last sample time. Two representative salivary melatonin profiles collected under the dim light conditions in our AD patients are given in Fig. 1.

### Measures

#### Questionnaires of outcome measures

##### Pittsburgh Sleep Quality Index (PSQI)25

The PSQI is an instrument to measure sleep quality in clinical populations and PSQI global scores derived by summing seven component scores ranged from 0 to 21. A patient with a global score above 5 was considered to have sleep disturbance.

##### Korean Version of the Neuropsychiatric Inventory Questionnaire (KNPI-Q)30

The severity score of the KNPI-Q (KNPI-Q severity, range 0–36) was used to assess the severity of a patient's behavioral symptoms. The Neuropsychiatric Inventory Questionnaire (NPI-Q) was a caregiver-based questionnaire measuring the presence and severity of 12 neuropsychiatric symptoms during the preceding month for patients with dementia (NPI-Q severity) as well as caregiver distress (NPI-Q distress).

##### Visual Analogue Scale (VAS)34

Subjective activation and mood were assessed by administering eight items in visual analogue scale (VAS), in which responses were indicated along 100-mm lines. Eight of these items included Global Vigor and Affect (GVA) visual analog scales validated by Monk34. The GVA has been shown to be sensitive to sleep loss35 and circadian variation36. GVA was calculated as follows:

\begin{aligned} & {\text{Global vigor }}\left( {{\text{GV}}} \right) = {{\left[ {\left( {{\text{alert}}} \right) + {3}00 - \left( {{\text{sleepy}}} \right) - \left( {{\text{effort}}} \right) - \left( {{\text{weary}}} \right)} \right]} \mathord{\left/ {\vphantom {{\left[ {\left( {{\text{alert}}} \right) + {3}00 - \left( {{\text{sleepy}}} \right) - \left( {{\text{effort}}} \right) - \left( {{\text{weary}}} \right)} \right]} 4}} \right. \kern-\nulldelimiterspace} 4}; \\ & {\text{Global affect }}\left( {{\text{GA}}} \right) = {{\left[ {\left( {{\text{happy}}} \right) + \left( {{\text{calm}}} \right) + {2}00 - \left( {{\text{sad}}} \right) - \left( {{\text{tense}}} \right)} \right]} \mathord{\left/ {\vphantom {{\left[ {\left( {{\text{happy}}} \right) + \left( {{\text{calm}}} \right) + {2}00 - \left( {{\text{sad}}} \right) - \left( {{\text{tense}}} \right)} \right]} 4}} \right. \kern-\nulldelimiterspace} 4}. \\ \end{aligned}
##### Korean version of the Cornell Scale for Depression in Dementia (CSDD-K)37

The CSDD-K was a 19-item tool designed to rate symptoms of depression for those with dementia. It had a cutoff score of 7 for depression in AD patients.

#### Caregiver burden questionnaires

##### Zarit Burden Interview (ZBI-K)29

Caregiver burden was measured using the ZBI-K which contained two subscales—one measuring psychological distress (termed Personal Strain, comprising 12 items), one measuring the impact of caregiving (termed Role Strain, comprising six items).

##### Korean Version of the Neuropsychiatric Inventory Questionnaire (KNPI-Q)30

The distress score of the KNPI-Q (KNPI-Q distress, range 0–60) was used to assess the level of caregiver distress for behavioral symptoms in patients with dementia during the preceding month.

### Statistical analysis

Demographic data, clinical data, sleep characteristics, and objective sleep parameters at baseline were compared between TG and CG using Chi-square test for categorical variables and independent t-test for continuous variables as appropriate. Data of outcome variables measured at three time points were collected, and they were compared using paired t-tests (T0 vs. T1; T0 vs. T2) in the TG and CG, respectively. Considering that some patients did not have complete dataset across three time points, generalized estimating equations analyses (GEE) were performed to examine associations between timed BLT and changes of outcome variables across three time points (T0, T1, T2). Normal as the distribution and Identity as the link function were used to estimate 95% confidence intervals. The model was fitted assuming an exchangeable correlation structure with robust standard errors. The main effects of group and time, and the interaction of Group × Time were analyzed with GEE. Outcome variables were being treated as a continuous variable in the GEE models.

All statistical analyses were performed using SPSS software package, version 18.0 (SPSS Inc., Chicago, IL, USA). Two-sided p values of less than 0.05 were considered statistically significant.

## Results

### Demographics data and baseline clinical characteristics

There was no significant difference in demographic data (including age, gender ratio, education level) or in clinical characteristics (including scores of ADL, GDS-K, and MMSE-KC, distribution of CDR scores) between TG and CG (Table 1).

### Changes in outcome variables across three time points

Means ± standard deviation of outcome variables for TG and CG at each time point (T0, T1, and T2) are shown in Tables 2 and 3. Significant results from paired t-test are presented on Fig. 3. In paired t-tests on the PSQI score, the TG showed its significant decrease immediately post-treatment and at 4-week follow-up of the LT, compared to those at baseline (p < 0.01). In paired t-tests on the ZBI-K score, the TG showed its significant decrease at 4-week follow-up of the LT (p < 0.05), and the CG did its significant decrease immediately post-treatment (p < 0.01).

### Generalized estimating equations (GEE) models for changes of outcome variables

GEE models were used to estimate changes of outcome variables at T1 and T2 with respect to T0 and differences in changes of those variables between TG and CG (Table 4).

Regarding subjective sleep quality, there was no significant main effect of time on PSQI scores at either T1 or T2 with respect to T0. However, there was a significant group-by-time interaction on PSQI scores at T2 with respect to T0, indicating that a reduction of PSQI score from baseline to 4-week follow-up of the LT in TG was significantly greater than that in CG (OR: 0.02, p = 0.026). Regarding objective sleep, there was no significant main effect of time on objective sleep parameters at either T1 or T2 with respect to T0 while there was significant main effect of group on SE, WASO and FI (p < 0.05 or 0.01). There was no significant group-by-time interaction on objective sleep parameters, indicating no significant effect of timed BLT on TG or CG. Regarding cognitive functions, there was a significant main effect of time on MMSE-KC scores at T2 with respect to T0 (OR: 9.10, p = 0.018), indicating a significant change of MMSE-KC score from baseline to 4-week follow-up of the LT. Regarding mood and behavior symptoms, there was no significant main effect of time on scores of CSDD-K, VAS-GV, VAS-GA, or KNPI-Q(s) at either T1 or T2 with respect to T0. There was no significant group-by-time interaction on these scores either, indicating no significant effect of timed BLT on TG or CG. Regarding caregiver burden, there was a significant main effect of time on ZBI-K scores at T2 with respect to T0 (OR ≤ 0.01, p = 0.018), indicating a significant change in ZBI-K score from baseline to 4-week follow-up of the LT.

## Discussion

The objective of this study was to determine whether timed BLT could improve sleep, cognition, mood, and behavior symptoms of home-dwelling patients with mild and moderate degree of AD and whether it could relieve the burden of their caregivers as a consequence of improving these outcomes. In a single-blind, placebo-controlled study with a between-within design, it was hypothesized that timed BLT could beneficially improve these outcome measurements in AD patients more than timed blue-attenuating LT as a control condition. It was also hypothesized that changes in these outcomes immediate post-treatment and at 4-week follow-up of the LT compared to baseline would depend on timed light interventions.