Participants. The Maine Sleep and Aging Study was conducted in accordance with the University Maine Institutional Review Board (Federalwide Assurance #: FWA00000479; IRB Organization (IORG) #: IORG0000642). All experimental protocols were approved by this IRB and all methods were carried out in accordance with relevant guidelines and regulations. Informed consent was obtained from all participants at the first home visit. Recruitment of independent community dwelling, aging adults were sought through advertisement within a 3-hour catchment area and through referral from Northern Light Mood and Memory Center. Inclusionary Criteria: Participants were ages 60-90 years who lived independently in the community, English speaking and possessed adequate vision with correction. Exclusionary Criteria:Any medical evident through exam or imaging of a neurological, psychiatric or medical disorder other than AD as a cause for aMCI (e.g. partial list: more than one cerebral infarct, poorly controlled diabetes, hypothyroidism, parkinsonism, parasomnia or REM sleep disorder, developmental disability, schizophrenia, etc.); acute symptom onset; depression score on CES-D≥16 (Center for Epidemiological Studies Depression Scale); Epworth Sleepiness Scale score≥10; restless or periodic leg syndrome; hypnotic/ psychotropic medication change (38). History of depression/anxiety disorders or > 5 years drug/ alcohol abuse but in recovery was allowed. Figure 1 (see Supplement) illustrates that 168 participants were recruited and 73 were excluded based on distance (n=22), exclusionary criteria (n=39), or withdrawal or failure to schedule (n=12). The final sample of 95 participants completed the home sleep testing protocol.
Included in the sample was a clinic-diagnosed aMCI patient group (MCI-DX; n=17) recruited from Northern Light Healthcare, Bangor, ME by co-author, C.S. The remaining cohort was evaluated for aMCI diagnosis by an expert consensus panel (co-authors C.S., F.A., J.A,) using demographics, full battery of neurocognitive testing, and comorbidities (MCI-Consensus group, MCI-CON; n=33). MCI-CON group met aMCI criteria for immediate and delayed verbal and visual memory tasks, but were normative on intelligence and cognitive reserve measures using Petersen (27) and the DSM-5 (3) criteria; Montreal Cognitive Assessment (MoCA) score of 26-19 (inclusive); delayed recall subtest score of 0 or 1 out of 5; normal circadian entrainment to nighttime sleep hours; and decision-making capacity to independently consent to research. The consensus panel identified pre-clinical aMCI (MCI-CON; n=33) and normal cognition groups (NC; n=45).
Instrumentation and Measures
SleepMove Mattress Device and System. The wireless sensor mattress pad is 1 mm in thickness within a sealed polyethylene cover that is placed within an antimicrobial-resistant fitted sheet positioned beneath the bottom bed sheet on the home bed for 2 nights. The sensor device design uses 32 distributed piezo-resistive pressure sensors to record movement and respiratory signals using 54 Hz average sampling rate. The receiver unit under the bed records data to a Micro SD card. System algorithms filter sleep movement (SM) bouts (periodicity circa 0.198 Hz) and respiration (periodicity circa 0.366 Hz) into two data streams. Time series algorithms identify SM-RR coupling using a concordance estimate of time lag (TL, ms) to capture the time delay between SM and RR, thereby calculating the strength of the coupling relationship between these two events. Respiratory frequency is indexed by segmentation of RR rate in 2 min bins. These data form the basis of artificial intelligence (AI) analyses described herein and in technical detail in Khosroazad, Abedi & Hayes (under review) (28).
Actigraphy. The participants wear Actiwatch 2 (Philips Respironics, Philips Actiware 6: v.6.0.9), a watch-like actigraphy monitor on the non-dominant wrist for 7 consecutive nights beginning on night 1 of the study. Actigraphy software provides standard sleep-wake metrics.
Self-Report. Stanford Sleepiness Scale (SSS): queries subjective sleepiness with a 7-point Likert scale for 7 consecutive days (29); Epworth Sleepiness Scale (ESS): asks about daytime sleepiness and situations (30); Pittsburgh Sleep Quality Index (PSQI): assesses sleep quality over a 1-month time interval with several subscales and a composite score (31). Center for Epidemiological Studies Depression Scale (CES-D): 10-item depression scale used for older adults (32).
Neurocognitive Testing. Montreal Cognitive Assessment (MoCA) screening tool for MCI that includes short term and working memory, visuospatial, attention, orientation and executive functioning (range score for MCI is 19-25) (33). Brief Visuospatial Memory Test-Revised (BVMT-R): (34). Hopkins Verbal Learning Test-Revised (HVLT-R): r (35). Boston Naming Test (BNT): confrontation naming task (36). Proxies for cognitive reserve were measured by the Vocabulary subtest of the Wechsler Adult Intelligence Scale (WAIS III, 37), and American National Adult Reading test (AMNART; 38), both of which are well-established means of estimating cognitive reserve.
Protocol. Figure 2 (supplement) shows the testing sequence. Participant eligibility was determined by phone interview screening. On study day 1 (night 1), two team members arrived at the home at approximately 1700 h to interview the participant; administer the MoCA and query: 1.) medical comorbidities; 2.) SES (socioeconomic status) through longest held career, e.g., secretarial, professional, homemaker, etc.; 3.) lifestyle and substance use using current and former intake of alcohol, tobacco and other use; weekly exercise and activities; 4.) obesity risk estimated through Body Mass Index; and 4.) self-report inventories of sleep problems (see Instrumentation and Measures).
The participants slept alone in the home bed on the mattress device for 2 consecutive nights. The signal acquisition hardware was placed under the bed connected to the mattress pad using a flat wire powered by battery. After night 2, the device was picked up and data transferred electronically to the laboratory. Participants reported bed and wake times and wore the actiwatch for 7 days. Neurocognitive testing was conducted in a home visit for approximately 90 minutes. The participant received a $100 Visa gift card.
Statistics and Data Processing. Multivariable, mixed model regression using group as a factor were used for data analysis for actigraphy, neurocognitive and sleep self-report data moderated by custom covariates as described (e.g., age, education, BMI) with IBM-SPSS, V.26. Demographics were analyzed with one-way ANOVA for continuous variables and Kruskal-Wallis h test for categorical variables. Imputation was used in <5% of the data.
For the SM-RR coupling output measure, a supervised Neural Network (NN) approach was used and is compared with Gaussian and Kernel approaches. SM and RR signals were separated in the frequency domain using FFT and band pass frequency filters as described in Instrumentation and Measures. SM-RR coupling Time Lag (TL) was examined in 10 minutes windows to determine the Probability Density Function (PDF) of the maximum time lags for each case. Receiver Operating Characteristic (ROC) statistics were applied to determine the sensitivity and specificity, AUC and c statistic with confidence interval. ROC plot illustrates the diagnostic binary classifier system, in this case, MCI vs. NC as its discrimination threshold is varied. Positive predictive values and negative predictive values were calculated to identify MCI status using sensitivity (true positives) and specificity (false positives). NN and the cross-validation method, Leave-One-Out Covariance (LOOCV) was used to train and test the network (39). Confusion matrices of true positive, true negative, false positive and false negative were calculated based on the verification group that was placed in an incorrect category by the trained network.