Table 2.

Distributed systems based solutions for Alzheimer’s disease.

Sr. No.	Study	Participants and Study Design	Evaluation Metric	Technology Used	Measurements	Key Findings	Study Limitations	User-Centred Design
Diagnosis
1	Akl et al., 2015 [27]	N = 97; Age ≥ 70 yrs; Longitudinal study (3 years)	MMSE, CDR (at start and end of each year) and a control group.	Passive infrared sensors	Walking speed, daily home activity, visitor visits and absence from home.	Automatic detection of MCI with AUROC = 0.906 and precision-recall = 0.93 (best results with 24-week window) using Random Forest and State Vector Machine.	The smart home sensors were placed in individual’s home where they lived independently. Not suitable for the multi-person setting. Moreover, if a week activity was missed, the whole 24-week window must be discarded.	No
2	Akl et al., 2017 [28]	N = 97; Age ≥ 70 yrs; Longitudinal study (3 years)	MMSE, CDR (at start and end of each year)	Passive infrared sensors	Walking speed, home activity	The walking speed of the elderly does not slow down while transitioning into MCI. Linear modelling. AUROC = 0.716 and precision-recall = 0.706 (12-week window)	Not suitable for the multi-person setting. A small population transitioned to MCI, for validation a larger population required.	No
3	Kaur et al., 2019 [29]	N = 374; Age = 60–90 yrs; (Cross-sectional study)	Interviews from 100 patients out of the 374. (Usability, accuracy, convenience).	Radiofrequency Identification Tags (RFID) on the active IoT devices such as mobile phones	Medical history, MMSE, patient location tracking, memory exercise games data.	Diagnoses AD patients using Ontology Bayesian network (AUROC = 0.76 and F1 score = 0.934).	To build the ontology system, data were collected from a doctor. Moreover, the system uses GPS for the patients; hence, if there is no net connection, the performance might get affected.	Yes
4	Alberdi et al., 2018 [31]	N = 29; Age = 73–97 yrs; Longitudinal study (2 years) (N = 13 healthy, N = 10 at risk and N = 6 had cognitive difficulties)	ARM Curl, Timed Up and Go Test, RMANS, PMRQ and GDS (at the start of the study and every 6 months).	Magnetic switch sensors (doors), RFID tags on objects at home, motion sensors	Mobility, cognition and mood changes were monitored using smart-home sensors.	Through regression analysis, mobility, depression and cognition can reliably be detected by in-home sensors and can predict the onset of AD. Results suggest that mobility is more related to behaviour abnormalities than to cognition. Sleep patterns and toilet visits can diagnose depression.	It is difficult to assess which data from all the sensors to use. Data storage and processing of useful features are a challenge. As indicated all features are not used, and there is class imbalance. Not suitable for the multi-person setting.	No
Continuous Monitoring
5	Alvarez et al., 2018 [36]	N = 18; Age = 55–94yrs; Longitudinal study (10 weeks)	Control group versus patient experimental group.	Wearable bracelet, Zenith Camera, Microsoft Kinect v2, Binary smart sensors, wireless sensor network beacons.	Physiological signals (Heart rate, skin temperature) and motion location tracking, gait, activity tracking using Kinect Microsoft and cameras.	Using multi-sensor data, the daily motion, night patterns, and detection of abnormal behaviour, fall, and gait abnormalities are evaluated. The distinction between abnormal and normal behaviour was detected with an accuracy of 98.4% using probabilistic models.	The use of cameras indicates privacy infringement, and the study included patients who did not experience cognitive impairment.	No
6	Karakostas et al., 2020 [37]	N = 109; (Mean Age = 69.8yrs) Cross-sectional	Healthy controls = 38 MCI = 44; AD = 27	Cameras, Wearable bracelets, Static sensors (on objects) and Smartphones	Camera videos to assess IADLs such as bill payment, making tea, making a phone call and walking.	Using Analysis of variance, the study shows that as the disease progresses the autonomy of the person reduces.	Use of cameras; limited IADLs and those were not a true reflection of real-life as the tasks were designed to be simple in the clinic.	No
7	Braley et al., 2019 [38]	N = 15; Age = 59–91yrs (Cross-sectional)	CDR > 0.5, Diagnosis and statistics of mental disorders (all met criteria)	Smart home sensors, speakers, actuators, TV screens	ADLs (cooking, washing, cleaning and using the telephone) were monitored	The elderly adults with more cognition problems faced difficulty while responding to voice instructions and also showed signs of stress.	Use of Cameras, limited experimenter view as only 3 cameras used, video viewing without voice, the possibility of highly subjective outcomes	No
8	Cavallo et al., 2015 [39]	N = 14 AD patients Age ≥ 80yrs (Cross-sectional) Duration depended on carer and patient availability	MMSE Qualitative assessment through experience reporting	Inertial sensors, pressure mats, Door magnetic switch sensors, Zigbee, GSM and GPS on a portable device	Home exit-stay; bed-chair monitoring, location tracking outdoor, social behaviour	Functional monitoring and alerts in case of emergency were given, but the acceptability and usability rates according to the patients and their caregivers were 50% and 30%, respectively	Not fixed duration of the experiments, all systems not utilized by all participants, low acceptability and utility rates	Yes
9	Radziszewski et al., 2017 [40]	N = 1; Age = 78yrs (Cross-sectional) (42 days)	MoCA and Dementia rating at the start and end of the study	Ambient sensors, wristwatch, Smart objects including LED bulbs, lamps, Actigraph light paths, Z-wave controllers	Day and night-time activities	K-means algorithm to detect activity and guide the patient back to bed using voice signals.	Not suitable for the multi-person setting. The study duration was insufficient to determine nocturnal patterns, limited no. of participants.	Yes
10	Lyon et al., 2015 [41]	N = 480; Age ≥ 70yrs; (Longitudinal) 8 years	MMSE, CDR	Motion sensors, magnetic switches, screens, internet, Zigbee	Mobility and activity patterns including socializing	Regression models showed that progression in cognitive decline and mood can be evaluated within six months. Wearables used as ground truth	480 single patient homes. Activity recognition not fine-grained such as in the kitchen, activity performed is washing, making tea or doing something else.	No
Therapy
11	Lazarou et al., 2019 [42]	N = 18; 12 MCI; 6 AD Longitudinal 4–12 months	Diagnostic and Statistical Manual of Mental Disorders (DSM-V) MMSE before and after study 3 control groups (4MCI 2 AD each)	Off-the-shelf IoT devices, (tags attached to drug-boxes), motion sensors, wearables, cameras and ambient sensors	Motion, Sleep, Activity patterns, physiological signals and observations using depth cameras.	The second group received intervention on a self-reported basis, and the third did not receive any alerts at all. The group that received automatic alerts and reminders showed significant improvement in cognition.	Only 6 participants received interventions. Limited but complex sensor data. Single patient per home.	Yes

Table Acronyms: AD: Alzheimer’s Disease; MCI: Mild Cognitive Decline; MMSE: Mini-Mental State Exam; CDR: Clinical Dementia Rating; MoCA: Montreal Cognitive Assessment; RBANS: Repeatable Battery for the Assessment of Neuropsychological Status (RBANS); PMRQ: Prospective and Retrospective Memory Slips; GDS: Geriatric Depression Scale; AUROC: Area Under Receiver Operating Characteristic Curve; HC: Healthy Controls; ADL: Activities of Daily Life; IADL: Instrumental Activities of Daily Life.