Activity and Behavioral Recognition Using Sensing Technology in Persons with Parkinson’s Disease or Dementia: An Umbrella Review of the Literature

1. INTRODUCTION

Currently there are almost one billion people over the age of 60 years worldwide. [1] Increased age is associated with increased prevalence of neurological diseases such as Parkinson’s disease (PD) and dementia, which affect 15% of the current population. [1,2,3] PD prevalence is increasing faster than any other neurological disorder worldwide [3] and dementia is estimated to affect 139 million people globally by 2050 [2]. In addition, these diseases cost global economies trillions of dollars each year. [1,2] Sensing technologies, such as wearables, are one available solution which can help address limitations and challenges of aging, however, also introduce complex topics such as the suitability of biometric applications within real-world healthcare scenarios. [4]

1.1. Physical Activity Metrics

Physical activity metrics such as step counts, energy expenditure, awake vs. sleep time, and intensity of various classified activities (light, moderate, vigorous), are the most studied digital biomarkers regarding activities. [5] Accuracy challenges of commercial wearables, such as Fitbit, in detection of physical activity biomarkers is well documented and previous studies conclude that beyond use for measurement of step counts in healthy individuals, researchers should use discretion when employing the devices for healthcare decisions. [5,6] Regardless of this, activity is a highly researched metric and decreased overall physical activity is associated with increased mortality risk in older adults. [7] Additionally, reduced capacity of ADLs is associated with increased severity of behavioral symptoms [7], making both key biomarkers for older persons with neurological diseases. This umbrella review defines activities using metrics for physical activity and functional activities of daily living (ADL) such as transitions, sitting, lying, standing, and sedentary activities.

1.2. Behavioral Symptoms of PD and Dementia

Although PD has classically been categorized and diagnosed based on the presence of cardinal motor-symptoms, such as slowness of movements (bradykinesia), rigidity and resting tremor, behavioral symptoms are currently being investigated as an important tool within both the prodromal (10-20 years prior to emergence of motor symptoms) and diagnostic stages (pre-motor and motor).[8] These behavioral symptoms go largely undiagnosed and are under-managed over disease progression. These symptoms can include constipation (50-60%), sleep disorders (60-90%), depression (35%), anxiety (40%), apathy (25-40%), hallucinations (33-42%), fatigue, pain (85%), orthostatic hypotension, urinary incontinence and dementia.[8,9] Likewise, behavioral and psychological symptoms of dementia, also referred to as neuropsychiatric symptoms, effect up to 90% of people with dementia over the course of the illness and include agitation, anxiety, apathy, euphoria, depression, hallucinations, and sleep disturbances.[10] This review focuses on behavioral symptoms which are most commonly measured with sensing technology and associated with a diagnosis of PD or dementia, for example; agitation, anxiety, apathy, depression, and sleep disturbances. Recent studies have shown that the use of sensing technologies, such as actigraphy and machine learning predictive models for behavioral symptoms are feasible and show positive correlations with traditional outcome measures for symptoms such as apathy and agitation. [11,12,13,14]

1.3. Human Activity and Behavioral Recognition in People with PD or Dementia

Human activity recognition (HAR) is a fast-growing field rooted within engineering and computer science. The field has changed rapidly over the last ten years, requiring a cutting-edge and broader approach.[15] A scoping review by Huhn et al. (2022) concluded that use of wearables in research has increased exponentially from 2013-2020, and has included almost 11 million total participants.[16] Combined with other technologies such as Internet of Things (IoT) and Artificial Intelligence (AI), HAR can give high accuracy, precision, and recall of activity classification. [17] HAR is classically defined as a process which identifies and classifies human activities over time based on measurements made by digital sensing devices, wearables, or non-wearables. [18] Gupta et al. (2022) further defines HAR as “the art of identifying and naming activities” and expands the traditional definition to include behavioral symptoms such as agitation, sleep disturbance, pain, and apathy. [19] Common sensors utilized for HAR include accelerometer, gyroscope, magnetometer, radar-sensors (wireless), and Global Positioning Systems (GPS).

According to the Cochrane Handbook for Systematic Reviews [20] an umbrella review, or an overview of reviews, should “use systematic methods to identify multiple systematic reviews on a related research question with the goal to extract and analyze results across outcomes”. We chose to conduct an umbrella review as there is a need to synthesize existing literature to inform future practice and research at this pivotal point in the field of HAR. We found that many systematic reviews and surveys have been published, especially within the last 5 years, regarding the use of sensing technologies for management and observation of activity and behavioral symptoms [21,22,23,24,25]. We chose to narrow this umbrella review to people of advanced age (>65 years) and with neurodegenerative disease (PD or dementia), as these populations continue to grow at unprecedented rates, presenting healthcare systems with immediate challenges to which these sensing technologies and techniques could be a viable solution.

1.4. Research Questions

• What are the current methods, sensing technologies, and AI techniques being used for activity and behavioral symptom recognition in older people with PD or dementia?
• Are statistical analyses and study protocols within the studies heterogeneous and/or reproducible?
• What gaps and possibilities exist in bringing research related to the use of sensing technologies for management and monitoring of human activity and behavioral symptoms into real-world settings and clinical practice for older adults with PD or dementia?

2. METHODS

This umbrella review follows standards set by Cochrane and the Johanna Briggs Institute Critical Appraisal Checklist for Systematic Reviews (JBI) [26] for methodology and reporting guidelines. We in addition use the rationale and guidelines for conducting an umbrella review provided by Choi et al. (2023).[27] To identify relevant and clinically useful studies for the purposes of classifying non-motor symptoms (i.e.: agitation, apathy, sleep disturbance) and functional activities of daily living in persons with PD or dementia, we have excluded systematic reviews primarily focused on gait and pure motor functions. In collaboration with the University of Bergen Medical Library, we developed a working search strategy, including PICO and inclusion/exclusion criteria, which are detailed in Table 1 and Table 2.

2.1. Search Strategy

Initial searches were conducted between September 15 – October 31, 2023, using both MeSH terms and free-text words such as: “wearable electronic devices” OR “sensor technology” OR “fitness track*” AND “complex chronic diseas*” OR “dementia” AND “human activity recognition” OR “activity recognition” AND “aged” OR “older adult*” AND “systematic review” OR “review”. The final systematic search was conducted in Medline Ovid, Embase Ovid, Web of Science, Cochrane Library, Epistemonikos, the Institute of Electrical and Electronics Engineers (IEEE Xplore), and ACM Digital Library databases on October 31, 2023. Restrictions included: publication date January 1, 2018 to 2024. Articles before 2018 were not included as we considered that there has been a substantial amount of newly investigated technologies over the last 5-year period, exacerbated by the pandemic and recent growth within the wearable industry [16], including research conducted using both research-grade and commercial wearables. A complete list of search terms for each database can be found within the supplemental material (Appendix 1).

In accordance with regulations set by the International Prospective Register of Systematic Reviews (PROSPERO), protocol registration was completed prior to data extraction of included articles and accepted in PROSPERO December 9, 2023 (CRD42023487121). Duplicates were removed using EndNote, and Rayyan [28,29] was used as a screening tool. Four reviewers (LDB, LG, MP, LS) conducted an independent, blind screening using Rayyan based on title and abstract from October 27 – November 10, 2023. An unblinded assessment of full-text articles identified as “conflicts” within Rayyan was conducted by four researchers (LDB, LG, BM, MP) between the dates of November 10 – December 10, 2023. PRISMA guidelines [30] were used for reporting of inclusions and study selection processes.

2.2. Data Extraction

Data was extracted by two researchers (LDB, LG). Disagreements were settled by discussion and resolved by a third party (BM) if agreement was not reached. Data was recorded in MS Excel spreadsheets and included categories for: author, date of publication, journal type, methods, technologies, primary articles within each systematic review (behavioral symptoms), AI, algorithms, statistical methods, thresholds, traditional measures as validators to digital biomarkers, aims, results, outcomes, datasets, and gaps in research or future directions.

2.3. Risk of Bias (Quality) Assessment

Assessment of risk and bias was conducted using the JBI Critical Appraisal Checklist for Systematic Reviews and Research Syntheses. [26] The JBI consists of eleven total questions, resulting in an overall appraisal decision for: 1) inclusion, 2) exclusion, 3) seek further info (Appendix 2). The results of the assessment were used to inform data extraction and synthesis and as assessment for the overall decision and quality for inclusions (Appendix 3). Three reviewers (LG, BM, MP) were involved in the quality assessments. Disagreements between reviewers were resolved by discussion and if agreement was not reached, a fourth researcher (LDB) settled all disputes.

2.4. Data Synthesis

50% of all systematic reviews do not incorporate meta-analysis for various reasons [31,32] and none of the included systematic reviews offered a meta-analysis. Meta-analysis was therefore not possible in this umbrella review due to heterogeneity of metrics and analyses used within the included systematic reviews, and a narrative synthesis of the included articles was subsequently conducted. Comprised content from the articles was grouped according to:

• Type of journal (technical vs. medical), authors, country, year, number & demographics of participants, study setting (real world vs. laboratory)
• Relevancy to HAR used for digital phenotyping and/or classification of behavioral symptoms
• Sensors, devices, AI, datasets, gold standard outcome measures, biomarkers, validation
• Reproducibility (i.e.: transparency and clarity of algorithms and technical details, inclusion of important demographic details such as age and diagnosis)
• Inclusion of ethical considerations, data protection
• Future recommendations from studies
• Studies conducted for an advanced stage of disease or end-of-life
• Consent procedure (informed vs. presumed)

3. RESULTS

A total of 1458 articles were identified and after removal of duplicates (n=392), 1066 articles were independently reviewed based on title and abstract. Two additional systematic reviews were identified by snowballing and full text review was conducted on a total of 20 articles. Articles were excluded based on the following reasons: not a true systematic review, focus on diagnosis of disease, HAR used for only motor symptoms of PD, fall prevention or prediction specific, did not include persons with PD or dementia, focused on commercial app development, study featured sensing technology for telehealth purposes, and one article did not meet the critical appraisal tool standards and was therefore excluded. A total of 1058 articles were excluded, resulting in 9 included high quality systematic reviews; Figure 1 presents the PRISMA flow diagram, detailing the process, including reasons for exclusion.

3.1. Included Systematic Review Characteristics

In total, within the nine included systematic reviews (Table 3), 402 individual primary studies were assessed. The included systematic reviews represent research from eleven countries. Most of the nine included systematic reviews offered evidence for a combination of management of activities and behavioral symptoms. We provide a summary of the included 9 systematic reviews in Table 3. As recommended by Faulkner et al. (2021)[33], this umbrella review sought to describe the primary studies in a more meaningful way. We therefore provide the reader a detailed summary (Appendix 6-7) of 28 primary studies within the total 402 individual primary studies assessed which were focused on the growing research topic of behavioral symptoms such as agitation, apathy, and sleep disturbances to enhance knowledge of metrics and methods used for measurement of these less researched digital biomarkers for future studies. Within the 9 included systematic reviews we identified 10 overlapping primary studies.

Sample sizes for the primary studies ranged between 1-2063 participants. The included systematic review by Morgan et al. (2020) [34] found that approximately 68% of publications had less than 50 participants and similarly, Esquer-Rochin et al. [35] stated that studies using existing databases included an average of 10 or less participants, and an average of 40 participants where ad-hoc datasets were utilized. Mean ages of included participants ranged from 22 to 95 years; five of the nine studies did not include a summary of mean age. All included systematic reviews included people with dementia, Alzheimer's disease, or PD as a primary or secondary diagnosis.

3.2. Sensing Technology, Devices, and Current Use in Research

All of the 9 included systematic reviews combined findings for accelerometry for detection of both activities and behavioral symptoms [34,35,36,37,38,39,40,41,42]. In addition, non-wearables [34,35], pressure sensors [35,40], robots [35], smart devices [34], GPS [35,37] triaxial sensors (including accelerometer, gyroscope and magnetometer) [34,35,40,41] and ambient home sensors [34,36,39] also made the stage; combining multiple wearable sensors (wrist and low back placement being the most common) and video or diary input for analysis and confirmation of data (i.e.: agitation events) was noted regarding efforts for classification of activities and behavioral symptoms [34,39,40] (Table 3).

Included systematic reviews by McArdel (2023)[37] and Breasail et al. (2021)[38] describe the use of sensing technologies to measure volume, intensity, pattern and variability of physical activity in persons with dementia, both stating that this is currently the most common use of accelerometry (actigraphy) within this population of patients. In agreeance with the included systematic review from Mughal et al. (2022) [41], we found that motor symptoms of PD are well studied and that research on sensing technologies for detection of behavioral symptoms is currently growing. An example of this primary research focus of motor symptoms of PD within our findings is the review by Sica et al. (2021) [42] who investigated home monitoring possibilities for persons with PD using sensing technologies, however included only one primary study dedicated to behavioral symptoms within their review, emphasizing the need for more synthesized literature on this topic. (Appendix 6-7)

3.3. Traditional Outcome Measures

Traditional “gold standard" assessment tools most frequently used within the included systematic reviews were Cohen-Mansfield Agitation Inventory (agitation) [43], Mini-Mental Status Exam (cognition) [44], Unified PD rating scale (staging of disease and function) [45], Montreal Cognitive Assessment (cognition) [46] and the Neuropsychiatric Inventory (behavioral symptoms) [[47]. This umbrella review had similar findings to Khan et al. (2018) (43% did not include traditional assessments) [40] and found that 22% of the included systematic reviews did not include information regarding gold standard assessment comparisons and that 67% included some information, but was incomplete. Digital biomarkers within the studies were also vast, with the most utilized being physical activity parameters such as sleep vs. wake time, step counts, activity counts, intensity of activities (low-high), activity amplitude, sedentary time, and classification of activities of daily living such as upright posture, sitting, standing, and walking (Table 3). Sleep disturbance and agitation were the most investigated behavioral symptoms.

3.4. Study Protocols and Methods

The protocols for data collection using the sensing technologies varied greatly from several hours to as long as one year, the most common metric being 7 days [36,37,38,39] followed by 48-72 hours [36,37,39,41]. In 33% of the included systematic reviews [34,39,40], the protocol was either not mentioned, or not given in full description for each study. Statistical analyses, protocols, and methods used within the articles investigated in the systematic reviews were heterogeneous and subsequently, meta-analysis was not possible or performed. A summary of statistical analyses utilized within the included primary studies, concentrated on articles covering behavioral symptoms, is provided in the supplementary materials (Appendix 6-7).

3.5. AI and Algorithms

A wide range of algorithms and AI techniques are mentioned in the included studies, among which are: fuzzy logic, time-frequency analysis, forward feature selection, random forest, k-nearest neighbors, support vector machines, artificial and deep neural networks [34,38,39,41]. Support vector machines were the most prevalent across all studies, while 55% of the studies [33,35,36,37,41] did not provide information on algorithms/AI for extracting activity/behavior information from sensor data. The systematic review by Sica et al. (2021) [42] provides a robust synthesis of AI, signal processing information, and performance validity of the included primary studies, reporting that algorithm based classification methods using digital biomarkers for bradykinesia, transitions (turning), gait and tremor resulted in moderate to high accuracy and validity compared to corresponding symptom diaries, and reflected upon the potential of AI and digital biomarker-based algorithms in efforts to enhance meaningful data gathered using commercial devices in a real-world setting. [42]

3.6. Datasets

One [35] out of the nine included systematic reviews highlighted current datasets that were used within simulated, laboratory-based studies. Esquer-Rochin et al. included a total of 104 primary articles within their systematic review, providing a thorough description of current datasets.[35] Of the existing mentioned datasets, ADNI, PD Telemonitoring, AZTIAHO, Parkinson’s dataset and Daphnet, include participants that have a diagnosis of Alzheimer’s or PD. [35] Three others additionally include older adults, OASIS, CASAS, and PUCK, however details of the demographics are not clearly stated. [35] Only 2 out of the 15 total datasets were directly applicable to use for HAR for persons with PD or dementia. [35] A description of all included datasets can be found in Appendix 4.

3.7. Quality & Bias Assessment

We conducted a critical appraisal of quality and potential biases using the Johanna Briggs Institute (JBI) Critical Appraisal Checklist for systematic reviews and found a score of eight out of eleven (moderate-good) for all included studies, except for Anderson et al. 2020, where only 6 out 11 questions were answered with a “yes”. A detailed list of questions on the JBI and conducted review can be found in the supplementary materials (Appendix 3-4). We excluded the umbrella review by Andersson et al. (2020) based upon its JBI score. Assessment scores were most affected because the studies did not provide a clear statement as to whether a critical appraisal was performed, and what tools were used.

3.8. Recommendations from Included Systematic Reviews

The future perspectives provided by the articles published within the technical vs. medical journals differed greatly. An example can be seen in the summary of suggestions from Esquer-Rochin et al. (2023) [35], concluding that collection of more data is the highest priority among the 104 technical primary papers included in their systematic review. In contrast, the highest-ranking future priorities within the articles published within the medical journals were ethical concerns, development of best practices for data management, and technology which is best suited to the participants and research needs (validation of technology).(Appendix 5) Included systematic reviews from Ardelean and Redolat (2023) [36] and Khan et al. (2018) [40] discuss ethical and safety issues, including unforeseen consequences of technology, and conclude that these topics should be highlighted in future research. Johansson et al. (2018) [39] emphasizes the need to take a “leap of faith” in the application of HAR and similar technologies for clinical trials and implementation into real-world settings. Breasail et al. (2021) [38] further discusses commercial vs. research grade sensing technologies and the need to best discriminate between which is most appropriate for future use in studies and real-life application, emphasizing the need to improve and validate devices and algorithms.

4. DISCUSSION

This umbrella review sought to answer several research questions and found that: 1) methods, sensing technologies, algorithms and AI techniques being used in HAR for persons with PD or dementia are diverse, and that standardization of the metrics and strategies in the field, across disciplines is a necessary step forward to allow potential application of HAR in clinical and real-life settings; 2) statistical analyses and study protocols are heterogeneous and introduce potential bias, decreasing generalization and reproducibility of study results; and 3) there are many opportunities for furthering research withing the field of HAR for persons with PD and dementia, including the topics/gaps of behavioral symptom management, use of HAR throughout the progression of disease including the end-of-life, improved data mining techniques and management of big data (minimalization), validation of technologies, and encouraging cross-collaboration and true interdisciplinary work.

The included systematic reviews by Khan [40] Mughal [41], and Ardelean [36] et al. found that the most common current use of HAR for behavioral symptoms is for detection and management of depression, apathy, agitation and sleep deprivation, however, we agree most with the systematic review by Mughal et al. (2022) [41] which stated that behavioral symptoms are often ignored and currently not well studied. We recommend a future systematic review be performed on the latest HAR methods for behavioral symptoms as a continuation and update of this topic, especially considering the rapidly evolving nature of the field.

Sample size is a common theme throughout the included systematic reviews. We will note that when it comes to the use of sensors, dataset size assessments are fourfold: unique sources (number of participants), datapoints (number of measurements), variety (number of action types), and modality (number of sensor types). Digital sensors usually generate a significant number of datapoints, from 24 per day (e.g., actigraphy measures) to approx. 8.3 million per day (e.g., 3-axis accelerometer at 32 Hz).

4.1. Underrepresentation and Ethical Considerations

Many people with dementia and other advanced neurological conditions are excluded from research based on structural discrimination, measures designed to protect them from harm (i.e.: informed consent), or inadvertently through lack of awareness of participants’ needs and difficulties. [76] Study demographics, including age and disease stage, can have a big impact on results. [77,78] In 2020 Alzheimer Europe published a report titled Overcoming ethical challenges affecting the involvement of people with dementia in research: recognizing diversity and promoting inclusive research [79]; the report states that ethics in dementia research extends from traditional standards of “do no harm” to include empowerment, rights, respect, equity and well-being and emphasizes that appropriate adaptations must be made to ensure that people with dementia have the same opportunities to take part in research. Likewise, future studies involving these vulnerable participants should consider possible unforeseen challenges and consequences of introducing these devices at later stages of disease, and acceptance should be addressed within the studies, as emphasized within the included systematic review by Johannesen et al. (2018)[39].

Khan et al. (2018)[40] brings to light the important topic of ethics related to the use of sensing technologies with vulnerable groups of patients and considers this a gap within the literature as only 7% of the included studies within their systematic review addressed ethical challenges, similar to our findings across all the included reviews. Rubeis et al. (2020) explored the “disruptive power” of AI in elderly care and identified four main risks in the use of AI and phenotypes for detecting and managing neurodegeneration in older adults: depersonalization, discrimination (including ageism), dehumanization, and disciplining (i.e.: enforcing norms) through the collection of big data.[80] The author discusses issues such as the use of AI for clinical decision making and reliance on algorithms for prediction of health outcomes, assessment of risks, and choice of appropriate interventions, which have a large impact on clinical responsibility, accountability and trust [80].

4.2. Future Directions

We identified several intersections between the medical and technical literature included in this umbrella review for future research direction: more research in real-world settings, validation of technologies for monitoring of symptoms of PD and dementia, more research conducted at late stages of neurodegenerative diseases, and adaptive technology to individual progression (precision models). Standardization of metrics, protocols, statistical methods, and use of AI and algorithms is necessary for real-life, real-time application of sensing technologies. A scoping review by Alfalahi et. al (2023) [81] introduced a novel paradigm the authors call “explainable digital phenotyping” influenced by brain, body, and social behavioral decline, based on validation of digital biomarkers and reliant on multidisciplinary co-creation combining AI and knowledge from healthcare professionals; the authors highlight the complexities of using digital sensing technologies as an unbiased data-driven approach for behavioral quantification, including the need for empirical consensus on standardization, reliability, and interpretation.

Morgan et al. (2023)[82], in agreeance with previous articles [83], describes dataset information related to “REal world Mobility Activities in PD” and details 20 total existing datasets, with only two including participants >44 years of age. The article further provides foundational research for future datasets with a detailed opened dataset named REMAP featuring transitional ADLs and participants with PD (61.25 years +-8.5).[82] Morgan et al.’s (2023)[82]results support our findings in the included systematic review by Esquer-Rochin et al. (2023)[35] that current datasets are largely unrelated to persons with PD or dementia, and future work should provide more robust datasets featuring transparent representation of older people with neurodegenerative diseases.

The included systematic review by McArdel et al. (2023)[37] revealed that over 50% of the included primary studies did not include information about validation of the technology used. Morgan et al. (2020)[34] offers the reader a future suggestion that testing the technology against itself using test-retest repeatability may be the best way to validate results in future studies. Traditional methods for validation, such as comparison to a gold-standard outcome measurement, may not correlate to the measurements taken by the device, and this may not always be a negative scenario. Measurements taken through HAR sensing methods should in theory be able to pull out details that subjective traditional outcome measures, such as the NPI, do not capture. This makes it difficult to conduct validation comparisons in the traditional sense. With the challenge of rapid obsolescence of new technologies, future studies may need to shift validation efforts away from specific device brand and model and towards characteristics of the devices that are a best fit for patient care scenarios and develop new methods for validation.

5. LIMITATIONS & STRENGTHS

This systematic umbrella review has several limitations. Gaps in the evidence are possible should the included systematic reviews not comprehensively cover the intended topic, which was found to be true in this case as most of the included reviews reported results for our topic within a broader context, encompassing both motor, activity and behavioral uses for HAR. As a result of our search results, we identified a gap in literature surrounding the topic of HAR for management and monitoring of behavioral symptoms. To address this gap, we included for the reader further analysis of the primary studies from within the included systematic reviews which were directly related to behavioral symptoms use for HAR (Appendix 6-7) in efforts to expand knowledge on this growing topic of interest. The overlap of included primary studies was assessed and reported to further reduce bias. The validity of any umbrella review depends highly on the quality of the included studies and existing systematic reviews. To assess quality and further reduce bias, three researchers performed a critical appraisal of all included reviews. As with standard systematic reviews, there is potential for missed studies, small study bias, missed outcomes, selective reporting bias, and study publication bias, which can influence results and effect the validity of the review. [84] Only 33% of the included systematic reviews deployed a search strategy including one or more technical database (i.e.: IEEE or ACM), creating potential for missed studies within the fields of engineering and data science [34,40,41]. To reduce bias between technical vs. medical literature, we included four multi-disciplinary reviewers for initial selection based on title and abstract and final full-text selection of articles. We also take into consideration that we have a narrow focus within this umbrella review on specific vulnerable groups (older adults, PD and dementia) and therefore may have missed studies which include HAR for younger participants and more general purposes. We further note, however, that by using the narrow scope of older adults, PD and dementia, we uncover true gaps within this specific and important sector. Lastly, another limitation of this umbrella review was that a meta-analysis was not possible as the included quantitative literature did not include a meta-analysis because measurements were clinically heterogeneous and used inconsistent metrics for analyses.

6. CONCLUSIONS

HAR has the potential to enrich knowledge of digital biomarkers which can help guide clinical decision making for people with PD and dementia. However, for HAR to be sustainable for real-world use in persons with neurodegenerative diseases, such as PD, a foundation of new interdisciplinary collaborations and a culture for technology complemented care is necessary. Researchers should improve knowledge and understanding of sample size relativity within the field of HAR, meaning that depending on the type of analysis employed, data quantities within smaller sample sizes might be sufficient if there are enough for both training of models and validation of technologies.

Future perspectives of HAR for monitoring activities and behavioral symptoms include the application of AI for prediction and observation, strategies for complementing knowledge and clinical differentiation of health professionals, and the empowerment of research participants encouraging ethical inclusion of participants in later stages of disease. Further accountability from the developers of the technologies used for HAR to discourage obsolescence and improve support, and more research friendly agreements with universities must be realized for future progress in this field.

7. DECLARATIONS