Technology-based assessment of motor and nonmotor phenomena in Parkinson disease.

ABSTRACT Introduction: The increasing development and availability of portable and wearable technologies is rapidly expanding the field of technology-based objective measures (TOMs) in neurological disorders, including Parkinson disease (PD). Substantial challenges remain in the recognition of disease phenomena relevant to patients and clinicians, as well as in the identification of the most appropriate devices to carry out these measurements. Areas covered: The authors systematically reviewed PubMed for studies employing technology as outcome measures in the assessment of PD-associated motor and nonmotor abnormalities. Expert commentary: TOMs minimize intra- and inter-rater variability in clinical assessments of motor and nonmotor phenomena in PD, improving the accuracy of clinical endpoints. Critical unmet needs for the integration of TOMs into clinical and research practice are the identification and validation of relevant endpoints for individual patients, the capture of motor and nonmotor activities from an ecologically valid environment, the integration of various sensor data into an open-access, common-language platforms, and the definition of a regulatory pathway for approval of TOMs. The current lack of multidomain, multisensor, smart technologies to measure in real time a wide scope of relevant changes remain a significant limitation for the integration of technology into the assessment of PD motor and nonmotor functional disability.


Introduction
Parkinson disease (PD) is a multisystem neurodegenerative disorder resulting in a complex pattern of disability due to the impairment of both motor (i.e., tremor, bradykinesia, rigidity) and nonmotor (i.e., cognition, sleep, autonomic) functional systems [1]. PD-associated clinical features are usually quantified by clinicians using validated clinical scales, such as the Movement Disorder Society Unified Parkinson's disease Rating Scale (MDS-UPDRS) [2], and the Non-Motor Symptoms Scale for Parkinson's disease (NMSS) [3]. These instruments, however, are prone to limitations such as subjectivity, inter-rater variability, and limited accuracy in capturing small variations between and within patients.
Technology advancements have expanded the application of a new generation of technology-based objective measures (TOMs) to detect and monitor a functional range critical for the comprehensive characterization and long-term monitoring of patients with PD [4,5]. TOMs may capture multiple motor activities, such as frequency and amplitude of movements, severity of tremor and dyskinesia, and extent of gait and postural impairment [6][7][8]. In addition, TOMs may characterize nonmotor phenomena that cannot be captured by conventional clinical scales, such as sleep architecture, respiratory rate, beat-to-beat blood pressure changes, heart rhythm variability (HRV), and electroencephalographic (EEG) activity [9,10].
By reducing the standard deviation of clinical endpoints and minimizing intra-and inter-rater variability in clinical assessments, TOMs have the potential to improve the quality of diagnostic definitions [11]. Multiple challenges, however, limit the integration of TOMs into the clinical and research practice, including standardization of extracted parameters, cost of technology, patient compliance [12], and risk of producing outcome measures that have little practical meaning to end-users [13,14]. In order to be purposeful, any measures of function need to represent variables that are important to patients and can be amenable to interventions by clinicians and researchers [15].
We sought to systematically review studies that employed technology for the evaluation of motor and nonmotor phenomena in PD, appraising the extent of current integration of TOMs into the assessment of functional disability.

Selection of studies
Abstracts were independently reviewed for eligibility criteria by two investigators (A.S. and S.S.). Disagreements were anticipated to be settled by consensus among the authors. Duplicated studies were identified and excluded. The reference lists of selected articles were additionally screened for pertinent studies not included in the original search strategy.

Data extraction and assessment of risk of bias
The following data were extracted from eligible studies using a standardized form: (a) year of publication; (b) study design; (c) study population; (d) inclusion and exclusion criteria; (e) primary and secondary outcome measures; (f) results; and (g) study limitations. Given the heterogeneity of study designs, we followed the Cochrane handbook recommendations and assessed the risk of bias of individual studies utilizing the National Heart, Lung, and Blood Institute tools (NHLBI) [17], as per the Cochrane handbook recommendations [18]. These tools are tailored to study types (i.e., cross-sectional, casecontrol, interventional) and include a qualitative, internal validity checklist ('Yes,' 'No,' and 'Nonapplicable') for domains such as methodological pertinence, potential sources of bias, confounding, and adequacy of results for quality classification as 'good,' 'fair,' or 'poor.' In general, a 'good' rating applies to studies with low risk of bias whose results are deemed valid; 'fair' to studies susceptible to some biases but deemed insufficient to invalidate their results; and 'poor' to studies with significant risk of bias.

Data analysis
Included studies were categorized per functional domain investigated (motor vs. nonmotor) and sorted per year of publication and technology employed. Results were summarized in tables and discussed in the text.

Results
The search strategy resulted in the identification of 2941 studies published between 1980 and 2018. A total of 2817 studies did not meet all inclusion criteria or were considered duplicates ( Figure 1). Thus, we included 124 studies (106 crosssectional, 11 case-control, and 7 prospective cohorts) which underwent data extraction and individual appraisal of the quality of evidence and risk of bias (Tables e-2 and e-3, online).
Kinematic and EMG measurements were employed to assess turning and freezing of gait (FOG), evaluating both spatial and temporal gait parameters [30,50,56], as well as the pattern of axial muscle activation during turns [33]. Freezers showed greater variability in stride length, stride time, and cadence compared to nonfreezers [30,50,56] along with reduced thoracic adaptation to hip movements during gait and turns [31,33,34], increased number of steps, prolonged turning time [57], and decreased range of motion in the ankle and hip joints immediately before FOG episodes [35].
There were similar reaction times and movement lengths in PD patients vs. healthy controls, but lower maximum speed and, consequently, longer execution time in PD [19,39]. TOMs accurately captured changes in transport time, wrist velocity, and arm acceleration during reach-to-grasp motor tasks [26,62]. Tablet-based measurements objectively quantified amplitude, velocity [64], and motor blocks during handwriting [63]. Wearable sensors were used to assess the effect of dopaminergic medications on speed and amplitude of movements, showing a more pronounced effect on the former [71]. Finally, optokinetic analyses were employed to assess orofacial movements, such as vertical jaw movements during speech [66] and hypomimia [68,69], showing sufficient accuracy in objectively capturing differences between PD patients and healthy controls (Table 2).
A study evaluated PD-associated trunk rigidity using a dynamometer to measure resistive torques during passive trunk flexion and extension [54]. The internal validity of this approach was confirmed by different authors reporting a direct correlation between dynamometer-based rigidity assessment and health-related quality of life in PD [70]. Unlike this study, however, most studies did not evaluate the extent to which changes measured by specific TOMs correlated with relevant changes as perceived by patients.
TOMs were also employed in the detection of subclinical tremor [73] and the analysis of the different tremor components during resting, movement initiation, and decelerating phase of movement [72]. Innovative machine-learning algorithms have been recently developed to evaluate the variability of tremor in the time-domain during resting and motor activities [76,77] (Table 2).

Speech
Multidimensional voice software programs for acoustic analysis (i.e., Praat, a freeware developed by the University of Amsterdam) have been employed for the quantitative assessment of amplitude, prosody, speed, grammar, and fluency of speech during sustained-vowels phonation, alternating and sequential motion rates, and normal reading [81]. TOMs proved useful in sensitively capturing differences in speech between PD patients and healthy controls, including maximal phonation time, phonation quotient, percent jitter, percent shimmer, and noise-to-harmonic ratio [81][82][83], suggesting their employment in the diagnostic assessment of PD [84], monitoring of PD-associated functional disability [85], and prognostic assessment of functionally relevant outcomes such as cognitive decline [86]. An automated speech assessment has been proposed as part of a battery of test, including also posture analysis, gait assessment, finger tapping, and response time, to monitor PD symptoms at the home environment using commercially available smartphone applications [87] (Table 3).
Fragmented sleep architecture and reduced quality of sleep were found both in mild and advanced PD [108], with strong correlation between sleep efficiency, as measured by the PSG, and clinical measures of sleep quality, such as the Epworth sleepiness scale (EPSS), the PD sleep scale (PDSS), and the Pittsburgh Sleep Quality Index (PSQI) [109]. Additional studies evaluated the effect of pharmacological and nonpharmacological treatments on PD-associated quality of sleep, reporting an objective improvement in the architecture of sleep with advanced therapeutic options such as levodopa/carbidopa intestinal gel infusion (LCIG) [106] and subthalamic nucleus deep brain stimulation [107]. (Table 4).

Autonomic function
Cardiovascular and sweating autonomic testing, as well as HRV and 24-hour ambulatory blood pressure monitoring have been employed in the assessment of sympathetic [131,132] parasympathetic [132,133], and cholinergic autonomic function [10]. Significant differences were observed between PD and healthy controls suggesting a role for autonomic testing in the diagnostic classification of PD [134], characterization of patients at risk of poor functional outcome [135] or higher risk of dementia [136], and in distinguishing PD from atypical Parkinsonian syndromes [137,138]. Recent studies also suggested that failure at the autonomic function testing might predict disease progression and survival in PD [139], as well as assist in the identification of patients expected to respond differentially to a range of treatments [140] (Table 5).

Conclusion
This systematic review showed that an increasing number of studies employed TOMs for the assessment of PD-associated motor and nonmotor phenomena over the last two decades ( Figure 5). A range of technologies were used to evaluate motor endpoints such as gait, balance, bradykinesia, tremor, rigidity, and speech, as well as nonmotor endpoints such as sleep, cognition, and autonomic function. TOMs demonstrated the potential of capturing motor and nonmotor phenomena with greater accuracy and reduced intra-and inter-rater variability than clinical scales and self-administered questionnaires. However, only a few studies correlated TOMs with patientcentered clinical scales, quality of life questionnaires, or handicap index. In addition, minimal clinically important differences have been estimated for a limited number of TOMs.
A possible limitation of our study consists of a searching strategy limited to published studies. We did not conduct searches in multiple databases, as well as Grey Literature or ongoing trials (e.g., clinicaltrials.gov). Thus, our conclusions do not take into consideration currently ongoing research endeavors.

Expert commentary
While substantially improving the accuracy of both motor and nonmotor clinical endpoints in PD, ultimately resulting in improved diagnostics and monitoring of functional disability [11], the integration of TOMs into randomized controlled trials and routine clinical practice remains limited by several unresolved issues [141]. An important roadblock is the lack of a clear regulatory pathway from the FDA and the EMA for the routine employment of TOMs in both clinical and research settings. Less than 3% of ongoing clinical trials of neurodegenerative disorders have employed TOMs as an outcome measure. However, a survey from medical directors from pharmaceutical companies indicated that the majority of them are considering using TOMs in future clinical trials within the next five years [141].]. Also, a smartwatch for the monitoring of epileptic seizures was recently approved by regulatory agencies based on data demonstrating their accuracy and usefulness in clinical practice [142,143]. This preliminary experience encourages similar studies in the field of movement disorders.
The ideal outcome measure would be objective, exhibit minimal intra-and inter-operator variability, continuously capture relevant data in the patient's home environment, and sensitively capture small but meaningful changes over a prolonged period of time. Available TOMs meet some of these criteria but fail others, such as capturing motor and nonmotor activities from an ecologically valid environment. Currently, most of the gait and balance measures rely on tests assessing the patient's functional capacity (e.g., timed-up-and-go) rather than functional activity (e.g., continuous recording of natural unrestricted gait). In addition, the resolution of biomechanical sensors remains restricted to the anatomical area on which the sensors are applied, possibly yielding low quantitative agreement with the broader range of motor disability, quality of life, and other measurable patient-relevant endpoints [11,144].
The assessment of nonmotor symptoms poses even more significant challenges due to the frequent use of cumbersome technologies such as PSG, EEG, or tilt table, which highlight the need for a tradeoff between the comprehensiveness of the assessment and their ecological validity. Simplified sleep mea- A: Infrared camera, computerized walkway for gait analysis, and wearable sensors worn of each extremity and on the chest can be used to sensitively capture aspects related to patient's mobility in the three axes of space, including neck, arm, and hip movements.B: Force plate and wearable sensors can be used to estimate postural stability, which can be represented as a line indicating the postural sway of the center of mass over a 30-second period of time. Irregular and wide oscillations of the center of mass are indicative of postural instability.        surements collected from an actigraphy may be preferred over the more accurate but less ecologically representative PSG [145]. A similar context applies to the evaluation of autonomic function. In a recent publication [146], we proposed that a 24h-ambulatory blood pressure monitoring might be effectively employed as a screening test for cardiovascular autonomic neuropathy, a disabling comorbidity in PD with relevant socio-economic impact [147]. Autonomic dysfunction remains underrecognized and undertreated in PD [148] in part because its ascertainment relies on cardiovascular autonomic testing available only in a few specialized laboratories. In conclusion, these findings highlight the urgent need for developing relatively simple and unobtrusive systems to monitor motor and nonmotor endpoints in the home and community settings rather than during in-hospital evaluations. A significant limitation consists of the lack of a multisensor, open-access, common-language platform combining the results of different sensors into a multidimensional TOM expressing a global measure of PD-associated functional disability. Although this unmet need has been reiterated by the Movement Disorders Society (MDS) in various international meetings and position papers [149], diagnostic and monitoring systems developed by different manufacturers continue to remain incompatible with one another. As a result, it is difficult or impossible to combine data gathered by different TOMs. This point represents one of the most critical areas of need, identified by the MDS Task Force for the Integration of Technology in PD as requiring further development. Only few studies have employed a smartphone application that integrates the capture of voice, posture, gait, finger tapping, and response time in the patient home environment, with high patient participation as well as sensitivity and specificity in the collected outcome measures [87].

Five-year view
Continuous improvements in technology are creating increasing opportunities for TOMs to improve self-management options and overall healthcare outcomes in PD. Thus, their integration into research and practice is expected to grow in the next five years. Critical challenges consist of validation of measures with patient-centered relevant endpoints, standardization of procedures, and approval by regulatory authorities.

Key issues
• Clinical scales for the assessment of Parkinson disease (PD) symptoms are prone to limitations such as subjectivity, inter-rater variability, and limited accuracy in capturing small variations within and between patients. A new generation of technology-based objective measures (TOMs) may provide a more accurate characterization of motor and nonmotor phenomena associated with PD. • We searched PubMed for human studies employing TOMs as primary, secondary, or exploratory outcomes for the qualitative or quantitative evaluation of PD-associated motor and nonmotor symptoms. There were 61 studies assessing motor phenomena such as gait and postural instability (n = 33 studies), bradykinesia (n = 13 studies), tremor (n = 8 studies), and rigidity (n = 7 studies), and 63 studies assessing nonmotor phenomena such as sleep disorders (n = 23 studies), cognitive impairment (n = 18 studies), dysautonomia (n = 12 studies), sensory deficits (n = 3 studies), and voice analysis (n = 7 studies). • Although TOMs have the potential to significantly improve the accuracy of both motor and nonmotor clinical endpoints, their integration into randomized controlled trials and routine clinical practice remains limited by several unresolved issues, including validation of patient-centered outcomes, standardization of measurements, and approval by regulatory authorities. • While TOMs have not yet been shown to be superior to the clinical evaluation, their integration into research and practice is expected to substantially increase in the next five years and translate into enhanced care, better self-management options for PD patients, and overall improved healthcare outcomes. A survey from 12 medical directors from pharmaceutical companies indicated that 83% of them are considering using TOMs in future clinical trials within the next five years. Wearables have the potential to capture multiple motor and nonmotor phenomena associated with Parkinson disease and transmit data using wireless Internet and Bluetooth connections. was an investigator in Chelsea-sponsored studies, has acted as a scientific advisor to Lundbeck, and is marketer for Droxidopa but has no financial interest in either company. They have received personal compensation as a consultant/advisory board member for Solvay, Abbott, Chelsea Therapeutics, TEVA, Impax, Merz, Lundbeck, and Eli Lilly; has received honoraria from TEVA, UCB, the AAN, and the Movement Disorder Society; and publishing royalties from Lippincott Williams & Wilkins, Cambridge University Press, and Springer. The authors have no further conflicts of interest to declare.

Reviewer disclosures
Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Funding
This paper was not funded.