Kinematic but not clinical measures predict falls in Parkinson-related orthostatic hypotension

Abstract

Objective:

We sought to tests the hypothesis that technology could predict the risk of falls in Parkinson disease (PD) patients with orthostatic hypotension (OH) with greater accuracy than inclinic assessment.

Methods:

Twenty-six consecutive PD patients with OH underwent clinical (including home-like assessments of activities of daily living) and kinematic evaluations of balance and gait as well as beat-to-beat blood pressure (BP) monitoring to estimate their association with the risk of falls. Fall frequency was captured by a diary collected prospectively over 6 months. When applicable, the sensitivity, specificity, and diagnostic accuracy were measured using the area under the receiver operating characteristics curve (AUC). Additional in-clinic assessments included the OH Symptom Assessment (OHSA), the OH Daily Activity Score (OHDAS), and the Movement Disorder Society Unified Parkinson Disease Rating Scale (MDS-UPDRS).

Results:

The prevalence of falls was 53.8% over six months. There was no association between the risk of falls and test of gait and postural stability (p≥ 0.22) or home-like activities of daily living (p> 0.08). Conversely, kinematic data (waist sway during time-up-and-go, jerkiness and centroidal frequency during postural sway with eyes-opened) predicted the risk of falls with high sensitivity and specificity (> 80%; AUC≥ 0.81). There was a trend for higher risk of falls in patients with orthostatic mean arterial pressure ≤75 mmHg.

Conclusions:

Kinematic but not clinical measures predicted falls in PD patients with OH. Orthostatic mean arterial pressure ≤75 mmHg may represent a hemodynamic threshold below which falls become more prevalent, supporting the aggressive deployment of corrective measures.

INTRODUCTION

Orthostatic hypotension (OH), defined as blood pressure (BP) drop of at least 20/10 mmHg (systolic/diastolic) within 3 minutes of standing from a supine position [1], is a frequent complication of Parkinson disease (PD) and related synucleinopathies [1]. In particular, OH has been associated with longer PD duration, a higher dosage of dopaminergic therapy, and early sympathetic cardiac denervation [2].

While representing a major source of functional disability and health care utilization [3]. OH has an insidious clinical presentation. Half of the patients complain of symptoms such as dizziness, lightheadedness, or blurred vision. The other half is asymptomatic or reports ostensibly unrelated symptoms such as confusion or difficulty concentrating [4]. Recent studies suggest that, regardless of its symptomatic status, OH represents a significant determinant of ambulatory capacity impairment, falls, and fractures, which are the leading causes for hospitalization and death in PD [4]. Critically, the introduction of small inertial sensors such as accelerometers and gyroscopes has provided a unique opportunity for understanding the pathophysiology of gait and postural instability in PD-associated OH [5]. These wearable technologies have shown a strong correlation with conventional kinematic systems for the analysis of gait and balance [6] and the possibility of capturing meaningful data during standardized clinical tasks and unrestricted motor activities.

Using a comprehensive battery of clinical testing, kinematic measurements, and continuous BP monitoring, we sought to test the hypothesis that technology might identify PD-OH subjects at risk of falls with greater accuracy than in-clinic assessment.

METHODS

Participants

Consecutive (n= 26) PD patients with OH were recruited from the University of Cincinnati between March 2018 and February 2019 and enrolled in a six-month, single-center, prospective cohort study (Supplementary material 1).

Inclusion criteria were: (a) Idiopathic PD meeting the United Kingdom Brain Bank criteria [7]; (b) Age between 30 and 85 years old; (c) OH, defined as a fall in systolic BP ≥ 20 mmHg or diastolic BP ≥ 10 mmHg within 3 minutes of standing after at least 5 minutes of rest in a supine position [8]; and (d) Stable dosage of dopaminergic medications and anti-OH medications for at least 4 weeks.

Exclusion criteria were: (a) Diabetes mellitus or other diseases potentially associated with cardiovascular autonomic dysfunction [9]; (b) Inability to perform activities of daily living (ADLs) and motor tasks due to cognitive impairment or motor limitations; and (c) Atypical signs lowering the diagnostic accuracy for idiopathic PD.

Study Procedures

Evaluations were conducted in a temperature-controlled room at 68°-74° Fahrenheit (20-23.5° Celsius) during a period of maximal efficacy of dopaminergic medications without troublesome dyskinesia (“best ON”), at least two hours after the last meal. The clinical assessment included three components: (a) lying-to-standing test (standing up without assistance after 10 minutes of supine resting and keeping the upright position for 5 minutes); (b) tests of gait and postural stability (time-up-and-go, two-minute-walk-test, and sway eyes-opened/closed - supplementary material 2); and (c) ADLs conducted in a standardized home-like environment (bedroom, kitchen, and work settings - supplementary material 2).

During each study session, a trained clinical examiner documented:

• Postural instability, defined as episodes of abnormal postural sway posing the patients at risk of falling, as per item 5 of the Tinetti balance assessment tool [10]. Subjects with at least one episode of postural instability during the assessment were classified as “postural instability+.”

• Orthostatic symptoms, as per the OH symptoms Assessment (OHSA) item 1 (lightheadedness, feeling faint, or feeling like you might blackout), 2 (problems with vision: blurring, seeing spots, tunnel vision), and 6 (head/neck discomfort) [11]. Subjects reporting orthostatic symptoms at least once during the assessment were classified as “symptomatic OH.”

Additional clinical tests included the Movement Disorder Society-Unified Parkinson’s Disease Rating Scale part III (MDS-UPDRS III; higher is worse) [12], the Unified Dyskinesia Rating Scale part III (UdysRS) (higher is worse) [13], the Tinetti balance assessment tool (lower is worse) [10], and the OH questionnaire (OHQ), which consists of two independently validated subscales, the OH Symptom Assessment (OHSA) and the OH Daily Activity Score (OHDAS) (higher is worse) [11].

At the end of the assessment, patients were provided with a falls diary for a prospective observational period of 6 months and instructed to note every falls, defined as per the World Health Organization (WHO) as an “event that inadvertently results in coming to rest on the ground, floor, or other lower-level” [14]. Patients reporting at least one fall over the six-month observational period were classified as “fallers”; patients with two or more falls were classified as “frequent fallers.”

Blood Pressure Assessment

Systolic, mean, and diastolic BP were measured for the entire duration of the clinical assessment using a wearable beat-to-beat continuous non-invasive blood pressure (c-NIBP) monitor (Caretaker 4 Wireless cNIBP monitor; Caretaker, VA, USA) [15] calibrated every 30 minutes with an arm-cuff automated oscillometric sphygmomanometer (HEM-7200 e Omron Healthcare Co., Kyoto, Japan). The cNIBP monitor consists of a small wearable device, which operates through an inflatable finger cuff plugged to a front-end box strapped to the subject’s wrist to transmit real-time data to a wireless device (supplementary materials 3, 4). After initial calibration with an arm-cuff sphygmomanometer, the air-inflated sensor worn on the proximal index finger phalange can detect the digital artery pulse wave at each heartbeat to calculate the BP using an algorithm of Pulse Decomposition Analysis [15].

The following BP parameters were collected: (a) Average systolic/diastolic BP and mean arterial pressure (MAP) in the lying, sitting, and standing positions; (b) Hemodynamically relevant OH, defined as patients with at least 1 episode of orthostatic MAP ≤75 mmHg [16]; (c) Hypotensive load, defined as the percentage of time spent with orthostatic MAP ≤75 mmHg [17].

The OH symptomatic status was based on presence/absence of orthostatic symptoms [4, 11]. Patients were divided into symptomatic vs. asymptomatic on whether they answered positively or negatively at the baseline evaluation to item 1 of the OHSA upon standing (“Dizziness, lightheadedness, feeling faint, or feeling like you might black out”).

Technology-based Balance and Gait Assessment

Balance and gait were evaluated using a full-body wearable motion sensor system (Mobility Lab, APDM wearable technologies Inc., Portland, OR, USA) consisting of six sensors (each one including a tri-axis accelerometer, gyroscope, and magnetometer) placed on the feet, wrists, sternum, and lumbar region [18]. An additional wearable system (Kinesia™, Great Lakes NeuroTechnologies Inc., Cleveland, OH, USA) consisting of two sensors placed on the waist and at the mid-thigh [19] was used to evaluate gait during both tests of functional capacity and homelike ADLs. The Kinesia system was employed to estimate the median waist postural sway (WS) (degree) during walking by measuring the magnitude of the rotation of the hip in the coronal and sagittal axes during each gait cycle (Supplementary material 5) [20, 21]. Data collected from all sensors were synchronized with performance of activities using a Matlab-based software platform (Matlab, Mathworks, Torrance, CA, USA) designed ad-hoc for this study.

Balance analysis

Using the APDM system, we estimated oscillations in the Center of Pressure (CoP), defined as the point of application of the ground reaction force vector [22], and the center of mass (CoM), defined as the body mass projection on the ground. Balance was evaluated during tasks of functional capacity with the sway test (^sway) with eyes opened (^EO) and eyes closed (^EC). The following measurements were collected: (a) JERK^sway (m²/sec⁵), an indicator of jerkiness; (b) root-mean-square (RMS^sway) (m/sec²), an indicator of the magnitude of the acceleration; and (c) centroidal frequency (CF^sway) (Hz), a measure of the frequency of oscillation around the CoM [23, 24].

Gait analysis

We used the APDM and the Kinesia systems to collect the following gait variables during the time-up-and-go (^TUG) and the two-minute walk (^walk) test: turn duration^TUG; total duration^TUG; peak turn velocity^TUG; WS^TU0; gait speed^walk; stride length^walk; number of steps during turning^walk; peak turn velocity^walk; upper limbs range of motion^walk; cadence^walk; WS^walk [24, 25]. Previous studies found good agreement between the APDM sensors and optokinetic infrared cameras [26, 27]. In addition, we used the Kinesia system to analyze the WS during home-like activities involving the bedroom (WS^bedroom) and kitchen (WS^kitchen) settings (Supplementary material 2).

Statistical Analysis and sample size justification

Clinical data analyses

Normal quantitative variables were summarized with mean ± standard deviation (SD) while median and interquartile range were used for skewed data. Categorical variables were reported with frequency and proportions. Spearman’s rank correlation coefficient was used for comparison between clusters of kinematic variables and the risk of postural instability, falls, BP parameters, and clinico-demographic characteristics. Comparisons between groups were carried out using the Fisher’s exact test/Chi-square and t-test/Wilcoxon rank-sum test as appropriate. Post-hoc analysis to evaluate differences between a specific test of functional activity and home-like ADLs in predicting postural instability and falls were carried out using the Fisher’s exact tests.

Kinematic data analysis

We first performed a cluster analysis to identify clusters of related variables (Table 1, supplementary materials 6 and 7). The variable within each cluster having least 1- R2 ratio was characterized as a representative cluster. The importance of clusters was determined using the proportion of variability explained by each cluster. Higher values of proportion explained by variability are indicative of more important clusters. Gait, balance, BP parameters, and clusters were analyzed comparing each variable to subjects with and without postural instability events during the in-clinic assessment, with falls occurring over a prospective observational period of 6 months (fallers vs. non-fallers), and symptomatic status (symptomatic vs. asymptomatic) using the Wilcoxon rank-sum test. McNemar’s test was used to compare the proportions of postural instability between functional capacity testing and home-like ADLs. The accuracy of wearable motion sensor instrumented parameters in predicting falls was obtained using the area under the receiver operating characteristics curve (AUC). Only parameters with AUC ≥0.8 were considered in our analysis.

Table 1.

Cluster Analysis

	Variables included in the cluster
Cluster 1	Upper limbs range of motionwalk, peak turn velocityTUG
Cluster 2	Gait speedwalk, stride lengthwalk, CFsway EC
Cluster 3	Cadencewalk, peak turn velocitywalk, durationTUG
Cluster 4	RMSsway EO, RMSsway EC, JERKsway EC, number of steps during turningwalk, WSbedroom, WSkitchen
Cluster 5	JERKsway EO, CFsway EO, WSTUG
Cluster 6	Turn durationTUG, WSwalk

Through a cluster analysis, we identified 6 clusters of kinematic variables describing the pattern of motor activities during tests of gait and postural stability and home-like activities. The representative cluster was identified based on the ratio of between 1-own R2 (proportion of variability) and the next R2.

Walk: two-minute walk test; TUG: time-up- and-go; WS: waist postural sway; EO: eyes opened; EC: eyes closed; RMS: root-mean-square; CF: centroidal frequency.

Sample size considerations

We assumed large effect sizes for kinematic over clinical measures to predict falls in the vulnerable PD-OH population, with a Cohen’s effect size of 0.8 for mean difference or 0.6 for correlation coefficients. Using these assumptions, we estimated that 26 PD-OH patients are needed to identify differences between kinematic and clinical measures at 5% level of significance with more than 80% power using two-sided t-test. Due to the exploratory nature of this study, we did not adjust the level of significance for multiple comparisons. All statistical analyses were conducted using STATA 15.1 while variable cluster analysis with SAS 9.4.

Ethics

The study protocol was approved by the Institutional Review Board of the University of Cincinnati (protocol number 2017-1547), and written informed consent was obtained from all participants.

RESULTS

Our cohort included 26 PD patients with OH (19 males and 7 females), age, 72.5 ± 9.9 years (range, 44-85) and disease duration, 7.0 ± 5.0 years (range 1-23) (Table 2). OH was symptomatic in 20 patients (76.9%) and asymptomatic in 6 (23.1%) (Table 3). The mean arterial pressure was 100.6 mmHg supine, 92.6 mmHg sitting, and 86.0 mmHg standing (Table 4). Over the six-month follow-up period, there were 14 fallers, yielding a falls prevalence of 53.8% (Table 2). No measurable differences were noted between fallers and frequent fallers (Supplementary material 8).

Table 2.

Clinical and Demographic Data

Variable	Sample (n=26)	Postural instability+ (n=21)	Postural instability− (n=5)	P-value	Falls+ (n=14)	Falls− (n=12)	P-value
Age (years)	72.5 ± 9.9	74.6 ± 7.4	63.8 ± 15.2	0.027	74.5 ± 12.1	70.2 ± 7.6	0.28
PD (years)	7.0 ± 5.0	5 [3, 8]*	8 [5, 10]*	0.31	6.5[4, 10]*	4.5[3.5, 8]*	0.31
Sex (M:F)	19:7	15:6	4: 1	1.00	10:4	9:3	1.00
LEDD (mg)	954.6 ± 522.5	1050 [600, 1300]*	900 [581, 1175]*	0.63	1079 [600, 1575]*	850 [565.5, 1137.5]*	0.26
MDS-UPDRS III Total score	40.1 ± 13.4	41.0 ± 14.2	36.4 ± 9.2	0.50	42.6 ± 14.6	37.2 ± 11.7	0.31
MDS-UPDRS III Rigidity subscore	5.5 ± 1.9	5.2 ± 1.8	6.6 ± 2.1	0.14	5.5 ± 2.1	5.4 ± 1.7	0.91
MDS-UPDRS III Freezing subscore	0.5 ± 0.9	0.6 ± 0.9	0 ± 0	0.19	0.6 ± 1	0.3 ± 0.6	0.25
MDS-UPDRS III Pull test subscore	1.6 ± 1.2	1.8 ± 1.3	0.8 ±0.5	0.12	1.6 ± 1.4	1.6 ± 1.1	0.98
MDS-UPDRS III Bradykinesia subscore	1.9 ± 0.7	2 ± 0.7	1.8 ± 0.8	0.59	2 ± 0.7	1.9 ± 0.8	0.78
MDS-UPDRS III Tremor subscore	4.5 ± 4.6	5.1 ± 4.8	2 ± 2.8	0.18	5.9 ± 4.4	2.9 ± 4.6	0.10
UDysRS part III	0.2 ± 0.4	0.1 ± 0.3	0.4 ± 0.6	0.097	0.2 ± 0.4	0.1 ± 0.3	0.19
Tinetti Total score	19.1 ± 6.4	18.7 ± 6.8	20.6 ± 4.8	0.57	18.2 ± 7.1	20.1 ± 5.7	0.47
Tinetti Balance subscore	11.6 ± 3.7	11.4 ± 4.0	12.6 ± 2.8	0.54	11.1 ± 4.2	12.3 ± 3.2	0.40
Tinetti Gait subscore	7.4 ± 2.8	7.3 ± 3.0	7.8 ± 1.9	0.72	7.1 ± 3.1	7.7 ± 2.6	0.65
OHQ	42.2 ± 18.0	42.3 ± 18.4	41.8 ± 18.6	0.96	40.9 ± 16.51	43.6 ± 20.2	0.72
OHDAS	18.8 ± 9.7	18.9 ± 10.5	18 ± 6.8	0.85	17.3 ± 10.0	20.3 ± 9.6	0.45
OHSA	23.4 ± 10.2	23.4 ± 10.0	23.8 ± 12.4	0.93	23.6 ± 9.5	23.3 ± 11.4	0.93

Data are expressed as mean ± standard deviation unless specified differently.

^*identifies data expressed as median [interquartile range].

PD: Parkinson disease; LEDD: Levodopa equivalent daily doses; M: Male; F: Female; MDS-UPDRS: Movement Disorder Society Unified Parkinson’s Disease Rating Scale; UDysRS: Unified Dyskinesia Rating Scale; OH: Orthostatic hypotension; OHQ: OH questionnaire; OHS A: OH Symptom Assessment; OHDAS: OH Daily Activity Score; MAP: Mean arterial pressure supine; mg: Milligrams. +indicates presence of condition while – indicates absence of condition. BP: Blood pressure; MAP: Mean arterial pressure supine; mmHg: Millimeters of mercury MDS-UPDRS III rigidity subscore: item 3.3 (range: 0-20); MDS-UPDRS III freezing subscore: item 3.11 (range: 0-4); MDS-UPDRS III pull test subscore: item 3.12 (range: 0-4); MDS-UPDRS III body bradykinesia: item 3.14 (range: 0-4); MDS-UPDRS III tremor subscore: items 3.15-3.18 (range: 0-36). Tinetti balance subscore: items 1-9 (range: 0-16); Tinetti gait subscore: items 10-16 (range: 0-12). OHQ: OHDAS + OHSA (range: 0-100); OHDAS (range 0-40); OHSA (range 0-60).

Table 3.

Symptomatic vs. Asymptomatic OH

Variable	Symptomatic OH (n= 20)	Asymptomatic OH (n= 6)	P-value
Age (years)	74.2 ± 8.1	66.7 ± 13.9	0.10
PD duration (years)	5 [3.5, 8]*	9 [5, 10]*	0.27
Sex (M:F)	14:6	5:1	1.00
LEDD (mg)	950 [600, 1237.5]*	1150 [581, 1275]*	0.69
MDS-UPDRS III	38.7 ± 13.0	44.7 ± 14.8	0.35
MDS-UPDRS III Rigidity subscore	5.2 ± 1.6	6.3 ± 2.6	0.21
MDS-UPDRS III Freezing subscore	0.6 ± 0.9	0.2 ± 0.4	0.35
MDS-UPDRS III Pull test subscore	1.7 ± 1.2	1 ± 1.1	0.20
MDS-UPDRS III Bradykinesia subscore	1.9 ± 0.7	2 ± 0.9	0.88
MDS-UPDRS III Tremor subscore	3.7 ± 4.7	7.3 ± 3.4	0.093
UDysRS part III	0.2 ± 0.4	0.2 ± 0.4	0.92
Tinetti	18.7 ± 6.9	20.5 ± 5.0	0.55
Tinetti balance subscore	11.5 ± 4.1	12.3 ± 2.5	0.62
Tinetti gait subscore	7.1 ± 2.9	8.2 ± 2.6	0.45
OHQ	44.4 ± 18.1	35.3 ± 17.4	0.29
OHDAS subscore	20.5 ± 9.1	13.2 ± 10.3	0.11
OHSA subscore	23.8 ± 11.0	22.2 ± 8.1	0.73
BP supine (mmHg)
- Systolic	138.0 [124.0,144.0]*	124.5 [121.0, 139.0]*	0.41
- Diastolic	81.0 [77.0, 87.0]*	76.0 [66.0, 84.0]*	0.39
- MAP	98.0 [93.0, 109.0]*	93.0 [88.0, 103.0]*	0.37
BP sitting (mmHg)
- Systolic	123.5 [113.0, 131.5]*	125.0 [108.0, 127.0]*	0.85
- Diastolic	75.0 [72.5, 79.0]*	72.5 [67.0, 78.0]*	0.43
- MAP	90.5 [86.5, 97.5]*	92.0 [84.0, 94.0]*	0.93
BP standing (mmHg)
- Systolic	114.5 [107.0,122.5]*	110.0 [99.0, 116.0]*	0.48
- Diastolic	72.0 [67.5, 74.0]*	65.5 [62.0, 72.0]*	0.15
- MAP	86.5 [82.0, 90.0]*	65.5 [62.0, 72.0]*	0.15
MAP ≤75 mmHg (prevalence)	55%	83%	0.21
Falls (prevalence)	50.0%	66.7%	0.65
Postural instability (prevalence)	85%	66.7%	0.56

Data are expressed as mean ± standard deviation unless specified differently.

^*identifies data expressed as median [interquartile range].

PD: Parkinson disease; LEDD: Levodopa equivalent daily doses; M: male; F: female; MDS-UPDRS: Movement Disorder Society Unified Parkinson’s Disease Rating Scale; UDysRS: Unified Dyskinesia Rating Scale; OH: Orthostatic hypotension; OHQ: OH questionnaire; OHS A: OH symptoms with the OH Symptom Assessment; OHDAS: OH Daily Activity Score; MAP: Mean arterial pressure; mmHg: Millimeters of mercury; mg: Milligrams.

MDS-UPDRS III rigidity subscore: item 3.3 (range: 0-20); MDS-UPDRS III freezing subscore: item 3.11 (range: 0-4); MDS-UPDRS III pull test subscore: item 3.12 (range: 0-4); MDS-UPDRS III body bradykinesia: item 3.14 (range: 0-4); MDS-UPDRS III tremor subscore: items 3.15-3.18 (range: 0-36). Tinetti balance subscore: items 1-9 (range: 0-6); Tinetti gait subscore: items 10-16 (range: 0-12). OHQ: OHDAS + OHSA (range: 0-100); OHDAS (range 0-40); OHSA (range 0-60).

Table 4.

Hemodynamic Variables

Variable	Sample (n=26)	Postura1 instability+ (n=21)	Postural instability− (n=5)	P-value	Falls+ (n=14)	Falls− (n=12)	P-value
BP supine (mmHg)
- Systolic	134.4 ± 15.8	139.0 [122.0,145.0]	126.0 [124.0,127.0]	0.16	136.5 [123.0,141.0]	130.0 [119.0,146.0]	0.98
- Diastolic	82.1±12.4	81.5 [75.0,87.0]	81.0 [79.0,83.0]	0.97	82.0 [75.0,86.0]	81.0 [75.0,93.0]	0.64
- MAP	100.6±13.5	100.5 [92.0,109.0]	97.0 [93.0,98.0]	0.71	98.0 [93.0,108.0]	96.0 [91.0,112.0]	0.74
BP sitting (mmHg)
- Systolic	123.5±15.8	126.0 [115.0,132.0]	115.0 [108.0,122.0]	0.12	123.5 [115.0,129.0]	125.5 [109.0,134.5]	0.64
- Diastolic	75.8±11.1	75.0 [72.0,81.0]	73.0 [72.0,76.0]	0.43	74.0 [72.0,77.0]	75.0 [70.5,81.0]	0.45
- MAP	92.6±12.4	91.0 (87.0,99.0)	90.0 [84.0,90.0]	0.30	90.0 [87.0,94.0]	92.5 [84.5,99.5]	0.61
BP standing (mmHg)
- Systolic	114.0±14.8	116.0 [106.0,123.0]	108.0 [106.0,114.0]	0.36	112.5 [106.0,120.0]	117.0 [108.0,122.5]	0.57
- Diastolic	71.0±10.6	72.0 [65.0,74.0]	72.0 [67.0,73.0]	0.95	71.0 [65.0,73.0]	72.5 [66.0,74.5]	0.34
- MAP	86.0±11.2	87.0 [81.0,90.0]	85.0 [84.0,86.0]	0.58	84.0 [81.0,90.0]	86.5 [83.5,89.5]	0.47
Hypotensive load (%)	9.8 ± 16.1	1.4 [0.2,17.0]	6.4 [3.7,13.2]	0.67	4.3 [1.1,17.0]	0.6 [0.0,14.3]	0.20
MAP ≤75 mmHg (prevalence)	61.5%	57%	80%	0.35	79%	42%	0.054
Anti-OH treatments	8 (30.8%)	16 (76%)	2 (40%)	0.28	7 (50%)	1 (8.3%)	0.036

BP: Blood pressure; MAP: Mean arterial pressure supine; mmHg: Millimeters of mercury.

+indicates presence of condition while – indicates of condition.

Data are expressed as mean +/− standard deviation or median [interquartile range] as appropriate.

Prediction of falls by clinical assessment

No association was observed between falls and in-clinic episodes of postural instability during tests of gait and postural stability (p ≥ 0.22), or home-like activities (p > 0.08) (Supplementary material 9). Anti-OH treatment was more prevalent in the group of fallers (p=0.036) (Table 4).

Prediction of falls by kinematic assessment

We identified six clusters of kinematic variables describing the pattern of motor activities during instrumented clinical tests and home-like activities (Table 1; Supplementary materials 6 and 7). Falls were associated with CF^{sway EO} (p ≤0.01) and with the cluster of variables including JERK^{sway EO}, CF^{sway EO}, and WS^TUG (cluster 5; p= 0.013) (Supplementary material 10). At the ROC analysis, values ≥ −0.21 of cluster 5 (JERK^{sway EO}, CF^{sway EO}, WS^TUG) predicted the risk of falling with 77.8% sensitivity, 71.4% specificity, and 0.87 AUC. Values ≥ 0.88 Hertz of CF^{sway EO} predicted the risk of falls with 84.6% sensitivity, 83.3% specificity, and 0.81 AUC (Figure 1).

Fig 1. Accuracy of Technology-based Objective Measures in the Prediction of Falls.

ROC curve showing sensitivity and specificity for cluster 5, which includes JERK^{sway EO}, CF^{sway EO}, and WS^TUG (Panel A) and CF^{sway EO} (Panel B) in the prediction of falls. AUC: Area under the curve; Se: sensitivity; Sp: specificity; CC: Correct classification; LR+: Positive likelihood ratio; LR−: Negative likelihood ratio; CF: centroidal frequency; EO: Eyes opened; WS: waist postural sway; TUG: time-up- and-go.

Prediction of falls by hemodynamic data

There was a trend towards higher risk of falls in patients with orthostatic mean BP ≤75 mmHg (p=0.054) (Table 4). There was no independent association for symptomatic or asymptomatic OH status (p=0.65), hypotensive load (p= 0.20), or average BP values in the supine, sitting, and standing position (Tables 3, 4). Also, there were no differences in the supine, sitting, and standing BP in patients with symptomatic vs. asymptomatic OH (Table 3).

Association between kinematic and clinical variables

Postural instability was associated with older age (p= 0.027) and decreased WS^walk(11.6 vs. 18 degrees: p= 0.014) (Table 2, Supplementary material 10). Clusters 1 (Upper limbs range of motion^walk, peak turn velocity^TUG) and 2 (Gait speed^walk, stride length^walk, CF^{sway EC}) were correlated with lower bradykinesia subscore (p= 0.065 and p= 0.014, respectively), cluster 2 with lower number of postural instability episodes and higher Tinetti score, cluster 5 (JERK^{sway EO}, CF^{sway EO}, WS^TUG) with longer PD duration (p= 0.015), cluster 6 (Turn duration^TUG, WS^walk) with older age (p= 0.032), higher pull test score (p= 0.025), higher bradykinesia subscore (p= 0.002), higher MDS-UPDRS III score (p= 0.011), and lower Tinetti score (p≤ 0.02) (Supplementary material 11). Asymptomatic OH was associated with higher turn duration^TUG (p=0.035) and WS^walk (p=0.036) (Supplementary material 10).

DISCUSSION

While we found no association between falls and clinical tests of gait and postural stability or home-like activities, output from kinematic data predicted the risk of falls with high sensitivity and specificity (>80%) and diagnostic accuracy (AUC ≥ 0.80). Furthermore, we confirmed that an orthostatic MAP ≤75 mmHg increased the risk of falls, reinforcing the importance of this hemodynamic threshold in patients with OH [16]. Although these findings were obtained in a small cohort, the 53% prevalence of falls in this population mirrors that previously reported for this population [4].

We identified six clusters of kinematic variables describing motor behaviors of PD patients with OH with falls correlating with balance-based parameters, such as CF, JERK, and WS during tests of postural sway or time-up-and-go. Also, we confirmed the critical importance of hip compensatory movements on postural stability [28] and documented a new kinematic measure, the WS, which is an index of waist oscillatory activity during the cycle of gait, associated with both postural instability and motor symptoms severity. In summary, we found that sway kinematic parameters may be useful for predicting falls, while the time-up-and-go and the two-minute walk may be useful tests for predicting postural instability, correlating with PD motor symptoms severity.

Kinematic data obtained through wearables showed high accuracy in the prediction of falls, with sensitivity and specificity superior to that of the Hoehn and Yahr scale, considered the most robust clinical predictor of falls in PD [29]. This finding suggests that specific kinematic variables, CF^{sway EO} and JERK^{sway EO}. in particular, may be used as markers of postural instability to predict the risk of falls in PD patients with OH and to monitor the efficacy of therapeutic interventions aiming at improving balance and gait. CF has been tested in a recent study evaluating the effect of interactive cognitive motor training on gait and balance among older adults [30]. JERK is an established indicator of postural instability, already described in patients with PD [23]. Our study also suggests that WS may represent a sensitive marker of postural instability and motor symptoms severity, with the potential of capturing critical aspects of PD symptoms and postural instability using a single motion sensor placed on the belt.

From a hemodynamic standpoint, there was a trend toward a higher risk of falls in patients with orthostatic MAP ≤75 mmHg. These findings confirm the notion that orthostatic MAP may be more relevant than BP drop at the laying-to-standing test [16], and that asymptomatic OH does not mean “benign” OH. In fact, the “lack” of symptoms, classically restricted to postural lightheadedness, may result in failure to adopt precautionary behaviors and result in a greater frequency of falls. This hypothesis was confirmed by kinematic variables showing relevant differences in the compensatory strategies used by patients with and without orthostatic symptoms, while no differences were observed in the prevalence of postural instability, falls, or hemodynamic variables between the two groups. Our findings are in agreement with prior studies reporting a similar prevalence of cognitive [31] and ADL impairment [4] in patients with symptomatic and asymptomatic OH. The finding of a higher turn duration during TUG and hip oscillation at the 2-minute walking test in those with asymptomatic OH may represent precautionary or compensatory behaviors. In fact, the effects of OH on postural stability and balance are complex, multifactorial, and incompletely understood. OH may directly cause postural instability through cerebral hypoperfusion of the frontal areas responsible for balance and gait [32] and widespread compound neurodegeneration involving brainstem cholinergic regions, which are associated with postural stability [33]. The role of dopaminergic therapy in the induction or exacerbation of OH is debated. The intake of dopaminergic medications has been associated with limited catecholamine synthesis, reduced cardiac output, and decreased systemic vascular resistances [34, 35]. However, several studies have failed to show an association between dopaminergic therapy and the prevalence or severity of OH [36, 37]. In this study, the overall dopaminergic replacement was similar among groups. Thereby it is unlikely that dopaminergic therapies might have played affected the BP data. On the contrary, we found an association between falls and anti-OH therapy. While it is unlikely that anti-OH medications can induce or facilitate falls, it seems reasonable to hypothesize that patients on anti-OH treatment were the ones with more severe OH and, therefore, at higher risk of falling.

To the best of our knowledge, this is the first study employing a comprehensive kinematic assessment to evaluate postural instability and falls in PD patients with OH. Also, we identified specific technology-based biomarkers of postural instability and falls, which may be used as surrogate measures in future clinical trials of symptomatic and disease-modifying interventions. To minimize biases associated with artificial in-clinic evaluations, we extended the kinematic assessment to home-like activities of daily living and found fewer events of postural instability in the latter, likely reflecting compensatory strategies during the execution of home-like motor tasks. Of relevance, only home-like activities of daily living showed a trend towards falls, supporting the notion that the assessment of functional activities is more sensitive when using tasks similar to those performed during routine real-life activities [38].

Some limitations should also be considered in the interpretation of our findings. First, although our sample size calculations justified the scope of recruitment, the relatively small cohort precluded statistical power for evaluating secondary endpoints and additional factors associated with the risk of falls. Second, the higher prevalence of males might limit the generalizability of our findings. Third, available wearable technologies are currently limited in the possibility of collecting accurate ADL data in an unrestricted environment. Moreover, some parameters such as WS warrant additional clinical validation. Our finding that both static balance (CF^{sway EO}) and a combination of static balance and waist postural sway (cluster 5= JERK^{sway EO}, CF^{sway EO}, and WS^TUG) are associated with an increased risk of falls will require validation in an independent cohort of PD patients with and without OH to determine the extent to which these kinematic parameters are specific of PD vs. OH. Despite the evaluation of patients in a home-like environment, this only partially recapitulated the full complexity of real-life activities of daily living. Fourth, we cannot discount the possibility that the devices themselves may have affected the patients’ performance. Fifth, we used selected parameters. For instance, the JERK variable was calculated based on the acceleration, which restricts the information from the medio-lateral and anterior-posterior directions to one single measurement. Finally, due to the exploratory nature of this study, we did not adjust for multiple comparisons. This might have potentially inflated the magnitude of some findings.

Despite these limitations, our results suggest that available wearable technologies can be used to identify patients at risk of falling with sensitivity and specificity superior to that of in-clinic evaluations. We have also provided further support toward conceiving OH, with or without symptoms of postural lightheadedness, as a reversible source of postural instability and falls in PD patients, particularly in those with an orthostatic MAP ≤75 mmHg. These findings are in keeping with the goals of furthering the integration of technology into the routine care of patients with movement disorder as proposed by the International Parkinson Disease and Movement Disorders Society Task Force on Technology [39] and may inform objective endpoints for future clinical trials.