After Alzheimer's, Parkinson's disease (PD) is the most common neurodegenerative disease worldwide, affecting around 1% of the population over 65, and its increase in prevalence has been documented in the past decades [1], [2]. PD cardinal features consist of tremor at rest, rigidity, akinesia (absence of movement) or bradykinesia (slowness of movement) and postural instability, although it is important to point out that the symptoms and the rate of progression of this disease differ between individuals [3], [4]. It is very common for patients with PD to go through gait and postural complications, such as freezing of gait, dysrhythmic and slow gait, flexed posture and decreased postural responses [4]. Postural instability is one of the most disabling motor symptoms and a major contributor to falls, and as a result, it threatens the independence and well-being of patients, diminishing their quality of life. Usually, this postural-related impairment is only discernible later in the course of the disease, but can be present at diagnosis and, as the disease progresses, becomes more prominent [2], [3], [4], [5].

Postural stability can be described as the ability to keep the body's centre of mass (CoM) within limits of stability (LoS), through static and dynamic conditions [5]. Indeed, postural stability is crucial to autonomously accomplish basic daily tasks as walking or even standing, and it can be achieved by the interaction of the visual, vestibular and somatosensory systems [5]. Therefore, postural instability can be characterized by a decrease in the LoS, the magnitude of postural responses and postural reflexes, which affect the performance of both static and dynamic activities [1], [2], [4].

To continuously monitor patients' postural stability and implement an adequate treatment to improve postural reflexes and reduce the risk of falls, it is important for physicians to quantify gait and balance deficits and to perceive motor changes that lead to postural complications [1], [4]. Generally, the clinical examination used to assess postural instability is a test denominated by the pull test, also known as the retropulsion test [5]. To perform the test, the examiner pulls the patient backwards by the shoulders with enough strength to displace the centre of gravity and make the patient take a step back. After that, the examiner grades the corrective postural response, based on the number of steps or falls, using a five-point scale (0-4), as shown in Table 1 [2], [5]. The direction of the perturbation used to perform the pull test is considered to be one of the most destabilizing. Nevertheless, healthy subjects are able to regain their balance with two or even one large step, with the help of the hip and the arms to counterweight, while PD patients with balance impairment are forced to take more steps, considering their balance correcting steps are often rapid and shorter and that they have reduced arm swing due to the increased muscle rigidity [2], [6].

The pull test is widely used since it is easy to administer, does not require specific instruments and marks a transition in the Hoehn and Yahr (H&Y) scale to stage 3, allowing to distinguish between a milder stage of PD (H&Y 1 and 2) and a moderate or severe stage (H&Y 3 to 5). However, pull test presents several limitations which prevent the physicians to monitor the progression of the disease accurately. In fact, it is difficult to standardize this test considering: (i) the variability of the pull force applied, existing an inherent subjectivity to interpret and score the examination; (ii) has a very limited scaling; (iii) lacks sensitivity to detect small but significant changes in balance and consequently the patients that are at a higher risk of falling; (iv) has poor reliability and is limited by physician's bias [5], [7], [8]. In addition, the pull test cannot distinguish between healthy people and PD patients in which the impairment of postural responses is not evident, considering these patients are typically able to regain balance within two steps. In fact, several studies have shown that, in the early stages of Parkinson's, patients already exhibit decreased postural responses and can be destabilized more easily than healthy subjects due to inefficient postural adjustments [2], [9].

To overcome the pull test as a standard examination of patients' postural instability, instrumented analysis has been proposed with the use of force platforms and motion capture technologies [2], [5]. The force platform is a technology that allows the measurement of the centre of pressure (CoP), defined as the location of the resultant vertical force on the force platform surface. Through the displacement of this measure, it is possible to obtain postural movements made by an individual while trying to maintain a balanced position [5]. Motion capture technologies use cameras (and other devices) to identify the position of the joints to perform an analysis, such as Vicon® or Qualisys® systems. Both equipments increase the objectivity of the evaluation of postural instability and place minimum constraints on the subject [2], [5], [8]. However, they have limited clinical adoption because they are expensive, require a lot of space and time to set up and lack portability [10], [11].

Wearable sensors emerged as promising solutions to better capture postural instability and eliminate the subjectivity of postural-associated clinical examinations, besides being easily accepted by patients given their low weight and size [3], [11], [12]. Being mainly comprising inertial sensors, such as accelerometers or gyroscopes, they are able to acquire data that can be applied to mathematical empirical calculations or more intelligent algorithms to estimate relevant postural metrics [5]. Despite being frequently used in motion assessment research and showing promising results, it is required to gather more scientific knowledge and further exploration about wearable sensor applications in clinical settings to monitor the postural stability of patients with PD. In fact, wearable sensors have not yet been adopted in hospitals or clinical institutions as a default tool to complement clinical examinations. A possible reason includes the lack of consensus in terms of sensor configurations (number and placement), but above all, about the correct protocol and measures to be considered in postural assessment [13], [14], [15]. Several studies have assessed patients' postural control during quiet stance by extracting sway measures and concluded that these metrics are relevant to characterising postural stability and identifying individuals with these impairments [8], [11], [12], [15]. However, gait and other dynamic basic activities can also provide valuable parameters to assess postural stability [16], [17], [18]. Taking all this into consideration, it becomes crucial to extract gait and postural-related features from data acquired while performing dynamic basic daily tasks to produce reliable and validated objective markers of PD postural stability [15]. This research takes a systematic approach to address this challenge and brings a step advance in the knowledge about motor metrics as an objective assessment of pull test scores in PD under dynamic conditions.

This research aimed to study the ability of gait and postural-associated metrics estimated by inertial-driven data to mark the different levels of postural instability assessed by pull test scores. It was performed a new cross-sectional study to conduct extensive statistical research to determine (i) which gait and postural-associated metrics are considered significant to distinguish between the different levels of pull test, and (ii) which gait and postural metrics are more correlated with the pull test score. This study is expected to further verify the hypothesis that inertial data captured from wearable inertial sensors can be used to measure motor metrics while patients perform their day-to-day tasks, that can contribute for an objective assessment of pull test scores in PD. To the best knowledge of the authors, this is the first study which accomplished this analysis on patients' postural stability in several motion tasks, as sitting, lying, walking together with pull test.