Gait Kinematic Parameters in Parkinson’s Disease: A Systematic Review

Literature search

We searched CENTRAL, MEDLINE, and PEDro from their inception to September 2019 using “Parkinson*”, “Gait”, “Walking”, “Accelerometer”, “Algorithm” and “Body-fix sensor” as key words. Reference lists from the identified articles were cross-checked to identify any further potentially eligible studies.

Study selection

We included all observational and experimental studies, or study protocols, conducted in PD patients or atypical parkinsonisms, that included a technology-based gait analysis focused on continuous gait disturbances and that specified which parameters had been studied. There were no restrictions regarding the type of intervention in the active and control arms.

We excluded reviews and studies written in languages other than English, French, Spanish, and Portuguese. All retrieved abstracts were independently screened by two authors. The full texts of potentially relevant articles were retrieved for further assessment. Disagreements were resolved by consensus.

Data extraction

Five pre-defined domains of items were extracted: general information (year and journal of publication, aim of the study, study design, population, intervention, time point assessments, technology development phase), gait assessment supplies (equipment, type of sensor, type of assessment), gait assessment procedures (protocol, medication status, and other outcome tools) and gait parameters values.

According to Maetzler’s classification [6], we classified studies according to their technology development phase, which covered three phases: i) preclinical development and testing (those studies focused on how to measure, i.e., testing algorithms or validating a new gait assessment system), ii) clinical development and testing phase (studies focused on the parameters that can be measured and on their clinical relevance) and iii) clinical validation (experimental and observational studies that use gait analysis as an outcome).

We also used an adaptation of the conceptual model of gait presented by Del Din, 2016 [9] to present and analyze the gait parameters reported in the included studies. Parameters that were only reported in one study, and not fitting the model, were included in the “other parameters” section. Data were extracted by two independent authors. Discrepancies were resolved through discussion.

Data analysis

We summarized the publication characteristics using frequencies and percentages. Review Manager software (v 5.3; Cochrane Collaboration) was used for calculating pooled mean difference (MD) and the 95% confidence interval (CI). Heterogeneity was assessed using the Q test and I² statistic. An I² value of <25% was chosen to represent low heterogeneity and an I² value of >75% to indicate high heterogeneity. A random-effects model was used to pool all outcomes. A p-value of <0.05 was considered to be statistically significant.

Gait parameters measured with non-wearable sensors

Table 3 lists the gait parameters using NWS reported in the included studies; the most frequently used unit of measurement and the mean and standard deviations of the reported values are also listed.

The most frequently reported parameters (≥20% of the studies) were gait velocity (81.1%, n = 30, PD mean value = 0.99±0.24 m/s), stride length (56.8%, n = 21, PD mean value = 1.06±0.18 m), cadence (48.7%, n = 18, PD mean value = 102.71±10.50 steps/min), step length (46.0%, n = 17, PD mean value = 0.58±0.13 m), double support phase (27.0%, n = 10, PD mean value = 25.89±7.23%) and step width (24.3%, n = 9, PD mean value = 0.13±0.02 m).

Gait parameters measured with wearable sensors

Table 2 lists the gait parameters assessed with a WS reported in the included studies; the most frequently used unit of measurement and the mean and standard deviations of the reported values are also listed.

The more frequently reported parameters (≥20% of the studies) were gait velocity (60.9%, n = 39, PD mean value = 1.01±0.26 m/s), stride length (37.5%, n = 24, PD mean value = 1.14±0.25 m), stride time (28.1%, n = 18, PD mean value = 1.18±0.18 s), cadence (28.1%, n = 18, PD mean value = 106.42±19.60 steps/min), step length (23.4%, n = 15, PD mean value = 0.60±0.06 m), step time (21.9%, n = 14, PD mean value = 0.55±0.03 s), stride time variability (21.9%, n = 14, PD mean value = 4.33±2.81% of the coefficient of variation (% CV)) and step time variability (20.3%, n = 13, PD mean value = 0.02±0.00 s).

Three studies evaluated gait in a controlled environment and nine in a free-living context. Due to both the low number of studies presenting a value for this parameter and the heterogeneity of the measurement units, we did not summarize the data nor present a reference value.

PD patients versus healthy controls

We were able to perform a forest plot analysis comparing the mean values of PD patients versus healthy controls (HC) for the following gait parameters: gait velocity, cadence, stride length, stride time, stride time variability, step length, step time, swing time, and double support time. All, except step time using WS, presented a statistically significant difference between groups. For gait velocity and stride length, a statistically significant difference between groups was found in WS assessment, but not in the assessment using NWS (Supplementary Material 2).

Wearable versus non-wearable sensors assessment

Comparison between the two types of devices was possible for gait velocity, stride, and step length. While gait velocity presented a statistically significant difference (p = 0.04, I² = 76.7%), there was no difference between WS and NWS in stride (p = 0.35, I² = 0%) or step length (p = 0.14, I² = 55%) (Supplementary Material 2).

Type of wearable sensor

The use of an accelerometer was compared with the use of other types of sensors for gait velocity. The subgroup analysis was not statistically significant (p = 0.18 and I² = 44.7%). Both groups showed a statistically significant difference between PD and HC (p≤0.05). The available data did not allow other comparisons for this topic (Supplementary Material 2).

Sensor location

The impact of sensor location (lower back versus feet versus other locations) was studied for gait velocity, stride time, and stride time variability. No differences between groups were registered. Heterogeneity (I²) ranged between 0–52.9%. All the parameters, except for stride time variability, using the sensor in the lower back, showed a statistically significant difference between PD and HC (p≤0.05) (Supplementary Material 2).

Sample characteristics

General characteristics of technology-based gait analysis in PD

From the 95 included studies, according to the technology development phase classification: 24.2% of the studies (n = 23) were in the preclinical development and testing phase, 31.6% (n = 30) were in the clinical development and testing phase and 44.2% (n = 42) belong to the clinical validation phase.

Protocol details

Table 4 shows the characteristics of the gait assessment protocol. The most frequently used distance in laboratory assessments was 10 meters (n = 23), the shortest distance reported was 3 meters and the longest 500 meters. Table 5 compares PD patients’ gait velocity using a gait assessment protocol with less than 10 meters, 10 meters and more than 10 meters. Due to the heterogeneity of the data, this comparison was only performed for gait velocity and a forest plot analysis was not possible.

The mean number of trials was 4.3±2.9. In 46.1% of the studies (n = 41), gait assessment was performed at a self-selected comfortable speed. In free-living assessments, the most common duration of data collection was 7 days (58.8%, n = 10).

In 58.5% of studies (n = 48), patients were in an “ON-state” during the assessment, in 7.4% (n = 7) in an “OFF-state” and in 4.2% of the studies (n = 4) the assessment was performed in both conditions (Table 4). Table 6 compares the PD mean values with and without having into account the “ON/OFF” medication state. Due to the low number of studies assessing gait in “OFF” state medication (n = 11, 11.6%) and the heterogeneity of the data, this analysis was only possible to perform for some gait parameters and did not allow for a forest plot analysis. Except for stride time variability, all the mean values of the studies only including an “On” state medication assessment, were closer to those from the HC group.

1) What should be measured?

The most frequently reported parameters in the included studies were gait velocity, stride and step length, and cadence. Compared to HC, PD patients had decreased velocity, reduced stride and step length, decreased swing time, increased stride time, stride time variability and dual support time (p < 0.05). These differences are in line with the usual description of PD gait impairments, i.e., a slow, short-stepped, shuffling, with a forward-stooped posture and asymmetrical arm swing [7, 13].

Beyond this, a large number of different, or differently measured gait parameters, were found in the included studies. From a clinical point of view, not every parameter that can be measured should be measured [6, 8]. The collection and interpretation of the data must lead to justified outcomes, i.e., those with an impact on activities of daily living, displayed in a visually intuitive format that covers the clinical information needs of the stakeholders (health professionals, patients, and caregivers) [6, 8]. For this, gait parameters should be correlated with robust measures of clinical meaningfulness, such as the MDS-UPDRS motor score or the Timed Up and Go Test (TUG). Once the most suitable parameters to measure PD gait impairments in different contexts are established, then the minimal clinically important differences should be addressed for each [6, 8]. Other measures emerging from the nonlinear analysis of human variability (e.g., entropy, fractals, and others) can give us a more accurate angle of patients’ gait dynamics in a real-life environment. However, work is needed to make them more intuitive and clinically informative [6, 8].

Although currently, sensor-based gait analysis has demonstrated feasibility and applicability for objectively assess PD gait impairments, differences still exist measuring the same parameter, with different devices or devices from different manufacturers [3, 15]. This highlights the difficulty of accurately measuring the spatiotemporal gait parameters and the need to continue developing valid and reliable mathematical algorithms. Despite the major technological advances and the current possibility of capturing and store extremely high amounts of data with TOMs, the ability to algorithmically analyze (eliminating the noise) and summarize the clinically relevant data to stakeholders remains limited. [3]

2) Which devices should be used?

The comparison between assessments using WS and NWS was investigated for gait velocity, stride and step length parameters. A statistically significant difference between groups was found in gait velocity (p = 0.04). Although it was the analysis with the highest number of studies (n = 18), due to the level of heterogeneity (I² = 76.7%), the results should be interpreted with caution. We believe that the differences in the type of devices and in the assessment protocols of the included studies might have contributed to this result.

No statistically significant difference was found in the two other parameters (stride length – p = 0.35, step length – p = 0.14). Taking into account the low value for heterogeneity (I² = 0%, p < 0.001), we believe that wearable sensors can be used in place of NWS (the gold standard of gait analysis).

WS have the added value of enabling the assessment of gait during activities of daily living in the patients’ actual environment. However, more studies exploring how gait parameters behave in a real-world context are needed [4].

It was only possible to explore the impact of the type of WS for gait velocity. This was undertaken by comparing the use of accelerometer (used in 67.2% of the WS) with all other types of sensors.

Accelerometers allow the measurement of dynamic accelerations of a body, when submitted to an external force, and provide information about the device orientation related to gravity [3, 15]. They are frequently combined with a gyroscope, which allow for the measurement of angular velocities [3, 15]. In some devices, a 3D-magnetometer is also added for orientation purposes.

Since no difference was found in this subgroup analysis (accelerometer versus all other types of sensors) and both groups were able to detect a statistically significant difference between PD and HC, we believe that for an accurate assessment and monitorization of PD patients’ gait impairments, the use of a single accelerometer is feasible. However, for the assessment of turns or of a more complex movement that requires the information captured by angular velocity, wearable devices including at least a gyroscope, seem more suitable.

In the included studies, only one study used an isolated magnetometer for gait analysis. Since magnetometers are very sensitive to magnetic changes (e.g., those produced by proximity with ferromagnetic objects) and therefore to many external interferences, they are more frequently used as a complement to accelerometers and gyroscopes, than as a single sensor [3, 15].

3) Where to place the sensor?

Our results showed that in 46.9% (n = 30) of the studies using WS, the sensor was used on the lower back, between the second and the fifth lumbar vertebra. Although it was only possible to investigate the impact of sensor location for three parameters, it was limited to the comparison between lower back, feet and all other locations, the results consistently show no statistically significant difference between groups. Stride time variability measured with the sensor in the lower back was the only parameter that did not show a statistically significant difference between PD and HC. However, a heterogeneity (I²) of 82% was found, whereby these results should be interpreted carefully.

Several gait analyses protocols have been used. However, an optimal and standardized method remains for establishing [15]. The number and location of the sensors are key aspects for the success of assessments with TOMs, especially in a free-living context [8, 16]. To increase wearing compliance without hindering the precision of data collection the number of sensors should be kept to a minimum, and the least obtrusive devices preferred [8, 16]. Today, although the lower back is not considered the most comfortable and unobtrusive location, it has been shown that a single sensor (accelerometer) in this location is able to capture with precision, physical activity and gait parameters in a laboratory and free-living context [16, 17]. Recently, there has been a move toward using sensors on the wrist or embedded in smartphones. However, problems still exist when collecting data. Kim et al., 2019 [16] report that sensors used on the wrist tend to overestimate the number of steps and the time spent at different intensities of activity. Höchsmann et al., 2018 [18] compared the accuracy of step detection of a smartphone (placed in a trouser pocket, shoulder bag, and backpack) with a WS used on the wrist and waist. At a gait velocity of 4.8 km/h (shoulder bag and backpack) and 6.0 km/h (all positions), smartphones did not exceed a 1% error deviation from the gold standard (threshold to be considered an accurate measurement). However, for a gait velocity of 1.6 km/h, a 3% error was found. In a free-living context, smartphones underestimate the number of steps [18]. Another limitation of free-living assessment with smartphones is the place where it is used. While for men a trouser pocket is a commonly preferred position, for women it is more likely to be the purse or backpack [18]. In the search for a solution for a smartphone-based body location the magnetometer sensor will most certainly be a crucial sensor to consider when dealing with the device’s orientation.

4) Which gait assessment protocol

The comparison between all the included studies and those that only used an assessment in “ON” state medication, revealed that PD gait parameters under the effect of the medication are closer to the HC. Only stride time variability did not follow this pattern. According to the literature [12], stride time variability is increased in PD patients and diminishes in response to dopaminergic medication. In our analysis, we found that the difference between PD on and HC increased when only studies assessing gait in “ON” state medication, were taking into account. However, this result should be interpreted with caution, since this was only a basic comparison of means and gait protocols differentiated substantially in the included studies.

The distance covered during gait analysis varied in the included studies. According to the analysis performed, the distance doesn’t seem to have a high impact on gait velocity tested in a controlled environment. However, the data from the included studies doesn’t allow us to conclude on this topic. More studies are needed to understand the implications of gait protocol length in PD gait parameters.

Almost half of the included studies (43.2%, n = 41) used only a self-selected comfortable speed, during gait assessment. Since some of the gait parameters, like stride length and cadence, are sensitive to velocity and to the presence of concurrent attention demands, gait assessment protocols should include different velocities and both single- and dual-task activities [19]. The most common duration of free-living assessment data collection was seven days, varying between three and ten days. Based on our results, we cannot conclude if this is the best option. These are challenging assessments due to the heterogeneity of ambulatory activity within habitual environments. We believe that the duration of data collection during free-living assessments should be a balance between not performing a burdensome assessment and the ability to collect enough and precise data to obtain a pattern of patients’ performance during the day [8]. As a fluctuating disease, the duration applied in other research fields, may not be appropriate. This topic should be addressed in future studies.

Conclusion

Our results support previous descriptions of PD gait impairments when compared with HC. No statistically significant differences were found for the impact of different types of devices (WS vs NWS), or different types or locations of wearable sensors during assessments. Future studies should test the reported gait parameters against validated clinical meaningful outcome measures in PD to select those most suitable for evaluating and monitoring the progression of gait impairments in PD. More studies are also needed to explore gait parameter behavior in a free-living context, with more complex movements (e.g., including turns, sequences of movements and others).