Table 1 Summary of literature on predicting future falls in Parkinson’s disease using digital devices

Pelicioni et al. (2019)²⁹

Contrast fall rates between PIGD and non-PIGD PD subtypes. Clinical assessment is used to classify phenotypes.

113 PD (67 PIGD, 46 non-PIGD) Prospective study; recording of gait-related features using triaxial accelerometer; also measured other disease and clinically relevant features.

PIGD more likely to fall overall with more falls related to FOG, balance-related falls, and at home.

12

Shah et al. (2023)⁷⁰

To investigate if digital measures of gait collected passively over a week of daily activities in people with PD increases the discriminative ability to predict future falls compared to fall history alone. Mixed group of fallers and non-fallers at time of recruitment.

34 PD (17 fallers, 17 non-fallers). 3 IMU sensors for one week of passive gait monitoring. Followed up by email every 2 weeks for a year for self-reported falls.

Inertial sensors worn on the feet and lumbar level for 7 days provided measures of gait pace, variability and turning that increased the ability to predict future falls in addition to history of previous falls

12

Ullrich et al. (2023)⁷¹

To compare different data aggregation approaches and machine learning models for the prospective prediction of fall risk using gait parameters derived either from continuous real-world recordings or from unsupervised gait tests. (FallRiskPD study). Mixed fallers and non-fallers.

35 PD patients with foot-worn sensors performing unsupervised 4 x 10 m walking tests over two weeks. Falls were self-reported for 3 months.

The highest accuracy (74%) was achieved with a Random Forest Classifier applied to the passive monitoring gait data, when aggregating all walking bouts and days of each participant

3

Tsai et al. (2023)⁴⁷

To evaluate the feasibility of combining disease-specific and balance-related measures as risk predictors for future falls in patients with PD. Mixed fallers and non-fallers.

95 patients underwent postural sway measurements and clinical functional scores. Followed up patients to determine if fall occurred after 6 months.

Fall history, Tinetti balance, sway length, velocity, and gait score associated with future falls.

6

Weiss et al. (2014)³²

Whether metrics derived from three-day continuous recordings were associated with fall risk. Non-fallers followed for one year to evaluate predictors of transition from non-faller to faller.

67 non-falling PD patients wore an accelerometer on lower back for 3 days.

Higher than median gait variability was associated with higher risk of progressing to fall in the follow-up period

12

Ma et al. (2022)⁷²

To evaluate gait features associated with higher risk of falls. Mixed fallers and non-fallers (both groups had history of falls)

51 PD patients assessed using six wearable gyroscopes and accelerometers during 7 m timed up and go task

Increased gait variability was associated with increased risk of falls

6

Sturchio et al (2021)⁷³

Whether kinematic features can predict increased risk of falls. Most participants already falling at the time of recruitment.

26 PD patients with orthostatic hypotension assessed using two-minute walk test, timed up and go and postural sway tasks

Waist sway, jerkiness, centroidal frequency predicted increased risk of falls

6

Greene et al. (2021)⁷⁴

Whether the number of falls can be estimated from kinematic parameters. Mixed datasets of falling and non-falling participants were used.

71 patients with PD compared with healthy older population. Participants were assessed using a three-metre timed up and go test with two sensors placed on shins

There was a moderate correlation between sensor data and actual fall counts

Up to 6