Skip to main content

Table 1 Study design, sample characteristics and main results

From: Sensor-based fall risk assessment in older adults with or without cognitive impairment: a systematic review

Author, year

Study design, sample including number of participants, mean age (SD) and sex

Cognitive Status

Record of falls/ fall history

Main findings

Bautmans, 2011 [40]

Cross-sectional

Community-based

Total n = 121, 80 (5), 50% female; Younger adults n = 40, 22(1), 50% female

Cognitively intact according to MMSE (MMSE≥24)

Retrospective 6 months, Tinetti Assessment Tool, Timed-Up and Go

HFR n = 40, LFR n = 41

- Participants with HFR showed slower gait speed (p < 0.05)

- With cut-off value 1.58 m/s gait speed discriminates between HFR and LFR with 78% sensitivity and 76% specificity

Bizovska, 2018 [43]

Prospective study

Community-based

Total n = 131, 71 (6), 82% female

CI as exclusion criterion

Prospective 1 year

SF = 35, MF = 15, NF = 81

- Trunk medial-lateral acceleration in short-term Lyapunov exponent differed between MF and NF (p < 0.05) but not after Bonferroni correction;

- Poor MF predictive ability of trunk medio-lateral short-term Lyapunov exponent but results improved when combining with clinical examination

Brodie, 2017 [59]

Cross-sectional

Community-based

Total n = 96, 75 (8), 59% female

CI as exclusion criterion according to MiniCog

Retrospective 12 months

F = 33, NF = 63

- Fallers showed significantly reduced gait endurance and increased within-walk variability (p < 0.05)

Brodie, 2015 [47]

Cross-sectional

Community-based

Total n = 96, 80 (4), 67% female

No information about CI

Retrospective 1 year, Physiological Profile Assessment Tool

F = 35, NF = 61

- 8-step mediolateral harmonic ratio identified significant differences in between F and NF based on age, walking speed and physiology (p < 0.05)

Buckinx, 2015 [48]

Prospective study

Nursing homes

Total n = 100, 86 (6), 80% female

No information about CI

Prospective 2 years

F = 75, NF = 25

- Gait characteristics were not predictive of long-term falls

Buisseret, 2020 [64]

Prospective study

Nursing homes

Total n = 73, 83 (8)

62% female

CI included,

16% with dementia

Prospective 6 months

F = 23, NF = 50

- When the Timed-Up and Go test results are coupled with indicators of gait variability measured during a six-minute walk test, accuracy of fall prediction improved from 68 to 76%

Ejupi, 2017 [60]

Cross-sectional

Community-based

Total n = 94, 80 (7), 68% female

CI as exclusion criterion according to MiniCog and MMSE

Retrospective 12 months

F = 34, NF = 64

- F showed significantly lower maximum acceleration, velocity and power during sit-to-stand movements compared to NF (p < 0.05)

Gietzelt, 2014 [36]

Cohort-study

Nursing homes

Total n = 40, 76 (8), 50% female

CI included (MMSE 9.3 ± 8.0)

Prospective for 2, 4 and 8 months

F = 13, NF = 27

- It is possible to classify gait episodes of F and NF for mid-term monitoring (4 months) during daily life using body-worn sensors (75.0% accuracy)

Greene, 2012 [55]

Prospective study

Community-based

Total n = 226, 72 (7), 73% female

CI as exclusion criterion

Prospective 2 years

F = 83, NF = 143

- Sensor-derived features yielded a mean classification accuracy of 79.69% for 2-year prospective falls

Howcroft, 2016 [56]

Cross-sectional

Community-based

Total n = 100, 76 (7), 56% female

CI as exclusion criterion according to self-reports

Retrospective 6 months

F = 24, NF = 76

- Best fall classification model using pressure-sensing insoles and head, pelvis and shank accelerometers (84.0% accuracy)

- Best single-sensor model with parameters derived from a head sensor during single task (84.0% accuracy)

Howcroft, 2018 [57]

Prospective study

Community-based

Total n = 75, 75 (7), 59% female

CI as exclusion criterion according to self-reports

Prospective 6 months

F = 28, NF = 47

- F had significantly lower dual-task head anterior-posterior Fast Fourier Transform first quartile, single-task left shank medial-lateral Fast Fourier Transform first quartile, and single-task right shank superior maximum acceleration (p < 0.05)

Hua, 2018 [41]

Cross-sectional

Community-based

Total n = 67, 76 (6), 100% female

No information about CI

Retrospective 1 year, Short Physical Performance Battery

HFR = 19, LFR = 48

- Coefficient of variance, cross-correlation with anteroposterior accelerations, and mean acceleration were the top features for classification in HFR and LFR group

Ihlen, 2018 [44]

Prospective study

Community-based

Total n = 303, 76 (7), 50%

Including CI (MMSE≥19)

Prospective 6 months

SF = 58, MF = 46, NF = 199

- Higher phase-dependent multiscale entropy of trunk acceleration at 60% of step cycle in F compared to NF (p < 0.05)

- PGME has predictive ability of falls among SF

Ihlen, 2016 [49]

Cross-sectional

Community-based

Total n = 71, 78 (5), 65% female

Cognitively intact according to MMSE score (≥24)

Retrospective 12 months

F = 32, NF = 39

- Refined composite multiscale entropy and refined multiscale permutation entropy of trunk velocity and trunk acceleration can distinguish between daily-life walking of F and NF (75.0–88.0% sensitivity, 85.0–90.0% specificity)

Iluz, 2016 [35]

Cross-sectional

Community-based

Older adults total n = 71, 78 (5), 65% females;

Younger adults Total n = 30, 28 (4), 57% female

Cognitively intact according to MMSE score (≥24)

Retrospective 1 year

F = 33, NF = 38

- Temporal and distribution-related features from sit-to-walk and sit-to-stand transitions during daily-life differed significantly between F and NF

- Mean classification accuracy was at 88.0% and better than traditional laboratory assessment

Mancini, 2016 [45]

Cross-sectional, prospective

Community-based

Total n = 35. 85 (5), 66% female

Dementia as exclusion criterion according to Clinical Dementia Rating Scale and/or MMSE

Retrospective 12 months, prospective 6 months

Retrospective analysis: SF = 12, RF = 7, NF = 16

Prospective analysis:

F = 7, NF = 28

- Quality of turning (mean turn duration, mean peak speed of turning, mean number of steps to complete a turn) were significantly compromised in RF compared to NF (p < 0.05)

Marschollek, 2009 [61]

Cross-sectional

Geriatric setting

Total n = 110, 80 (−), 74% female

no information about CI

Retrospective n/a

F = 26, NF = 84

- Pelvic sway while walking, step length and number of steps in TUG differed significantly between F and NF (p < 0.05)

- Adding sensor-based gait parameters to geriatric assessment improves specificity in fall prediction from 97.6 to 100.0%

Marschollek, 2011 [62]

Prospective

Geriatric setting

Total n = 46, 81 (−), −

No information about CI

Prospective 1 year

n/a

- Sensor-derived parameters can be used to assess individual fall-risk (58% sensitivity, 78% specificity) and identified more persons at fall risk than a conventional clinical assessment tool

Pozaic, 2016 [63]

Cross-sectional

Community-based

Total n = 136, 73 (6), 69% female

CI as exclusion criterion according to Screening of Somatoform Disorders (> 10)

Prospective 1 month

Fn = 13, NF = 123

- Time and frequency domain-based features derived from a wrist-worn accelerometer on the dominant and non-dominant hand can significantly distinguish between F and NF (p < 0.05)

Qui, 2018 [50]

Cross-sectional

Community-based

Total n = 196, 72 (4), 100% female

No information about CI

Retrospective 5 years

F = 82, NF = 114

- Sensor-based data distinguished accurately between F and NF (89.4% accuracy)

Rivolta, 2019 [19]

Cross-sectional

Hospital setting

Older adults total n = 79, 69 (17), −

Younger adults total n = 11, 35 (−), −

No information about CI

Tinetti Assessment Tool

HFR = 33, LFR = 46

- Sensor-based balance and gait features assessed during Tinetti Test differed significantly between individuals with HFR and LFR (p < 0.05)

- Linear model and artificial neural network had a misclassification error of 0.21 and 0.11, respectively, in predicting Tinetti outcome

Sample, 2017 [58]

Cross-sectional

Community-based

Total n = 150, 76 (9), 59% female

No information about CI

Retrospective 12 months

F = 59, NF = 91

- Sensor-based data collected during Timed-Up and Go resulted in a more sensitive model (48.1% sensitivity, 82.1% specificity) than including Timed-Up and Go time duration only (18.2% sensitivity, 93.1% specificity)

Senden, 2012 [51]

Cross-sectional

Community-based

Total n = 100, 77 (6), 56% female

CI as exclusion criterion

Tinetti Assessment Tool

HFR = 19, LFR = 31, NFR  = 50

- Walking speed, step length and root mean square had high discriminative power to classify the sample according to the Tinetti scale (76.0% sensitivity, 70.0% specificity).

van Schooten, 2015 [52]

Cross-sectional, prospective

Community and residential care home

Total n = 169, 75 (7), 54% female

CI included (MMSE≥18)

Retrospective 6 months; prospective 6 months

Retrospective analysis:

F = 60, NF = 109

Prospective analysis:

F = 59, NF = 110

- Sensor-derived parameters of the amount of gait (number of strides), gait quality (complexity, intensity, smoothness) and their interactions can predict prospective falls (67.9% sensitivity, 66.3% specificity).

Wang, 2017 [46]

Prospective

Community-based

Total n = 81, 84 (4), 44% female

No information about CI

Prospective 12 months

MF = 11, NF = 70

- Rate in stair descent was higher in MF than in NF (p < 0.05).

Weiss, 2011 [53]

Cross-sectional

Community-based

Total n = 41, 72 (7), 66% female

Cognitively intact according to MMSE score (≥24)

Retrospective 1 year

F n = 23, NF n = 18

- Sensor-derived Timed-Up and Go duration was significantly higher in F compared to NF (p < 0.05)

- Jerk Sit-to-Stand, SD and average step duration correctly classify 87.8% of F and NF (91.3% sensitivity, 83.3% specificity)

Weiss, 2013 [54]

Prospective

Community-based

Total n = 71, 78 (5), 65% female

Cognitively intact according to MMSE score (≥24)

Prospective 6 months

F = 39, NF = 32

- Gait variability differed significantly between F and NF (p < 0.05);

Zakaria, 2015 [42]

Cross-sectional

Hospital setting

Total n = 38, 67 (7), 47% female

No information about CI

Timed-Up and Go

HFR = 21, LFR = 17

- Sensor-derived parameters of Timed-Up and Go phases can classify into people at HFR and people at LFR.

  1. SD: standard deviation, n: number, MMSE: Mini-Mental State Examination, HFR: high fall risk, LFR: low fall risk, CI: cognitive impairment, SF: single faller, MF: multiple faller, NF: non-faller, F: faller, NFR: no fall risk