Background
Methods
Search strategy and criteria
Results
Study selection
Study analysis
Author/Year | Faller Identification Method | Population/Sample Size/Age (Mean ± SD) | Technology | Sensor placement if applicable | Test Protocol | Outcome Measures | Model | Model validation | Accuracy | Specificity | Sensitivity | AUC |
---|---|---|---|---|---|---|---|---|---|---|---|---|
Bautmans et al. 2011 [34] | Fall history (> = 1) in past 6- month, or TUG >15 s, or Tinetti <=24 | F (n = 40, 80.6 ± 5.4), NF (n = 41, 79.1 ± 4.9), YA (n = 40, 21.6 ± 1.9) | A single Tri-axial Accelerometer | Sacrum | Straight line walking | Gait Speed, Step time symmetry, step/stride regularity | Logistic regression. Only gait speed was effective for discrimination analysis | NA | 77 | 78 | 78 | 0.83 |
Caby et al. 2011 [35] | Fall history (> = 1) in past 1-year, with additional physician screening | F (n = 15, 80.1 ± 5.3), NF (n = 5, 83.2 ± 4.3) | 10 Tri-axial Accelerometer sensor network | Knee, Ankle, Elbow, Wrist, Shoulder | Straight line walking | 67 gait acceleration features extracted(temporal, frequency, power, and correlation between sensors) | RBFN SVM, KNN, NB | Leave-one-out cross validation | 75-100 | 40-100 | 93-100 | |
Jansen et al. 2011 [36] | Fall history (> = 1) unknown length, or TUG >15 s, or Tinetti <=24 | F (n = 40, 80.6 ± 5.4), NF (n = 40, 79.0 ± 5.0) | A single Tri-axial Accelerometer | Sacrum | Straight line walking | 22 acceleration features 5 groups (step count, step time, step length, step symmetry and step RMS) | NB,MLP,SVM, LWL, Decision Tree, NEAT | Ten-fold cross validation (Max value) | 61-82 | 62-84 | 58-80 | |
Liu et al. 2011 [37] | Fall history (> = 1) in past 1-year | OA (n = 68, 80.1 ± 4.4; MF/NMF = 9/59) MF (> = 2 falls) | A Tri-axial Accelerometer | Waist | TUG,AST,STS5 | 126 features (temporal, energy, spectral) | Linear multiple regression, | Leave-one-out cross validation | 78 | 90 | 59 | |
Marschollek et al. 2011 [38] | 1-year prospective fall occurrence (> = 1) | OA (n = 46, 81.3; F/NF = 19/27) | A Tri-axial Accelerometer | Waist | TUG, Straight line walking | Kinetic Energy, Pelvis Sway, Gait variability, Step time/length, number of steps for TUG, spectral density parameters | Decision tree, logistic regression | Ten-fold cross validation (mean value) | 65-80 | 78-96 | 42-74 | 0.65-0.87 |
Paterson et al. 2011 [40] | 1-year prospective fall occurrence (> = 1) | F (n = 54, 69.0 ± 6.9) NF (n = 43, 68.4 ± 7.3) | Two Tri-axial Accelerometers | Foot mount | 7 min walking on a circuit | Stride dynamic (Fractal Scaling Index) | Logistic regression | NA | 67 | 58.1 | 74.1 | |
Weiss et al. 2011 [39] | Fall history, past 1-year (> = 2) | F (n = 23, 76.0 ± 3.9) NF (n = 18, 68.3 ± 9.1) | A Tri-axial Accelerometer | Lower back | TUG | Duration of TUG and subtasks, acceleration range and Jerk. Number of steps for TUG, gait speed | Logistic regression | NA | 63.4-87.8 | 50.0-83.3 | 65.2-91.3 | |
Yamada et al. 2011 [20] | Fall history (> = 1) in past 1-year | F (n = 16, 84.8 ± 10.1) NF (n = 29, 80.2 ± 6.4) | Wii Balance Board | NA | Game-based measure in seated/standing | Game score | Discriminate analysis | NA | 88.6 | |||
Greene et al. 2012 [21] | 2-year prospective fall occurrence (> = 2) | F (n = 83, 71.8 ± 6.9) NF (n = 143, 71.4 ± 6.6) | Two Tri-axial Inertial sensors (accelerometer/gyroscope) | Shank | TUG | 44 features (spatial/temporal gait parameters, angular velocity parameters, turn parameters) | Discriminate classifier | Ten-fold cross validation (mean value) | 73-83 | 73-96 | 56-90 | 0.74-0.85 |
Greene et al. 2012 [22] | Fall history (> = 2, or one fall requiring medical attention) in past 1-year | F (n = 65, 74.0 ± 5.8) NF (n = 55, 73.3 ± 5.8) | A Tri-axial Inertial sensor (accelerometer/gyroscope) | Lower back L3 | Standing balance (EO/semi- tandem, EC/narrow stance) | RMS of AP/ML acceleration, frequency variability, spectral entropy | SVM | Ten-fold cross validation (mean value) | 63-72 | 58-82 | 59-67 | |
Schwesig et al. 2012 [23] | 1 year prospective fall occurrence (> = 1) | OA (n = 141, 82.7; MF/NMF = 17/124, MF (> = 3 falls) | Two Tri-axial Inertial sensors (accelerometer/gyroscope) | Shoe-mounted | Straight line walking | Temporal gait parameters | Logistic regression, ROC curve | NA | 42-61 | 63-100 | 0.66-0.7 | |
Senden et al. 2012 [24] | Tinetti <=24 | F (n = 50, 79 ± 6) NF (n = 50, 74 ± 5) | A Tri-axial Accelerometer | Sacrum | Straight line walking | spatial-temporal gait parameters, step time symmetry, harmonic ratio, inter-stride variability, RMS acceleration | Linear regression, ROC curve | NA | 0.67-0.85 | |||
Doheny et al. 2013 [25] | Fall history, past 1-year (> = 2, or one fall requiring medical attention) | F (n = 19, 74.9 ± 7.0) NF (n = 20, 68.4 ± 6.2) | Two Tri-axial Inertial sensors (accelerometer/gyroscope) | Sternum, Thigh | STS5 | Total Time, Sub-phase time, Spectral Edge Frequency, postural sway (RMS acceleration), | Logistic regression | Leave-one-out cross validation | 74.4 | 80 | 68.7 | 0.70 |
Doi et al. 2013 [26] | 1 year prospective fall occurrence (> = 1) | F (n = 16, 84.8 ± 5.9) NF (n = 57, 79.7 ± 8.2) | Two Tri-axial Accelerometer | Upper/lower trunk | Straight line walking | Harmonic Ratio | Logistic regression, ROC curve | NA | 84.2 | 68.8 | 0.81 | |
Riva et al. 2013 [28] | Fall history (> = 1) in past 1-year | F (n = 44, 63.3 ± 6.4) NF (n = 90, 62.0 ± 6.1) | Tri-axial Accelerometer | Lower back | Treadmill walking | Harmonic Ratio, Index of harmonicity, Multiscale Entropy, Recurrence quantification analysis parameters | Logistic regression | NA | 71-72.5 | 96.6 | 16.7-21.4 | |
Nishiguchi et al. 2013 [27] | Fall history (> = 1) in past 1-year | F (n = 41, 75.4 ± 4.6) NF (n = 111, 73.5 ± 4.6) | Laser Range Finder | NA | Choice Stepping Test | Step reaction time, error rate, stepping –response score | Logistic regression, ROC curve | NA | 69.7 | 73.0 | 0.73 | |
Colagiorgio et al. 2014 [29] | Combination of (Tinetti + BBS + BESTest) < 29 / 33 | OA (n = 66, 76 ± 10, F/NF = 22/44) YA (n = 13, 26 ± 5) | Microsoft Kinect | NA | Standing balance(EO,EC, Nudged on firm surface or foam surface), Reaching forward, Stand-to -Sit, Sit-to- Stand, AST | 80 features (COM postural sway, Chest Pitch Angle, velocity of transition, velocity of stepping) | Majority Classifier, Decision Tree, SVM,KNN, NB | .632 bootstrap technique | 47.9-84.3 | 47.8-91.3 | 47.7-83.1 | |
Simila et al. 2014 [31] | BBS < =49 | OA (n = 20, 76.8 ± 5.6) YA (n = 19, 27.5 ± 4.4) NP (n = 15, 55.2 ± 7.3) | Tri-axial Accelerometer | Lower back | BBS, straight line walking | Resultant acceleration in each task, gait pattern as measured by averaged acceleration in each step | KNN, ROC curve | NA | 60.8-87.2 | 62-96.6 | 42.1-89.5 | 0.66-0.89 |
Kargar et al. 2014 [30] | Physician examination | OA (n = 12, 65 -90; F/NF = 7/5) | Microsoft Kinect | NA | TUG | Number of steps of TUG, step time, turn duration | SVM | Leave-one-out cross validation | 67.4 | 67.5 | 67.3 | |
Kwok et al. 2015 [32] | 1 year prospective fall occurrence (> = 1) | F (n = 18, 70.7 ± 5.2) NF (n = 55, 69.7 ± 7.8) | Wii balance board | NA | Standing balance (EO) | Mean sway velocity | Logistic regression, ROC curve | NA | 0.67-0.71 | |||
Howcroft et al. 2016 [10] | Fall history (> = 1) in past 6-month | F (n = 24, 76.3 ± 7.0) NF (n = 76, 75.2 ± 6.6) | Pressure sensing insole, Tri-axial Accelerometers | Head, Pelvis, Shank, Shoe | Single/Dual task straight line walking | COP path parameters, temporal gait parameters, Harmonic Ratio, Maximum Lyapunov exponent(local dynamic stability) | MLP, NB, SVM | Hold out method (75% training set, 25% test set) | 72-84 | 73.7-100 | 33.3-100 | |
Howcroft et al. 2017 [33] | 6- month prospective fall occurrence (> = 1) | F (n = 28, 75.0 ± 8.2) NF (n = 47, 75.3 ± 5.5) | Pressure sensing insole, Tri-axial Accelerometers | Head, Pelvis, Shank, Shoe | Single/Dual task straight line walking | COP path parameters, temporal gait parameters, Harmonic Ratio, Maximum Lyapunov exponent(local dynamic stability) | MLP, NB, SVM | Hold out method (75% training set, 25% test set) | 49.2-56.5 | 52.7-66.6 | 27.0-46.3 |