Journal of Biomechanics 48 (2015) 911–920 Contents lists available at ScienceDirect Journal of Biomechanics journal homepage: www.elsevier.com/locate/jbiomech www.JBiomech.com Kinematic analysis of video-captured falls experienced by older adults in long-term care W.J Choi a,n, J.M Wakeling b, S.N Robinovitch b,c a Department of Physical Therapy, Chapman University, 9401 Jeronimo Rd, Irvine, CA 92618, USA Department of Biomedical Physiology and Kinesiology, Simon Fraser University, Burnaby, BC, Canada c School of Engineering Science, Simon Fraser University, Burnaby, BC, Canada b art ic l e i nf o a b s t r a c t Article history: Accepted 15 February 2015 Falls cause 95% of hip and wrist fractures and 60% of head injuries in older adults Risk for such injuries depends in part on velocity at contact, and the time available during the fall to generate protective responses However, we have no information on the impact velocities and durations of falls in older adults We addressed this barrier through kinematic analysis of 25 real-life falls (experienced by 23 individuals of mean age 80 years (SD¼ 9.8)) captured on video in two long-term facilities All 25 falls involved impact to the pelvis, 12 involved head impact, and 21 involved hand impact We determined time-varying positions by digitizing each video, using direct linear transformations calibrated for each fall, and impact velocities through differentiation The vertical impact velocity averaged 2.14 m/s (SD¼ 0.63) for the pelvis, 2.91 m/s (SD¼ 0.86) for the head, and 2.87 m/s (SD¼ 1.60) for the hand These values are 38%, 28%, and 4% lower, respectively, than predictions from an inverted pendulum model Furthermore, the average pelvis impact velocity was 16% lower than values reported previously for young individuals in laboratory falling experiments The average fall duration was 1271 ms (SD¼ 648) from the initiation of imbalance to pelvis impact, and 583 ms (SD¼255) from the start of descent to pelvis impact These first measures of the kinematics of falls in older adults can inform the design and testing of fall injury prevention interventions (e.g., hip protectors, helmets, and flooring) & 2015 Elsevier Ltd All rights reserved Keywords: Falls Biomechanics Older adults Hip fracture Impact velocity Injury Kinematic analysis Video Introduction Falls are the number one cause of injuries in older adults, including at least 90% of hip fractures and wrist fractures, and 60% of head injuries (Grisso et al., 1990; Harvey and Close, 2012; Palvanen et al., 2000) The risk for injury during a fall depends in part on the velocity at contact (or “impact velocity”) of the impacting body parts (Majumder et al., 2008; Robinovitch et al., 1991) Accordingly, impact velocity is a key input parameter for biomechanical testing of fall injury prevention technology (e.g., hip protectors (Mills, 1996; Minns et al., 2004a; Robinovitch et al., 2009), helmets (ASTM, 2007), and compliant flooring (Knoefel et al., 2013; Laing and Robinovitch, 2009; Minns et al., 2004b)) Risk for injury during a fall may also depend on the time duration of the fall, which governs the faller's ability to initiate and execute protective responses, such as arresting the fall with the upper limbs (DeGoede et al., 2001; Robinovitch et al., 2005) n Corresponding author E-mail address: wchoi@chapman.edu (W.J Choi) http://dx.doi.org/10.1016/j.jbiomech.2015.02.025 0021-9290/& 2015 Elsevier Ltd All rights reserved Our current knowledge of the impact velocities and durations associated with falls is limited to the results of laboratory studies with young adults falling (from standing height) onto gym mats (Feldman and Robinovitch, 2007; Hsiao and Robinovitch, 1998; Robinovitch et al., 2003) However, the kinematic patterns observed in lab-based falling experiments with young adults may differ substantially from real-life falls in older adults, due to differences in the situational and environmental context of falls, or age-related changes in physiological factors such as physical and cognitive function, medication use, and disease On the one hand, real-life falls in older adults may generate impact velocities that are higher than those observed in laboratory-based falls in healthy young adults, due to age-related declines in sensorimotor and cognitive function, and a corresponding absence or declines in balance recovery responses (e.g., stepping or grasping) and fall protective responses (e.g., upper limb fall arrest) On the other hand, the perturbation conditions associated with real-life falls, which tend to be caused by internal versus external perturbations (Robinovitch et al., 2013), may cause them to less severe than laboratory-based falls which have used large external perturbations to overcome participants' ability to recover balance 912 W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 Our purpose in the current study was to document the impact velocities of key body sites (hip, head and hand) during real-life falls in older adults We collected and conducted kinematic analysis of video footage of real-life falls experienced by older adults in longterm care (LTC) facilities We then determined the impact velocities of the hip, head and hand, and the time duration of the falls We compare our results to previous studies with young adults, and to theoretical predictions from simple mathematical models 2.2 Fall duration The questionnaire required the team to estimate the exact video frames corresponding to the onset of imbalance leading to the fall, the initiation of the descent stage of the fall (defined as one video frame after the foot contacted the ground in the last recovery step, if any), and the first occurrence of impact to the hand (s), head and pelvis We report two estimates of fall duration before impact to the body parts: the “total fall duration”, defined by the interval between the onset of imbalance and initial impact to the body part, and the “descent duration,” defined as the interval between the onset of fall initiation and impact Methods 2.3 Impact velocities 2.1 Real-life fall library To estimate impact velocities, we manually digitized landmark of the pelvis (anterior superior iliac spine), head (ear or forehead) and hand (palm), using a Matlab routine developed by Hedrick (2008) We digitized each frame over the interval starting one frame before fall initiation and ending one frame after impact of the corresponding body part (Fig 1(a)) We then applied a two-dimensional direct linear transform (2D DLT) to reconstruct those points as position coordinates in the object space (Hedrick, 2008; Meershoek, 1977) Finally, we used finite difference to estimate time-varying vertical and horizontal velocities The resulting velocity—time traces were fit with a fifth-order polynomial (Fig 2) using Matlab's polyfit function, the approach used in falling experiments by van den Kroonenberg et al (1996) to fit vertical displacement versus time curves for the hip during falls from standing The vertical impact velocity was estimated as the maximum value of the curve fit, based on previous observations (and theoretical considerations) that the peak downward velocity occurs very near to the instant of contact (Feldman and Robinovitch, 2007; Hsiao and Robinovitch, 1998) We also report values of the peak horizontal velocity and the magnitude of horizontal velocity at the instant of peak vertical velocity While our 2D DLT procedure corrected for lens distortion, an important limitation of the technique is the potential for “perspective errors” arising when the digitized points of interest move outside the calibrated image plane In an attempt to minimize these errors, we determined DLT calibration coefficients specific to each fall video, by visiting the site of each fall, and recording images of a flat calibration panel from the surveillance camera that captured the fall This study builds on recent descriptive reports by our team on the circumstances of falls captured on video cameras in two long-term care (LTC) facilities (Robinovitch et al., 2013; Yang et al., 2013) Between April 2007 and February 2013, we partnered with two LTC facilities in the Vancouver area (Delta View, a 312-bed facility in Delta, BC, and New Vista, a 236-bed facility in Burnaby, BC) to capture video footage of real-life falls in older adults (Robinovitch et al., 2013) Delta View had a network of 216 digital cameras, while New Vista had 48 Cameras were located in common areas (dining rooms, hallways, and lounges) and not bedrooms or bathrooms Each video was recorded at 30 frames per second and a resolution of 640 by 480 pixels or 720 by 480 pixels (Fig 1) The study was approved by the Office of Research Ethics at SFU Each resident provided written consent to the facilities for video capture of their images, and these data were shared as secondary data with the research team Additional written consent was secured from individuals to share their images Each fall video was initially analyzed by a team of three experts (research assistants and graduate students trained by co-author SNR) using a structured, validated questionnaire (Yang et al., 2013) The questionnaire probed the cause of fall, the activity at the time of the fall, the initial direction of the fall (forward, backward, sideways, or straight down), the landing configuration (forward, backward, or sideways), the occurrence (if any) of stepping responses, and the occurrence of impact to the hand(s), knee(s), head and pelvis Fig Video snapshots showing: (a) forward fall by older adult in long-term care (LTC); (b) forward fall by young adult in the laboratory environment; (c) backward fall by older adult in LTC; and (d) backward fall in young adult in the lab The far-right panel illustrates the 25-marker calibration panel, placed at the exact location of the fall (in the lab or LTC facility), and oriented in the plane of the fall The white dots in Fig 1(a) indicate digitized pelvis (ASIS) landmarks W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 913 Fig Traces from typical trials of the velocity of the pelvis versus time for: (a) forward fall by an older adult in long-term care (LTC); (b) backward fall by an older adult in LTC; (c) forward fall by a young adult in the lab; and (d) backward fall by a young adult in the lab Vertical velocities are shown in dashed lines/ circles, and horizontal velocities in solid lines/squares (see far-right images in Fig 1) The panel had dimensions 160 cm  160 cm, and contained a  grid of 10 cm diameter circular markers spaced 40 cm apart The panel was placed with the bottom surface flush to the ground, centered at the midpoint of the faller's feet (at the moment of fall initiation), and oriented in the plane of the fall (defined by the line connecting the mid-point of the feet and the location of the head at the moment of pelvis impact) 2.4 Laboratory measures of accuracy We tested the accuracy of our velocity estimates through laboratory falls with an inverted pendulum and a human participant Each trial was captured with an 8-camera motion capture system recording at 250 Hz (Motion Analysis Corp., Santa Rosa, CA, USA), and a single surveillance camera identical to the type used in our partnering LTC sites recording at 30 Hz (model WVC210, Cisco Systems, San Jose, CA) The surveillance camera was placed at a height of 2.6 m and horizontal distance of m from the site of the fall, which was typical of the falls we captured in the two long term care facilities (although there was variability in this distance for the real-life falls) The pendulum consisted of a 1.57 m long aluminum rod of uniform mass distribution, connected to a low-friction hinge joint at the floor Reflective markers were placed on the midpoint (representing the pelvis) and top end (head) of the rod The pendulum was released from vertical and descended over a 901 arc before impacting the ground Human falls were conducted for three falling directions: forward, backward, and sideways (Fig 1) In all trials, participants self-initiated the fall, and were instructed to fall naturally Reflective markers were placed on the anterior superior iliac spines, greater trochanters, sacrum, wrists, elbows, shoulders and forehead Trials were conducted at five falling angles with respect to the axis of the surveillance camera: 301 (nearly toward the camera), 601, 901 (perpendicular to the camera axis, providing a sideways view), 1201, and 1501 (nearly away from the camera) A single trial was acquired for each camera angle, in (for human falls) each of the three fall directions standard deviation in the offset error across the five camera angles We regarded the technique as acceptable for a given fall direction if the standard deviation in the offset error was 0.35 m/s or less, which would reflect a 95% confidence interval less than or equal to 0.7 m/s in the estimated impact velocity We regarded the mean offset error (for a given direction) as a fixed bias, and subtracted direction-specific mean errors from peak velocities estimated from video analysis We also excluded falls with movements that were clearly different than those involved in our calibration trials These included falls with significant rotation during descent, falls directly toward or away from the camera, falls directed straight down, and falls involving impact to objects other than the floor (e.g., walls or furniture) We also excluded falls not involving pelvis impact, or where the pelvis was occluded from camera view during descent or impact, since our primary objective at the onset of the study was characterizing the severity of impact to the pelvis during falls 2.6 Analysis of video-captured falls in older adults We provide descriptive results (means and standard deviations) for the vertical and horizontal impact velocity of the pelvis, head and hand We focus our attention more on vertical than horizontal velocity, as the stronger indicator of risk for serious injury (van den Kroonenberg et al., 1996) We also report total fall durations and descent durations for each body part We compare measured vertical velocities toffi theoretical pffiffiffiffiffiffiffiffi estimates based on free fall of a falling mass (where impact velocity¼ 2gh, where g is gravitational acceleration of 9.81 m/s , and h is the vertical descent distance of the body part), andpan inverted pendulum with uniformly-distributed mass (where impact ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi velocity¼ ð3=2Þgh) We also conducted regression analysis to test whether impact velocity is associated with fall height Results 2.5 Inclusion criteria for video-captured falls in older adults 3.1 Laboratory falls We used the results from our laboratory-based human falls to guide the selection of video-captured falls in older adults for analysis, based on fall direction In particular, for each trial, we calculated the offset error, defined as the difference between the impact velocity estimate from 3D motion capture and the 2D single video camera images We then determined, for a given fall direction, the mean and In our laboratory experiments with the inverted pendulum (Fig 3), the average difference (across the five falling angles) between the 2D DLT and 3D motion capture techniques was 0.008 m/s (SD¼0.04) for the vertical impact velocity, and À 0.11 m/s (SD¼0.07) for the 914 W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 Fig Agreement from laboratory experiments between the 2D DLT technique and 3D motion capture for vertical and horizontal impact velocities Results are shown for the mid-point of a 1.57 m length pendulum, and for the pelvis for a human participant falling in the forward, backward, and sideways directions (at five different camera angles; see text for explanation) Sideways falls in human exhibited larger variability between camera angles (vertical velocity SD ¼ 0.59 m/s, horizontal velocity SD ¼0.54 m/s) than forward and backward falls The individual plots clearly show that SD is less than 0.7 m/s in all but the sideways direction W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 horizontal impact velocity (where a negative mean reflects overestimation from the 2D DLT technique) In laboratory-based human falls (Fig 3), the DLT technique met our criteria for acceptable accuracy for forward and backward falls, but not sideways falls The mean difference in vertical impact velocity between the 2D DLT and 3D motion capture techniques was À 0.26 m/s (SD¼ 0.21) for forward falls, À0.06 m/s (SD¼0.21) for backward falls, and À 1.01 m/s (SD was 0.59) for sideways falls The mean difference in horizontal impact velocity was À 0.15 m/s (SD¼0.32) for forward falls, À 0.16 m/s (SD¼ 0.18) for backward falls, and À 0.13 m/s (SD¼0.54) m/s for sideways falls For both human and inverted pendulum falls, the accuracy was not influenced by the falling angle with respect to the camera axis (p40.5 by linear regression) For human falls, the mean error was not different (p40.1) between forward and backward falls, and was equal to À0.16 m/s for both vertical and horizontal velocity The pooled SD for forward and backward falls was 0.23 m/s for vertical velocity and 0.24 m/s for horizontal velocity 3.2 Falls by older adults Between April 2007 and February 2013, we captured 813 falls experienced by 306 individuals (Fig 4) Based on our acceptance criteria and the results from our laboratory-based falling experiments, we excluded cases where the initial fall direction was sideways (n¼ 152) We excluded an additional 636 cases based on other exclusion criteria, as described in Fig Our final analysis included 25 falls (23 from Delta View, from New Vista) experienced by 23 older adults (Tables and 2) There were 21 backward falls and forward falls, all of which involved pelvis impact The average age of the faller was 80.3 yrs (SD¼9.8), and 61% (n¼ 14) were female The most common cause of imbalance 915 was incorrect weight shifting (12 of 25 cases), followed by hit/bump (7 cases) Trips, collapses, and loss-of-support each accounted for falls The most common activity at the time of falling was standing (14 of 25 cases), followed by walking (8 cases), and transferring from standing to sitting (3 cases) Two falls occurred while using a walker (video IDs 23 and 24) Stepping after the onset of imbalance occurred in 16 of 25 falls Head impact occurred in 48% of cases (n ¼12; all forward falls, and of 21 backward falls), and hand impact occurred in 84% (n ¼21; all forward falls, and 17 of 21 backward falls) In 76% of cases involving hand impact (n ¼ 16), the hand impacted before the pelvis or head All four forward falls involved impact to the knee (s) before the pelvis Table reports estimated impact velocities for each fall, after subtracting direction-specific mean offset errors (for vertical velocity: 0.26 m/s for forward falls and 0.06 m/s backward falls; for horizontal velocity: 0.15 m/s for forward falls and 0.16 m/s backward falls) Over the 25 falls by older adults, the vertical impact velocity averaged 2.14 m/s (SD¼ 0.63) for the pelvis, 2.91 m/s (SD¼ 0.86) for the head, and 2.87 m/s (SD¼1.60) for the hand (Table and Figs and 6) For eight backward falls involving impact to both the pelvis and head, the vertical impact velocity was 2.67 (SD¼0.82) for the head and 1.98 (SD¼0.45) m/s for the pelvis The horizontal impact velocity averaged 1.16 m/s (SD¼1.42) for the pelvis, 2.64 m/s (SD¼1.12) for the head, and 1.52 m/s (SD¼1.14) for the hand The total fall duration averaged 1271 ms (SD¼648) for the pelvis, 1730 ms (SD¼805) for the head, and 1188 ms (SD¼702) for the hand The descent duration averaged 593 ms (SD¼255) for the pelvis, 757 ms (SD¼217) for the head, and 479 ms (SD¼230) for the hand The vertical impact velocity of the pelvis averaged 2.19 m/s (SD ¼0.61) in trials where the hand(s) impacted before the pelvis, compared to 2.41 m/s (SD ¼ 0.85) in falls not involving hand Fig Fall video selection process Among 813 fall videos captured, 25 forward and backward falls were selected for analysis 916 Table Participant characteristics and descriptive data for 25 falls from 23 older adults Faller ID Age Sex Body mass (kg) 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 10 11 12 13 14 15 16 17 18 18 19 20 21 22 23 84 93 71 70 69 64 90 82 84 63 84 93 87 72 69 88 86 74 75 75 86 100 79 91 79 F F M M F M M F M M M F F F F F M F M M M F F F F Parkinson's disease Alzheimer's disease Stroke Hypertension COPD Diabetes Fall direction √ √ √ √ √ 57.6 58.1 72.5 √ √ √ 61.1 √ √ 85.2 58.1 68.4 50.9 79.1 65.7 88.9 68 45.4 √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ F F F F B B B B B B B B B B B B B B B B B B B B B Pelvis impact Head impact Hand impact Knee impact Cause of fall Activity at the time of fall Stepping response √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ T T B IT B B LOS B IT IT IT IT C IT B LOS IT IT C IT IT IT IT B B W W W W S S S S S T T S S W S T W W S S W S S S S No Yes Yes No No Yes No Yes Yes No No Yes Yes Yes No Yes Yes Yes No Yes No Yes Yes Yes Yes √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ √ Fall direction: F ¼forward, B ¼backward; Cause of fall: T ¼ trip/stumble, B¼ hit/bump, IT ¼incorrect transfer, LOS¼loss of support, C¼ leg collapse; Activity at the time of fall: W ¼ walking, S¼standing, T¼ transferring; data; COPD ¼ chronic obstructive pulmonary disease Injury noted Head Head Head Head Head ¼missing W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 Video ID Table Impact velocities, fall durations, and fall heights for 25 falls from 23 older adults Video ID Faller ID Head impact Peak horizontal velocity (m/s) Total fall Horizontal duration velocity at peak vertical (ms) velocity (m/s) Descent duration (ms) Fall height (cm) 1.68 1.29 1.70 1.98 7.50 1.68 4.32 1.39 2.34 1.42 3.63 1.39 1370 1290 3067 1097 1370 645 533 1097 125.3 99.6 72.7 99.8 10 11 12 13 14 15 16 17 18 18 19 20 21 22 23 1.18 1.74 3.51 3.37 2.13 2.46 2.02 3.03 1.68 1.71 1.75 1.29 2.62 1.80 3.19 2.06 2.05 2.50 1.89 2.29 2.58 2.14 0.63 0.49 1.99 7.23 2.18 0.63 0.02 1.10 0.77 0.65 1.10 0.16 0.84 1.31 0.95 0.17 0.09 1.60 0.64 0.82 1.15 0.96 1.59 1.95 0.38 0.07 6.72 2.12 0.61 0.09 1.09 0.40 0.49 1.10 0.14 0.79 0.86 0.95 0.07 0.08 1.13 0.22 0.82 1.07 0.96 1.16 1.42 500 1125 848 1074 3063 1000 467 1233 1067 1333 600 621 1677 1400 933 1167 800 1833 1433 1033 1733 1271 648 500 625 848 815 500 571 467 400 167 367 600 621 323 533 367 767 800 400 467 567 467 593 255 55.1 78.7 103.6 106.2 72.3 94.4 91.0 83.3 57.2 63.9 84.2 66.0 79.3 85.3 77.7 74.3 88.7 62.0 67.4 84.7 65.2 81.5 17.0 Peak vertical velocity (m/s) Peak horizontal velocity (m/s) Hand impact Total fall Horizontal duration velocity at peak vertical (ms) velocity (m/s) Descent duration (ms) Fall height (cm) Peak vertical velocity (m/s) Peak horizontal velocity (m/s) Total fall Horizontal duration velocity at peak vertical (ms) velocity (m/s) Fall height (cm) a a a a a a a a a a a 3.52 2.80 4.25 2.68 4.36 1.87 2.68 3.86 1.73 1194 3067 1129 548 533 1129 84.9 100.7 164.0 2.47 3.00 2.84 3.95 4.42 2.20 3.95 2.89 2.18 774 2833 968 129 300 968 28.1 67.6 101.0 3.29 3.49 2.57 1125 625 96.4 7.38 1.78 4.25 2.24 1.19 3.20 2.24 0.77 0.57 406 1000 636 406 500 636 34.5 65.7 134.0 2.12 3.14 2.89 3313 750 86.0 1.50 1.35 1.32 3063 500 66.7 2.67 2.00 2.30 1.85 4.47 1.83 1.77 1.29 1.30 1.37 1.64 1.22 0.60 1.19 0.52 400 1167 1033 1267 667 400 333 133 300 667 51.8 59.9 54.4 45.6 107.7 2.12 2.07 4.31 3.04 1.25 2.32 4.20 4.28 0.31 2.11 3.80 4.11 1133 862 1871 1633 1133 862 516 767 109.4 95.7 139.1 129.1 1.96 2.49 1.84 1767 800 110.7 2.48 2.91 0.86 3.12 3.02 1.02 3.10 2.64 1.12 1933 1730 805 667 757 217 90.5 109.7 24.9 a a a a a a 2.28 5.56 1.03 3.83 3.22 0.33 0.10 3.50 3.22 0.33 0.48 3.50 1333 900 1267 833 467 333 867 833 89.3 80.3 61.5 65.4 1.69 1.81 1.76 2.87 1.60 0.76 2.03 1.25 1.96 1.20 0.67 0.75 0.88 1.52 1.14 1433 967 1633 1188 702 467 500 367 479 230 57.8 40.6 55.1 66.7 26.2 Total fall duration ¼time interval between the moment of imbalance and impact; Descent duration ¼time interval between fall initiation and impact; Fall height¼ vertical descent distance from fall initiation to impact; a Descent duration (ms) ¼data not available W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 Peak vertical velocity (m/s) Forward falls Backward falls 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Mean Standard deviation Pelvis impact Not able to digitize 917 918 W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 Fig Mean values from the 25 falls in older adults (with standard deviations shown as error bars) of: (a) vertical impact velocity; (b) horizontal impact velocity; (c) total fall duration; and (d) descent duration In each case, values of shown for each of the pelvis, head and hand impact, and 2.19 m/s (SD ¼0.58) in trials where steps occurred after imbalance versus 2.31 m/s (SD ¼0.81) in falls not involving steps (Table 2) When compared to theoretical predictions based on free-fall from each measured fall height (Table and Fig 7), our vertical pelvis impact velocities averaged 46.0% (SD¼14.95) lower than predictions from a falling mass model, and 38.0% (SD¼17.3) lower than predictions from an inverted pendulum model Similarly, our vertical head and hand impact velocities averaged 37.4% (SD¼15.3) and16.9% (SD¼ 55.4) lower than free-fall predictions, and 27.7% (SD¼17.7) and 4.0% (SD¼ 63.9) lower than pendulum fall predictions, respectively Furthermore, from regression analysis (SPSS, Version 18.0), we found that fall height associated with the vertical impact velocity of the head (v¼0.022nhþ0.5, R2 ¼0.403, p¼ 0.036), but not of the pelvis (p¼0.19) or hand (p¼0.41) Discussion This is the first study to our knowledge to report impact velocities and fall duration from real-life falls in older adults Our results provide important baseline measures of fall severity for the design and assessment (through mechanical testing systems or mathematical models) of interventions for fall injury prevention, including wearable hip protectors, helmets and compliant flooring Impact velocity is a measure of fall severity that is important for the design and testing of injury prevention strategies Our measured vertical impact velocities for older adults averaged 16% lower than the mean value for the pelvis (of 2.55 m/s (SD¼ 0.85)) and 9% greater than the mean value for the wrist (2.64 m/s (SD¼0.66)) reported by Hsiao and Robinovitch (1998) from a laboratory study with young adults falling on gym mats after receiving a sudden (slip) perturbation This previous study included 20 backward and 11 sideways falls in the analysis, and only reported average values for all trials, without separating the results by fall direction (Hsiao and Robinovitch, 1998) Furthermore, our vertical impact velocities averaged 38% lower for the pelvis, and 28% lower for the head, than theoretical predictions from an inverted pendulum model, based on fall height Moreover, the fall height associated with vertical impact velocity of the head, but not of the pelvis or hand These results suggest that, in the falls we analyzed, older adults utilized mechanisms to absorb energy during descent, and reduce their impact velocity (and risk for injury) These included attempts to recover balance by stepping (which occurred in 64% of falls), and impacting the ground with the hands before the pelvis or head (which occurred in 84% of falls) In previous falling experiments with young adults, pelvis impact velocities were decreased 22% by taking a step after imbalance, and 18% by impacting the hands before the pelvis (Feldman and Robinovitch, 2007) While our small sample precluded meaningful statistical analysis, our trends agree with these findings Pelvis impact velocities averaged 9% lower in falls involving hand impact (compared to no hand impact), and 5% lower in falls involving stepping (compared to no stepping) Additional mechanisms beyond the scope of this study may have contributed to velocity reduction, including squatting during descent and contacting the pelvis with the trunk relatively upright (Robinovitch et al., 2004) The fall duration is of interest since it reflects the time available for the faller to initiate and execute protective responses to avoid injury during landing Our study shows that a duration of nearly 1200 ms is available to initiate protective responses between the moment of imbalance, and the instant of impact to the pelvis (which occurred on average at 1271 ms) or hand (which occurred on average at 1188 ms) This is considerably longer than the fall durations reported by Hsiao and Robinovitch (1998) in their laboratory falls due to sudden slip perturbations, where the average interval between the onset of the perturbation and pelvic contact was 715 ms (SD¼160), while wrist contact was 680 ms (SD¼116) This reflects that real-life falls in older adults occur over a considerably slower time interval than falls W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 919 Fig Sample traces of the vertical and horizontal velocity of the pelvis, head, and hand versus time for (a) forward fall and (b) backward fall by older adults in long-term care Both falls resulted in impact to the hand, pelvis, and head In the forward fall, impact occurred first to the hand This seemed to reduce the subsequent vertical velocity of the pelvis and head, which impacted the ground near-simultaneously During the backward fall, impact first occurred near-simultaneously to the pelvis and hand This seemed to reduce the vertical velocity of the head, before a final rapid increase as the trunk rotated downward recorded in laboratory experiments with young adults, where a rather severe, sudden perturbation is necessary to overcome balance recovery responses Another study (Robinovitch et al., 2005) found that the time required for older adults to move their hands into a protective position to arrest a fall averages 615 ms (SD¼ 88) This is well below our average value of total fall duration but similar to our mean descent duration In our study, upper limb protective responses were typically initiated soon after the onset of imbalance (Fig 6), which likely contributed to the observation of hand impact in 84% of falls There are important limitations to our study Our results are based on falls experienced by residents in LTC, and may not apply to healthier community dwelling older adults, or young adults Our small sample size prevented us from examining how falls associate with intrinsic factors such as physical and cognitive function or medications Larger studies are required to relate the kinematics of falls to the clinical context We only included falls that involved pelvis impact, leading us to exclude many forward falls Our video footage was collected at 30 fps, and therefore our resolution in detecting fall initiation and impact times was limited to the duration of one frame of the video (33 ms), or about one-fifteenth (7%) of the shortest descent duration we report (479 ms for the hand) We calculated velocities after fitting displacement versus time traces with a fifthorder polynomial, which may have resulted in filtering or loss of valid kinematic information However, our approach was similar to that used in previous video-based laboratory measures of fall impact velocities in humans (van den Kroonenberg et al., 1996) Furthermore, we found that a fifth-order polynomial provided the best agreement to velocity estimates from 3D motion capture (recording at 250 Hz) in our inverted pendulum calibration tests Furthermore, analyzing the complex movements of falls from planar video is challenging, due to the out-of-plane motions of the body segments that often accompany during descent In our laboratory falls, we found acceptable accuracy in our velocity estimates for forward and backward falls, where the trajectory of body parts tended to remain parallel to the calibration plane However, measurement accuracy was unacceptable for sideways falls (where knee and trunk flexion often caused out-of-plane movement of the pelvis), which were excluded from analysis The exclusion of sideways falls was unfortunate given that hip fractures are most likely to occur from sideways falls (Greenspan et al., 1994; Nevitt and Cummings, 1993) Vertical impact velocities for the pelvis averaged 2.23 for backward falls and 1.66 m/s for forward falls, and one might expect similar values for sideways falls Future studies might test this hypothesis by capturing sideways falls with 3D cameras, or with multiple cameras and 3D analysis techniques In summary, based on analysis of 25 video-captured falls experienced by 23 older adults in long-term care, we found that the vertical impact velocity averaged 2.14 m/s for the pelvis, and 2.91 m/s for the head These values are 38% and 28% lower, respectively, than theoretical predictions from an inverted pendulum model based on fall height Furthermore, the average vertical impact velocity of the pelvis was 16% lower than values reported for young individuals in laboratory falling experiments The duration of the fall averaged 1271 ms from the moment of imbalance, and 583 ms from the start of descent, to the instant of pelvis impact These first measures of the kinematic 920 W.J Choi et al / Journal of Biomechanics 48 (2015) 911–920 influence this work, including employment, consultancies, stock ownership, honoraria, paid expert testimony, patent applications/ registrations, and grants or other funding Acknowledgments This research was supported by the operating grants from the Canadian Institutes for Health Research (funding reference nos AMG-100487 and TIR-103945) References Fig Comparison for the 25 falls by older adults between measured and model predictions of vertical impact velocities (based on fall height) for (a) pelvis, (b) head, and (c) hand On average, the measured vertical impact velocities for the pelvis were 48.0% (SD ¼ 14.2) lower than free-fall model predictions and 40.0% (SD ¼16.4) lower than inverted pendulum predictions The vertical impact velocities for the head were 38.4% (SD¼ 16.5) and 28.8% (SD¼ 19.0) lower than free-fall and inverted pendulum model predictions, respectively, and the vertical hand impact velocities of the hand were 18.8% (SD ¼ 55.3) and 6.3% (SD¼ 63.9) lower than free-fall and inverted pendulum model predictions, respectively profiles of real-life falls in older adults should inform the development and testing of fall prevention technology, including wearable hip protectors, helmets, and compliant flooring, and contribute to the design of exercise programs to train fall protective responses Conflict of interest statement None of the authors above have any financial or personal relationships with other people or organizations that could inappropriately ASTM A.S.f.T.a.M., 2007 Standard Performance Specification for Ice Hockey Helmets DeGoede, K.M., Ashton-Miller, J.A., Liao, J.M., Alexander, N.B., 2001 How quickly can healthy adults move their hands to intercept an approaching object? Age and gender effects J Gerontol Ser A Biol Sci Med Sci 56, 584–588 Feldman, F., Robinovitch, S.N., 2007 Reducing hip fracture risk during sideways falls: evidence in young adults of the protective effects of impact to the hands and stepping J Biomech 40, 2612–2618 Greenspan, S.L., Myers, E.R., Maitland, L.A., Resnick, N.M., Hayes, W.C., 1994 Fall severity and bone mineral density as risk factors for hip fracture in ambulatory elderly J Am Med Assoc 271, 128–133 Grisso, J.A., Schwarz, D.F., Wishner, A.R., Weene, B., Holmes, J.H., Sutton, R.L., 1990 Injuries in an elderly inner-city population J Am Geriatr Soc 38, 1326–1331 Harvey, L.A., Close, J.C., 2012 Traumatic brain injury in older adults: characteristics, causes and consequences Injury 43, 1821–1826 Hedrick, T.L., 2008 Software techniques for two- and three-dimensional kinematic measurements of biological and biomimetic systems Bioinspir Biomim 3, 034001 Hsiao, E.T., Robinovitch, S.N., 1998 Common protective movements govern unexpected falls from standing height J Biomech 31, 1–9 Knoefel, F., Patrick, L., Taylor, J., Goubran, R., 2013 Dual-stiffness flooring: can it reduce fracture rates associated with falls? J Am Med Dir Assoc 14, 303–305 Laing, A.C., Robinovitch, S.N., 2009 Low stiffness floors can attenuate fall-related femoral impact forces by up to 50% without substantially impairing balance in older women Accid Anal Prev 41, 642–650 Majumder, S., Roychowdhury, A., Pal, S., 2008 Effects of trochanteric soft tissue thickness and hip impact velocity on hip fracture in sideways fall through 3D finite element simulations J Biomech 41, 2834–2842 Meershoek, L., 1977 Matlab Routines for 2D Analysis with Non-Perpendicular Camera Angle International Society of Biomechanics Mills, N.J., 1996 The biomechanics of hip protectors Proc Inst Mech Eng Part H J Eng Med 210, 259–266 Minns, J., Dodd, C., Gardner, R., Bamford, J., Nabhani, F., 2004a Assessing the safety and effectiveness of hip protectors Nurs Stand 18, 33–38 Minns, J., Nabhani, F., Bamford, J.S., 2004b Can flooring and underlay materials reduce hip fractures in older people? Nurs Older People 16, 16–20 Nevitt, M.C., Cummings, S.R., 1993 Type of fall and risk of hip and wrist fractures: the study of osteoporotic fractures The Study of Osteoporotic Fractures Research Group J Am Geriatr Soc 41, 1226–1234 Palvanen, M., Kannus, P., Parkkari, J., Pitkajarvi, T., Pasanen, M., Vuori, I., Jarvinen, M., 2000 The injury mechanisms of osteoporotic upper extremity fractures among older adults: a controlled study of 287 consecutive patients and their 108 controls Osteoporos Int 11, 822–831 Robinovitch, S.N., Brumer, R., Maurer, J., 2004 Effect of the “squat protective response” on impact velocity during backward falls J Biomech 37, 1329–1337 Robinovitch, S.N., Evans, S.L., Minns, J., Laing, A.C., Kannus, P., Cripton, P.A., Derler, S., Birge, S.J., Plant, D., Cameron, I.D., Kiel, D.P., Howland, J., Khan, K., Lauritzen, J.B., 2009 Hip protectors: recommendations for biomechanical testing-an international consensus statement (Part I) Osteoporos Int 20, 1977–1988 Robinovitch, S.N., Feldman, F., Yang, Y., Schonnop, R., Leung, P.M., Sarraf, T., Sims-Gould, J., Loughin, M., 2013 Video capture of the circumstances of falls in elderly people residing in long-term care: an observational study Lancet 381, 47–54 Robinovitch, S.N., Hayes, W.C., McMahon, T.A., 1991 Prediction of femoral impact forces in falls on the hip J Biomech Eng 113, 366–374 Robinovitch, S.N., Inkster, L., Maurer, J., Warnick, B., 2003 Strategies for avoiding hip impact during sideways falls J Bone Miner Res 18, 1267–1273 Robinovitch, S.N., Normandin, S.C., Stotz, P., Maurer, J.D., 2005 Time requirement for young and elderly women to move into a position for breaking a fall with outstretched hands J Gerontol Ser A Biol Sci Med Sci 60, 1553–1557 van den Kroonenberg, A.J., Hayes, W.C., McMahon, T.A., 1996 Hip impact velocities and body configurations for voluntary falls from standing height J Biomech 29, 807–811 Yang, Y., Schonnop, R., Feldman, F., Robinovitch, S.N., 2013 Development and validation of a questionnaire for analyzing real-life falls in long-term care captured on video BMC Geriatr 13, 40