Five of the six transfemoral prosthesis users fell at least once in this study, while none of the seven healthy control participants fell in [14, 24]. These results substantiate the high fall prevalence relative to healthy adults reported in retrospective studies [1–4, 7, 41, 42], highlighting that stumbles to the sound limb should not be overlooked when considering interventions for fall prevention.
The following subsections provide an in-depth discussion of results with respect to the stumble conditions of (1) swing phase, (2) participant age, and (3) prosthesis type, as well as a commentary on arm motion. Finally, based on these observations, several interventions for improving recovery and mitigating falls for the transfemoral prosthesis user population are considered.
Considering swing phase
Transfemoral prosthesis users who recovered in early swing (P2, P3, P4, P6) used an elevating strategy, as HC did. This was accomplished via substantial sound-side thigh and knee flexion to complete the elevating step, and a successful initiation (thigh flexion) and completion (extended knee angle) of the next prosthetic-side step (Fig. 4); this lower-limb motion allowed prosthesis users to land with a greater step length, more anterior foot position relative to COM, less trunk flexion, and more negative trunk velocity at each of the recovery foot-strikes (Fig. 5), aligning with the work of Grabiner et al. which has shown that an improvement in these metrics (i.e., trending in the direction described) indicates a decreased fall risk [31–34, 43]. However, transfemoral prosthesis users who recovered exhibited more support limb (prosthetic-side) thigh abduction in the next step and increased trunk flexion and flexion velocity compared to healthy control data (Fig. 13). Thigh abduction is likely a compensation mechanism in order to facilitate swing phase during the non-cyclic activity of stumble recovery (see Considering Prosthesis Type subsection). Increased trunk flexion has been reported as an indicator of increased fall likelihood [31, 32] and may be due to lack of moment generation of the passive prosthetic limb joints (See Considering Age, Commentary on Arm Motion, and Interventions for Consideration subsections). Likewise, contralateral ankle plantarflexion employed by HC during the elevating step (not observed for prosthesis users) may have helped reduce forward angular momentum and thus limit trunk flexion during recovery [21, 44, 45] (see Considering Age and Interventions for Consideration subsections).
Transfemoral prosthesis users who fell in early swing (P1, P5) either inadequately performed the elevating strategy or attempted a delayed lowering strategy but could not successfully initiate a contralateral (prosthetic-side) step. In the seven-participant healthy adult study, all 86 stumbles that occurred before 40% swing phase resulted in elevating strategies (i.e., none abandoned elevating to perform the delayed lowering strategy) [14, 24]. In this study, the transfemoral prosthesis users who used the delayed lowering strategy (i.e., abandoned elevating) exhibited substantially high forward trunk flexion velocity at initial sound-side foot-strike, and only one could complete the second prosthetic-side step, suggesting that employing the delayed lowering strategy in early swing may increase the body’s forward angular momentum enough to contribute to falls. This increase in trunk flexion and forward trunk flexion velocity at recovery foot-strikes aligns with the work of Grabiner et al. [12, 31–34, 40, 43], which has used these metrics as fall indicators for other populations.
These early swing findings are inconsistent with the only other study of sound-side, early swing perturbations with transfemoral prosthesis users, in which majority lowering strategies and no falls were reported . This may be due to the ability of participants in that study to use the handrails and/or to the lack of a physical obstacle to clear in the rope-blocking apparatus.
Collectively these results suggest that properly performing the elevating strategy is key for recovering from early swing stumbles.
Transfemoral prosthesis users who recovered in late swing (P2, P3, P4, P6) used a lowering strategy, as HC did. This was accomplished via substantial thigh flexion and a safe landing configuration for the prosthetic-side step as well as substantial sound-side thigh and knee flexion in the subsequent step to clear the obstacle (Fig. 4); this lower-limb motion allowed users to land with a larger step length, more anterior foot position relative to COM, and more negative trunk flexion velocity at the prosthetic-side foot-strike and subsequent sound-side foot-strike (Fig. 8), again aligning with trends seen in [31–40, 43, 46].
However, lowering strategy recoveries involved more trunk flexion and more prosthetic-side thigh abduction compared to healthy controls (Fig. 13) likely in part due to deficiencies in stance release of their prescribed knee prosthesis (see Considering Prosthesis Type subsection).
The transfemoral prosthesis users who fell in late swing (P1, P5) all attempted a lowering strategy but could not initiate and/or complete the next (prosthetic-side) step. This was evidenced by the lack of prosthetic-side thigh flexion or a flexed knee at prosthetic-side foot-strike (Fig. 4), which contributed to the smaller step length, posterior foot position relative to COM, and more forward trunk flexion velocity at prosthetic-side foot-strike (Fig. 8), metrics that agree with previously reported fall indicators. Note that these lowering strategy falls were also characterized by a longer time to initial loading (first sound-side foot-strike), which is consistent with findings of .
These late swing findings are mostly consistent with Shirota et al. , who also observed lowering strategies, as well as with Crenshaw et al. , who observed one elevating (fall), one lowering with hopping (recovery), and one lowering (fall), and suggested that lack of knee control contributed to the difficulty of executing the recovery step with the prosthetic limb.
Collectively these results suggest that properly initiating swing after the lowering step, landing in a safe (i.e., extended knee) configuration on the prosthetic limb, and using sufficient thigh/knee flexion to clear the obstacle in the subsequent sound-side step are key for recovering from late swing stumbles.
Transfemoral prosthesis users who recovered in mid swing (P4, P6) used an elevating or lowering with hopping strategy. The trends observed in early swing elevating strategies and late swing lowering strategies extend to mid swing regarding lower-limb dynamics (Fig. 11) and foot-strike states (Fig. 12).
These mid swing findings are mostly consistent with Shirota et al.  in terms of strategy selection. However, while Shirota et al. observed no falls, this mid swing region presented the highest likelihood of falls across participants (i.e., five of six fell) in the present study, suggesting that this region may warrant particular attention when considering interventions. The lowering with hopping response was qualitatively similar to the hopping strategy discussed in previous works [29, 30]. Crenshaw et al. 2013  notes that this strategy may be difficult for individuals with plantarflexion weakness; indeed, in this study only one participant successfully recovered in this way.
Collectively these results suggest that either properly performing the elevating strategy or properly initiating prosthetic-side swing after the lowering step (for a more normative, without hopping, lowering/delayed lowering strategy) is key for recovering from mid swing stumbles.
It is notable that the two oldest participants in this case series (P1 and P5) were the only two participants to fall after every perturbation. These results are consistent with several studies of healthy (non-prosthesis user) older and younger adults, in which the older adults fell more [19, 41] which was attributed to delayed and/or diminished muscle responses [16, 19]. Pijnappels et al.  highlights that older adults have problems meeting the requirements for adequate balance recovery, since muscle strength, reaction time, and coordination decline with age. Specifically, Pijnappels et al.  reported significantly longer rise times of electromyography (EMG) amplitudes of the biceps femoris, gastrocnemius medialis, and soleus muscles in the support limb of the older adults, which reduces rate of force generation. This finding is likely compounded in the present study by the fact that in the case of elevating strategies, the support limb is the prosthetic limb, in which these three muscles have been altered or removed due to amputation/congenital limb difference (discussed more in the Commentary on Arm Motion and Interventions for Consideration subsections). Pijnappels et al. and others conclude that a combination of resistance/strength training and task-specific motor skill training has the potential to improve responses [25, 34], which are discussed in the Interventions for Consideration section.
Pavol et al. 1999  reported in their overground stumble study of older adults that the participants who took more rapid steps had significantly increased likelihood of falling after a trip. Additionally, gait asymmetry metrics have been used to determine and quantify gait pathologies . Thus the cadence and prosthetic-to-sound limb swing time symmetry were computed for each participant. P5 walked with a substantially increased cadence (i.e., more rapid steps, at 115 steps/min) compared to the remaining participants (87–97 steps/min). P1 and P5 also walked with higher swing-time asymmetry (ratio of 1.52 and 1.41, respectively) compared to the remaining participants (1.23–1.31). Therefore, age of participant may be a proxy for other step metrics that indicate higher fall risk.
Considering prosthesis type
Four of the transfemoral prosthesis user participants wore an Ottobock C-Leg (P1, P3, P5, P6), which is a hydraulic-based microprocessor-controlled knee (MPK). Two participants wore hydraulic non-MPKs: P2 wore an Ottobock 3R80, a single-axis, rotary-hydraulic knee; P4 wore a Blatchford KX06, a four-bar, hydraulic knee. All six are hydraulic knees with stance control function (i.e., high resistance against flexion during stance). Stance resistance against flexion is initiated at heel-strike in the non-MPKs, while it is initiated during swing extension in the C-Leg. The C-Leg and 3R80 are both single-axis knees, while the KX06 employs 4-bar kinematics that increase toe clearance during swing phase and also enhance stability during stance phase.
For early and late swing perturbations, there was no clear trend in prosthesis type contributing to falls versus recoveries. Recall that the two transfemoral prosthesis users who fell were also the oldest participants, which was discussed in the Considering Age subsection. However, the transfemoral prosthesis users who recovered (regardless of prosthesis type) used substantial thigh abduction during the subsequent prosthetic-side step (Fig. 13), likely to help initiate swing at a time when the prescribed device was not in the proper mode. Both non-MPKs and MPKs rely on various sensing thresholds to allow for stance release, as well as the ballistic coordination with the thigh for swing phase motion, both of which are reliable during the cyclic motion of walking; however, these prostheses do not account for the interruption of inertial swing dynamics that occurs during sound-side stumbles.
For mid swing perturbations, the participant who recovered using an elevating strategy wore a non-MPK (P4 with KX06). The participant who recovered using a lowering strategy wore an MPK (P6 with C-Leg). However, it is unclear whether the devices themselves helped facilitate recovery; especially in the case of P6, the C-Leg’s lack of stance release likely led to the hopping response after lowering that required substantial thigh abduction to facilitate swing and involved increased trunk flexion. Thus, the recoveries in mid swing are potentially attributed to the ability of the prosthesis user rather than the prosthesis itself. For the delayed lowering responses (all falls), the non-MPK users (P2 and P4) were able to initiate a subsequent step with their prosthetic side after the lowering step (Fig. 10), while the MPK users could not initiate this subsequent step. Thus, the non-MPKs may have some advantage at more quickly initiating swing; however, their capabilities were not robust enough to successfully complete the step.
There was one instance of prosthetic knee buckling, in which P5 did not take a full prosthetic-side step after his delayed lowering step and landed with the knee substantially flexed (Figs. 10 and 11).
From these observations, for sound-side stumble recovery there does not seem to be a clear advantage of knee prosthesis type. Instead, both demonstrated similar deficits: namely, the inability to initiate swing and/or complete swing during the next step. All participants wore similar energy storage-and-return type prosthetic feet, and all showed similar deficits in terms of lack of active plantarflexion and less range of motion compared to HC. The effect of prosthetic foot model was beyond the scope of this work but could be investigated in future studies.
Commentary on arm motion
Roos et al. 2008  concluded that arm movements contribute to stumble recovery for healthy adults by both elevating the body’s center of mass and reducing its forward angular momentum, which provide more time for the positioning of the recovery limb. Pijnappels et al. 2010  concluded that arm responses counteracted transverse plane body rotation which helped with recovery foot positioning. For all three strategies, but particularly with elevating, transfemoral prosthesis user recoveries were characterized by substantially more vertical, anterior, and medial/lateral deviation of the forearm from its position at perturbation (prior to returning to that position) as well as increased trunk flexion compared to HC (Figs. 3, 6, 9, 10, and 13). As discussed in the Considering Swing Phase and Interventions for Consideration subsections, Pijnappels et al.  conclude that the reactive torques of the support limb of healthy adults enable the necessary push-off reaction, thus reducing the forward angular momentum of the body and providing more time for positioning of the elevating limb. Given that the support limb was the prosthetic limb (i.e., no active ankle or knee power and compromised muscles) for elevating strategies in this study, perhaps the exaggerated arm motion observed among transfemoral prosthesis users is necessary to counter the lack of moment generation of the support limb, allowing for reduced forward angular momentum and/or better foot placement to avoid falling.
Interestingly, the transfemoral prosthesis users who fell after every perturbation were the two oldest participants, and they employed arm motion similar to those found in the older participants of Roos et al. 2008 ; described as a more “protective” strategy, these arm movements were characterized by more anterior displacement (a forward reaching-like motion that suggested bracing themselves for an expected fall), rather than the “preventive” strategy employed by younger adults to counteract loss of balance (Figs. 3, 6, 9, and 10).
Interventions for consideration
What causes falls?
In this case series, fall incidence was related to swing percentage (the mid swing stumbles resulted in the most falls) and age (the oldest prosthesis users fell after every stumble), but not with prosthesis type (MPK versus non-MPK), as discussed in the previous discussion subsections.
Falls for this population can be attributed to two main deficiencies in the transfemoral prosthesis users’ responses after a stumble. First, falls occurred if the tripped (sound) limb did not reach ample thigh and knee flexion to sufficiently clear the obstacle in the elevating step (Fig. 4 and 11A and C). The inability to perform the elevating strategy may come from a decrease in strength or reaction time of the tripped (sound) limb (see Considering Age subsection), and/or a lack of counteracting moment generation in the support (prosthetic) limb (see Considering Swing Phase and Commentary on Arm Motion subsections). Recall that even the elevating strategy recoveries involved more trunk flexion, forward trunk flexion velocity, and arm motion than healthy controls, also likely due to these two factors.
Second, falls occurred if the prosthetic limb did not facilitate a successful step response (Fig. 7 and 11F and H) after the initial elevating or lowering sound-side step; specifically, either prosthetic swing was not initiated (as evidenced by lack of thigh flexion before falling) or swing was initiated but the prosthesis landed in an unsafe configuration (i.e., knee flexed). The inability to perform this prosthetic-side step is likely due to their prescribed prostheses’ control scheme that does not decrease flexion resistance until stance-release thresholds are met, as well as its passive behavior that relies on ballistic coordination with the thigh for full swing phase motion, both of which are compromised when the cyclic motion of gait is interrupted during a stumble (see Considering Prosthesis Type subsection). Recall that even the elevating and lowering strategy recoveries involved more thigh abduction in the prosthetic-side step, also likely due to this factor.
Both of these deficiencies in lower-limb dynamics (highlighted in Figs. 4, 7, and 11) overall led to shorter steps, less anterior foot placement relative to COM, and more trunk flexion/flexion velocity at each of the response foot-strikes (highlighted in Figs. 5, 8, and 12) for fallers compared to non-fallers, confirming that these previously reported fall indicator metrics [12, 31–39] translate to the transfemoral prosthesis user population.
These observations suggest that appropriate interventions to decrease fall risk for transfemoral prosthesis users may be to assist in properly performing the elevating strategy and/or to assist in initiating and safely completing the prosthetic-side step. Such interventions could be accomplished with some combination of training and external assistance, discussed below.
Strength training and/or task-specific motor skill training may help prosthesis users recover by targeting the two aforementioned deficiencies.
First, strength training targeted at the sound (tripped) limb’s hip and knee flexors may improve success in elevating over the obstacle. (Note one could potentially also accomplish this with a powered exoskeleton on the sound limb.) Training targeted at the prosthetic-side (support limb) hip flexors could also help by providing counteracting torques during the elevating step; however, transfemoral prosthesis users still may lack in the necessary knee or ankle torques due to their passive prostheses. Additionally, training targeted at the prosthetic-side hip could help initiate swing phase; however, the transfemoral prosthesis users who recovered in this study employed substantial thigh abduction to accomplish the step due to prosthesis constraints, likely necessitating external interventions discussed subsequently. Strength training has shown potential for improving responses for fall-prone populations ; future work is needed to investigate the feasibility of such training for prosthesis users for sound-side stumbles.
Second, a task-specific training protocol in the form of repeatedly introduced treadmill acceleration disturbances that require a stepping response may also improve recovery outcomes. In particular, the work of Grabiner et al. has shown that this task-specific training can improve the metrics at response foot-strikes linked to increased fall risk (reported in Figs. 5, 8, and 12 [34,
49]), and ultimately reduce fall incidence for some populations . Future work is needed to investigate the feasibility and lasting effect of such training for transfemoral prosthesis users for sound-side stumbles.
Powered prostheses (e.g., [50, 51, 52]) may reduce fall risk by addressing the deficiencies of the typically prescribed passive prosthetic knees in responses to stumbles.
First, a powered prosthesis could improve the sound limb elevating response by providing the necessary counteracting ankle and/or knee torques in the support limb, allowing for a more normative push-off to facilitate the elevating step and reduce the body’s forward angular momentum [19, 30].
Second, a powered prosthesis could presumably sense a stumble and initiate powered swing phase for the subsequent step without requiring ballistic coordination with the thigh, as passive prostheses do. In other words, users could more easily initiate and complete swing phase in the following step, addressing one of the leading causes of falls among study participants. Additionally, powered prostheses can provide robust stance support even if the prosthesis lands with a flexed knee. Some energetically passive knees could provide robust stance support as well, although in general, any prosthesis that does so must implement this as a stumble-specific behavior, since otherwise doing so would interfere with stance knee yielding during slope or stair descent.
Such improvements in the lower-limb deficiencies observed may improve the body’s state at prosthetic-side foot-strike (i.e., larger step length, more anterior foot placement, less trunk flexion, and more negative trunk flexion velocity).
Despite this potential for improvement, to date there have been no prosthetic interventions developed or tested to address sound-side stumbles for transfemoral prosthesis users. Future work is needed to investigate the feasibility of such interventions.
These suggestions may not only reduce falls but also generally improve responses for transfemoral prosthesis users who did not fall. For example, transfemoral prosthesis users still exhibited responses that indicated increased risk of falling (increased trunk flexion/flexion velocity) and compensation mechanisms (increased thigh abduction and arm motion) relative to healthy controls; thus, training and/or assistive device interventions may improve these metrics, ultimately helping to reduce the need to abduct and help control forward angular momentum to reduce these deficiencies.
There are several limitations to this work. First, a sample size of more participants would have improved the confidence of the observations enumerated here. More non-MPK users would have improved the discussion on the effect of prosthesis type. Regardless, given the inherent heterogeneity of the prosthesis user population (i.e., varying ages, activity levels, comorbidities, years of prosthesis use, prosthesis types), a case series characterization such as that presented here is arguably more representative than one in which a singular averaged result is reported .
Second, while substantial efforts were made experimentally to ensure the obstacle was not perceived prior to contacting the participant’s foot (see Methods and Results of ), the participants did know that they would be stumbled at some point during the trial.
Third, there is potential that requiring a cognitive task during the trials may alter balance outcomes; however, a recent stumble study found that performing Serial Sevens (i.e., the cognitive task chosen in this work) did not alter the participants’ recovery response . Furthermore, in the real world tripping often occurs when individuals are distracted/not paying attention [55, 56], so the task is not entirely unlike real-life situations.
Fourth, rest was not standardized across participants; instead, participants were given the opportunity to rest as much as desired in between each stumble based on their own comfort and energy levels. This rest period was based on personal preference rather than physiological monitoring, so there is potential that fatigue could have affected responses. However, there were no observed trends in fall/recovery outcomes with order of trials (e.g., no evidence that more falls occurred towards the end of a session) across participants.
Finally, the authors note that there are other factors that may have played a role in responses: socket attachment, residual muscle condition, and physical fitness level. Though this type of analysis was beyond the scope of the present work, future studies investigating these factors are encouraged.