Introduction

Falls are a leading cause of fatal and non-fatal injuries in older adults in the U.S1. Around one in three older adults fall each year2, leading to a heavy quality of life impact and over $50 billion in medical costs annually3. Up to half of all walking steps taken daily are part of turns, depending on the environment4. Unfortunately, falls during turns were found to be 7.9 times more likely to result in a hip fracture than falls during straight line walking5, and hip fractures in older adults lead to a severe decline in independence and quality of life6,7. Turning is a dynamic task that requires changing the body’s facing and travel directions while keeping the center of mass (COM) controlled with respect to the base of support (BOS)8,9,10.

Aging creates unique challenges that can inhibit effective turning while walking, and the strategies that are used to maintain balance during turning tend to change with age. Older adults have been shown to turn more slowly and use different movement strategies when compared with younger adults11,12,13. For example, Thigpen et al. (2000) showed that during 180° turns, older adults tended to take more steps to execute the turn in place of a pivot turn, which otherwise would rotate the body in one movement over the planted foot12. Similarly, Dixon et al. (2018) showed that, unlike young adults, older adults do not tend to “lean” into the new walking direction when performing a turn13. Although turning can be challenging for older adults and poses higher risk of falls and injury, there is minimal research focusing on how older adults control their balance during turning while walking.

This study focuses on frontal plane balance metrics. Previous studies indicate that the ability to control balance in the frontal plane is particularly important towards fall mitigation14,15. Gait studies have indicated that active control is necessary to adjust mediolateral balance in the frontal plane, but is not as demanding in the sagittal plane16,17,18,19. Additionally, mediolateral instability tends to be associated with an increased risk of falls and fall-related injuries for older adults20,21,22. When balance is lost in the lateral direction during single support, a “crossover” corrective step - stepping across the body - could be needed to prevent a fall. This is a more difficult corrective maneuver for older adults than losing balance medially during single support, or in the antero-posterior direction20.

Angular momentum has been emphasized as a useful balance metric associated with fall-risk in older adults23,24,25,26,27. Frontal-plane angular momentum (Hf) is a measure of rotational speed of the body around the whole body’s COM in the frontal-plane. Older adults tend to have a larger Hf range, representing Hf extrema, during straight-line gait than younger adults23,28, which has been associated with a lesser ability to control Hf and a greater risk of falls29,30. Previous research with young adults has shown that Hf range tends to increase during turns when compared with straight-line gait8. However, this has not yet been explored in older adults. Additionally, as a measure in the speed-domain, Hf often reaches maximum values during gait when the body is relatively upright and nears zero at times that body sway changes its direction (i.e., at extrema of body tilt orientations). Thus, angular momentum values do not indicate the states of body alignment or orientation, but they may associate with extreme values of body sway if Hf is not adequately controlled (i.e., attenuated after an extreme value). Hf is predominantly controlled during gait by the moment or torque generated about the body’s COM from the ground reaction forces footfall to footfall. For example, taking a wider step (given the same ground reaction force patterns) may increase the moment arm for that ground reaction force to act to arrest the body’s rotation, thus avoiding an unstable body orientation and quickly attenuating any previous extreme Hf values23,31. Thus, this study aims to better understand the regulation of not only Hf used by older adults during turns, but also the relevant (and independent) control of the COM’s alignment with its BOS.

An additional balance metric, lateral distance (LD), has been helpful in describing balance states during turns8,32. LD is the frontal-plane distance between the body’s COM location to the closest lateral edge of the BOS. Therefore, it describes the proximity of the COM to the lateral BOS edge and is negative if the COM is lateral of the lateral BOS edge. Previous research with young adults and children found that LD minima tend to be smaller or more negative during turns as compared to straight-line gait8,32. This indicates that people can experience more complex balance states during turns, which often includes times when the COM is lateral to the BOS. LD during turning has not yet been explored with older adults, so this study aims to better understand how older adults regulate their balance states to complete a successful and safe turn.

Turning takes place in a variety of contexts in daily life. This study looks at 90˚ turns in two contexts: pre-planned 90˚ turns and late-cued 90˚ turns. Pre-planned turns mimic scenarios where one can prepare adequately with the steps leading up to the turn. Late-cued turns are meant to represent an unexpected sudden turn, where there is no preparation leading up to the turn. Prior research shows that late-cued turns can pose increased challenges to maintain balance while redirecting momentum in the new direction of travel8,13,33,34. Understanding the way in which these challenges are managed are relevant to understand for future fall mitigation practices.

Describing how Hf and LD are regulated in these two turn types provides important rotational and positional balance regulation contexts that may relate to clinical measures used to assess fall risk. Current clinical practice includes a variety of established balance, cognitive and psychosocial assessments to indicate fall risk in older adults35,36,37,38,39,40,[41](https://www.nature.com/articles/s41598-025-22800-x#ref-CR41 “CDC. Clinical Resources [Internet]. STEADI - Older Adult Fall Prev. 2024 [cited 2025 Jan 29]. Available from: https://www.cdc.gov/steadi/hcp/clinical-resources/index.html

“). Therefore, towards the goal of better understanding and improving fall mitigation in older adults, this study also preliminarily explores how clinical fall-risk assessments may associate with biomechanical frontal plane balance measures during turns.

The purpose of this study is to better understand how older adults regulate their frontal plane balance during 90 degree pre-planned and late-cued turns. Based on previous research with young adults, we hypothesized that older adults would have (1) larger range of Hf during both pre-planned and late-cued turns as compared to straight-line gait, and (2) larger range of Hf during late-cued turns as compared to pre-planned turns. We also hypothesized that older adults would have (3) smaller LD minima during pre-planned and late-cued turns as compared to straight-line gait, and (4) smaller LD minima during pre-planned turns as compared to late-cued turns.

Methods

Participants

This study included 16 healthy older adults (14 female, age 71.3 ± 4.93 years, mass 70.8 ± 11.5 kg, height 1.63 ± 0.078 m) who provided informed consent to voluntarily participate in this study. The protocol was approved by the Stevens Institute of Technology Institutional Review Board and carried out in accordance with relevant guidelines, regulations, and the Declaration of Helsinki. Informed consent was obtained from all participants. Participants self-reported that they: were free of lower extremity injuries or pain, were able to walk for a quarter of a mile unassisted, were able to hear without the use of a hearing assistive device, had no diagnosed balance deficit, and were not currently in a physical therapy program. The Dynamic Gait Index (DGI), an eight-item clinical tool used to assess balance and fall risk, was administered to determine potential balance deficits. All participants scored 19 or greater on the DGI42 as part of the inclusion criteria. The Montreal Cognitive Assessment (MoCA) was administered to rule out cognitive impairment, and all participants scored 23 or greater36 as an additional inclusion criterion. Additionally, all participants demonstrated decisional ability via the University of California, San Diego Brief Assessment of Capacity to Consent, which was customized for this study43.

All 16 participants performed straight-line gait and pre-planned turns, but only eight participants performed late-cued turns due to time limitations, as the protocol required at least one hour remaining to complete the late-cued turns. In addition, out of an abundance of caution about fall-risk, the study physical therapist recommended against one participant completing the late cued turns due to fatigue and concern about understanding the task, even though there was adequate time (participant nine in subsequent figures).

Experimental setup

Any time an older adult participant was in the lab, a physical therapist was present to guard their balance and provide verbal instructions following a printed script. All participants were screened for severe fall-risk and cognitive health as eligibility requirements. Although not an eligibility requirement, each participant self-reported their fall history over the prior 12 months44 and completed the International Falls-Efficacy Scale (FES-I) to indicate their concern of falling during every-day activities38. Participants also completed the Timed Up and Go (TUG) and Dual Timed Up and Go (Dual-TUG) assessments39,40.

A grocery store aisle intersection was simulated using a T-shaped walkway indicated with bright blue tape boundaries on the black floor (Fig. 1). The walkway was 0.91 m wide[45](https://www.nature.com/articles/s41598-025-22800-x#ref-CR45 “U.S. Access Board - Chap. 4: Accessible Routes [Internet]. [cited 2025 Jan 29]. Available from: https://www.access-board.gov/ada/guides/chapter-4-accessible-routes/

“), including a 10 m straight-way with a 90˚ turn in the center leading to a 5 m straight-way, as previously described8. Vertical poles were located at the entry to the turning aisle (Fig. 1). A screen (2.03 m diagonal) at the end of the 5 m intersection mimicked a grocery aisle signage for the turns. Three tasks were performed in an order of increased challenge as a safety precaution: straight-line gait, followed by pre-planned turns, and then late-cued turns. There was a five-minute instructional period before each task followed by two practice trials. For each trial, a starting foot was randomly prescribed with a 50% chance for the left and right starting foot. Participants were asked to walk as if they were not in a rush, but people were behind them such that they were not able to stop abruptly.

Participants performed 10–14 trials of straight-line gait followed by 10–14 trials of pre-planned turns (Fig. 1). For straight-line gait, they were asked to imagine they were in a grocery store walking straight down an aisle at their regular pace. For pre-planned turns, they were told to pretend that they were in a familiar grocery store, and they were shopping for broccoli. They knew which aisle to turn into, and a large picture of broccoli was shown on the screen to indicate the grocery store signage. They approached the intersection and turned to walk towards the screen at the end of the walkway.

Fig. 1

Three walking tasks. A Straight-line gait for 10 m. B Pre-planned left turns, participants are told to pretend they are in a grocery store looking for broccoli. They walk 5 m to the intersection and then turn down the simulated grocery store aisle and walk another 5 m. C Late-cued turns, participants are told they are in an unfamiliar grocery store and only know if they should turn after they reach the intersection and look down the aisle at the cue on the screen. When green broccoli was displayed on the screen, it indicated that they should turn, and when a red circle with a line through it appeared, it indicated that they should continue walking straight. Poles were placed at the intersection corners to simulate an aisle.

Finally, participants who had at least one hour remaining performed 20–24 trials of late-cued turns: 10–12 late-cued turn trials and 10–12 late-cued catch trials (Fig. 1C) that were randomly ordered. For this task, participants were told that they were in an unfamiliar grocery store looking for broccoli and, when they reached the intersection, they should look at the screen to determine if the broccoli was down that aisle or not, knowing that there was a 50% chance that it was in that aisle. Once the participants reached the intersection, a broccoli or a “no” symbol (red circle with a line through it) was shown on the screen. The cue was initiated manually by the researcher viewing when the foot position passed into the intersection on the motion capture computer screen, which displayed a video view aligned with the intersection for this purpose. If the broccoli was displayed, participants performed a 90˚ turn down the intersection (a late-cued turn trial). If the “no” symbol was shown, then the participant continued walking straight until the end of the 10 m walkway (a catch trial). The screen was blank until the participant reached the intersection (to avoid looking at the screen prior to the intersection).

Kinematic analysis

A 13-segment whole-body kinematic model46 was built using optical motion capture data (250 fps, OptiTrack USA). Rigid tacking marker clusters were placed on the head, torso, pelvis, left and right upper arms, forearms, thighs, shanks, and feet. Virtual anatomical markers were identified (3D digitization) using a calibration pointing device (Probe Kit, Optitrack, Corvallis, OR), which were tracked through time and space using rigid tracking clusters (Fig. 2)46. All marker data were gap-filled and smoothed using a cubic spline filter (MATLAB ‘csaps’ function with the smoothing value set to 0.0005).

Fig. 2

Front and back of participant with tracking markers on the head, torso, pelvis, left and right upper arms, forearms, thighs, shanks, and feet. 3D digitized virtual markers marked by red dots and listed.

Frontal plane angular momentum

Whole-body COM location was computed using a weighted average of each segment’s center of mass. Whole-body angular momentum (H) about the COM was computed as follows8,23:

$$:\overrightarrow{\varvec{H}}=:{\sum:}_{i=1}^{n}[\left({\overrightarrow{\varvec{r}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{i}}}-{\overrightarrow{\varvec{r}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{w}\varvec{b}}}\right):\text{x}:{m}_{i}\left({\overrightarrow{\varvec{v}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{i}}}-{\overrightarrow{\varvec{v}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{w}\varvec{b}}}\right)+:{\varvec{I}}_{\varvec{i}}{\overrightarrow{\varvec{\omega:}}}_{\varvec{i}}]$$

where (:{\overrightarrow{\varvec{r}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{i}}}) and (:{\overrightarrow{\varvec{v}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{i}}}) represent the (:{i}^{{th}) segment’s COM position and velocity vectors respectively, (:{\overrightarrow{\varvec{r}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{w}\varvec{b}}}) and (:{\overrightarrow{\varvec{v}}}_{{\varvec{C}\varvec{O}\varvec{M}}_{\varvec{w}\varvec{b}}}) represent the whole-body COM position and velocity vectors respectively. (:{m}_{i}), (:{\varvec{I}}_{\varvec{i}}), and (:{\overrightarrow{\varvec{\omega:}}}_{\varvec{i}}) represent the (:{i}}{th}) segment’s mass, moment of inertia, and angular momentum vector respectively, and n represents the number of segments. The frontal plane component of the whole-body angular momentum was computed as the component in the frontal plane defined by the pelvis orientation. The vector normal to the frontal plane was the horizontal-plane component of the anterior axis of the pelvis (i.e., so any anterior or posterior pelvic tilt would not affect its computation). Thus, using the “right-hand rule”, when viewed posteriorly, Hf is positive during clockwise body rotation (i.e., the body above the COM rotating rightward) and negative during counterclockwise body rotation (i.e., the body above the COM rotating leftward). The maxima values and range within the phase of interest (described shortly) were computed for each trial. To ease comparisons with previous research and across participants, Hf was normalized to a unitless form by dividing Hf by the product of each participant’s height ((:l)), mass, and (:\sqrt{g*l}), where (:g) is gravitational acceleration8,23.

Lateral distance

LD was calculated by determining the horizontal distance between the total body COM and the lateral edge of the BOS. The lateral direction was defined using the horizontal plane component of the pelvis’s mediolateral axis. The distance along this vector from the COM to the leading foot’s lateral edge was used as LD. The BOS was defined by the foot markers (i.e., all markers below the ankle in Fig. 2) below a height threshold established during the static calibration trial (when both feet were in contact with the ground) plus a 1 cm tolerance. As the COM approaches the lateral edge of the BOS, the LD value decreases. If the COM is lateral of the lateral edge of the BOS, the LD value is negative. To ease comparisons with previous research and visualization across participants, LD was normalized by dividing LD by each participant’s leg length8. Minima LD values were computed within the phase of interest for each trial.

Phases of interest

The phase of interest for straight-line gait was defined by the heel strike before and after the center 4 m of the walkway, to capture steady state walking. For the turning trials, the phase of interest was defined by a person-specific threshold determined by the average pelvis heading angle during straight-line gait plus or minus three times the standard deviation8,47,48,49. The turn phase began at the heel strike before the pelvis heading angle exceeded the threshold and ended at the heel strike after the pelvis heading angle returned below the threshold relative to the body’s new direction of travel8.

Additional contextual kinematic measures

Mean gait speed was computed during each task’s phase of interest. Additional contextual kinematic computations included mean step and stride length, and stride width using the methods described by Huxham et al. (2006)50, as well as mean stride duration, phase duration, and number of footfalls per each task’s phase of interest.

Baseline assessments

In addition to the biomechanical balance metrics described, this study explores preliminary associations between clinical, cognitive and psychosocial assessments and balance during walking and turning. The DGI is an 8-item clinical tool used to assess balance and fall-risk, with higher scores indicating better balance35. The MOCA is a 16-item cognitive assessment, with higher scores indicating better cognitive abilities36,37. The FES-I is a self-administered 16-item questionnaire to determine fear of falling during daily activities, with higher scores indicating a greater fear of falling38. The TUG is a clinical assessment that measures the amount of time it takes to stand, walk 10 feet, return to the chair, and sit back down39. The Dual-TUG includes an additional task of counting backwards by three while performing the TUG assessment. For both, faster times indicate better balance and gait40,[41](https://www.nature.com/articles/s41598-025-22800-x#ref-CR41 “CDC. Clinical Resources [Internet]. STEADI - Older Adult Fall Prev. 2024 [cited 2025 Jan 29]. Available from: https://www.cdc.gov/steadi/hcp/clinical-resources/index.html

“).

Statistical analyses

Differences in Hf, LD, and kinematic measures across straight-line gait and turning tasks were examined using linear mixed models that included random intercepts for study participants, random slopes for trial number nested within study task and study participants, and fixed effects for study task. Mixed models were chosen because they allowed for the appropriate handling of repeated measurements within study participants and innately maximize the information from subjects with missing observations. Pairwise comparisons between study tasks were estimated within the regression models via orthogonal contrasts. Residuals were examined for all models to ensure that all assumptions were satisfied. The Holm test was used to correct for multiple comparisons and to maintain a two-tailed familywise alpha at 0.05 across hypotheses tested. An adjusted p-value < 0.05 was used to determine statistical significance51.

Exploratory analyses were conducted (1) to examine whether assessment scores were related to Hf or LD and (2) to examine whether assessment scores moderate the relationship between study task and Hf or LD. Mixed models that included random intercepts for study participants and random slopes for trial number nested within study task and study participants were used for both exploratory aims. To explore whether assessment scores related to Hf or LD, the mixed model also included a fixed effect for assessment score. To explore if the assessments scores moderate relationships between study task and Hf or LD, the mixed model also included fixed effects for study tasks, assessment score, and an assessment score X study task interaction term. Results were presented both overall and by study task. Post hoc tests were conducted within the model to test whether the relationships differed across study tasks.

An a priori power analysis was completed using the key dependent variables in this study (Hf range and LD minima) using a published study with the same structure8. Assuming the sample size of 16 participants in the straight line gait and pre-planned turns tasks and 8 participants in the late-cued turns task yielded a power of at least 80% to detect a 1.5 standard deviation difference in these variables between experimental tasks (i.e., our primary analysis) while assuming a within-participant correlation of at least 0.5 and accommodating the Bonferroni corrected alpha necessary to maintain a 2-tailed familywise alpha of 0.05 (p < 0.0167) across the planned comparisons. For within-participant analyses, a minimum number of 10 trials per participant per study condition yielded 88.7% power to detect a 1.5 standard deviation difference in key study metrics between study tasks for the participant-level analysis, while accommodating the Bonferroni adjusted per-comparison alpha (p < 0.0167) necessary to maintain an overall alpha of 0.05 across the planned comparisons. A Bonferroni adjustment was chosen for the purposes of power analysis because it is highly conservative and the Holm test, which was used in all analyses, is based on the empirical p-value attained at the time of actual data analysis. (Supplemental Document 1 includes full statistical result tables).

Results

Group-level results for Hf and LD across tasks are displayed in Fig. 3; Table 1 and participant-level results for Hf and LD across tasks are displayed in Fig. 4 and Supplemental Document 1,** Tables S2 and S3**.

Fig. 3

Group-level statistical results for A angular momentum range (Hf) and B lateral distance minima (LD). Each participant’s average Hf range and LD minima values across trials per task are indicated by dots and connected with lines. Group-level average per task indicated by bars. Brackets and p-values indicate significant differences.

Fig. 4

Participant-level statistical results. Data points represent min, max and range for A Hf (positive values indicate clockwise rotation, when viewed posteriorly) and B LD for all trials across tasks and participants. Statistical analysis for Hf indicates changes in range, statistical significance for LD indicates changes in minima. Black bars indicate significant differences that align with the group-level significant findings, red bars indicate significant differences that oppose the group-level significant findings. Spin turns are indicated by triangles (when the left foot was closest to center of intersection) and step turns are indicated by circles (when the right foot was closest to center of intersection).

Frontal plane angular momentum range

Using group-level statistical analyses, Hf range was smaller during straight-line gait than pre-planned (p < 0.0001) or late-cued turns (p = 0.002) (Table 1; Fig. 3). There was no significant difference between Hf range during pre-planned turns and late-cued turns at a group-level.

Participant-specific analyses revealed that not all participants followed the group-level statistical findings for Hf range comparisons across tasks (Fig. 4, Supplemental Document 1, Table S2). For the group finding that Hf range was smaller during straight-line gait than pre-planned turns, ten of 16 demonstrated this result statistically and one statistically demonstrated the opposite result. For the group finding that Hf range was smaller during straight-line gait than it was during late-cued turns, two participants demonstrated this result and two statistically demonstrated the opposite result. For the group finding that there was no statistical difference in Hf range between pre-planned and late-cued turns, six of eight participants demonstrated this result and two participants had larger Hf ranges during pre-planned turns than late-cued turns.

Lateral distance minima

Using group-level statistical analyses, LD minima was larger during straight-line gait than during either pre-planned (p < 0.0001) or late-cued (p < 0.0001) turns, and LD minima was larger during late-cued turns than during pre-planned turns (p < 0.0001) (Table 1; Fig. 3).

Participant-specific analyses revealed that all participants demonstrated the group finding of larger LD minima during straight-line gait than during pre-planned turns (Fig. 4, Supplemental Document 1,** Table S3**). Four of eight participants demonstrated the group finding of larger LD minima during straight-line gait than during late-cued turns. Seven of eight participants demonstrated the group finding of larger LD minima during late-cued turns than during pre-planned turns.

Additional contextual kinematic measures

Using group-level statistical analysis, mean gait speed per task phase of interest was greater during straight-line gait than both pre-planned (p < 0.0001) and late-cued turns (p < 0.0001) and greater during pre-planned turns than late-cued turns (p < 0.0001) (Table 2). The effect of task on Hf range (p = 0.02) significantly differed by gait speed. The effect of task on LD minima did not significantly differ by gait speed (p = 0.36). Additional kinematic results are included in Table 2 and significantly differed across tasks (p < 0.0001). Supplemental Document 1,** Tables S4-S6** include additional gait speed and turn strategy context.

Exploration of possible associations with baseline assessments

Table 3 shows associations between baseline assessment scores and Hf range and LD minima. Generalized linear model results revealed that better scores (higher DGI, MOCA; lower FES-I, TUG, dual-TUG) for all baseline assessments were associated with greater Hf range for combined tasks and during late-cued turns. Better DGI and TUG scores had significant associations with greater Hf range during pre-planned turns. A post-hoc analysis showed that associations with Hf range between DGI, FES-I, TUG, and dual-TUG were stronger for late-cued turns than either during pre-planned turns or during straight line gait.

Better scores for FES-I, TUG, and dual-TUG had significant associations with smaller LD minima for combined tasks. Better scores for MOCA, FES-I, TUG, and dual-TUG had significant associations with smaller LD minima during pre-planned turns. Better scores for TUG and dual-TUG had significant associations with smaller LD during late-cued turns. A post-hoc analysis showed that associations with LD minima between MOCA, TUG, and dual-TUG were stronger during pre-planned turns than straight line gait, and associations between TUG and dual-TUG were stronger during late-cued turns than straight line gait.

Discussion

The purpose of this study was to compare frontal-plane balance measures during walking and turning performed by healthy older adults. As hypothesized, as a group, older adults used larger Hf ranges during pre-planned and late-cued turns than during straight-line gait. Based on prior research with young adults, we hypothesized that older adults would have a larger Hf range during late-cued turns when compared to pre-planned turns. However, this was not supported in our cohort of eight older adults who performed late-cued turns; there was no significant difference in Hf range between pre-planned and late-cued turns at the group-level. Additionally, as hypothesized, older adults used larger LD minima during straight-line gait as compared to pre-planned or late-cued turns, and larger LD minima during late-cued turns than pre-planned turns. For both Hf range and LD minima, within participant statistical analyses revealed marked differences across individuals in this cohort of healthy older adults. Finally, an additional post-hoc exploration between participant baseline assessments and their balance metrics was included to better understand across-participant differences to guide future research.

For Hf, most participants followed the group-level statistical result that Hf range was smaller during straight-line gait than during either pre-planned or late-cued turns. This group-level result follows what has been found in younger adults performing the same task8. However, some older adults used smaller Hf ranges