Twenty healthy male participants (age: 23.8 ± 2.8 years; height: 180 ± 4.7 cm; weight: 73.8 ± 6.4 kg) took part in the study. They self-reported 16.7 ± 2.4 years of experience in multidirectional sports. All of them reported playing soccer as their main (n = 17) or secondary sports (n = 3), training 2.7 times per week on average. Some had additional COD experience from tennis and volleyball. Additionally, six participants regularly performed strength training. They reported no lower limb injuries within the last six months. Past but healed injuries included pulled hamstrings, ankle sprains, medial collateral ligament and meniscal injuries, but no ACL injuries. All participants were students at the University of Innsbruck and recruited from the University's recreational soccer teams. Players were not able to partake if they had recent injuries to the lower limbs, did not have prior experience in COD sports or if they were not enrolled as a student (due to insurance reasons).

Informed written consent was obtained after each participant received verbal and written information on the procedure. The study protocol was approved by the ethics committee of the University of Innsbruck (Board for Ethical Issues, Review Board Sport Science, ID 77/2022), following the principles of the Declaration of Helsinki.

An a-priori sample-size estimation with a significance level of 0.05, a desired power of 0.8, and an expected large effect size (d = 1.15) for the comparison of knee flexion angles of unplanned and preplanned CODs (36) resulted in a required sample size of nine participants. This estimation was performed using G*Power (v.3.1.9.7) (37) for t-tests analyzing the difference of two dependent means. A secondary estimation specific to linear mixed models [GLIMMPSE (38)] confirmed the required sample size for the statistical analysis used in the current study. To account for possible dropouts, the sample size was raised to ten and doubled to investigate the additional effect of COD sequence and angle, resulting in n = 20.

All statistical procedures were completed in Jamovi 2.3 (42) using the GAMLj module for linear models (43). All dependent variables (six kinematic variables at IC and two performance variables) were investigated using linear mixed models with the random factor “participant”. Random effects included a participant-specific intercept to account for baseline differences between participants. Adding a random slope did not improve fit over the random-intercept-only model based on lower AIC/BIC values for the random-slope model and a non-significant likelihood ratio test (LRT) based on the log-likelihood (example for knee flexion: p = 0.068). Fixed-effect estimates and 95% confidence intervals were stable across supported random-effects structures. We therefore retained the random-intercept-only model for parsimony. The model used the fixed factors “condition” with four levels (first unplanned COD, first preplanned COD, last unplanned COD, last preplanned COD) and “COD angle” with three levels (45°, 90°, 135°) and investigated their interaction (condition × COD angle). Pelvis orientation data and run-up speed were only available for the first COD of a sequence. Therefore, a separate model was developed for those variables, which only regarded data of the first CODs and therefore used only two levels for the fixed factor “condition” (first unplanned, first preplanned) and the same three-level fixed factor for “COD angle”. Completion time was only available for the whole sequence. Therefore, the factor “COD angle” was adjusted to represent the combination of the two observed angles, which resulted in four possible angle sums: 90°, 135° 180° and 225°, while the factor “condition” included the levels unplanned and planned. Models were estimated using the restricted maximum likelihood (REML) method. We quantified effect sizes for fixed effects in our linear mixed models as partial η² computed from the F-statistics and their numerator and denominator degrees of freedom. This metric expresses the proportion of residual variance uniquely attributable to a given fixed effect after accounting for all other fixed effects and the specified random-effects structure. For interpretation, partial η² > 0.01 is considered a small effect, partial η² > 0.06 a medium effect and partial η² > 0.14 a large effect (44).

In the presence of significant interaction effects, post-hoc tests were conducted. Significant post-hoc tests were Bonferroni-corrected by multiplying the p-value by the amount of post-hoc comparisons relevant for the hypotheses tested. To test H1, nine post-hoc comparisons were regarded: comparisons of first preplanned vs. first unplanned CODs for every COD angle (three tests), comparisons between COD angles within the first unplanned COD trials (three tests) and within the first preplanned trials (three tests). For H2, the following nine post-hoc tests were of importance: comparisons of the first to last COD of the unplanned trials for every COD angle (three tests), comparisons between COD angles within the first unplanned trials (three tests) and within the last unplanned trials (three tests). The significance level for all statistical analyses was set to alpha = 0.05. Descriptive statistics are expressed as mean (M) and standard error (SE). Missing data was simply excluded; no imputation was performed. The linear mixed-effects model accommodates unequal numbers of observations per participant, so unbalanced data were permissible. Due to occasional foot sensor problems, 15 out of 480 ankle angles had to be excluded.

A significant main effect for condition [F(1,215) = 113.05, p < 0.001, partial η² = 0.345] as well as for COD angle [F(3,215) = 67.02, p < 0.001, partial η² = 0.483] was found for completion time (Figure 2). Participants performed unplanned COD sequences (M = 5.5 s, SE = 0.05 s) significantly slower than preplanned sequences (M = 5.14 s, SE = 0.05 s). Completion times significantly increased with sharper COD angle combinations. No significant interaction effect (condition × COD angle) was found for completion time [F(3,218) = 1.48, p = 0.222, partial η² = 0.02].

There were no significant effects of condition or COD angle on run-up speed (condition: F(1,215) = 0.719, p = 0.397, partial η² = 0.003, COD angle: F(2,216) = 0.234, p = 0.791, partial η² = 0.002, condition × COD angle: F(2,216) = 2.190, p = 0.114, partial η² = 0.02). The mean run-up speed was M = 3.74 m/s (SE = 0.107 m/s) in unplanned conditions and M = 3.81 m/s (SE = 0.107 m/s) in preplanned conditions.

Kinematic patterns related to ACL injury risk could not solely be attributed to one of the anticipation conditions. Rather, the movement patterns found in preplanned CODs show both risk-mitigating as well as injury-provoking mechanisms. While trunk and pelvis show “safer” transversal motion in line with our hypothesis, the biomechanical patterns of the wider hip angle in the frontal plane and the more extended knee in the sagittal plane during preplanned scenarios are more in line with common ACL injury mechanisms, contradicting our hypothesis. In contrast, all kinematic characteristics observed in the preplanned condition can be related to COD techniques enhancing performance. These findings can be explained in the context of the performance-injury conflict (31, 32). Participants performed preplanned COD sequences faster than unplanned COD sequences. To exclude the effect of differences in the run up on completion time or kinematics at IC (45), run-up speed was included in the analysis and did not differ systematically between the conditions. Therefore, the performance of the COD sidestep itself and subsequent reacceleration must be enhanced in preplanned movements compared to unplanned movements. We allocate this performance advantage to appropriate movement planning, such as earlier trunk rotation and pelvis orientation.

A pelvis rotation towards the new intended movement direction, as seen in the preplanned condition, is considered an efficient strategy for both COD performance as well as injury prevention by initiating whole-body rotation before ground contact and therefore reducing rotational demands for the lower limbs during the stance phase (46). Due to reduced preparation, this technique for reducing risk and enhancing performance was not achievable in unplanned CODs. Additionally, in unplanned CODs at angles exceeding 90° we observed a more pronounced increase in trunk rotation away from the movement direction. This misalignment is linked to higher injury risk (32), possibly by increasing internal rotation moments in the knee (47) contributing to ACL loading (13). This highlights the demands for trunk stabilization at sharper COD angles. From a performance view, proper trunk rotation towards the new movement direction improves cutting speed (48). Both adequate pelvis and trunk positioning were lacking in unplanned scenarios, resulting in higher-risk biomechanical patterns in support of our first hypothesis.

Participants employed other movement patterns enhancing performance at the potential cost of higher knee loads (less knee flexion, more hip abduction) in preplanned CODs. On the other hand, unplanned scenarios showed less “risky” angles in hip and knee joints. Participants landed with lower hip abduction angles in unplanned scenarios, while there was a stronger increase in hip abduction in preplanned conditions from 45° to sharper angles. High hip abduction and the resulting wide foot placement is considered to increase COD performance by reducing contact times and facilitating acceleration into the new movement direction (32). However, it is also closely related to higher KAM, by placing the center of mass (COM) further from the center of pressure (COP), increasing injury risk (47). A reduction of hip abduction by employing a narrower foot placement is considered a technique to lower ACL injury risk (8). As a caveat, we would like to mention, that, while the validity of IMU-derived hip abduction has been found to be variable across participants (20), we still included hip abduction as an outcome variable in the interest of gaining insight into the lower-body frontal plane in ecologically valid testing.

Additionally, a more extended knee was observed in the preplanned scenarios, which benefits performance by reducing contact times and generating lateral forces for a push off into the new direction (32, 49). The extent of the sagittal knee movements' contribution to the ACL injury mechanism has been debated (21, 50) but there is evidence that a nearly extended knee (0–40°; 22) and a stiff knee landing increases injury risk by facilitating anterior tibial translation contributing to ACL load (51, 52). A bent knee, on the other hand, can facilitate the hamstrings to actively prevent anterior tibial translation (53). In fact, a “soft landing” using a higher knee flexion range of motion (ROM) to absorb and dissipate ground reaction forces is a common technique instruction to reduce injury risk (54). This technique, however, comes at the cost of performance and, therefore, is unappealing for coaches and athletes (32). In unplanned scenarios with less preparation time, participants exhibited greater knee flexion. This finding was consistent with earlier research. With some exceptions (55), the majority of previous studies have reported higher knee flexion angles in tasks with reduced preparation times (21, 56), even though unplanned tasks are generally considered riskier and produced higher KAM (23). The flexed knee likely reflects an effort of athletes to lower their COM in response to the task's unpredictability similar to what has been observed when individuals run on unstable surfaces (57). So, while the lacking trunk and pelvis positioning might be a sign of missing preparation time, the lower knee flexion and hip abduction angle could be regarded as means of readiness and robust motor control strategy in uncertain situations, preparing for the unknown.

It should be noted, that, with an average approach speed of 3.74 m/s in the unplanned condition and a cue 3 m before the COD target, participants had about 800 ms to react to the cue. Other studies investigating unplanned COD movements used shorter available time to react (ATR). While shorter times such as 300 ms are recommended to reveal unanticipated responses, anticipation effects were found in conditions with ATR up to 600 ms (24, 58). In contrast to the classic left vs. right COD protocols for one given COD angle, our protocol included six COD options (2 directions×3 angles), which increases the cognitive demand for preparing a movement response to the directional cue (24). Further, we observed that for our protocol shorter ATRs often lead to undesired compensatory movements delaying the actual response, such as small quick steps instead of a sidestep. Therefore, our ATR was chosen specifically for our COD protocol and the skill level of our participants, who were recreational soccer players. Since our task still revealed kinematic differences between the anticipation conditions, we conclude that the chosen ATR was appropriate but we aim to include shorter cue times for comparison in following studies.

Although protective measures and a robust control strategy seem appropriate for uncertain scenarios, such strategies are typically not observed in realistic game situations and realistic injury scenarios, which report e.g., low knee flexion angles just before ACL injuries (3). This discrepancy could come from divergent COD kinematics between isolated CODs, which are typically investigated in biomechanical studies, and CODs within a movement sequence, representative of real-world scenarios. To employ a more game-like situation, where CODs follow other cuts or sprints, we incorporated a sequence of CODs. Results showed that protective measures seen in unplanned isolated CODs were not found for CODs at the end of a sequence of unplanned movements. While an isolated unplanned COD resulted in some injury-protective kinematics (H1), CODs later in a sequence of movements showed increased ACL injury risk patterns.

A simple sequential COD task was able to provoke and expose potentially risky movement patterns in unplanned movements. In the last COD, participants touched down with a more extended knee at COD angles sharper than 90°, which is considered to increase ACL injury risk (as explained above; 52). Similarly, they rotated their trunk further away from the intended new movement direction in the last COD in unplanned sequences and showed an increase in hip abduction at sharper COD angles, again potentially increasing injury risk (as explained above; 32). These high-risk kinematic characteristics, e.g., the low knee flexion angle, were comparable to angles observed in isolated preplanned COD scenarios. Therefore, performing multiple CODs in a row seems to “overwrite” the protective strategy observed in isolated unplanned scenarios. While in preplanned movements, higher knee loads through kinematics such as an extended knee might be tolerable due to appropriate motor planning and adequate muscle activation, these movements might be potentially harmful, when they occur in an unplanned scenario. Future research should be focused on comparing knee joint loading and muscle activation between isolated and sequential CODs to investigate these assumptions.

The adaptation between first and last COD are likely due to increased cognitive and physical demands of the last COD, including spatial reorientation within the six-option COD course and reacceleration demands. A straight run up, as before the first COD, allowed participants to focus and prepare for the uncertainty of the first task and (unconsciously) select movement strategies balancing load and injury risk. However, the spatial reorientation and reacceleration demands before the last COD may have occupied the sensorimotor system enough to affect motor planning, resulting in higher-risk movements. It is known that cognitive and attentional demands influence decision making and motor planning capabilities. And as task complexity and cognitive demands increase, movement planning further deteriorates (24). Therefore, the added cognitive demand and further limited preparation time due to the preceding high-intensity 180° cut might have influenced movement planning and execution negatively. Similarly, secondary tasks during cutting (e.g., counting backwards) divide focus and attention and have been shown to deteriorate movement execution (24, 59). And while little is known about the demands of sequential CODs, COD research including more complex COD tasks or small sided games have shown a similar effect (18). CODs in game situations, and hence in injury situations, often occur as part of a chain of movements (linear sprints, acceleration/deceleration, stops, multiple CODs; 34) with elevated cognitive demands. Additionally, ACL injuries often occur following neurocognitive errors (28) and links were found between low neurocognitive functions and ACL injuries (60, 61). Training of neurocognitive functions, playing a critical part in motor control, could therefore have benefits for knee loading and injury prevention (29). COD assessments including cognitive and attentional challenges are a first step into understanding this relationship. By recreating real-life situations, we get a better insight into the circumstances, which contribute to the occurrence of high-risk movements and high knee loading.

One factor to consider when comparing the first and last COD of a sequence is a potential difference in run up speed (45), which was not measured nor controlled before the last COD. While the first COD is following a linear sprint from a starting position, the last COD is preceded by a 180° COD. If anything, we assume participants approached the last COD slower due to the preceding de- and reacceleration phase of the 180° COD. Prior work indicates that slower approaches generally yield COD movement characteristics associated with lower ACL injury risk (62). Thus, a speed difference alone would be expected to bias the last COD toward “safer” COD movements. In contrast, our findings indicate injury-provoking kinematics at the last COD, suggesting that we measured an effect of sequential-task demands rather than an effect of approach speed. We acknowledge the absence of last-COD speed measurements as a limitation for our interpretation. Where possible, future studies should estimate approach speed (e.g., from IMUs using pelvis velocity as a proxy) and/or include approach speed as a covariate in the analyses.

The findings encourage the use of more complex COD tasks for realistic assessments of movement strategies. A sequence of unplanned CODs seems more suitable to provoke injury-prone COD kinematics, which can then be the basis for risk screening or intervention assessments under more realistic conditions. As a note of caution, we would like to stress that all COD assessments must carefully balance the sensitivity of the protocol for exposing potentially harmful movement patterns with the actual risk of injury during the assessment. Given our thorough warm-up routine and the absence of opponents or balls, we were confident to not harm our participants.

As a side note, a main effect of the COD angle (across all anticipation conditions) was significant for all observed joint angles. This highlights the different demands when performing CODs of different angles, resulting in distinct biomechanical patterns. This angle-dependency is congruent with earlier COD research (63). Therefore, results of CODs assessments can only be adequately compared within the same COD angles. Additionally, when investigating ACL injury mechanisms in COD movements, it can be important to examine multiple angles, as specific risk-increasing characteristics (such as trunk rotation) seem to be most prominent for certain angles (e.g., in angles sharper than 90°).

We recognize a few limitations in our study, such as shifting weather and field conditions, along with the lack of kinetic data. In future research, we plan to incorporate estimates of joint moments and loading. In regard to our outcome variables, we mention that IMU-derived kinematics may be affected by sensor drift, soft-tissue artifact, and calibration/alignment errors, which can reduce accuracy but were kept to a minimum by following the system manufacturer's guidelines. While we understand the challenges of replicating real-life sports situations, we also see the value in doing so. Despite these obstacles, our method successfully highlighted how athletes adjust their movements during unexpected CODs with various angle choices, all within a realistic on-field environment. The findings of this study are currently limited to male athletes. Due to high incidences of ACL injuries in female soccer players (64), replicating the protocol with a female population could give insight into sex-specific biomechanics and motor control of CODs.

In conclusion, this study highlights how both task anticipation and the sequence position within a series of COD maneuvers significantly influence lower limb, pelvis and trunk kinematics, with direct implications for ACL injury risk. These results support the idea that performance and injury risk often exist in conflict, with athletes intuitively prioritizing performance-enhancing kinematics in preplanned conditions over injury risk-mitigating movement strategies. Unplanned CODs induced a mix of protective and risk-associated movement adaptations, characterized by safer sagittal plane knee and frontal plane hip kinematics but less favorable trunk and pelvis alignment. However, these protective mechanisms were not found when observing the last COD of a sequence of unplanned CODs. Distinct kinematic differences between the first and last COD maneuver of an unplanned sequence revealed more risk-associated kinematic characteristics in the final COD movement of an unplanned sequence, marked by a more extended knee, higher hip abduction in sharper angles and misaligned trunk rotation. This suggests combined cognitive and physical load as a contributing factor to movement deterioration and highlights the need for prevention programs addressing both factors: Besides strengthening and technique exercises that also maintain performance, training programs should include reactive tasks which train quick decision-making, selective attention, and spatial orientation skills to facilitate motor planning in cognitively demanding situations. Collectively, these findings emphasize the role of limited preparation time and increased neurocognitive demands in constraining motor planning during COD movements, leading to altered biomechanical patterns that may increase the risk for ACL injuries.