The purpose of this prospective observational study was to examine the ability of attention- demanding and community-relevant mobility tasks (obstacle crossing and dual-task walking) at hospital discharge to estimate fall risk and predict walking activity 3-months after hospital discharge post stroke. The findings supported our central hypothesis that performance on attention- demanding and community-relevant mobility tasks would estimate fall risk and be associated with walking activity post discharge.
Incidence of falls in subacute stroke
The prospective recording of falls in this study revealed that, for individuals who are ambulatory requiring minimal to no assistance at hospital discharge post-stroke, the incidence of falls is high. Almost half of our participants fell in the first 3 months post hospital discharge, which is consistent with previous reports during this time period.35,49,52-55,187 Our data also support previous findings that individuals are most likely to fall at home and during walking activities and transfers.37,51,53,54 Similar to previous reports,37,49,51,54 we found that fallers had longer hospital lengths of stay, lower mean self-efficacy for balance, higher-self reported impact of stroke on walking, slower gait speed, more impaired balance and lower extremity motor control, greater disability level and dependence in daily activities compared to non-fallers. Our sample differed from previous cohorts in terms of age, disability level, cognitive level, and presence and
severity of depression. 35,37,49,50,188 This may have been attributed to our population being relatively young compared to typical samples. Where nearly three-quarters of all strokes in the United States occur in people over the age of 65,189,190 the mean age of our sample was only 59.8 years (SD11.7). In addition, persons living in the southeastern United States have the highest stroke prevalence with more than one-third of all stroke hospitalizations in North Carolina occurring in people younger than 65.191
Obstacle-crossing ability and fall-risk in subacute stroke
In this study, we prospectively examined whether measures of obstacle-crossing ability (pass/fail), gait performance and situational awareness during obstacle crossing at hospital discharge post-stroke were associated with fall risk in the first 3 months at home. Our findings support our primary hypothesis that people who failed the obstacle task on at least one trial would be more likely to experience a fall within the 3-month follow up period than those who passed the obstacle task. Participants who failed the obstacle crossing task had higher odds of falling (OR = 10.00) and had a 3.00 times the risk (200% increase in risk) of falling than those that passed the obstacle-crossing task. In the only other study examining an obstacle-crossing task to predict falls in subacute stroke, Said et al15 found a higher risk ratio of 5.83 compared to our risk ratio of 3.00. The difference in risk ratio is likely due to differences in the population demographics and methodology. Said et al15 had a smaller sample size (20 versus 45 in the current study), fallers who were older (with a mean age of 70.4 (SD 8.7) years15 compared to our population with mean age of 61.3 (SD 11.7) years), population with longer median post stroke onset (median of 53 days versus 14 days in the current study), and time spent prospectively monitoring falls (6 months versus 3 months in the current study). The difference in the observed fall risk ratio between studies may
have been due to a smaller sample size limiting the population exposed, a population of older individuals increasing their samples fall risk, and monitoring falls for 3 months longer increasing the time frame during which falls could have occurred.
The relationship between obstacle-crossing ability at hospital discharge and falls during the first 3 months after hospital discharge was confounded by age, sex, and time post stroke. Interestingly, we did not find a significant difference between fallers and non-fallers based on age or a significant bivariate association between age and falling but entered age as a covariate based on previous research which found that older individuals post stroke fall more frequently.15 This noncongruent finding is likely due to our population being younger overall. The only statistically significant covariate in our model was time post stroke. In our sample, individuals with greater time post stroke required acute inpatient rehabilitation (AIR) and had greater impairment and disability. More participants being discharged from AIR failed the obstacle-crossing task (56.7%) at hospital discharge and fell (33%) in the first 3 months post discharge than participants discharged from acute care. Therefore, the obstacle-crossing task may be better suited for those being discharged home from AIR. Based on these findings, time post stroke should be considered as a covariate when validating this model.
We also identified that one’s self perception of how stroke impacts their walking influenced the relationship between obstacle-crossing ability at hospital discharge and falls during the first 3 months after hospital discharge. Our study substantiates prior evidence that in subacute and chronic stroke populations, self-efficacy for walking, balance and/or falls have a strong relationship with gait and balance performance,192,193 mediate the association between physical performance capacity and activity and participation in a chronic stroke population,82 and have an
association with falls.194,195 Therefore, including an assessment of self-efficacy for walking post stroke should be considered at hospital discharge when evaluating risk for falls.
Our findings partially supported our hypotheses that greater pre-obstacle step length variability and slower obstacle-crossing gait speed at hospital discharge would estimate fall status at 3 months post hospital discharge. Based on our desire to maintain a parsimonious model, we entered only one of these variables in the regression model, pre-obstacle step length variability. Pre-obstacle step length variability did not estimate fall status in our population, even though there was a significant difference between fallers and non-fallers and bivariate associations with obstacle crossing and fall status. Our finding that fallers had significantly greater pre-obstacle step-length variability is consistent with the findings reported by Said et al.15 The difference in pre-obstacle step length variability in our sample was driven by the late pre-obstacle phase. Entering pre- obstacle step length variability from the late phase of the obstacle approach versus pre-obstacle step length variability from the entire obstacle approach may have changed the outcome and should be considered in future model development and assessment with a larger sample size. Furthermore, if the approach to the obstacle was shorter (e.g., consisting of the late phase only), the difference in step length variability between fallers and non-fallers may have been higher, leading to an estimation of fall status. Further analysis of the relationship between pre-obstacle step length variability and falls is warranted based on our results.
Obstacle-crossing gait speed was not entered in our model based on having a strong relationship with other variables entered. Despite this, we found that obstacle crossing gait speed was significantly difference between fallers and non-fallers and had bivariate associations with obstacle crossing status and fall status. Our findings that fallers had significantly slower obstacle crossing gait speed than non-fallers is consistent with the findings reported by Said et al.15 In our
sample, obstacle-crossing gait speed for fallers and non-fallers was slower than previously reported,15 with fallers in our study demonstrating a median obstacle-crossing gait speed of 0.25 m/s (IQR 0.12, 0.49) versus the previously reported gait speed of 0.42 m/s (IQR 0.14, 0.68) and non-fallers obstacle-crossing gait speed of 0.59 m/s (IQR 0.27, 0.93) versus 0.92 (IQR 0.72, 1.03 m/s). This is likely related to the differences in the time of assessment post stroke (median of 14 days versus 53 days) providing more time for rehabilitation and recovery of walking mobility in other samples compared to ours. Clinically, measuring obstacle crossing ability as one measure of fall risk at hospital discharge would be practical and is recommended based on previous research and our current findings.
Our findings did not support our remaining hypothesis that longer obstacle-crossing paretic swing duration would estimate fall status at 3 months post hospital discharge. Our obstacle- crossing paretic swing duration data was not consistent with previous findings that paretic limb obstacle crossing duration was longer in those who failed the obstacle crossing task and those who fall.15 The difference in findings for this variable may have resulted from the fact that we had 6 participants who either could not complete the obstacle-crossing task due to an inability to clear the obstacle on all 4 attempts (n=2) or who chose to clear the obstacle leading with the non-paretic limb only during all 4 trials (n=4); therefore reducing our sample size for this analysis. A lower obstacle height and/or a greater number of trials may have provided each participant an opportunity to successfully cross the obstacle leading with paretic limb to allow for acquisition of this data for all trials and all participants. Increasing the number of trials may not be feasible in an AIR setting, making this measure of obstacle-crossing ability less practical.
The lack of ability for pre-obstacle step length variability and paretic swing limb duration to estimate fall risk may have been related to the methodological procedures used during our
obstacle-crossing task. For example, our participants were not blinded to the obstacle task set-up prior to the start of the trial and we do not know how that time period may have impacted their planning. Further, the length of the obstacle approach, the obstacle height and width, and the number of trials provided may have impacted performance. In the only other study that has investigated the relationship between obstacle crossing ability and fall risk after stroke,15 the participants were provided 8 trials (versus our 4) and the obstacle was only 4 cm high and 1.5 mm width versus our average of 8.8 cm and 13 mm width. Providing a greater number of trials may have provided more opportunity for the participant to fail the task and placing the obstacle a shorter distance from the start and either lowering or raising the height of the obstacle may have changed performance. As noted above, increasing the number of trials may not be feasible in an AIR setting, making these 3 measures of obstacle-crossing ability less feasible and practical.
In contrast to our hypothesis, time spent fixating on the obstacle prior to obstacle crossing at hospital discharge was not associated with fall status at 3 months post hospital discharge. In addition, there were no significant differences in the visual parameters between fallers and non- fallers and no significant relationships between any of the visual parameters during obstacle crossing and fall status. Therefore, based on our established methodology, we did not enter any of the visual parameters into the regression models. One variable, percent dwell time on the obstacle, demonstrated a trend towards significance and therefore warrants further investigation. The lack of significance at this time may be due to an insufficient sample size (n=42) for detecting differences in visual tracking parameters, as the study was not powered for the visual tracking parameters due to the exploratory aspect of this measure. Since this is the first study to examine visual tracking during obstacle crossing post stroke, there are no other data to compare our findings with at this time. In addition, since our participants were not blinded to the obstacle task set-up
prior to the start of the trial, we do not know how that time period may have impacted their visual tracking prior to or during obstacle-crossing. Further, the length of the obstacle approach and the number of trials provided may have impacted visual tracking performance. Based on our preliminary results, we think that it will be important to modify future study methodology and clinical assessment procedures by blinding the participant to the obstacle-crossing environment prior to the start of the assessment to allow for visual tracking during both the planning and obstacle-crossing time periods. Due to the important role that vision plays in successful obstacle crossing, further investigation of visual eye tracking behavior on fall risk is warranted. However, the actual integration of eye tracking technology in the clinic will require the development of a practical and economically feasible assessment tool, including the hardware and software.
As an exploratory aim, we examined whether our observed estimates of association between obstacle crossing ability and fall status at 3-months post hospital discharge were different based on whether therapy was received post hospital discharge and the location of discharge (acute care or AIR). Provision of rehabilitation services did not impact the observed associations between failing an obstacle-crossing task and fall status. Conversely, being discharged from AIR impacted the relationship between obstacle crossing ability and fall status. Therefore, as discussed previously, the use of an obstacle-crossing task to estimate fall risk post stroke would be best integrated into a falls risk-assessment in an AIR setting.
Obstacle-crossing as a screening tool for fall risk in subacute stroke
Based on our analysis of model performance, assessment of obstacle-crossing ability has potential value as a screening tool at hospital discharge for persons post stroke being discharged to the community. The approximate pre-test probability of falls post stroke has been reported to
be around 50%,55,196 Among participants who failed the obstacle-crossing task in our study, the probability of falling at least one time was 78%. Among participants who passed the obstacle- crossing task, the probability of not falling was 74%. Furthermore, the sensitivity, specificity, +LLR, -LL, and AUC of our unadjusted model were equal to or higher than other reported tests aimed at predicting the binary outcome of fall status in subacute populations.35,55,161,197
In summary, our study demonstrated that the use of an obstacle-crossing task with the simple binary exposure of obstacle crossing ability (pass or fail) at hospital discharge can be used as a feasible, practical, and safe falls-risk assessment tool for individuals post stroke being discharged home from either acute care or AIR. The obstacle crossing task was able to identify people at risk of even a single fall, which is critical due to the high frequency of falls during this time period post stroke and the greater risk of injury post stroke. Even though our study used a sophisticated set-up and research-grade equipment, based on our findings, a clinician could determine the crossing-ability status (pass/fail) and an obstacle-crossing gait speed (combined pre- obstacle and obstacle-crossing gait speed) that was using a stop watch. Therefore, the assessment could be completed by one therapist with minimal and inexpensive equipment.
Dual-task interference in subacute stroke
All of the participants completed the baseline dual-task walking tasks without incident which confirms that dual-task testing at hospital discharge is feasible and safe. Compared to single- task gait speed (0.50 m/s), our participants had significantly slower dual-task gait speed (0.38 m/s), with the magnitude of the dual-task deterioration in gait speed in line with chronic stroke populations ranging from 0.06 to 0.20 m/s.62,198,199 The relative mean DTE on gait speed in our participants (-22.15%) was similar to what has been previously reported in a population 1-6 months
post stroke for the same category fluency task (-19.72%).150 Our participants did not demonstrate a significant difference in correct response rate between the single and dual-task trials. Comparison between studies investigating dual-task walking is difficult due to the limited number of studies performed in the acute and subacute time frame post stroke, the variation in cognitive tasks (e.g., calculations, auditory Stroop task, clock task, spontaneous speech, working memory, etc.), variation in instructions provided (no instruction, motor prioritization, or cognitive prioritization), and a lack of similarity in reporting of outcome measures (e.g., absolute and relative DTE on both the motor and cognitive performance).
Our study findings provide evidence that individuals at hospital discharge post stroke display similar patterns of cognitive-motor interference to subacute (1-6 months) and chronic (> 6 months) stroke populations. Our post stroke sample at hospital discharge demonstrated primarily dual-task costs on gait and cognitive performance (mutual interference) suggesting inadequate attentional resources for dual-task walking. In addition, our post stroke sample demonstrated dual- task costs on gait performance alone (gait interference) suggesting that participants were prioritizing postural control during gait.32 To a lesser extent, our participants demonstrated dual- task costs on gait with concurrent improvements in cognitive performance (cognitive-priority trade-off). The patterns observed in our study were similar to what has been reported in multiple study samples greater than 6 months post stroke32 including one study that reported cognitive- priority trade-off as the most common pattern for a narrative speech dual-task.88 Clinically, our findings and previous research support the idea that monitoring cognitive and motor performance together are important with the expectation that patterns will differ at baseline and potentially during recovery, especially when gait is automatic and does not require focused attention.200,201 The observed differences in the type of interference pattern post stroke continues to challenge our
understanding of and the ability to quantify the relationship between gait and cognitive performance post stroke.
There are a number of factors that may have influenced the magnitude and pattern of cognitive-motor interference observed in our participants. For example, the type and difficulty of the cognitive tasks can affect gait speed differently post stroke.62 In the current study, performance on the category naming task may have been affected by the level of task difficulty relative to the expertise and abilities of the participants (e.g., category tools or types of flowers),31 resulting in the variability observed in the dual-task effect on cognition (DTEc). Furthermore, the individual’s walking and cognitive capacity appear to influence the pattern as evidenced by our finding that stroke severity differentiated between those who demonstrated mutual inference with those who demonstrated cognitive-priority trade-off. In our sample, greater stroke severity limited the capacity for both motor and cognitive performance (mutual interference) during dual-tasking. In addition, distractions in the testing environment may have captured the participant’s attention (e.g., searching for answers to category naming task or other pedestrian traffic in testing area, etc.) and influenced motor and/or cognitive performance. To limit the influence of environmental factors on our outcomes, we tested in a quiet, consistent environment using standardized instructions.
Based on our findings, a cognitive-motor dual-task assessment is feasible, safe, and practical for individuals that are ambulatory and have minimal cognitive-linguistic involvement in an AIR setting. Yet, even though we followed the recommendations for standardization of cognitive-motor dual task assessment,200 there was a tremendous amount of variability in gait and cognitive performance (cognitive greater than gait) and the cognitive-motor interference patterns demonstrated in our sample. Based on the recommendations, we did not explicitly provide instructions for prioritization. A modification to the instructions provided may have resulted in a