Susceptibility to the sound-induced flash illusion is associated with gait speed in a large sample of middle-aged and older adults

Background: Multisensory integration is the ability to appropriately merge information from different senses for the purpose of perceiving and acting in the environment. During walking, information from multiple senses must be integrated appropriately to coordinate effective movements. We tested the association between a well characterised multisensory task, the Sound-Induced Flash Illusion (SIFI), and gait speed in 3255 participants from The Irish Longitudinal Study on Ageing. High susceptibility to this illusion at longer stimulus onset asyn- chronies characterises older adults, and has been associated with cognitive and functional impairments, therefore it should be associated with slower gait speed. Method: Gait was measured under three conditions; usual pace, cognitive dual tasking, and maximal walking speed. A separate logistic mixed effects regression model was run for 1) gait at usual pace, 2) change in gait speed for the cognitive dual tasking relative to usual pace and 3) change in maximal walking speed relative to usual pace. In all cases a binary response indicating a correct/incorrect response to each SIFI trial was the dependent variable. The model controlled for covariates including age, sex, education, vision and hearing abilities, Body Mass Index, and cognitive function. Results: Slower gait was associated with more illusions, particularly at longer temporal intervals between the flash-beep pair and the second beep, indicating that those who integrated incongruent sensory inputs over longer intervals, also walked slower. The relative changes in gait speed for cognitive dual tasking and maximal walking speed were also significantly associated with SIFI at longer SOAs. Conclusions: These findings support growing evidence that mobility, susceptibility to falling and balance control are associated with multisensory processing in ageing.


Introduction
Multisensory integration defines the ability to utilise information from different sensory modalities to perceive and act in the world. This ability underpins many daily functions including balance and walking (Paraskevoudi et al., 2018), both of which are essential factors in maintaining independence, physical and cognitive health in older age.
Appropriately weighting sensory information based on their relative reliability is fundamental in gait and balance maintenance (Peterka, 2018) and it is one of the factors underpinning multisensory integration in ageing (Jones and Noppeney, 2021). Ageing brings a degradation of sensory inputs, including vision (Fozard and Gordon-Salant, 2001;Owsley, 2011), hearing (Jayakody et al., 2018) and proprioception (Campos et al., 2018), and, thus, a need to adapt to altered reliability of each of the senses (Peterka, 2002). Additionally, age-related changes occur in the way information from different senses is integrated (Murray et al., 2016). For example, the benefit on response time by presenting congruent stimulation across modalities is greater in older than younger adults. These benefits have been associated with reduced susceptibility to falling and faster gait in a visual-somatosensory task . The mechanisms underlying these associations could be related to cortical synchronization (Bollimunta et al., 2011) associated with cortico-thalamic gating, as well as to temporal discrimination abilities across the senses (Paraskevoudi et al., 2018). Mahoney and colleagues  have argued that both gait and multisensory integration are associated with overlapping neural circuits including the Superior Temporal Sulcus at cortical level and the Superior Colliculus at subcortical level. In the present study, we utilise the Sound-Induced Flash Illusion (SIFI) (Shams et al., 2000), a multisensory test which depends on the temporal synchronization between sensory inputs (Hirst et al., 2020), to deepen our understanding of the association between multisensory integration and gait. This is important as temporal multisensory perception can, to some extent, be trained in older adults (e.g. Jones and Noppeney, 2021;McGovern et al., 2022;O'Brien et al., 2020;Setti et al., 2014) and, in turn, physical exercise can modify multisensory integration (O'Brien et al., 2017), although the mechanisms behind it are yet to be fully elucidated.
The SIFI occurs when one brief visual stimulus (flash) is presented with two auditory stimuli (beeps) within a short temporal interval (Stimulus Onset Asynchrony, SOA), and participants report seeing two flashes. Measures of SIFI susceptibility have proven fruitful in discriminating between healthy and pathological ageing. Specifically, older adults with a history of falling (Setti et al., 2011), increased sway (Merriman et al., 2015;Stapleton et al., 2014) or poorer cognitive abilities (Chan et al., 2015;Hernández et al., 2019) are more susceptible to this illusion over longer SOAs than healthy older adults, indicating a wider Temporal Binding Window (TBW), for review see Hirst et al. (2020). An exceedingly wide TBW could be due to adaptation to slower or less accurate temporal perception, and to difficulty inhibiting irrelevant sensory inputs (Basharat et al., 2018), potentially linked to GABAergic mediated cortical hyper-excitability in the ageing brain, or to higher reliance on previous perceptual experience, i.e. perceptual priors (Chan et al., 2021). In turn, these characteristics of multisensory perception in ageing appear related to balance maintenance and walking. For example, SIFI susceptibility is associated with mean sway while standing , and SIFI susceptibility is also correlated with a change in Berg balance score after a multisensory balance training intervention (Merriman et al., 2015). Balance maintenance in ageing is associated with gait speed in healthy ageing (Xie et al., 2017) and in dementia sufferers (Lee et al., 2020), and both functions are associated with the Cerebellum, which plays a role in cognition and timing. Recently, an association between larger grey matter volume in the right Angular Gyrus and lower susceptibility to SIFI was found in older adults (71-87) (Hirst et al., 2021). The Angular Gyrus is also functionally associated with gait and balance (Tian et al., 2017). We therefore hypothesised that gait speed should also be associated with susceptibility to this illusion; as proposed by Paraskevoudi et al. (2018), slower walking speed should be associated with a less precise temporal integration of stimuli across different senses, as reflected by a higher susceptibility to the SIFI, particularly at longer SOAs.
Gait can be measured through different parameters some that are more susceptible to controlled processing, such as speed, stride length, and step base width, whereas cadence, swing and stance time are more associated to automated processing . We focused on gait speed as it is a good predictor of falls risk (Cesari et al., 2005;Verghese et al., 2009), which, in turn, is associated with multisensory integration .
In the present study, we utilise the data from The Irish Longitudinal Study on Ageing (TILDA) to test the hypothesis that faster gait speed is negatively associated with SIFI susceptibility, i.e., the faster gait speed, the less susceptible the individual is to the SIFI, particularly at longer SOAs. This hypothesis was tested under three conditions: usual pace, cognitive dual tasking, and maximal walking speed. Gait during dual tasking and gait at maximal speed association with SIFI was tested utilising relative change, i.e. the difference between gait speed at normal pace and the other conditions, normalised by gait speed at normal pace. This is to assess whether a greater decrease in speed during dual tasking is associated with being more susceptible to SIFI at longer SOAs and a larger change between speed at a usual pace and maximal speed is associated with reduced susceptibility at longer SOAs due to changes in multisensory integration (Jones and Noppeney, 2021) and motor processes related to multisensory integration in ageing (Turgeon et al., 2011).

Participants
Participants were drawn from the third wave of the TILDA study. TILDA includes a population-representative sample of individuals aged 50+ years. The study was approved by the Trinity College Faculty of Health Sciences Ethics Committee, and testing protocols conformed to the Declaration of Helsinki. All participants provided informed consent at waves 1 and 3, only wave 3 is utilised in this analysis. Demographic information is presented in Table 1.
The SIFI task was included at wave 3 of TILDA in the health assessment and 4014 participants took part in the SIFI task, 3965 of which were aged over 50 years, and therefore eligible for inclusion. Participants were not included if they had a Montreal Cognitive Assessment (MoCA) score lower than 23 (n = 466), were missing Max Gait speed measures (n = 165), missing MoCA (n = 2) missing cognitive walk (n = 29) or normal walk (n = 1), missing age (n = 1), registered legally blind (n = 2) missing self-reported hearing (n = 2), missing BMI (n = 2), missing visual acuity (n = 10), missing measures of depression (n = 12) or missing Cognitive Trials 2 performance (n = 18). This left a total sample of 3255, the demographics of this sample are shown in Table 1.

Material and measures
TILDA comprises both a Computer Assisted Personal Interview (CAPI) carried out by trained interviewers in the participants home and a health assessment which includes the SIFI, visual acuity, gait, and tests of cognitive functioning carried out by a research nurse in a dedicated health centre (Kearney et al., 2011).
SIFI: In the illusion condition, named 2B1F (i.e. 2 Beeps + 1 Flash), a white disc ('flash', subtending a visual angle of approximately 1.5 degrees in size was briefly (16 ms) presented against a black background and positioned approximately 5 degrees below central fixation. The visual stimulus was accompanied by two auditory 'beeps' (3500 Hz, 10 ms, 1 ms ramp), with the volume set at maximum loudness on the computers (approximately 80 dB). One flash was always presented simultaneously with one beep, the second beep could either precede (2B1F-) or follow (2B1F+) the flash-beep pair. Congruent multisensory conditions were also presented (2B2F and 1B1F). Each of the multisensory conditions 2B2F and 2B1F was presented twice with SOAs of 70, 150, and 230 ms. Four visual-only trials (two 0B2F and two 0B1F) were presented with an SOA of 70 ms. The multisensory and visual only trials were presented in one block in randomised order. The task was to report the number of flashes perceived. The auditory-only, 2 beeps (2B0F), trials were presented in one single block presented at SOAs of 70, 150 and 230 ms. A practice phase, comprising one trial from each of the multisensory and visual conditions was presented prior to the test.
Gait: Gait was measured with a computerised gait assessment mat measuring 4.88 m in length (GAITRite®, CIR Systems Inc., New York, USA). Gait speed was measured in cm/s and calculated by averaging performance over two trials in each walking condition.

SIFI
The SIFI took approximately 6 min to complete as part of the 3 hour health assessment. A Standard Operating Procedure was implemented throughout the testing. For the SIFI specifically, participants were seated in front of a computer (DELL Latitude E6400 with Intel core 2 Duo CPU, 2Gb RAM using Windows 7 Professional OS) and the task was explained to them. Participants were presented with a practice block and were then invited to start the test. For each trial, they were instructed to look at the fixation cross presented at the centre of the screen for 1000 ms and report the number of flashes (or beeps in the auditory-only block) appearing, which was then recorded by the nurse, who pressed the corresponding key on the computer keyboard. The test was self-paced.

Gait
Participants were asked to walk along the mat twice for each walking condition: at their usual pace (normal gait); usual pace while saying out loud every second letter of the English alphabet (cognitive dual tasking), or as fast as they could (maximal gait). Each walk started 2.5 m before and stopped 2 m beyond the edge of the mat to allow for acceleration and deceleration, so that a relatively steady pace could be measured. The three conditions, normal gait, cognitive dual tasking and maximum gait speed were administered in this order.

Selection of dependent and independent variables
The experiment has a crossed design where each participant undertook two trials of each of four SIFI conditions (2B1F+, 2B1F-, 2B0F, 2B2F) at each of three SOAs (70, 150, 230 ms). Each participant therefore has 24 corresponding measurements (two trials of four experimental conditions measured at three SOAs). The dependent variable was the binary response indicating whether the correct number of flashes was identified or not for each SIFI trial. Gait was collected separately to the SIFI with three conditions per participant (usual, maximum walking speed and dual walking/cognitive task). The difference in speed when the participant was asked to walk at their fastest pace compared to their usual pace (here on referred to as "delta max") and the difference between their usual pace and their gait speed while performing the cognitive dual task hereon referred to as "delta dual task" was calculated as follows:

Delta Max = (Maximal Gait Speed − Usual Gait Speed)/Usual Gait Speed
Delta Dual Task = (Usual Gait Speed − Dual task speed)/Usual Gait Speed Therefore, for maximal speed one would expect those that perform better at SIFI, i.e. to be less susceptible to the illusion, to have a higher delta max, indicating the ability to walk much faster than the usual pace if required, and lower delta dual task values, indicating that the secondary task was not very impactful on gait speed.

The model
As mentioned, the response variable was a binary variable indicating whether the correct number of flashes was identified in each SIFI trial. To model this, a mixed effects logistic regression was implemented using the glmer function in the lme4 package in R. A random intercept term for participants and a random intercept for experimental condition were included to capture variability at the participant level and variability due to experimental condition (similar to Baayen et al. (2008) mixedeffects modelling with crossed random effects for subjects and items).
The interaction between SOA and gait was of particular interest, in line with our hypothesis that those who walk slower would be also less correct in identifying one flash correctly in the illusion condition, particularly at longer SOAs.
The following factors were controlled for in the models: age, sex, SOA, education, self-reported hearing, visual acuity, Colour Trail Making Test (version 2, i.e. alternate two colours), MoCA, BMI, Depression (CESD score ≥ 9), the number chronic conditions (from a total of 6 conditions) and the number of cardiovascular conditions (from a total of 6 conditions). To control for response bias and unisensory visual performance, the model also controlled for the conditions 1B1F and 0B2F. Education was measured as "Primary level or less" (0-8 years of education); "Secondary level" (13-14 years) and "Third level" which includes college graduate or higher (≥14 years). Visual acuity was measured using the Early Treatment Diabetic Retinopathy Study (ETDRS) logMAR chart and the score from the best eye used in the model. Self-reported hearing was recorded as excellent, very good, good, fair or poor. All continuous variables were standardised before analysis and so the resulting coefficients in the following section for continuous variables refer to the odds ratio for each one standard deviation increase in the relevant covariate.

Results
Below we present the results for each gait task. Figs. 1-3 show the adjusted marginal probabilities of identifying the correct number of flashes for each SOA under a range of values of the identified gait condition ( Fig. 1: Usual gait speed, Fig. 2: Delta Cog. gait; Fig. 3: Delta Max Gait speed). In each case the range of values chosen for the specified gait condition were chosen to fall within the 5th and 95th percentiles of the observed values in the dataset.
Gait at usual walking pace: There was a significant interaction between gait and SOA (p < 0.001), whereby increased gait speed increased the odds of a correct response at SOA 150 and 230 ms i.e. faster gait speed was associated with lower susceptibility to the SIFI at longer SOAs. There was also a significant interaction between 2B1F-and gait speed (p < 0.001), whereby faster gait was associated with a lower odds of being correct when one beep preceded the beep-flash pair compared to when it followed it (across all SOAs), which is associated with an asymmetric window of integration (see Fig. 1).
Additionally, a significant interaction was observed between gait speed and performance in the 2B2F control condition (p < 0.001), whereby faster gait was associated with a lower odds of being correct (Fig. 1). The interaction between the unimodal 2B0F condition and gait speed did not reach significance. Gait during cognitive dual task condition: The interaction between delta cognitive dual task gait and SIFI SOA was significant for the SOA 230 ms (p < 0.001). Slower walking while dual tasking compared with the walk at usual pace without dual tasking (i.e. larger values of Delta Cog.) was associated with greater SIFI susceptibility at the longest SOA. This interaction failed to reach significance at the 150 ms condition (p = 0.71). See Table S2. There was an interaction with the 2B1F-condition and the Delta Cog. score (p = 0.03), whereby larger Delta Cog. values (i. e. larger difference between the cognitive gait speed and the normal gait speed) were associated with higher SIFI susceptibility at longer SOAs. In general, the 2B1F-condition was associated with higher SIFI susceptibility (i.e. lower probability of a correct response) when compared to the 2B1F+ condition regardless of gait speed and SOA, indicating an asymmetry in the binding window (Fig. 2). Performance in the other control conditions (2B0F and 2B2F) did not significantly interact with dual task walk speed (p = 0.12 and 0.18 respectively).
Maximum walking speed: As with the other gait tasks participants were more susceptible to the SIFI under the 2B1F-condition when compared to the 2B1F+ condition regardless of SOA and gait speed (Fig. 3). There was also a significant effect for Delta Max (p < 0.001) i.e. the normalised difference between maximum walking speed and walking at the usual pace whereby those with higher delta max values had higher odds to be correct at SIFI, more specifically each standard deviation increase in Delta Max gait was associated with increased odds of 2.34 95 % CI (1.56, 3.53) of a correct SIFI response see Table S3. There was also a significant interaction between this measure and SOA, at both 150 (p < 0.001) and 230 ms (p < 0.001). Indicating that those who were able to increase their speed more when asked to do so (maximal speed walk) had a higher increase in accuracy in these conditions (Fig. 3). The impact of gait speed on the 2B1F-condition did not significantly differ from the 2B1F+ condition (p = 0.11), as both conditions had a similar increase in accuracy under increasing maximal gait speed (Fig. 3), however as stated there was a systematically lower accuracy for the 2B1F-condition regardless of SOA and gait.
As an additional sensitivity analysis, history of falling, operationalised as any self-reported incidence of a fall from waves 1, 2 and 3 (current wave) which included any type of fall (i.e. accidental, injurious or recurrent) was also included in the above models but was not found to be associated with SIFI for any of the gait conditions analysed and its addition did not change the results discussed above (OR 95 % CI for model with usual gait speed (0.94, 1.1) p = 0.786; OR 95 % CI for model with cognitive walk (0.94, 1.1) p = 0.776; 95 % CI for model with maximal gait speed (0.94, 1.1) p = 0.811).

Discussion
We tested the association between multisensory processing and gait , the unimodal auditory 2B0F condition (c) and the control bimodal condition 2B2F (d). Faster usual walking speed was associated with increased probability of a correct response at longer SOAs in the 2B1F+ condition. For corresponding statistics see Supplementary Table S1.
(measured during usual pace walking, and relative change during dual tasking and maximum walking speed) in a large cohort of middle-aged and older adults. The SIFI is based on the temporal synchrony between the auditory and visual stimuli and depends on the TBW of the individual, which changes with ageing. We argued that, because the SIFI is sensitive to the temporal dynamics of multisensory integration, SIFI susceptibility would be associated with gait speed. Speed was chosen as it had previously been associated to functional abilities in older adults. This hypothesis was supported by recent studies showing that those that benefit more from multisensory stimuli are also less susceptible to falling  and walk faster . This hypothesis was verified: higher susceptibility (i.e. lower accuracy) at long SOAs in the SIFI illusion condition was associated with slower gait, slower walking while performing a cognitive dual task and less quickening of pace in when asked to walk quickly, relative to normal pace (i.e. delta max). The association between performance on the illusory conditions and gait in all three walking conditions suggest that there is the predicted association between gait speed and the ability to integrate (or segregate) signals from different senses. Although the SIFI and gait tasks were not performed simultaneously in TILDA, the association between SIFI susceptibility and gait speed is consistent with previous results suggesting that SIFI performance is associated with balance control (measured as mean sway during standing) in older adults . Walking is a complex activity requiring the coordination of inputs from the visual, vestibular, proprioceptive (Paraskevoudi et al., 2018), and auditory system (Campos et al., 2018) and recruiting multisensory brain areas, at sub-cortical and cortical levels required for cortico-thalamic gating and sensory reweighting . Efficiency of these processes may underlie the association between faster gait and lower susceptibility to the SIFI at longer SOAs. In addition, susceptibility to SIFI was higher across conditions when one beep preceded the beep/flash pair (2B1F-) than when the second beep followed it, with an interaction with gate speed in walking while dual tasking. This is in line with previous studies showing an asymmetry in SIFI susceptibility and TBW (McGovern et al., 2022), a pattern that differs from younger adults, whereby the TBW tends to be narrower with an auditory lead (Stevenson et al., 2012). It is known that the dynamics of multisensory integration change with ageing in a way that is not fully understood (Basharat et al., 2018). An auditory lead is an unlikely situation compared with a visual lead, where, in a real-world scenario, a lag between vision and audition is justified by spatial distance (depth). Older adults, who in general benefit more from multisensory integration (see Laurienti et al., 2006), may be less accurate, or less conservative, in a situation whereby their prior experience acquired through life does not apply. However, these considerations remain speculative, and more research is needed. Across conditions the predicted interaction between gait speed and performance on the illusory condition is the most consistent result.
A limitation of this study is that it is based on cross-sectional data and on a simplified version of the SIFI paradigm, however longitudinal data Adjusted marginal probability of making a correct response in the Sound Induced Flash Illusion (SIFI) task as influenced by cognitive dual task gait speed relative to usual walking speed i.e. standardised gait speed. Effects are shown for three Stimulus Onset Asynchronies (SOAs), 70 ms, 150 ms and 230 ms, and for four SIFI conditions, the 2B1F+ condition (a), the 2B1F-condition (b), the unimodal auditory 2B0F condition (c) and the control bimodal condition 2B2F (d). Larger cognitive walk speed relative to usual walking speed was associated with decreased probability of a correct response at longer SOAs. For corresponding statistics see Supplementary Table S2. in future studies may elucidate whether multisensory integration, as measured using the SIFI and gait-related impairment are causally related. The small number of trials used in this version of the SIFI, due to the constraints of a large population study, undoubtedly limits the type of analyses that can be conducted, e.g., perceptual sensitivity analyses (see e.g. Setti et al., 2014), however it is reassuring that the patterns of response found in TILDA are comparable to other smaller studies with larger numbers of trials and a more controlled set-up (see Hirst et al., 2020). An alternative explanation of these findings is that gait and SIFI do not share underling mechanisms based on sensory integration, but are both associated with a generalised brain decline, which partly explains the overlap between sensory and cognitive decline (e.g. Anstey et al., 2001). Although a common cause hypothesis cannot be excluded, the inclusion of cognitive performance within the current model should partly control for this, as cognitive ability should be impacted by a generalised brain decline. Our results therefore speak in favour of a specific association between gait and multisensory processing and their causative mechanisms, opening potential avenues for rehabilitation. The promising, although still preliminary, results of training interventions (see O'Brien et al., 2023 for a review), both computer-based (e.g. Setti et al., 2014;O'Brien et al., 2020;McGovern et al., 2022) and real-life, e. g. physical exercise (O'Brien et al. 2017) present an interesting avenue to train multisensory perception with potential benefits on mobility and function. However, many questions, both theoretical and applied, remain open (see Hirst et al., 2020). In particular, the exact mechanisms that are at play in the ageing brain remains to be detailed, these being changes in reliance on perceptual priors across different senses (e.g. Chan et al., 2021), gating and inhibition (Paraskevoudi et al., 2018;Balz et al., 2016), or general changes to the internal clock (e.g. Jablonska et al., 2022). On applied grounds, it remains to be established whether paradigms such as SIFI, presenting incongruent stimuli across the senses, can provide additional diagnostic value for issues such as incidents of falling, where diagnostic tools based on congruency of inputs across the senses exist  and inform more tailored rehabilitation strategies.

Funding
The work was supported by the Health Research Board grant nos. ILP-PHR-2017-014 (awarded to Fiona N. Newell, Annalisa Setti and Rose Anne Kenny) and ILP-HSR-2017-021 (awarded to Rose Anne Kenny). This publication has also emanated from research supported in part by a Grant from Science Foundation Ireland under grant number 18/FRL/6188 awarded to Belinda Hernandez. TILDA is funded by the Irish Government, the Atlantic Philanthropies and Irish Life plc. These funders were not involved in the study design, collection, analysis and interpretation of data, writing of the paper or submission for publication. Any views expressed in this report are not necessarily those of the Department of Health and Children or of the Minister for Health. Marginal effects of SIFI and Delta Maximum Gait Speed on Probability of Correct Response Fig. 3. Adjusted probability of making a correct response in the Sound Induced Flash Illusion (SIFI) task as influenced by maximal gait speed relative to usual gait speed i.e. delta maximum gait Speed. Effects are shown for three Stimulus Onset Asynchronies (SOAs), 70 ms, 150 ms and 230 ms, and for four SIFI conditions, the 2B1F+ condition (a), the 2B1F-condition (b), the unimodal auditory 2B0F condition (c) and the control bimodal condition 2B2F (d). Larger maximum relative to normal walk speed was associated with increased probability of a correct response at longer SOAs. For corresponding statistics see Supplementary Table S3.