The Relationship Between Spatio-temporal Gait Parameters and Cognitive Function: Protocol for a Cross-sectional Study

20 Background : Motor dysfunctions, such as slower walking speed, precede the occurrence of dementia 21 and mild cognitive impairment, suggesting that walking parameters may be effective biomarkers for 22 detecting early sub-clinical cognitive risk. In fact, while our preliminary study had a small sample, we 23 found several walking parameters obtained by three-dimensional motion capture system, to be 24 correlated with computer-based assessments of various cognitive function modalities. The Cognitive- 25 Gait (CoGait) Database Project, described in the current protocol, aims to establish a database of multi- 26 dimensional walking and cognitive performance data, collected from a large sample of healthy participants, crucial for detecting early sub-clinical cognitive risk. 28 Methods: The study will recruit healthy volunteers, 20 years or older, without any neurological or 29 musculoskeletal disorders. The estimated sample size is 450 participants, including a 10% attrition 30 rate. Using computer-based cognitive assessments, all participants will perform six tasks: (i) the simple 31 reaction time task, (ii) Go/No-Go task, (iii) Stroop Color–Word Test, (iv) N-back test, (v) Trail making 32 test, and (vi) Digit Span test. Gait will be measured through joint kinematics and global positioning in 33 participants’ lower legs, using pants with an inertial measurement unit-based three-dimensional 34 motion capture system, while walking at a comfortable and faster pace. Finally, we will establish a 35 prediction model for various cognitive performance modalities, based on walking performance. 36 Discussion: This will be the first study to reveal the relationship between walking and cognitive 37 performance using multi-dimensional data collected from a large sample of healthy adults, from the 38 general population. Although there are several methodological limitations, such as accuracy of 39 measurements, the CoGait database is expected to be the standard value for both walking and cognitive 40 functions, supporting the evaluation of psychomotor function in early sub-clinical cognitive risk, 41 including motoric-cognitive risk syndrome. 42

disorders [10][11][12][13][14][15]. In a recent survey, the incidence rate for MCR in people in their 60s was 54.9 per 55 1,000 persons, which strongly suggests that preventive activities for these cognitive disorders should 56 be started at working age, before the initial presentation of early cognitive decline [16]. 57 There are many risk factors for dementia, such as genetic factors, lifestyle habits, sleep 58 quality, education, and physical and social activities, even in the absence of detectable cognitive risk 59 [6,7,10,11,[17][18][19]. Thus, we argue that cognitive risk screening in healthy participants requires 60 additional multimodal parameters. Establishing a database of multimodal parameters that includes 61 these risk factors is necessary to distinguish individuals with higher cognitive risks from the general 62

population. 63
Specifically, walking performance is the most notable marker of cognitive risk [20,21]. In 64 fact, our preliminary observation, using a three-dimensional motion capture system and computer-65 based cognitive assessments in a small sample, showed a significant relationship between some 66 cognitive and walking performance modalities in healthy volunteers. This suggests the feasibility of a 67 motion capture system to predict and screen for a decline in cognitive function (see Supplementary 68 Figure S1, Additional File 1). Our study will collect the detailed walking parameters and cognitive 69 assessment scores of a sample of healthy volunteers, aged 20 years or older, to establish more accurate 70 predictive models for cognitive function. 71

Study design 73
The current study, called the Cognitive-Gait Database (CoGait) project, will follow a cross-74 sectional design, aimed at elucidating the relationship between walking and cognitive performance in 75 healthy adults. The sample size was estimated to be 410 persons, calculated using the effect size (f = 76 0.074) of a similar study [20] under the following conditions: multiple linear model; df = 24, a = 0.01, 77 1-b = 0.8. The upwardly corrected sample size of 450 people, accounts for an attrition rate of 10%. 78 The sample size calculation was performed using G*power [22,23]. 79

Study setting and recruitment 80
Healthy volunteers, older than 20 years and without any neurological or musculoskeletal 81 disorders, will be recruited for the study. The research team consists of a research scientist at Xenoma 82 Inc. (TF), as well as faculty members at the Tokyo University of Technology (YW, TK, and AA), and 83 Dokkyo Medical University (SI). Recruitment will be handled by the research team members. 84

Inclusion criteria 86
Men or women older than 20 years, without any neurological or musculoskeletal disorders 87 potentially affecting walking and cognitive functions, will be eligible for participation. Informed 88 consent will be sought from all participants before they declare their medical histories. In the absence 89 of exclusion criteria (see Exclusion criteria), participants' walking and cognitive performance will be 90 measured (see Measures). 91

Exclusion criteria 92
Participants will be excluded from the study if their medical histories include the disorders 93 or conditions listed in Table 1. Participants with visible abnormalities in walking function (i.e., a 94 mobility function score of <7 on the Functional Independent Measure), as assessed by a skilled 95 physician, physiotherapist, nurse, or research scientist, will also be excluded [24]. Only native 96 Japanese speakers will be recruited. 97 Table 1. Exclusion criteria 98

Exclusion criteria
People who cannot walk independently.
People with any amputations.
People with disabilities in vision, hearing, and/or equilibrium.
People who cannot use a tablet owing to disabilities in their upper limbs.
People at high risk of falling.
People who cannot understand the experimental instructions in Japanese.

Measures 99
General procedure 100 First, we will obtain written informed consent from all participants. Second, we will obtain 101 their personal information and medical histories (Tables 1 and 2). Unique personal identities (IDs) will 102 be generated for participants who do not meet the exclusion criteria, and printed as QR codes, required 103 during registration, for the walking and cognitive assessments. Upon receiving their IDs, participants' 104 walking will be measured (see Gait measurement). After a 10-minute break, they will participate in 105 the cognitive assessment (see Cognitive assessment). The walking and cognitive function datasets will 106 be securely stored in online cloud storage. An overview of the study procedures and measurements is 107 presented in Figure 1. 108  To calibrate the three-dimensional model calculation prior to gait measurement, each 123 participant will be asked to adopt two postures: leaning forward with their hands pressed against a 124 wall ( Figure 2D), and standing upright ( Figure 2E). Next, participants will be asked to walk in a 125 straight line on a 16 m walkway, including 3 m inlet zones, at the start and end points. The 126 measurements will be conducted under two conditions: fast (maximum speed) and comfortable (self-127 selected speed). Participants will practice the walking task under each condition, several times before 128 the measurements, to ensure they understand the requirements of the experimental tasks. In the fast 129 condition, we will instruct participants to walk at their maximum speed, without running or falling. In 130 the comfortable condition, we will instruct them to walk at their regular, comfortable speed. 131 The measurement datasets will comprise raw IMU sensor signals (acceleration and angular 132 velocity), global positioning of each sensor and anatomical landmark, and joint angles in the pelvis, 133 hip, knee, and ankle (347 parameters in total). The datasets will be automatically uploaded to cloud 134 storage ( Figure 1). The data processing methods are described below (see Data Analysis). 135

Cognitive assessment 136
Cognitive assessments will be conducted using a custom-developed web-based software 137 application. Computer-based cognitive assessment covers a wide range of cognitive functions and 138 minimizes floor and ceiling effects [26]. Moreover, such assessments can collect data not only in terms 139 of the accuracy of each task, but also in temporal, spatial, and spatio-temporal domains, differentiating 140 it from conventional paper-pencil-based cognitive assessments [27,28]. 141 The software was coded using JavaScript® and runs on a web browser (Safari, Apple, 142 Cupertino, CA). To ensure visual conformity, all tests will be conducted using tablets with the same 143 Making Test (TMT), and (vi) Digit span (DS) test. During the tests, the tablets will be positioned in a 146 landscape orientation and tilted at 20 °. Participants' right index fingers will be placed 2.0 cm behind 147 the tablet. Before any tests, all participants will practice the tasks at least twice, with verbal instructions 148 from the expert staff, using a tablet. 149

(i) Simple reaction time task 150
The flow of the SRT task is illustrated in Figure 3A. Participants will be asked to fix their 151 gaze on the center of the white cross (fixation point), after the warning signal (1000 Hz, 50 ms). The 152 target signal (red circle), will appear on a black background at random timing (1-3 seconds after the 153 warning signal). Participants will be asked to press the "はい (Yes)" button with their right index 154 finger when the signal appears [29]. The SRT task comprises 10 trials. 155 The flow of the Go/No-Go task is shown in Figure 3B. The procedure of this task is similar 157 to the SRT task, but the target signal is either a red circle, red triangle, or red cross, on a black 158 background with a 2 s presentation time. Participants will be asked to respond only when the red circle 159 appears on the display. The task consists of 10 trials each, for the "Go" and "No-Go" paradigms [30]. 160 The Go/No-Go ratio was determined such that the reaction time will be prolonged compared to SRT

(iii) Stroop Color-Word Test 164
The Stroop Color-Word Test was translated from a previous study [31] and converted into 165 a digitalized test, using a tablet. The color words are displayed as target signals after the warning signal. The N-back task is a major approach used for assessing working memory capacity [33,34]. 176 Single digit numbers will be displayed on the tablet as target signals ( Figure 4A), with a presentation 177 time of 2 s for each target signal. 178 In the one-back condition, participants will be asked to respond only when the target signal 179 is the same as the last digit number (congruent condition). In the two-and three-back conditions, they 180 should respond only when the target signals are the same as the second-and third-digit numbers, 181 respectively. The frequency of the congruent condition will be set to 44% of the target signals, and the

(v) Trail Making Test 186
We adapted the Japanese version of the paper-based TMT, so that the data could be uploaded 187 to our cloud storage [35]. The TMT consists of the TMT-A and TMT-B. The TMT-A contains 25 circled 188 numbers, ranging from 1 to 25; participants will be asked to tap the circled numbers in order, from 1 189 to 25 ( Figure 4B). The TMT-B contains 13 circled numbers, ranging from 1 to 13, and 12 circled 190 Japanese kana letters; participants will be asked to tap the numbers and letters following the rule 1- individually and sequentially, and participants will be asked to remember the sequence of the 201 presentation. Next, participants will be asked to recall the sequence, using the numeric keypad. The 202 DS test will be conducted under forward and backward conditions. Participants will be required to 203 recall the sequences in forward or backward direction, depending on the task condition; sequence 204 length ranges, from two to nine numerical digits, in both tasks. Participants will be required to perform 205 the test under both conditions, and the trials will be repeated three times for each sequence length. 206 When participants record three mistakes in the same sequence, the DS test will be completed. In our 207 preliminary experiment with healthy volunteers (n = 4; 29.8±2.68 yrs), the matching rate for the 208 backward condition was relatively low, with larger sizes of target digit numbers (Supplementary 209 Figure S3E, Additional File 1). 210

Gait analysis 212
From the gait measurement datasets, we will calculate the general walking parameters, such 213 as stride length and minimum toe clearance, using the built-in software (e-skin LETS WALK, Xenoma 214 Inc., Tokyo, Japan). The general parameters are presented in Table 3. The sweeps of raw signals, such 215 as IMU data, joint angle, and global positioning of each sensor or anatomical landmark, will be 216 averaged with the time normalized by the percentage of the stride cycle. When the stride cycles cannot 217 be defined because of poor data quality, the dataset will be excluded. 218 Table 3. Items related to the general walking parameters 219

Stride length Cadence Speed
Base width Stance-swing ratio Step width Double support time Minimum toe clearance Joint angles (hip, knee, and ankle) Left-right asymmetry 220

Psychological analysis 221
Cognitive performance is evaluated by reaction time, process time, and task accuracy. The 222 reaction time will be calculated for the SRT, Go/No-Go task, Stroop Color-Words Test, and N-back 223 test. The reaction time is defined as the interval between the onset of the target signal and the 224 participant's response. The process time is defined as the interval between the task onset and 225 completion in the TMT and DS test. 226

Statistical analysis 227
In this study, we established statistical models for predicting cognitive functions based on 228 walking characteristics. Thus, the dependent variables are reaction time, process time, and task 229 accuracy, and the independent variables are walking parameters and averaged signal traces, in both 230 the fast and comfortable conditions. Prior to the substitution of the independent variables in the 231 statistical model, we will sift the variables, to prevent problems related to multi-covariance [40]. 232 The walking parameters will be reduced to less than 25 dimensions, by principal component 233 analysis (PCA), as needed. The statistical model will be used for multiple linear regression, with and 234 without the random sample consensus (RANSAC) algorithm. The accuracy of the statistical model 235 will be evaluated using Akaike's information criterion (AIC) [41]. All statistical analyses will be 236 performed using the Python script, with the scikit-learn library. 237

DISCUSSION 238
To the best of our knowledge, this will be the first study to reveal the relationship between 239 walking and cognitive functions using a three-dimensional motion capture system for the general 240 population. From a methodological perspective, the accuracy of the algorithm of three-dimensional 241 bone modeling in e-skin MEVA was already confirmed in a previous study, and was also supported by  paper-based testing [46]. In particular, our custom-made application could better correct the higher 252 dimensional cognitive performances, including the temporal, spatial, and spatio-temporal domains, 253 than conventional applications, which is more suitable for these feasibility studies.  For gait measurement, we considered that this study focuses on revealing the relationship 271 between walking and cognitive functions in healthy participants. Thus, the possibility of mass 272 screening of both functions, is more important than the accuracy of the parameters. In fact, the more 273 accurate three-dimensional motion capture, using the conventional optical system, required a much 274 longer time for data collection, which is not feasible for our purposes [54]. In contrast, the gait analysis 275 system, e-skin MEVA and LETS WALK, simply consists of IMU pants and a computer for recording 276 data, and has already been applied in clinical practice, such as rehabilitation for spinal cord injury 277 patients (Higashibaba and Irie, in submission). At the same time, we are conducting several studies on 278 the system's reliability and its relationship to conventional assessment tools in motion capture (Sonobe 279 and Amano, in preparation). 280 For the cognitive assessments, our web-based applications were created, based on a well-281 established psychological paradigm that is sensitive to early cognitive decline, even in healthy Finally, a few participants with MCI or MCR will be included in this study, owing to our 288 inclusion criteria. However, the database developed in this study could provide standard values for 289 both walking and cognitive functions, which would support the evaluation of psychomotor function, 290 including the MCR syndrome. Our research team also plans to conduct a cohort study for participants 291 in this study, to define the risks of MCR and MCI. This CoGait project database will be developed for 292 use as a worldwide platform, for cross-sectional and longitudinal studies on cognition, walking, and 293 frailty, as well as other studies in the field of geriatrics. in accordance with the Declaration of Helsinki. Prior to conducting our study, we will obtain 312 participants' written informed consent and assure them of their right to withdraw their consent at any 313 time, for any reason. 314

Consent for publication 315
At registration, we will obtain participants' consent for publication. When a participant 316 withdraws consent, we will exclude their data from the published data. 317

Availability of data and materials 318
After publication, the database of walking and cognitive parameters developed in this study 319 will be published as a public repository. Access permissions will be managed by the Tokyo University 320 of Technology and Xenoma Inc. 321

Competing interests 322
The authors have no conflicts of interest to disclose. YW received research funding from  The experiments consist of four phases or sessions: informed consent, registration of 520 personal information (i.e., height, gender, and birthday), gait measurement with e-skin MEVA, and 521 cognitive assessment, using the tablet. The datasets were automatically uploaded to online cloud 522 storage. 523   to tap the numerical digits in ascending order. In the TMT-B, participants also tap the targets in 543 ascending order, but they should tap the numerical digits and kana letters alternately. C: Digit span 544 (DS) test. This test uses a numeric keypad and numeric number indicator. Participants are asked to 545 remember the target sequence of numerical digits individually shown in the indicator. After the 546 presentation, they are asked to recall the sequence in the same or reversed orientation, using the keypad. 547 After inputting the digit sequences, they tap the Enter (Etr) key. In addition, they can fix the input 548 using the All Clear (AC) and Delete (Del) keys. Description of data: We performed preliminary experiments before developing this study protocol. 555 Figure S1 depicts the result of a pilot observation of the relationship between the walking and cognitive 556 parameters in a small population. Figure S2 shows the result of comparisons of joint angles between 557 MEVA and VICON (conventional motion capture system). Figure S3 presents the results of 558 preliminary experiments of cognitive performance, using a custom-made web-based application. 559 Figure S4 shows the rule for the Japanese kana letters in the TMT-B test. Figure 1 Schematic diagram of the experimental procedure The experiments consist of four phases or sessions: informed consent, registration of personal information (i.e., height, gender, and birthday), gait measurement with e-skin MEVA, and cognitive assessment, using the tablet. The datasets were automatically uploaded to online cloud storage.  Cognitive assessments using the reaction time paradigms A: Simple reaction time (SRT) task: In addition to the presence of the xation point, the warning signal (WS) rings (1000 Hz, 50 ms). Participants should tap " (Yes)", as soon as possible after the WS. Feedback is presented after the responses. B: Go/No-Go task: The ow of the task was almost the same as that of the SRT. However, in this task, participants were asked not to respond to a non-target presentation (triangle and cross). C: Stroop Color-Word Test: All experimental paradigms and ows are similar to those in the Go/No-Go task. Participants responded only to the red Chinese letters.

Figure 4
Cognitive assessments using the process time paradigms A: N-back test. The N-back test is one of the most established assessments of working memory capacity. Participants are asked to respond when the target signal is the same as the last digit number. B: Trail Making Test (TMT). The TMT consists of TMT-A (numeric numbers alone) and TMT B (a mixture of numeric numbers and Japanese kana letters). In the TMT-A, participants are instructed to tap the numerical digits in ascending order. In the TMT-B, participants also tap the targets in ascending order, but they should tap the numerical digits and kana letters alternately. C: Digit span (DS) test. This test uses a numeric keypad and numeric number indicator. Participants are asked to remember the target sequence of numerical digits individually shown in the indicator. After the presentation, they are asked to recall the sequence in the same or reversed orientation, using the keypad. After inputting the digit sequences, they tap the Enter (Etr) key. In addition, they can x the input using the All Clear (AC) and Delete (Del) keys.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.