We performed a single-blind, 3-arm, parallel design, randomized feasibility trial with a 1:1:1 allocation ratio. The feasibility study was conducted in conjunction with an on-going trial (Clinicaltrials.gov identifier NCT04070703; 28/08/2019), with the research protocol approved by the institutional review board of the Oregon Research Institute. The study was performed between August 2020 and September 2021, during the 2020-2021 phase of the COVID-19 pandemic, with all research activities (recruitment, enrollment, intervention, and data collection/assessment) performed remotely via the use of telephones (for prescreening) and online conferencing software (Zoom.com) (for trial implementation and outcome assessment). All participants provided informed consent, completed electronically via online Qualtrics survey software (qualtrics.com). The feasibility-trial reporting follows the general guidelines described in the Consolidated Standards of Reporting Trials (CONSORT) 2010 statement: Extension to randomized pilot and feasibility trials .
Setting, Participants, Recruitment, and Procedures
The study was undertaken in home settings where the target population was community-dwelling, independent-living older adults. Initial enrollment efforts targeted individuals who lived in Oregon but was expanded to other states in the U.S. after the first wave of recruitment. Multiple methods of recruitment were used, including direct mailing, online promotions (e.g., research website, social media), word of mouth, and contact with volunteers from previous research projects.
The recruitment criterion was community-dwelling adults aged 65 years or older who met diagnostic criteria for having MCI . Evidence of MCI was established if individuals (a) self-reported a change or decline in memory (corroborated by an informant), (b) recorded a score of ≤0.5 on the Clinical Dementia Rating (CDR)  scale (with information derived from both the participant and an informant), and (c) showed no major impairment in cognitive function as indexed by a score of ≥24 on the Mini Mental State Evaluation (MMSE) . We excluded individuals who (a) had participated in a rigorous and structured physical activity or exercise program (including Tai Ji Quan) in the past 3 months, (b) showed signs of depression as indicated by a score of ≥5 on the Geriatric Depression Scale , (c) had major medical or physical conditions that precluded exercise, as determined by their healthcare practitioner, or (d) were unable to follow the consent process or sign the study consent form.
Recruitment was conducted primarily via initial phone communication, followed by a scheduled online Zoom meeting using a procedure established in a prior study . Briefly, a research assistant made initial contact via a telephone call with those who responded to the study promotion. This contact confirmed eligibility criteria regarding age, memory status, levels of habitual physical activity, and time availability. An independent interview was conducted with an informant who provided information regarding the respondent’s cognitive status based in part on a CDR assessment . Individuals were further evaluated, via an online Zoom session, on additional eligibility criteria, including administration of the CDR and MMSE. During each stage of prescreening, individuals were given sufficient time to ask questions and clarify issues related to risk, benefits, and commitment related to study participation. Those who met the study eligibility criteria signed the study consent form, and then underwent an online baseline assessment session conducted by an independent assessor. The recruitment for the feasibility study ended, as planned, in April 2021.
Randomization and Blinding
Eligible individuals were randomly assigned in a 1:1:1 ratio to receive one of the three active interventions, with a block size of 4 or 6. Randomization was not stratified. The randomization list was generated by a research data analyst with the use of nQuery software (www.statsols.com) and was kept in sealed envelopes. Allocation concealment was ensured since the randomization was performed by a research assistant who did not know the sequence of interventions and was not involved in the assessment or intervention. Randomization occurred after informed consent was obtained and baseline assessments were completed.
Participants were not blinded to the intervention group allocation but were requested not to disclose their group allocation to the study assessors during all follow-up assessments. Outcome assessors were masked to group allocation throughout the study period. Blinding was strictly maintained by (a) emphasizing to assessors the importance of minimizing assessment bias, (b) regular checking of the blinding status via inquiries made to randomly selected participants, and (c) randomly asking assessors to guess what intervention the participants had received. Every effort was made to keep the assessors from knowing the group allocation, including the use of identical assessment forms, assigned study participation identification numbers, and a designated database separate from the one used by unblinded study recruiters. In instances when blinding was broken, the assessor was immediately excused and replaced by another blind assessor.
Cognitively Enhanced Tai Ji Quan Training
The intervention was adapted from the Tai Ji Quan: Moving for Better Balance program [14, 28], augmented by a set of cognitively stimulating exercises practiced concomitantly with Tai Ji Quan to increase cognitive processing demands [27, 29]. The progressive training consisted of 5-15 minutes of preparatory (warm-up) exercises, 45-55 minutes of core training (learning, practicing, repeating) in the 8-form routine , and 1-2 minutes of cool-down exercise. During the initial 10 weeks, the core physical exercise involved primarily learning and repeatedly practicing the 8 forms (in a prespecified sequence of Moving a Ball, Part Wild Horse’s Mane, Repulse Monkey, Brush Knees, Fair Lady Works the Shuttle, Grasp Peacock’s Tail, Waving Hands like Clouds, and Single Whip), with training focusing on symmetrical and coordinated actions such as trunk rotation and weight shifting, controlled and coordinated displacement of the body’s center of mass over the base of support, dynamic eyes-head movements, and anterior-posterior and medial-lateral stepping. From week 10 on, training became fully integrated with prespecified cognitive exercises, described below, that targeted domains of memory, executive function, and processing speed.
The training approach used a mix of dual-task, task prioritization, and memory exercises intended to actively engage participants in learning and practicing Tai Ji Quan forms and variations of forms and encouraging instructor-participant interaction during practice. Specifics of integrated Tai Ji Quan exercises included (a) verbalizing step-by-step movements involved in a form, (b) responding to correct form instructions while inhibiting unwanted instructions, (c) exercising form recalls (including 2 or 3 forms back) either standing (with a narrow base of support) or moving, (d) practicing a sequence of forms while verbally associating form names with form numbers (or the reverse), (e) performing forms without verbal or visual instructional cues, and (f) performing word spelling (forward and backward), side-switching (alternating between lateral and bilateral sides), out-of-sequence form switching, and varying form sequences (forward, backward, oddly or evenly ordered).
Standard Tai Ji Quan Training
This intervention consisted of the identical 8-form routine training protocol  implemented for the cognitively enhanced Tai Ji Quan group, except that the cognitive exercises were not included. Instead, an integrated set of therapeutic movement exercises was added that focused on training of sensory integration, mobility, and limits of stability and recovery of near loss of balance. These exercises followed the general guidelines regarding balance training for older adults, with activities that emphasized progressively difficult postures with varying sizes of the base of support, postural perturbation that displaced the body away from equilibrium, weight-loading on lower-extremity postural muscle groups, and sensory compensation and integration .
The intervention followed a traditional Tai Ji Quan training approach, where instructors play a dominant and leading role in teaching and practicing, without instructor-participant interaction during exercise. Accordingly, each exercise session primarily involved teaching participants, via both verbal and visual cues, Tai Ji Quan forms and movements through repeated practice. The specifics of physical training activities included controlled, self-initiated Tai Ji Quan–based exercises with synchronized breathing, including center-of-gravity displacement using a dynamic interplay of stabilizing and self-induced destabilizing postural actions involving unilateral weight-bearing and weight-shifting movements, trunk and pelvic rotation, ankle sway, multidirectional stepping, and eye-head-hand movements . Similar to the protocol of the cognitively enhanced group, the exercise training in this intervention was taught progressively, with learning and practice of the 8 individual forms taking place during the first 10 weeks and repeated practice with variations thereafter.
This intervention, with the protocol adopted from our prior trial  and piloted in another study , consisted of various muscle stretches, performed in either seated or standing positions. The teaching strategy was identical to that of standard Tai Ji Quan training in that participants were asked to follow instructors’ instructions in performing all exercise activities specified in the program, and no instructor-participant interaction was encouraged. Each session began with light warm-up exercises involving arm, neck, and leg circles; trunk rotation; and light walking. The core part of the exercises involved stretches of the upper body (neck, arms, back, shoulders, and chest) and lower extremities (quadriceps, hamstrings, calves, and hips), along with slow and gentle trunk rotations. Each exercise session ended with breathing exercises and progressive relaxation of major muscle groups.
Online Intervention Delivery Protocol
The three active exercise interventions were delivered using Zoom, for 16 weeks, with the same length of time (60 minutes per class session) and frequency (twice weekly). On each class day, participants were given a secured Zoom link via e-mail that allowed them to access the scheduled online exercise session. Participants were encouraged to sign into the online class 15 minutes before it started and, as a human subjects protection feature, they were admitted to the class by either the class instructor or project staff. After signing in, participants were instructed to (a) turn on video and (b) use the Speaker View feature in Zoom for better viewability of exercise instruction. During the class, participants in the standard Tai Ji Quan and stretching classes were encouraged to mute their audio for better audio quality. This was not done for those in the cognitively enhanced Tai Ji Quan group because of the need for interaction between the instructor and participants.
Exercise instructors trained by the first author taught all exercise sessions. All instructors were recruited from the community and had at least 10 years of teaching exercise or Tai Ji Quan experience. For consistency, in all three interventions, each session consisted of warm-up exercises, core exercises, and a brief cool-down activity. Exercise intensity in each intervention group was monitored through a subjective measure of perceived exertion, ranging from light to moderate for both Tai Ji Quan groups and light for the stretching group.
We began the study with a maximum of 8 participants per class to optimize viewability in the Zoom environment. We gradually increased the maximum number of participants in a class to 12-15 in subsequent waves of classes. We set the instructor’s teaching space to an area approximately 4 feet by 4 feet square in order to provide (a) onscreen viewability, (b) instructional clarity, and (c) effective space utilization and movement safety. On the participant end, we set an optimum physical distance of 8–10 feet for personal computer users and 6–8 feet for iPad or smart phone users for better viewability. Participants in each group were asked to have an armless chair set aside for use during exercise.
Intervention fidelity. The intervention fidelity, as part of the quality control over intervention delivery, was defined as the extent to which interventionists successfully implemented the prespecified teaching protocol . Our fidelity evaluation included (a) successful completion of the 60-minute session delivered twice weekly over a 16-week period, (b) the instructor’s adherence to the teaching and training protocol described in detail in our predeveloped teaching manual, and (c) an overall class attendance rate of ≥75% (out of the total planned 32 intervention class sessions). The items in a and b were monitored weekly by means of observing class attendance and conducting a weekly online class fidelity check. Items in a and b were also monitored by means of class recordings, which included an evaluation of the instructor’s faithfulness in implementing the key components of the interventions, involving warm-up, core exercise training, cool-down, and clarity in verbal and visual instructions. The fidelity check outcomes were as follows: (a) Pass Satisfactorily (faithful implementation), (b) Needs Attention (signs of deviations or drift and, therefore, adjustments needed), or (c) No Pass (in-service training needed) (see below).
Monthly in-service training sessions conducted by the first author were planned to assure the quality of delivery. The training involved reviewing sessions taught in the previous weeks, correcting observed interventionist drift from the prespecified teaching activities/forms, reinforcing appropriate teaching techniques and steps (i.e., demonstration, learning/practice, repetitions, emphasis of key training points), doses (exercise sets, repetitions), and quality (visual and verbal instructions and presentation). The sessions also included a brief discussion of any safety issues related to the teaching and practicing of movements in the teaching manual.
Intervention attrition and study retention
. On the basis of our prior trials [27, 28, 32], we set a satisfactory a priori intervention attrition rate (i.e., participants who dropped out of the assigned intervention class before the 16-week termination) at ≤15%. A drop from the exercise class (intervention dropout) was defined as declining to continue participating in the class exercises either by notifying project staff (via e-mail or telephone) or not responding to any class participation reminders sent by project staff. We aimed at a 90% rate of retaining and assessing randomized participants with valid outcome data (i.e., with an expected ≤10% missing or incomplete on cognitive and dual-task gait performance outcomes) during the 16-week study period. The same definition for intervention dropout was applied to determine the study dropout rate.
. To assess the exercise program acceptability for the newly developed cognitively enhanced online Tai Ji Quan intervention, participants assigned to this group were asked to complete a 5-item attitudinal scale upon completing the class. The survey asked participants to rate their opinions about the class in the areas of exercise safety, challenge/enjoyment, appropriateness of exercise intensity, helpfulness in improving brain health, and overall program satisfaction. For the acceptability evaluation by the participants assigned to our cognitively enhanced Tai Ji Quan intervention group, we specified an a priori overall satisfaction level of ≥70%.
Adverse events monitoring
. Throughout the study period, both intervention- and non-intervention-related adverse events were closely monitored and recorded by research staff and adjudicated by the principal investigator of the study. We classified adverse events in three categories: Mild (i.e., events that required no medical treatment or were non-life threatening), Moderate (i.e., events that required medical treatment but were not immediate life-threatening conditions, and Serious (i.e., events that resulted in death or were life threatening and required medical treatment, including hospitalization, or significant disability/incapacity). For all events observed or reported, we further classified them into three categories in relation to the intervention: Unrelated (an event that was reported but not directly related to participation in the intervention), Possible (an event that was observed during a class that was considered likely to be associated with participation), or Definite (an event that was observed or reported during a class and was considered directly related to participation).
Data Collection of Secondary Study Outcomes
We included a set of cognitive outcome measures ranging from global cognitive function and dual-task gait performance to several domain-specific cognitive measures. All outcome measures, which are described below, were collected remotely via Zoom.
The Montreal Cognitive Assessment (MoCA)  measures global cognitive function that encompasses multiple domains (attention, executive functions, memory, language, conceptual thinking, calculations, visuospatial abilities, and orientation). MoCA scores range from 0 to 30, with higher scores representing better cognitive functioning. The final MoCA scores were adjusted for level of education (i.e., 1 point was added to the total score of individuals with ≤12 years of education) .
The Timed Up and Go (TUG) test  was used to assess gait under single-task and dual-task conditions . During the single-task condition, participants completed a 20-foot walk that involved (a) standing up from a chair, (b) walking (10 feet forward) at normal pace to a line on the floor, (c) turning, (d) walking back (10 feet) to the chair, and (e) sitting down. The 20-foot walk was then repeated under a dual-task condition where participants were asked to walk while performing an arithmetic task (i.e., starting at the number 81 and sequentially subtracting 5 from the resulting number). No specific verbal instructions were given for prioritization of one of the walking tasks during the dual-task walking trial. Using the average score (in seconds) of each walk, a dual-task cost measure was estimated through the following methods :
Dual-task costbaseline = (dual-task walking – single-task walking) / single-task walking x 100
Dual-task cost16_weeks = (dual-task walking – single-task walking) / single-task walking x 100
Change in dual-task cost = dual-task cost16_weeks – dual-task costbaseline
Change (measured in percentage) in dual-task cost from baseline was used to define dual-task cost at week 16, with positive values indicating deteriorated dual-task performance (i.e., dual-task cost), and negative values representing an improvement in dual-task performance with respect to single-task (i.e., dual-task benefit) .
The four domain-specific cognitive function measures were Trail Making Test B (TMT-B)  (in seconds), Category Fluency for Animals  (executive function), Forward Digit Span  (attention), and Backward Digit Span  (working memory). In TMT-B, participants were asked to connect numbers and letters in an alternating progressive sequence (i.e., 1-A-2-B-3-C, and so on) and scoring was expressed in terms of the time (in seconds) to completion. For Category Fluency, participants were asked to generate the names of as many animals as possible in 60 seconds. In assessing Forward and Backward Digit Span, participants were verbally presented with a series of digits (e.g., 6, 2, 9, 7) at a rate of one digit per second and were asked to repeat them verbatim. If participants succeeded, they were given a longer list (e.g., 5, 3, 8, 1, 6). The number of digits increased by one until the participant consecutively failed two trials of the same digit span length. In the Forward Digit-Span, participants were asked to repeat the digits in the given called order whereas in the Backward Digit-Span participants were asked to repeat the digits in reverse. The maximum number of points for each of these tests was 16 .
Modifications and Test Consistency
To accommodate online assessment, modifications were made in some of the cognitive measures. Specifically, for the MoCA measure , in the Visuospatial/Executive section, participants were asked to verbally connect each letter to the corresponding number for the Trail Making task and draw both the object (e.g., cube) and clock on a piece of paper and, once completed, show them to the assessor for evaluation. In completing the “Read list of letters” task in the Attention section, the letter-tapping task was replaced by asking participants to count the number of A’s in the list. For the TMT-B measure , the hand-drawing task was replaced by asking participants to verbally alternate the letters and numbers out loud, and the assessor recorded the time to complete the task.
To assess test-retest reliability, we assessed the cognitive measures twice, using a 2-week interval. Test-retest reliability of the TUG measure was reported previously .
In addition, we conducted a study health survey that contained participants’ demographic information, health or medical conditions, and levels of leisure physical activity  and depression . The survey was collected via either online Qualtrics (qualtrics.com) or regular mail. Other information included self-reported weight, height, and blood pressure, which were obtained during an online assessment.
Descriptive Data Analysis
Participant characteristics at baseline were summarized by intervention group using mean, standard deviation, or percentage. Intervention adherence was calculated as the percentage of preplanned intervention sessions (32 total) attended by participants. Responses on acceptability were tabulated to describe participants’ ratings of intervention acceptability in the cognitively enhanced Tai Ji Quan group. Similarly, observational or reported counts of any adverse events were tabulated by severity categories and their relationship to intervention. As a feasibility study, no inferential statistical analyses were performed . As a result, we planned a sample of 60 (20 in each group) on the basis of our prior similar work .
For descriptive purposes, baseline and 16-week data related to cognitive and dual-task outcomes are described for each intervention group, with means and standard deviations. Trends data (i.e., differences between baseline and 16-week data) are presented as change scores, along with 95% confidence intervals. The intent-to-treat principle as used with all available data presented at baseline and 16 weeks. We performed a test-retest reliability analysis using the intraclass correlation coefficient (ICC) values. All data were processed or analyzed, where appropriate, using SPSS version 20 (SPSS, Inc.).