A web-based lifestyle intervention program for Chinese college students: study protocol and baseline characteristics of a randomized placebo-controlled trial

This study was a randomized placebo-controlled trial, with a standard four-arm parallel and double-blinded design. The study consisted of three intervention groups (IGs) and one placebo-control group (PCG). IG-1 (PA-first arm) first addressed PA, then FVC. IG-2 (FVC-first arm) first addressed FVC, then PA. IG-3 (Simultaneous arm) addressed PA and FVC at the same time. The PCG was provided with 8-week placebo treatments, which were irrelevant to PA or FVC. The participant recruitment and online registration were conducted 1 month before the intervention start (T0), during which only sociodemographic information was collected. All participants were asked to complete questionnaires at four time-points, including baseline (at the beginning of intervention (T1), the end of 8-week intervention (T2), 3-month follow-up after intervention (T3), and 12-month follow-up after intervention (T4) (see Fig. 1).

Participants were recruited from one university in the central region of China. The recruitment information was verbally advertised to the college students by their PE lecturers in their first mandatory physical education (PE) course. In China, all college students, by law¹ are required to take PE courses. Therefore, it is feasible and advantageous to recruit participants by means of PE classes, as each PE class consists of students from different departments independent of their motivation. The nature and purpose of this program was explained to the PE lecturers (in a train-the-trainer set-up). The students were informed that there would be a health learning program, in which they could obtain health-related knowledge and earn 50 RMB (US $7.5) cash as an incentive if they participate in the study and complete the 4-wave data collection. Students were also asked to complete the online health sessions independently and not to consult with others about the contents of the health sessions. The group allocation was blinded for both students and PE lecturers. Once students expressed their interest to the PE lecturers, they were provided with the study consent inform, and were invited to complete the online registration. After registering and screening for eligibility, all qualified participants were allocated with equal probability (in a 1:1:1:1 ratio) to one of the four groups. Assignment to the four conditions was randomized and stratified for PE class type (e.g., dancing, qigong, table tennis, yoga, and Taichi). The randomization was conducted on the first page of the online health session with a permutated block design (block size was 8), according to the randomization block list generated from the backend management system of the website platform.

A randomization check indicated that there were no significant differences across the four groups at baseline with respect to the age (F _{(3, 552)} = 1.13, P = .34), gender (χ² = 4.79, P = .19), study year (χ² = 8.21, P = .51), and relationship status (χ² = 1.15, P = .76). Thus, the randomization was successful.

College students (≥18 years old) were eligible to participate in the study if they (1) were not collegiate athletes or vegetarians; (2) had no restrictions to physical mobility (e.g., heart diseases) or fruit-vegetable intake (fruit allergies or diabetes); and (3) were able to use a computer and a mobile phone, and had access to the internet on a regular basis.

In this study, the required sample size was estimated using G*Power 3.1 software. It was hypothesized that the three intervention groups would significantly increase PA and FVC relative to the control group at T2, T3 and T4. Based on the previous study [30], a maximum effect size of 0.45 on PA (Cohen’s d) was assumed. As such, 79 participants were required per group to provide 80% power with alpha of 0.05 to test the hypothesis. After factoring in drop out, around 25% of participants were added per group [28]. A total of 424 participants (106 per group) were determined to be recruited.

The intervention was guided by the theoretical backdrop of the Health Action Process Approach (HAPA) [31] and the Compensatory Carry-Over Action Model (CCAM) [15]. The HAPA model suggests that the process of behavior change can be divided into two distinctive phases: the motivational and volitional stages, during which individuals may experience a dynamic process from the formation of intention to the performance of behavior. At the beginning of behavioral change, called the motivational phase, an individual develops the intention to perform a specific health behavior. During this stage, specific crucial factors such as action self-efficacy, outcome expectancies, and risk perceptions can collectively contribute to intention formation. Subsequently, once a “good intention” is initiated, the individual enters the volitional phase. During this process, the individual benefits the most from planning (e.g., action planning and coping planning), which can bridge the gap between intention and action. Before the behavior becomes a stable habit, maintenance and recovery self-efficacies and other resources (e.g., social support) play irreplaceable roles in maintaining the behavior change and avoiding relapse. To guide the simultaneous and sequential intervention components, the CCAM was employed to support the transfer of one behavior to another.

The duration of the web-based lifestyle interventions was 8 weeks (see Fig. 1) [29, 32, 33]. For the PA-first arm, the content included first a 4-week treatment addressing PA, and a subsequent 4-week treatment addressing FVC. For the FVC-first arm, only the sequence of intervention delivery was changed, with FVC addressed first followed by PA. The simultaneous arm had the same amount of treatments for PA and FVC as the sequential arms, but addressed these two behaviors simultaneously for 8 weeks (see Fig. 2). The treatments for the three interventions groups focused on improving social-cognitive variables related to PA and FVC behavior change, including risk perception, outcome expectancies, goal setting, self-efficacy-beliefs, action planning, coping planning and social support (see Additional file 1).

Additionally, in order to facilitate the implementation and maintenance of health behavior, behavior change techniques (BCT; e.g., provide information about health consequences, provide instruction on how to perform the behavior, barrier identification, relapse prevention, prompt review of behavioral goals, facilitating social comparison; see Additional file 2) were used in the intervention sessions based on the 93-item BCT taxonomy v1 [34]. Participants were provided with two types of feedback, including individualized feedback on past behavior, and normative feedback on whether the participants met the criterion regarding the behavioral recommendations (see Fig. 3). Furthermore, in order to maximize retention rate, multiple reminder strategies were implemented. For example, the PE lecturers reminded all to click the weekly hyperlink of the health session and follow the online instructions. Meanwhile, SMS and WeChat (a prevalent mobile social media application in China) messages were distributed to the participants weekly, prior to each intervention session in order to remind students to participate in the weekly intervention (e.g., Dear student, the new health session will start tomorrow. We kindly remind you to engage in this week’s health session by clicking the hyperlink in your computer which has been sent to you via WeChat this morning. Have a nice day!).

In order to avoid the social desirability and expectation/Hawthorne effects, participants in the control group received placebo treatments which appeared in all respects to be identical to the intervention in the IGs (e.g., the intervention duration and procedure), but lacked the critical ingredients of PA and FVC treatment. In particular, participants were provided with general health information which was irrelevant for changing actual PA and FVC behaviors, such as an introduction to tourist attractions, tips on acupressure massage, and an introduction to some relaxing music and movies. All interventions were delivered on a newly updated web-site platform, through which all participants in the IGs and PCG were invited to attend the health session once per week. They were informed that the intervention would last for around 20 min per session.

Due to the randomization, only the website system could record the participants’ identity and group allocation. When students logged into the website, the system automatically linked them to the different modules according to their group allocation at baseline. With this technology, both intervention and control participants were blinded with respect to the group allocation and reminders.

The majority of the items in this study were measured via visual analogue scales (VAS) (example, see Fig. 4). VAS in web-based research has been suggested to be better than traditional Likert-type scales, because of its user friendly appearance and design, more accurate responses, and lower dropout rate. The validity of this has been supported by empirical and theoretical evidence [35].

Demographic items included gender, age, relationship, study year, and major. All data collection was completed on the newly established website with electronic questionnaires.

The data will be initially treated for missing values using the multiple imputation approach. After examining the distribution of the data, skewed data will be log-transformed and replaced with median values (interquartile range). All analyses will be performed using full intention-to-treat analyses, with scores on dependent variables for dropouts carried over using the multiple imputation method. The Independent samples t-tests, Chi-square tests and F-tests will be used to analyze differences between drop outs and completers regarding baseline characteristics. Statistical significance will be set at the 5% level (two-tailed).

To test the effectiveness of web-based interventions, linear mixed models with maximum likelihood estimation and generalized linear mixed models with weighted least square estimation will be performed. Regression estimates will be adjusted for participant differences in the number of observations contributing to the mixed models and for variances within subjects [46]. Although gender and age were balanced at baseline, the interaction effects of these variables and treatment should not be omitted [47]. Therefore, fixed effects of the linear mixed models will include testing for time (T1-T4) and treatment (IG-1, IG-2, IG-3 and PCG) effects, adjusted for baseline values, age, gender, and PE class type. In addition, − 2 log likelihood, Akaike and Bayesian information criteria will be used to assess the model fit.