Study design and participants
The study employed a cross-sectional design using a self-constructed online questionnaire and a dietary assessment tool for collecting data on in-season competitors. In-season competitors were recruited via purposive and snowball sampling from social media platforms, known physique coaches, professional colleagues, local gyms, and known in-season competitors throughout the US. To participate in the study, eligible participants had to be willing to complete the online questionnaire and four dietary recalls, be currently preparing for a bodybuilding competition, (i.e., in-season), in one of the physique divisions (i.e., bodybuilding, bikini, figure, fitness, physique, or wellness), be at least 18 years old and female, fluent in English, currently reside in the US, and have internet access. Data collection took place between July and November 2020. Interested participants were sent a link to the study website to complete an online screener via Qualtrics [20] to see whether they met the inclusion criteria to participate in the study. Those who qualified to participate in the study received a link to the online questionnaire via Qualtrics [20] and were asked to provide consent online prior to responding to the questionnaire. Participants who completed the questionnaire were emailed or texted a unique username and password to complete the four 24-h dietary recalls via Automated Self-Administered 24-Hour (ASA24®) Dietary Assessment Tool, version (2020), developed by the National Cancer Institute, Bethesda, MD [21]. Reminder text messages and emails were sent to ensure participants completed their dietary recalls. Participants were incentivized ($100) to complete the online questionnaire and four 24-h recalls.
G*Power 3.1.9.2 was used to determine the sample size for this study. Using multiple linear regression with a total of 15 predictor variables and covariates in a fixed model, and setting power at 80%, a medium effect size (f 2 = 0.11), an α = 0.05 for significance, and accounting for 10% missing data, approximately 114 participants were needed to detect a 10% increase in the variance explained by the predictor [22].
All procedures were conducted in accordance with the applicable guidelines and regulations, and study reporting conformed with the STROBE Statement for cross-sectional studies. All protocols, marketing materials, and study website were approved by Loma Linda Institutional Review Board (IRB# 5180399). This study is covered by a Certificate of Confidentiality from the National Institutes of Health (#CC-OD-20–527).
Instrument development and validation
An online questionnaire was developed to assess dietary supplement use, sociodemographic variables, bodybuilder variables, the TPB beliefs, i.e., behavioral, normative, and control beliefs, and their underlying constructs, i.e., attitude, subjective norm, perceived behavioral control and intention, in in-season competitors. Items for dietary supplement use, sociodemographic and bodybuilder information were created based on the limitations discussed in a systematic review [23], a sample dietary assessment questionnaire for bodybuilders [1], and a focus group. A 90-min in-person focus group of nine women bodybuilding competitors who met the inclusion criteria for the study helped identify and provide insight into the relevant dietary supplements, sociodemographic, and bodybuilder characteristics in in-season competitors. Details of the focus group methodology and analyses had been described previously [24]. Before incorporation into the final online questionnaire, items for the TPB beliefs section were separately developed and administered to 21 women bodybuilding competitors. Then, a content analysis was performed [25] to acquire a list of modal salient beliefs for each behavior to design the protein and dietary supplement beliefs items used in the final questionnaire.
The validation of the online questionnaire included assessing the reliability and validity of the TPB constructs; pilot-testing and content validation of the sociodemographic, bodybuilder, and dietary supplement variables (i.e., all non-TPB items); and cognitive testing of the final questionnaire. A confirmatory analysis was performed to ensure valid and reliable items were selected to assess the TPB constructs. A high degree of internal consistency was sought to ensure the TPB items selected assessed each of the constructs, [25] i.e., the attitude toward the behavior, subjective norm, perceived behavioral control and intention, for both protein intake and dietary supplement use. The Cronbach α values for the protein intake TPB items are as follows: attitude = 0.820 (strong reliability), subjective norms = 0.918 (strong reliability), perceived behavioral control = 0.717 (strong reliability), and intention = 0.880 (strong reliability). The Cronbach α values for the dietary supplement use TPB items are as follows: attitude = 0.645 (moderate reliability), subjective norms = 0.871 (strong reliability), perceived behavioral control = 0.694 (moderate reliability), and intention = 1.0 (strong reliability). Since the Cronbach α values for all the TPB items ranged from 0.65 to 1.0, indicating moderate to strong reliability, all items were retained in the final questionnaire.
Content validity for the non-TPB items was assessed using item-level and scale-level content validity indices (CVI) and the multi-rater kappa coefficient. Content validation experts [26] included individuals with subject matter expertise in sport nutrition, i.e., sports dietitians, from across the U.S. These experts were recruited via purposive and snowball sampling from known dietetic professionals and colleagues. The first iteration of content validation assessed the initial item-level CVI, and a second iteration evaluated the item-level CVI and the scale-level CVI. For the item-level CVI, each rater/expert rated each item for its relevance to the underlying construct using a four-point ordinal scale (1 = not relevant, 2 = somewhat relevant, 3 = quite relevant, and 4 = highly relevant), which was later collapsed into a binary scale (1 = quite / highly relevant vs. 0 = not / somewhat relevant). Afterwards, item-level CVI was calculated as the proportion of raters/experts rating the item as relevant on this binary scale.
To adjust for inter-rater agreement by chance, the modified multi-rater kappa was calculated using the modified kappa formula [24]. Scale level CVI was calculated as the average of all item-level CVIs for items in a given construct, e.g., sociodemographic or bodybuilder characteristics, by all raters/experts [24]. The first iteration of the questionnaire revealed 15 questions with an item-level CVI < 0.78. The second iteration revealed a scale-level CVI of 0.92 or higher and a modified kappa rating of excellent for all scales [27, 28].
The cognitive method of retrospective probing was performed to study the way women bodybuilding competitors process and respond to the non-TPB items in the online questionnaire [29]. This method enabled women bodybuilding competitors to successfully navigate the online questionnaire without interruption and allowed the researcher the ability to observe any technical difficulties that arose during the completion of the questionnaire [30]. Every competitor completed the questionnaire online prior to the retrospective probing. Retrospective probing was conducted on 2–6 women bodybuilding competitors at each 2-h cognitive testing session. This method required a total of 20 women bodybuilding competitors, i.e., bikini, figure, physique, and wellness, detecting at least 50% of the more serious problems affecting survey measurement error 50–90% of the time for our questionnaire [31]. The interviewer’s notes and audio recording transcription to the retrospective probing question’s responses were aggregated and summarized for each question to revise the questionnaire [30]. The revised questionnaire was compared to the previous version to demonstrate the revision had either fewer problems or eliminated problems within the questionnaire [29]. A total of five separate retrospective probing sessions was conducted to achieve saturation without discovering new high-impact problems [32].
The final validated online questionnaire has four main sections, which includes sociodemographics, bodybuilder, supplements, and the TPB—with a total of 549 items plus open-ended questions in the dietary supplements section. Items are primarily multiple-choice, and a few are open-ended. The TPB section has a 7-point bipolar adjective scale.
Measurement of TPB variables
The validated online questionnaire developed for this study was used to measure the TPB variables. The questionnaire items for attitude (five items), subjective norm (six item), perceived behavioral control (four items) and intention (three items) for both protein intake and dietary supplement use can be found in Supplementary Table 1 (see Additional file 1) and Supplementary Table 2 (see Additional file 2), respectively. Intention was the predictor of protein intake and dietary supplement use. The predictors for intention were attitude, subjective norm, and perceived behavioral control. In turn, behavioral beliefs, normative beliefs, and control beliefs were used to predict attitude, subjective norm, and perceived behavioral control. Attitude, subjective norm, and perceived behavioral control, for each outcome variable, i.e., protein intake and dietary supplement use, was assessed using the methods outlined in Ajzen [25]. In addition, the prediction of: (a) attitude by the product of behavioral beliefs and outcome evaluation, (b) subjective norm by the product of normative beliefs and intent to comply, and (c) perceived behavioral control by the product of control beliefs and perceived power was assessed using the methods outlined in Ajzen [25].
Measurement of dietary supplement use
In the dietary supplement use section of the validated online questionnaire, respondents are asked “Which of these dietary supplements have you used on a consistent basis during the last 12 months?”. The list of supplements includes nine vitamins/minerals, 16 for power and strength (e.g., beta-alanine and whey), 10 for weight loss (e.g., guarana), nine for endurance (e.g., caffeine), eight for immunity (e.g., antioxidants), six for joint health, and six herbals. Respondents were also allowed to list other dietary supplements that they use which are not listed in the questionnaire. Questionnaire items also asked about the frequency of intake per day during off-season, in-season, and peak week training periods, reasons for supplement use, observations about how the use of supplements affected their performance/activities, and others.
Measurement of protein intake
The dietary intake data was collected using the Automated Self-Administered 24-Hour (ASA24®) Dietary Assessment Tool, version (2020), developed by the National Cancer Institute, Bethesda, MD [21]. Protein intake was measured from the average of four non-consecutive 24-h dietary recalls (three weekdays and one weekend day). Participants were instructed to report all food, fluids, and dietary supplements they consumed from midnight of the previous day to 11:59 pm of the current day, and were provided the Participant Quick Start Guide for 24-Hour Recall using ASA24-2018 & ASA24 -2020 instructional materials to assist in completing the dietary recalls [33]. A registered dietitian reviewed and cleaned the data, including addressing known issues, based on the National Cancer Institute’s recommendations [34, 35].
Sociodemographic and bodybuilder variables
Sociodemographic and bodybuilder variables were collected using the validated online questionnaire. Sociodemographic variables included age, race, ethnicity, height, current weight, educational attainment, employment status, household income from all sources, and exercises as part of a competitor’s training protocol performed for at least 10-min over past 7-days. Exercises and associated metabolic equivalent (MET) values were obtained from the 2011 Compendium of Physical Activities to calculate total MET minutes per week, resistance training MET minutes per week, and aerobic training MET minutes per week [36]. Educational attainment, race and ethnicity questions were based on the 2020 Census [37, 38]. Employment status and household income from all sources items were adapted from the Behavioral Risk Factor Surveillance System 2020 questionnaire [39].
The following competitive women bodybuilder variables were assessed: bodybuilding divisions (i.e., bodybuilding, bikini, fitness, figure, physique, and wellness), competition status (i.e., amateur and professional), total number of years of competition, most recent top competition placing details (i.e., organization name, year, placing, professional/amateur competition, and whether professional status was awarded), competitor type (e.g., designated natural vs. all others), most recent competition weight, lowest off-season weight since last competition, highest off-season weight since last competition, total number of years competing, total number of competitions, time in weeks since the last competition, and weeks remaining till the next competition.
Data management and statistical analyses
The data collection tools were located on secure websites, with ASA24® having its own researcher site [40] from where collected data can be accessed. In the case of missing data or an unusual value being detected, the researcher attempted to contact the participant to rectify the issue. During the analyses, missing data was handled using multiple-imputation via the expectation–maximization algorithm as described by Graham [41]. Five imputations were used.
The IBM SPSS Statistics Version 28.0 (IBM Corp, Chicago, IL, USA) was used to perform all statistical analyses. All continuous variables were not normally distributed, thus they were 90% winsorized [42]. Descriptive statistics were used to describe the study’s population’s sociodemographic, training, bodybuilding, dietary intake, and supplement use. Repeated measures ANOVA with Sidak multiple comparison testing was used to determine differences between self-reported dietary supplement use across seasons while controlling for current weight, total activity, the number of days require to complete the four 24-h dietary recalls, and the completion of non-consecutive dietary recalls (yes/no).
Multiple linear regressions were performed to assess the relationship among: (a) each behavior (i.e., protein intake and dietary supplement use) with intention and perceived behavioral control, (b) intention with the TPB constructs (i.e., subjective norm, attitude, and perceived behavioral control), (c) the TPB constructs with all their respective belief items for behavioral, normative, and control beliefs. All dependent variables were linear as the standardized residuals were normally distributed and met the assumption of homoscedasticity, plus outliers were not influential as Cook’s distance scores were between -1 to + 1. Pearson’s correlations were utilized to assess the association among: (a) each behavior (i.e., protein intake and dietary supplement use) with intention and perceived behavioral control, (b) intention with the TPB constructs (i.e., subjective norm, attitude, and perceived behavioral control), (c) the TPB constructs with all their respective belief items for behavioral, normative, and control beliefs. In addition, Pearson’s correlations were run to assess potential confounders. Associations between dietary supplement use and all TPB outcome variables were controlled for age, current weight, total activity, the number of days required to complete the four 24-h dietary recalls, total number of years competing, total number of competitions, and the categorical variables completion of non-consecutive dietary recalls (yes or no) and employment status in all multiple linear regressions. Associations between protein intake and all TPB outcome variables were controlled for age, current weight, total activity, the number of days required to complete the four 24-h dietary recalls, and the categorical variables non-consecutive dietary recalls (yes or no), employment status, household income per month, education, and most recent top competition placing organization name in all multiple linear regressions. In addition, energy intake was also adjusted only in the analysis for protein intake.
Individual belief items for each behavioral belief, normative belief, and control belief were TPB main construct predicted scores created by regressing each TPB main construct, i.e., attitude, subjective norm, and perceived behavioral control, on their respective components of each belief item. The components of each belief item contain: (a) the behavioral belief, the outcome evaluation for that belief, and the interaction of those two; (b) the normative referent, the motivation to comply with that referent, and the interaction of those two; and (c) the control belief, perceived power for that control belief and the interaction of those two. An example of one regression equation in the calculation of the predicted score of attitude from the first behavioral belief item is written below with the following acronyms for behavioral belief (BB), outcome evaluation (OE), and their interaction (BB × OE):
$$\mathrm{Attitude }= {\upbeta }_{0} + {\upbeta }_{1} ({\mathrm{BB}}_{1}) + {\upbeta }_{2} ({\mathrm{OE}}_{1}) + {\upbeta }_{3} ({\mathrm{BB}}_{1} \times \mathrm{ O}{\mathrm{E}}_{1})$$
This process would be continued for all the belief items that make up each of the three beliefs, i.e., behavioral, normative, and control. Once the predicted scores from all the belief items were created, each TPB main construct was regressed on each of the predicted scores from the individual belief items contained within behavioral, normative, and control beliefs. For example, using protein intake, attitude was regressed on all 10 predicted scores from each item for behavioral belief. Significance for all analyses was set a priori at P < 0.05.
