Considering Patient Diet Preference to Optimize Weight Loss: Design Considerations of a Randomized Trial Investigating the Impact of Choice

Abstract

A variety of diet approaches achieve moderate weight loss in many individuals. Yet, most diet interventions fail to achieve meaningful weight loss in more than a few individuals, likely due to inadequate adherence to the diet. It is widely conjectured that targeting the diet to an individual's food preferences will enhance adherence, thereby improving weight loss. This article describes the design considerations of a study protocol aimed at testing this hypothesis.

The study is a 2-arm randomized trial recruiting 216 medical outpatients with BMI ≥30 kg/m² followed for 48 weeks. Participants in the experimental arm (Choice) select from two of the most widely studied diets for weight loss, a low-carbohydrate, calorie-unrestricted diet (LCD) or a low-fat, reduced-calorie diet (LFD). The participant's choice is informed by results from a validated food preference questionnaire and a discussion of diet options with trained personnel. Choice participants are given the option to switch to the other diet after three months, if desired. Participants in the Control arm are randomly assigned to follow one of the two diets for the duration of follow-up. The primary outcome is weight assessed every 2-4 weeks for 48 weeks. Secondary outcomes include adherence to diet by food frequency questionnaire and obesity-specific health-related quality of life.

If assisting patients to choose their diet enhances adherence and increases weight loss, the results will support the provision of diet options to patients who desire weight loss, and bring us one step closer to remediating the obesity epidemic faced by our healthcare systems.

Introduction

Identifying effective strategies for reducing the prevalence of overweight and obesity is challenging but critical to public health [1]. The prevalence of obesity has risen dramatically since the 1960s such that now approximately 35% of Americans have a body mass index ≥30 kg/m² [2]. A variety of diet approaches have proven successful in achieving moderate weight loss in many individuals [3, 4]. Yet, most diet interventions fail to achieve even moderate weight loss in a substantial minority of individuals, even in the setting of a controlled clinical trial with considerable resources and motivated participants [5]. Difficulty with adherence to diet recommendations is commonly felt to contribute to mediocre weight loss as well as to weight regain after weight loss [5-7].

An editorial accompanying the results from the largest and longest diet intervention trial to date, the Women's Health Initiative, argued that new methods are needed to increase adherence to diet interventions, and that matching diet composition to an individual's preferences holds promise for reversing the obesity epidemic [8]. This same recommendation was repeated by authors of two prominent research trials of multiple diet approaches [7, 9]. One article concluded:

...diets that are successful in causing weight loss can emphasize a range of fat, protein, and carbohydrate compositions that have beneficial effects on risk factors for cardiovascular disease and diabetes. Such diets can also be tailored to individual patients on the basis of their personal and cultural preferences and may therefore have the best chance for long-term success [7].

There are strong conceptual reasons for providing choices to the individual to enhance adherence to dietary change and, therefore, improve obesity treatment outcomes. Self-determination theory suggests there are three fundamental human needs: autonomy (i.e., freedom to determine one's outcomes), competence (i.e., feeling capable of performing the behavior), and relatedness (i.e., feelings of social integration) [10]. Autonomy is considerably enhanced by allowing an individual to choose but is considered less often than competence and relatedness in weight loss studies. It seems intuitive that allowing patients to choose from among treatment options based on their personal preferences would improve long-term adherence to the selected treatment. It is quite possible, however, that an obese individual's food preferences are precisely the reason for the development of obesity—that is, matching a diet to one's preferences might simply lead to further overeating. Currently, many weight loss clinics offer only one dietary approach which is problematic if a patient becomes dissatisfied or fails to lose weight. Those clinics that do offer different diet options might allow the patient to choose without formally assessing food preferences and without evidence that allowing choice improves outcomes. Therefore, advising patients to consider their diet preferences when choosing from diet options deserves rigorous testing. This article describes the design and design considerations of a study protocol that was conceived to address this issue.

MATERIAL AND METHODS

Overview

The study is a 2-arm randomized controlled trial (RCT) over 48 weeks in which participants in the test arm (Choice) select either a low-carbohydrate diet (LCD) or a low-fat, reduced calorie diet (LFD) based on personal food preferences assessed by a validated questionnaire. To further enhance their autonomy, Choice participants have the opportunity to switch to the other diet after 3 months if unsuccessful or dissatisfied with their initial selection. Participants in the Control arm are randomized to the LCD or LFD for the study duration. The primary aim of the study is to examine the difference in weight loss at 48 weeks among participants randomized to the Choice arm versus participants randomized to the Control arm. Secondary aims are to examine diet adherence and obesity-specific health-related quality of life (HRQOL) differences between the two arms. Exploratory aims will examine whether the constructs of autonomy orientation, competence, and relatedness moderate the effects of patient choice and whether different diet sequences impact weight loss in the Choice arm.

Conceptual Model to Explain How Choice Might Impact Weight Loss

That patient involvement in treatment decision making may lead to optimal outcomes is predicted by self-determination theory (SDT) [10]. According to SDT, humans have three fundamental psychological needs: autonomy (freedom to determine one's outcomes), competence (capable of performing a particular behavior; also known as self-efficacy), and relatedness (socially integrated). When these needs are met, intrinsic motivation is maximized, and, in turn, outcomes (e.g., psychological states, health, persistence, and creativity) are optimized.

When patients are instructed to follow a physician's recommendations without opportunity for collaborative decision making (i.e., decreased autonomy), motivation becomes less intrinsic and more extrinsic. When this happens, psychological reactance (i.e., motivation to reestablish freedom through emotions, attitudes, or behaviors) may occur [11]. Even if reactance does not occur, and patients instead agree to follow the recommendations, their behavior is less likely to persist because it is partially or fully motivated by sources external to the self, such as rewards, deadlines, approval from others, or social norms.

Research supports the associations posited by SDT in a variety of health contexts (e.g., substance abuse, exercise, diet) albeit with limitations. For example, in a study that involved 6 months of a very low-calorie diet followed by 23 months of follow-up, greater autonomy orientation and relatedness were associated with greater intrinsic motivation, which in turn was associated with more regular attendance of the weight loss program, greater initial weight loss, and greater maintenance of the weight loss at study end [12]. Important to note, actually having choice may not be as important as having the perception of having choice. In a pilot study of 15 overweight children, although all children received the same diet, 7 of the children were led to believe that they had chosen the diet they received, whereas 8 did not perceive that they had a choice. Children who perceived having a choice lost more weight than children who did not [13].

The conceptual model for the study is based on SDT (Figure 1). Autonomy is manipulated in this trial and is enhanced in half of the participants by allowing them to choose between two substantially different diet approaches. If participants choose the diet they feel more confident they can follow, then providing choice may also enhance perceived competence. Perceived competence is enhanced in both arms when the interventionist teaches participants the key features of the respective diets, how to integrate the new diet into their daily routines, and strategies for overcoming barriers and setbacks. Relatedness is enhanced in both study arms through emotional and instrumental support derived from interactions at group counseling sessions from interactions with the dietitian and with other patients. Although individuals present at baseline with differing levels of autonomy, competence, and relatedness, we hypothesized that the Choice intervention components will serve to enhance all 3 psychological needs (as compared with only 2 in the Control arm), which in turn should lead to greater dietary adherence and greater weight loss.

Figure 1.

Determinants of Weight Loss

Rationale for Study Design

For clinical trials considering patient preferences for interventions, several designs exist [14-16]. Of these, the Wennberg design and fully randomized preference trial fit best with our research question. In the Wennberg design, participants are randomized to a choice or a no choice arm. Within each arm, some participants receive the LCD and some the LFD. In the no choice arm, diet is determined by randomization. In the choice arm, diet is determined by participant preference. One limitation of the Wennberg design is that the preferences of participants in the no choice arm are never assessed, so it is not known if imbalances in preferences exist at baseline. In the fully randomized preference trial, preferences are assessed in all participants at baseline, then participants are randomized to the LCD or LFD intervention. Within each arm, some participants will be indifferent, some will prefer the LCD, and some will prefer the LFD. A major limitation to this design is that participants with a strong preference for one diet may not consent to enroll in the study or may adhere poorly to their assigned diet if it does not align with their preference.

We have combined important components of these designs to minimize the limitations inherent in each (Figure 2). Specifically, we assess food preferences in all participants at baseline, then randomize participants to either a Choice or Control (no choice). This will allow us to examine our primary question, namely, what is the effect of allowing choice on outcomes, but will also allow us to explore whether strong baseline preferences moderate the effect of choice versus no choice. For example, participants with strong baseline preferences for one diet approach (or strong preferences for certain types of foods) may do better if they are given the choice, whereas participants who are indifferent may do better if they are assigned to a specific diet. The subsequent randomization to LCD versus LFD among those assigned to Control helps to promote balance in baseline preference for the diet approaches among participants who enroll.

Figure 2.

Study Design

Rationale for Two Dietary Choices

In keeping with current opinion, we present Choice participants with two diet options to avoid the decision paralysis that may occur when more than two options are given [17]. In addition, having two options is a more feasible study design, logistically and economically, compared with having more options. We selected the LFD and LCD because they are the foremost competitors in the current scientific literature and popular culture, have the most evidence regarding effectiveness, and are recommended for weight reduction in clinical practice guidelines [18, 19]. From the research to date, it appears that the LCD and LFD may be comparable for weight loss. Clearer, however, is that both diets work better than a participant's baseline diet, as both diets consistently result in weight loss and improvements in cardiovascular disease risk factors [4]. In addition, the diets are substantially different enough that participants can easily distinguish them, enhancing the impression of choice.

Justification for Operationalization of Choice

In this study we operationalize choice as the opportunity to choose between the two diets at baseline and the opportunity to switch to the other diet at 3 months, using a shared decision-making process. Similar to the process a patient might undergo upon entering a weight management clinic, we assess food preferences, present diet options, and then allow the patient to choose his/her preferred diet. At baseline, Choice participants review their individual results from a validated food preference questionnaire as well as sample menus and other details of the diets with trained personnel so the participants can make an informed decision. At 3 months, Choice participants meet individually with trained personnel to discuss their progress and decide whether to switch to the alternative diet.

Although we could have operationalized choice as occurring at baseline only, we felt that offering the opportunity to switch after 3 months mimics real-world clinical conditions, in which the patient and clinician would assess success with the diet after some time and then decide whether to abandon it (in favor of another diet or no diet) or to continue. The importance of considering changes in preference over time was supported by a recent multi-site study comparing the low-carbohydrate diet to the low-fat diet [20]. At baseline, preference of diet was assessed on a Likert-type scale before patients were randomly assigned to one of the two diets, independent of preference. Diet preference was then assessed periodically during the 2-year study. Importantly, over 40% of participants changed their preference from baseline during the course of the study, highlighting the need to incorporate changes in preference when testing the impact of preference on behavior.

Study Population and Recruitment

The study enrollment target is 216 patients, enrolled in four waves. Participants are recruited by advertisement in outpatient clinics that are part of a Veterans Affairs hospital in Durham, NC using flyers, pamphlets, and messages transmitted to patients via television monitors located in the clinics. Also, health care personnel may refer interested patients for potential enrollment via a consult option in the electronic medical record. In addition, patients matching inclusion criteria are identified in the electronic medical record and invited to participate in a letter signed by the study PI, using an opt-in strategy with the assumption that participant motivation would be higher than if an opt-out strategy was used.

To assess eligibility of interested patients, study staff review the electronic medical record, perform a brief phone screen, and then schedule a screening appointment if the patient remains eligible. At the screening appointment, informed consent is obtained and participants complete a medical history, weight assessment, fasting blood tests and a validated questionnaire assessing their food preferences (described below). Final eligibility is determined following the inclusion and exclusion criteria (Table 1). Individuals excluded because of blood test or blood pressure levels, or who were eligible but had schedule conflicts with the planned sessions, can be rescreened for eligibility in future recruitment waves.

Table 1.

Inclusion and Exclusion Criteria

Inclusion criteria:
• BMI ≥30 kg/m2	• Primary care provider within the health system
• Stable health by medical history and lab tests	• Reliable transportation
• Access to a telephone
Exclusion criteria:
• Age ≥75 years old	• Transplant recipient
• Serum creatinine >1.5 mg/dL in men, >1.3 mg/dL in women	• Pregnancy, breastfeeding, or lack of birth control if premenopausal
• Type 1 diabetes, hemoglobin A1c ≥12%, or daily insulin use	• Dementia, psychiatric illness resistant to therapy, or substance abuse in the past year
• Cirrhosis, chronic viral hepatitis, or acute hepatitis of any etiology	• Pacemaker or defibrillator (bioelectric impedance assessment can interfere)
• Unstable angina or positive coronary ischemia workup in past 3 months	• Inability to complete all study measures
• Blood pressure ≥160/100 mm Hg, fasting triglyceride ≥600 mg/dL; fasting LDL-C ≥190 mg/dL	• Enrollment in another research study that might affect the main outcomes
• Weight loss therapy in the past month

Randomization

The study design involves two randomizations for some participants (Figure 2). In the first randomization, all study participants are randomly assigned to either the Choice or the Control arm, with probability equal to 1/2 (i.e., 1:1 randomization). The second randomization involves only participants in the Control arm who are randomly assigned to either the LCD or LFD, with probability equal to 1/2.

Participants are randomized to the Choice versus Control arms using a computerized random number generator in blocks of size <10 (all study personnel except the statisticians are blinded to block size). A stratified randomization is done with strata defined by gender, baseline BMI (<40 kg/m², ≥40 kg/m²), and presence or not of type 2 diabetes to achieve approximate balance across the two primary treatment groups with respect to these three important variables thought to be strongly correlated with post-treatment outcomes.

Participants are considered randomized when they learn of their randomization arm at the first group visit. Randomization arm is not revealed to the participant until arrival at the initial group visit to avoid participants starting a diet prior to the onset of the intervention.

Study Procedures

Administration of Geiselman Food Preference Questionnaire

The Geiselman Food Preference Questionnaire (FPQ) is completed by all participants during screening but used to help participants in the Choice arm only to choose their diet approach. FPQ results from Control participants are not used as part of the intervention. The 72-item FPQ assesses foods that represent 6 combinations of macronutrient composition (12 items for each of the 6 macronutrient combinations, Table 2) [21]. The FPQ instructions read, “Please mark the box how much you like the following foods at this moment.” The foods listed are common sources of macronutrients in the typical U.S. diet and/or are among the favorite foods of specific populations, such as obese men and women, and Whites and African-Americans. Examples include steaks, pot roasts, burgers, French fries, ice cream, and various preparations of eggs. The 72 items are rated on a 9-point Likert scale (1=dislike extremely; 5=neutral, neither like nor dislike; 9=like extremely). The questionnaire takes approximately 5 minutes to complete. The scoring system can calculate a mean hedonic rating for each of the 6 macronutrient combinations.

Table 2.

Geiselman Food Preference Questionnaire Macronutrient Combination Categories

	High Simple Sugar	High Complex Carbohydrate	Low Carbohydrate / High Protein
High Fat	HF/HS	HF/HCCHO	HF/LCHO/HP
Low Fat	LF/HS	LF/HCCHO	LF/LCHO/HP

Legend: HF=high fat; HS=high simple sugar; HCCHO= High Complex Carbohydrate; LCHO=Low Carbohydrate; HP= High Protein; LF=low fat

The FPQ has demonstrated high test-retest reliability for the 6 cells (r=0.82-0.96, all p's<0.01). Fat preference on the FPQ also correlated highly with percentage fat kilocalories consumed by observed dietary intake (r=0.86, p<0.003) and by self-report using a food frequency questionnaire (r=0.83, p<0.006). Importantly, in a sample of 204 subjects, preferences for high carbohydrate/low fat foods and high fat/high protein/low carbohydrate foods were approximately equally distributed.

Participants who score highest in the Low Carbohydrate/ High Protein column of Table 2 are advised that their baseline food preferences fit best with the LCD. Participants who score highest in the Low Fat/ High Simple Sugar or the Low Fat/ High Complex Carbohydrate cells are advised that their food preferences fit best with the LFD. Participants who score highest in the High Fat/ High Simple Sugar and/or the High Fat/ High Complex Carbohydrate cells are advised according to their next highest scoring cell. While the FPQ information may classify Choice participants according to their food preferences, it is ultimately their choice as to which diet to follow.

Choice Arm

Initial Encounter

A group of ~30 participants randomized to the Choice arm attends an informational session led by the study dietitian and physician at the first visit (Week 0). This group is larger than the planned small groups (approximately 15) because these individuals will split into two groups (one LCD group and one LFD group) after choosing which diet to follow. After informing patients that they have been randomized to the Choice arm, patients are given an overview of the study procedures and schedule. Next, participants receive hard copies of their individual results from the food preference questionnaire and are guided in interpreting their results.

After discussing the FPQ results, the key features of both the LFD and LCD, including foods allowed and foods disallowed, are explained, and participants are provided sample menus. Participants are instructed to review their handouts at home and consider their diet options carefully over the rest of the week. Participants are asked to refrain from talking with other group participants about their choice (to prevent influencing each other) but encouraged to discuss the decision with family, friends, and health care providers. The study dietitian calls each participant the following week to answer any questions, solicit the participant's diet choice, and administer a short questionnaire to examine the factors that contributed to the participant's choice (e.g., FPQ results, diet description, input from family). Participants are then scheduled for the next small group meeting, when they learn details about the chosen diet, and after which the diet is initiated.

Group Counseling

Once participants have chosen a diet, they begin group meetings with other Choice participants who have chosen the same diet. Choice participants following either the LCD or LFD are not mixed, nor are they mixed into groups of Control participants following the corresponding diet. The group sessions run in parallel to the Control group sessions, with one distinct difference: after 3 months, Choice participants have the opportunity to switch to the other diet if unsuccessful or dissatisfied with their initial selection.

Participants who switch diets participate in an individual counseling session regarding the basic tenets of the new diet and then join the corresponding Choice group for their new diet choice for subsequent meetings. These “new” participants are provided all educational materials that have been distributed up to that point. Because altering group membership may lead to different group dynamics and changes to perceived group cohesion and social support, measures of these constructs are taken at multiple time points to be used in exploratory analyses as time-varying covariates in predicting weight loss.

Control Arm

Participants are assigned to small groups (numbering 10-15) based on their randomly-determined diet assignment (LFD or LCD). At the first visit (Week 0), Control participants are informed of their diet assignment and provided with an overview of the study procedures and schedule. An overview of the assigned diet is provided, but participants are asked to wait to start the diet until after the following visit (i.e., in 2 weeks). This way, Control and Choice participants will initiate dietary modification at the same time. From this point forward, the diet interventions for Control-LCD participants parallel those of the Choice-LCD participants, as do those for Control-LFD and Choice-LFD, except that the Control participants do not have the opportunity to switch diets.

Intervention Components Common to Both Arms

Because the only difference between the two study arms is whether participants choose or are assigned to a diet, all other intervention components after the Week 0 visit are identical in both arms.

Group Visits

Groups meet every 2 weeks for the first 6 months, then monthly for the second 6 months, with the mid-month meeting replaced by a telephone call from the dietitian. Meetings last approximately 90 minutes and start with data collection by blinded study personnel. After measurements are completed, participants move to a separate room for behavioral and supportive counseling, sharing of food choices/recipes, and exercise recommendations, led by the dietitian or other unblinded personnel. Diet-related topics for these meetings include diet-specific instructions, grocery shopping, eating away from home, dealing with social situations, recipe makeovers, and maintaining weight loss, among others, and frequently incorporate behavioral techniques (e.g., self-monitoring, mindfulness eating, planning for high-risk situations) to aid adherence. Because the spouse may be the predominant food shopper and/or preparer for the study participant, spouses (or partners) are invited to attend group counseling sessions. Physical activity topics include overcoming barriers to physical activity and to demonstrating exercises that can be performed easily at home (or work), regardless of the presence of physical disabilities. Participants are also advised of the current recommendations to strive for 30 minutes of moderate-intensity aerobic physical activity 5 days per week [22]. Participants receive a pocket calorie, fat, and carbohydrate counting guide at the diet initiation visit (Week 2) to aid with adherence to their assigned diet [23]. They are encouraged to drink at least 6 glasses of water and take a basic multi-vitamin and –mineral supplement daily. Early in the study, participants set their weight loss goal for the study period with the help of study personnel. An unblinded study physician attends meetings to assess and adjust medications (for blood pressure or blood glucose) as needed and respond to medical questions.

Telephone Counseling

The dietitian places telephone calls every 4 weeks from weeks 26 to 46. The advantages of incorporating telephone counseling are reducing patient burden (rather than having to attend in-person visits) and focusing on individual goal-setting and problem-solving. If the dietitian is unable to contact a participant for two weeks surrounding the scheduled date, that month's telephone call is skipped, which does not interrupt the flow of the intervention because every call follows up on the previous call, regardless of when it occurred. Calls typically last 10-15 minutes.

The telephone counseling incorporates principles of motivational interviewing (MI), which capitalizes on patients’ motivation to helps them overcome barriers, an essential part of self-management [24]. During each call, the dietitian asks open-ended questions following a script and uses reflective listening to help the patient establish and rank possible goals. After goals have been established, the dietitian uses open-ended questions to elicit how the patient will attempt to make the behavior change (i.e., action plans). For each action plan, patients are asked to rate their self-efficacy to achieve their goal on a 10-point scale (not at all confident to very confident). Patients reporting <7 are asked to revise their goals and action plans because people who score ≥7 are more likely to accomplish their goals [25]. Patients are asked to record their goals and action plans. During follow-up telephone calls, the dietitian assesses participants’ progress toward reaching personal goals. Patients revise or establish new goals and action plans according to their level of success in achieving the previous goals.

Dietary interventions

Using a lay press diet book [26] (provided at the week 2 diet initiation visit) and handouts (provided at every visit), details about the respective diets are provided. For the LCD, carbohydrate intake is initially restricted to 20 grams per day to induce a change in physiology such that lipolysis occurs, producing fatty acids and ketone bodies for energy use. Caloric intake is not specified, as ad libitum instruction in an LCD typically leads to spontaneous calorie reduction [27, 28]. Artificial sweeteners can be used without restriction but participants are asked to decrease intake of caffeine and alcohol to 0-2 servings daily. Carbohydrate intake is increased gradually and methodically as the participant reaches his/her weight loss goal or when cravings threaten adherence.

Due to the diuresis that occurs at LCD initiation, low-dose diuretics (hydrochlorothiazide 25 mg daily or less, furosemide 20 mg daily or less, or equivalent doses of other medications) are discontinued and higher doses of diuretics are reduced. Medications are restarted during the study if blood pressure or peripheral edema warrants. To avoid hypoglycemia, diabetes medications other than metformin are discontinued or decreased at diet initiation, depending on the baseline level of blood glucose control. These decisions are made by the study physician using an algorithm as a guide, and communicated to the primary care practitioner. Such medication adjustments are needed predominantly in the first 1-2 months of the diet.

The LFD follows the diet recommendations by the National Institutes for Health for weight loss using a book written by the American Heart Association and handouts with specific instructions and recipes for those following a low-fat diet [18, 29]. Total fat intake is restricted to less than 30% of daily energy intake, and, saturated fat will be restricted to less than 10% of daily energy intake. Cholesterol intake is restricted to less than 300 mg per day. Recommendations regarding caffeine and alcohol intake are similar to those for the LCD. The goal for total caloric intake is calculated by subtracting 500 calories per day from the maintenance energy requirement for each individual, calculated using the Mifflin-St Jeor formula for resting metabolic rate multiplied by an activity factor corresponding to an individual's activity level [30]. If a participant experiences a prolonged plateau, then energy intake may be reduced further by 300-500 kcal per day as has been recommended when a dieter achieves new, lower maintenance energy intake requirements [18]. If a participant achieves goal weight before the end of the study, energy intake will be adjusted to maintenance-level calories. For participants following the LFD, blood pressure and diabetes medications are adjusted more gradually based on follow-up measurements.

Intervention Fidelity

Intervention fidelity is maintained in several ways throughout the study and monitored by the study investigators. Lessons taught at group visits were planned in advance by the dietitian with input and training from the study investigators. Handouts for every visit were finalized prior to the study and are used to guide the content of the group meeting. A study physician attends the group meetings to monitor balance of counseling and procedures across arms, diets, and cohorts. During the study, investigators listen to a sampling of audiotapes of the dietitian's calls and, when necessary, provide additional training or suggestions to the dietitian to maximize adherence fidelity.

Outcome Measures

Study measurements are performed or collected by blinded research personnel (RAs) with the exception of the diet knowledge questionnaire, which is specific to the followed diet and therefore administered by the dietitian to maintain blinding by the RAs (Table 3). All questionnaires are completed at baseline and every 3 months, with the exception of group cohesion and diet knowledge, which are not assessed at baseline before participants have experienced the group meetings or learned about their diet. Participants who wish to discontinue the group meetings are asked to return at the 6- and 12-month time points, during which all specified measurements are made.

Table 3.

Summary of Outcome Measures^*

Outcome	Method (# items)	Scale	Reliability	Validity
Primary Outcome
Body weight	Calibrated scale	NA	NA	Correlates with health factors, HRQOL and mortality
Secondary Outcomes
Dietary adherence (% calories from fat, carbohydrate)	Block FFQ (98)	NA	Fat as percentage of energy ICC = .72	Associated with %fat and %carbohydrate intake from 24-hour recall[54]
Obesity-specific HRQOL	IWQOL-Lite (31)	1 (never true) to 5 (always true)	α =.90 to .94 on (subscales) and .96 (total); r = .81 to .88 (subscales) and .94 (total)	Associated with BMI and weight change[55]
Process Measures
Physical activity	IPAQ (27)	NA	r = .72-.88	Associated with accelerometry and fitness[37, 56]
Level of autonomy orientation	GCOS (12 vignettes)	1 (very unlikely) to 7 (very likely)	α = .75; r = .74	Associated with perceived competence in managing diabetes[39]
Level of intrinsic motivation	TSRQ-Diet (15)	1 (not at all true) to 7 (very true)	α = .81 to .84 for subscales	Associated with attendance at weight loss program and amount of weight loss[12]
Perceived competence	Self-report (4)	1 (not at all true) to 7 (very true)	α = .90	Associated with reductions in HbA1c[39]
Dietitian's support for autonomy	HCCQ-Diet (6)	1 (not at all true) to 7 (very true)	α = .82	Associated with reductions in HbA1c[39]
Spousal support for dietary change	Spousal Support for Healthy Eating (10)	5-point scale (never to very often)	α = .83 to .87	Predicts dietary goal achievement[41]
Group cohesion	GCS-R (25)	1 (strongly disagree) to 4 (strongly agree)	α =.48 to .89 (pre-test) and .77 to .90 (post-test); α = .79 (treatment midpoint) and .84 (endpoint); α = .56 (change score)	Associated with attendance at weight loss program and amount of weight loss[12]
Diet-specific knowledge assessment	LCD and LFD Knowledge Questionnaires (10)	various	NA	NA

^*All measures have a continuous distribution.

NA=not applicable; HRQOL=health-related quality of life; FFQ=food frequency questionnaire; ICC=intercorrelation coefficient; IWQOL-Lite=Impact of Weight on Quality of Life-Lite; BMI=body mass index; IPAQ=International Physical Activity Questionnaire; GCOS=General Causality Orientations Scale; TSRQ=Treatment Self-Regulation Questionnaire; HbA1c=hemoglobin A_1c; HCCQ=Health Care Climate Questionnaire; GCS-R=Group Cohesion Scale-Revised; LCD=low carbohydrate diet; LFD=low fat diet.

The primary outcome, body weight is measured at each visit (19 total visits), at approximately the same time of day, on a standardized digital scale, with participants wearing light clothing and shoes removed. Waist circumference is measured every 3 months using a non-elastic tape measure placed on the skin in a horizontal plane around the abdomen at the level of the iliac crest [18]. Blood pressure is measured twice at each visit using an automatic sphygmomanometer following guideline recommendations [31]. An average of repeated measurements is used for analyses.

Dietary adherence, one of the secondary outcomes, is assessed every 3 months with the Block Brief 2000 Food Frequency Questionnaire (FFQ), developed from the National Health and Nutrition Examination Survey (NHANES) III food intake data [32]. Despite its length (98 items, 15-20 minutes to complete), we chose the FFQ over 24-hour recall and 4-day food records because it lowers participant and research staff burden, improves completion rates, and eases data entry and analysis. To facilitate completion, the FFQ is mailed to the patient's home prior to the specified visit so that it can be completed prior to arrival. Extra copies are available at the visit in case a participant did not bring the completed FFQ.

Comparing adherence levels between the Choice and Control groups is complicated by the fact that participants will be following one of two widely divergent diets. Therefore, a summary measure of adherence was devised because we cannot simply compare mean carbohydrate (or fat) intake between the two groups. Instead, using data obtained from the FFQ, participant level of macronutrient intake (carbohydrate if following the LCD, fat if following the LFD) will be calculated followed by the absolute percentage deviation from the goal macronutrient intake. For the LFD, the goal for fat intake is straightforward; 30% or less of daily calories should come from fat. Participants who consume exactly 30% of calories from fat will have a 0% deviation whereas 33% of calories from fat would be a 3% deviation (less adherent) and 27% would be -3% (more adherent). In this latter case, we will allow negative deviation from the goal to reduce the potential for skewness of the data. The goal adherence level for the LCD is more difficult to determine because the initial goal is in absolute grams of carbohydrate, not percent of calories, and because goal level might increase as weight loss is achieved. For purposes of measuring adherence to the LCD in the proposed study, we have chosen the goal to be 10% of calories from carbohydrate, based on our prior study in which the mean carbohydrate intake ranged from 10% (SD=11.8) at 2 weeks to 14.6% (SD=13.3) at 48 weeks (for the other arm of that study, which followed an LFD, deviations from goal were similar, ranging from 31 to 33% of calories from fat). We have chosen the unit as % of calories from carbohydrate rather than grams of carbohydrate because the former adjusts for individual variation in calorie intake and can be back calculated to grams, and because we wanted units to be the same for the two diets. Adherence for both diets will be a continuous measure, with smaller deviations from goal considered to be greater adherence. Our main comparison for the adherence analysis will be based on comparing these mean deviations between the Choice and Control arms.

Health-related quality of life, the other secondary outcome, is assessed with the Impact of Weight on Quality of Life-Lite questionnaire (IWQOL-Lite) [33, 34], an obesity-specific HRQOL measure that is likely more responsive to weight change than generic measures [35, 36]. The IWQOL-Lite has a total score and 5 subscales, as indicated by a confirmatory factor analysis: Physical Function, Self-Esteem, Sexual Life, Public Distress, and Work, with higher subscale and total scores indicating lower HRQOL.

Daily physical activity is assessed with the International Physical Activity Questionnaire (IPAQ), which is indicated for use in young and middle-aged adults (15-69 years) [37]. The long version assesses activity over the past 7 days using 31 items in six domains: occupational, transport, yard/garden, household, leisure, and sitting. The IPAQ provides estimates of metabolic equivalent tasks (MET) energy expenditure and has been validated against accelerometry [37].

The following blood tests are performed by the hospital laboratory at baseline and every 6 months with the participant fasting 12 hours using standard methods: chemistry panel, fasting lipid profile, hemoglobin A_1c, and insulin. A non-fasting chemistry panel is performed at week 4 to monitor for electrolyte abnormalities and dehydration, and again at Week 14 for any participant in the Choice group who switches diets at Week 12. Results of lab tests are reported in the electronic medical record for the research personnel and the clinical providers to view. Certain participants are also asked to perform two self-monitoring biologic tests that are not used as outcomes. Participants with diabetes are instructed to self-monitor blood glucose by glucometer at least once a day to monitor for hypoglycemia. Participants following the LCD are given a supply of commercially-available urine ketone dipsticks (semi-quantitative) to use at home for feedback on their adherence to the LCD. More positive results correlate moderately with adherence at diet initiation.

Several psychological constructs are measured to examine factors potentially related to diet adherence as informed by the conceptual model. The General Causality Orientations Scale assesses the extent to which a person is oriented toward aspects of the environment that stimulate intrinsic motivation, are optimally challenging, and provide informational feedback.[38] It consists of 12 vignettes describing a typical social or achievement oriented situation followed by three types of responses--an autonomous, a controlled, and an impersonal type. Because we are only examining the effect of autonomy orientation, we only include the autonomy-type responses. The 15-item Treatment Self-Regulation for Diet questionnaire assesses the degree to which a person's motivation for dieting is intrinsic [12]. Perceived competence is assessed with a 4-item questionnaire that assesses the degree to which participants feel confident about being able to maintain a healthy diet [39]. The 6-item short form of the Health Care Climate Questionnaire-Diet assesses the extent to which patients perceive the intervention staff (dietitian, study physician) as autonomy supportive versus controlling for encouraging dietary change [40]. Instrumental spousal or other social support for dietary change is measured with the Evaluation of Social Support for Health Eating Habits Survey [41]. Two items specific to a low-fat diet were altered to correspond to the LCD for participants following that diet. The Group Cohesion Scale-Revised is used to assess group cohesion in terms of interaction and communication among group members (including domination and subordination), member retention, decision-making, vulnerability among group members, and consistency between group and individual goals [42]. Lastly, we developed diet-specific knowledge assessment questionnaires (one for each diet) to assess knowledge of the basic tenets of the diets.

Statistical Power and Sample Size

The sample size estimate for the study, n=108 subjects per group (total n=216), is based on the primary hypothesis that participants in the Choice arm will have greater weight loss from baseline to 48 weeks compared to participants in the Control arm. Our goal was to have sufficient sample size to detect an effect size as small as 0.20 for the primary hypothesis with 80% power and a type-I error rate (alpha) of 0.05. Based on previous data suggesting a standard deviation of 22 kg in weight at 48 weeks, an effect size of 0.20 translates to a minimum detectable difference of 4.4 kg in weight between the Choice arm and the Control arm. Due to the longitudinal nested study design (i.e., repeated weight measurements on patients nested within small intervention groups), clustering and with-in person correlations must be taken into account.[43] Based on previous data, we assumed an inter-cluster correlation coefficient (ICC) = 0.005, and a correlation ρ = 0.90 between baseline and 48-week weight measurements and a dropout rate of 25% by the end of the study.

Statistical Analysis

The main conclusions drawn from this trial will be based on the pre-specified primary and secondary hypotheses and will be tested with two-sided p-values at the standard 0.05 level. The primary and secondary analyses will be conducted on an intent-to-treat basis; patients will be analyzed in the group to which they were randomized, regardless of intervention adherence [44]. Statistical analyses will be performed using SAS for Windows (Version 9.2: SAS Institute, Cary, NC) and R (http://www.R-project.org).

Linear mixed models (LMM) will be used to address the primary and secondary study aims, which involve the continuous longitudinal outcomes weight, adherence, and HRQOL [45]. The model will include fixed effects of time (in weeks) and group assignment (Choice vs. Control) by time (in weeks) interaction and will also include random effects to implicitly account for two sources of variation: 1) the correlation between a participant's repeated outcome measurements over time; and 2) the correlation between participants who are in the same small meeting group. We will test the hypotheses of a differential change in outcome (weight, adherence, HRQOL) between the Choice and Control arms at 48 weeks using a linear contrast in the above model. LMM handle dropouts in a principled manner where all patients with partial data on the outcome variables are included in the analyses. However, depending on the type and scope of any missing data, we will also explore multiple imputation as a strategy to use in conjunction with our primary analytic tools [46]. As recommended in the Committee for Proprietary Medicinal Products guidelines, the primary (as well as secondary) analyses will include the race and sex stratification variables as fixed covariates in the main analytic models.

As secondary analyses, we will include changes in group cohesion and dietary knowledge as time-varying predictors to determine whether changes in these constructs have a meaningful impact on weight loss at 48 weeks. In these analyses we will follow the principals and strategy outlined by Singer and Willett (2003) [47]. In one mixed model, changes in group cohesion and knowledge from baseline to 3 months will be included as a covariate, and in another model, changes from 3 to 6 months will be included as a covariate. We will use a similar analysis plan to examine whether physical activity levels at interim time points partially explain the relationship between group assignment and weight loss at 12 months.

Our plans for examining moderators of weight loss are informed by the conceptual model (Figure 1). According to the model, autonomy, perceived competence, and relatedness are needed to optimize motivation and therefore should be positively related to weight loss, as should the ability to choose. Additionally, levels of social support present outside the intervention context should be positively related to weight loss. Therefore, we expect that participants presenting at baseline with higher levels of autonomy, perceived competence, relatedness, and spousal support will benefit more from having been assigned to the Choice arm versus the Control arm, as compared to participants with lower levels of these variables. Thus, we hypothesize that the effect of the intervention (i.e., choice versus no choice) on weight loss will be moderated by (i.e., depend on) baseline individual differences in these variables.

To test for effect moderation, we will include in the model for weight loss baseline (i.e., pre-randomization) values of each of the moderators as main effects, and all two-way interactions between the group assignment variable and the moderators. The putative baseline moderators—autonomy, perceived competence, relatedness, and spousal support —will be entered into these models on their original continuous scale. We will explore higher-ordered terms for this model (e.g., quadratic terms for the putative moderators) if suggested by exploratory analyses.

DISCUSSION

Multiple recent studies have demonstrated that several diet approaches are effective for weight loss and amelioration of risk factors [3, 4, 9, 28, 48-51], including low fat/low calorie diets, low glycemic index/load diets, high protein diets, low carbohydrate diets, and Mediterranean diets. The authors of some of these prior studies have concluded that dieters should be provided a choice of diets to enhance adherence. Surprisingly few studies, however, have explicitly tested the notion that allowing patients to choose among treatment options actually improves weight loss outcomes.

In a study with a similar design, PREFER, 176 participants with at least a moderate preference for either a traditional low-fat diet or a lacto-ovo-vegetarian (LOV) diet were randomized to a choice or no choice condition [52]. The results showed that participants who were assigned a diet lost more weight than those who could choose; no differences in weight loss were seen between the two different types of diet [53]. Our study differs from PREFER in a few key ways. First, in PREFER, each participant's preference between the two diets was assessed after an information session but prior to randomization, and to be included, participants had to express at least moderate preference for one of the diets. Then, a random sample was given the opportunity to choose their preferred diet and the rest were assigned one of the two diets. Because approximately 60% of participants randomized to the choice arm did not wish to follow the LOV diet, approximately 30% were removed from the study by a random selection process to keep the two diet groups balanced in the choice arm. In contrast, we assessed preferences for certain types of food with a questionnaire but did not elicit participants’ preferences for overall diet approach prior to randomization. During the screening process, in fact, we advised potential participants that they should not enroll in the study if they had a strong preference for one of the diet approaches after providing a brief description of each. Further, we expect an approximately equal distribution of choices between the diets offered in our study, based on data from validation of the Geiselman Food Preference Questionnaire and from piloting the choice process in our targeted patient population.

Another important difference is how we operationalized the choice process as a collaborative, informed decision-making process. After randomization to the choice arm, these participants spend their first group meeting receiving information from the dietitian and physician about the two diet options, including foods allowed, foods to avoid, and sample menus. They also receive their individual results from the validated FPQ indicating whether their food preferences fit with one of the diets. The participants are advised, however, to choose a diet based on the totality of information, not the FPQ results only. They are asked to avoid making a decision at the group meeting and to refrain from making comments that might reveal their preference or influence the decisions of other group members. Participants are also advised to seek input over the ensuing week from their family members, friends and health advisors after which the dietitian calls to answer any further questions before eliciting their choice. This process was designed to mimic a participatory decision-making process that might occur in a clinical setting. It was also designed to maximize the background information participants might need to make their decision but minimize the opportunity for participants to influence each others’ decision. Other distinct contrasts between the studies include the actual diet options (LCD vs. LOV) and the demographic mix of the participants--PREFER included 87% women and 70% White, whereas our patient population is primarily male and more racially mixed. This is important because men and minorities are underrepresented in weight loss trials.

Limitations of the study design include several anticipated procedural difficulties related to the somewhat complex design of the study. The double randomized design necessitated starting four diet groups in the same week: Choice-LCD, Choice-LFD, Control-LCD, and Control-LFD. A goal of approximately 15 participants per group meant recruiting a cohort of approximately 60 participants over a short duration (we aimed for 6-8 weeks). This required intensive advertising to patients and clinicians at the hospital and clinics, and mailing large batches of letters to potentially eligible patients. Recruitment was more strained once several cohorts were already enrolled and being followed for measurements. We also had to anticipate possible imbalance of the Choice groups after their first visit if more participants selected one of the diets. This meant that the meeting room might have to accommodate more than 15 participants, especially given that we encouraged participants to bring spouses or partners who might assist with their lifestyle changes.

There are several strengths to the design additional to the ones outlined above. For example, both the LCD and LFD were included as randomization options in the Control arm to enhance recruitment and reduce bias. If the Control arm consisted of only one diet, more participants with an underlying preference for that diet might enroll. This could then result in an imbalanced proportion of participants selecting that particular diet in the Choice arm. Having both diets in the Control arm also makes for a more valid comparison of dietary adherence between the Choice and Control arms if for some reason adherence to one diet was higher than to the other. Use of the FPQ is a strength in that it provides participants information about their food preferences that is more objective and quantitative than can be provided by simply describing the two diets and listing food choices that may or may not be preferred foods. Because the FPQ can be completed and analyzed in less than five minutes, it could be integrated feasibly into an outpatient clinical program that offers diet options.

CONCLUSIONS

Results from the current study could change the way weight loss clinics offer diet counseling to patients. If allowing the patient to choose from diet options improves weight loss, then dietitians, other clinicians, and weight management clinics should assess food preferences and offer more than one diet option to patients to enhance their chances of success. This strategy also has potential for increasing the reach and appeal of weight loss services, and improving the satisfaction and quality of life of patients seeking such services. If allowing the patient to choose from diet options does not enhance weight loss, then clinicians and researchers may need to investigate whether an ‘optimal’ diet can be chosen for people seeking dietary counseling for weight management. Regardless, the current literature supports that several dietary approaches should be included in our armamentarium for combating obesity.