Background:

For patients with obesity seeking weight loss, federal authorities recommend either calorie restriction of standard food choices/calorie counting (CC) or adopting the MyPlate (MyP) distillation of the Dietary Guidelines for Americans. MyP recommends increasing fruit and vegetable consumption, making half of grain choices whole grain, replacing sugary drinks with water, limiting sodium intake, and restricting empty calorie intake. Comparing the effectiveness of these 2 approaches in low-income patients to reduce excess body fat long-term is innovative.

Objectives:

Conduct a comparative effectiveness trial of the CC and MyP approaches. Primary patient-centered hypothesis: The MyP approach to weight loss will yield greater satiety (indicators: feeling hungry, meal satisfaction, feeling full). Primary medical hypothesis: Both approaches will yield similar reductions in body fat (indicators: weight, waist circumference) at 12-month follow-up. Secondary hypotheses: Mental health, satisfaction with program, and quality of life (QOL) will increase more in MyP participants than in CC participants. Systolic blood pressure will decrease more in MyP participants.

Methods:

A total of 261 study participants were randomly assigned to the CC (n = 130) or MyP (n = 131) conditions. For the 6-month intervention phase, all participants had the opportunity to participate in two 1-hour home health education sessions, two 1-hour group education sessions, and seven 20-minute telephone coaching sessions. Additionally, MyP participants were invited to attend two 1-hour cooking demonstrations. Trained bilingual community health workers delivered the interventions. The study population was predominantly low-income Latino and African American patients recruited from a federally qualified health center in Long Beach, California. They were 95% female, 86% Latino, 8% African American, and 4% White. The mean participant age was 41 years. Assessment periods were baseline, 6-month follow-up, and 12-month follow-up. Assessments included questionnaire measures, anthropometry, and food frequency questionnaires. Intervention sessions and assessments were conducted in English or Spanish, depending on participant preference. Key outcome analyses involved random intercept mixed-effects modeling of repeated measures across the 3 assessments.

Results:

Study retention was 77% at 12-month follow-up. The MyP and CC conditions both yielded improved satiety on 2 measures; only the CC condition yielded reduced hunger, contrary to prediction. Both conditions yielded reduced waist circumference for overall sample and for female and Latino participant subgroups but neither condition yielded significant weight loss. MyP yielded reduced systolic blood pressure at 6 months but not at 12-month follow-up; CC participants experienced no change in blood pressure. Both conditions yielded improvements in mental health, health-related QOL, and satisfaction with their respective weight loss program.

Conclusions:

Both intervention approaches yielded beneficial changes in satiety, QOL, and reduction in excess body fat. Patient satisfaction with the program was high in both conditions. For a predominantly low-income, Latino patient population, the simpler MyP approach to reducing excess body fat may be as efficacious as the more complex traditional calorie restriction approach to reducing excess body fat.

Limitations and Subpopulation Considerations:

Many participants missed ≥5 intervention sessions, which diminished intervention impact. Acculturation was an important moderating influence on outcomes, with the least-acculturated participants experiencing less intervention benefit than more acculturated participants.

Patient Representatives

Maria Chandler, MD, MBA, chief medical officer of The Children's Clinic of Long Beach (TCC), California, agreed that TCC would participate as a partner in the proposed comparative effectiveness trial of 2 government-sanctioned approaches to weight control. She identified 2 patient representatives familiar with TCC. One was Ms Evangelina Ramirez, a Spanish-bilingual community organizer and member of TCC's community advisory board (CAB). The other was Ms Stephanie Love, a clinic manager/vocational nurse and articulate representative of the local African American community. Their input was critical to the design of the intervention protocol. For example, at their insistence childcare was included as part of the group education offerings, to make it possible for low-income mothers to attend the group education sessions while somebody else looked after their child/children. They participated in the study's CAB meetings and were available as needed throughout the study. When accrual was lagging relative to what was expected, they were instrumental in identifying a second recruitment site and reassuring the research assistance staff that nothing about the recruitment protocol was inadvertently alienating possible study recruits. They helped address the perennial problem of study retention with their respective patient populations and were important advocates for higher incentives to boost participant completion of the 12-month follow-up assessment. Their input enabled the research staff to achieve 80.1% retention (when including 9 women who became pregnant and were therefore not included in analyses) at 12-month follow-up, thereby optimizing the study's internal validity by minimizing selection bias. They actively participated toward the end of the study in interpreting the results and suggesting practical strategies for disseminating the results to their respective communities.

Community Advisory Board

With input from Dr Chandler, the study patient representatives, and TCC's director of community outreach, the investigators identified 15 members who constituted the CAB. In addition to Dr Chandler and the patient representatives, the CAB included 15 members representing a broad cross-section of the Long Beach, California, community, including physicians; dietitians; community gardening specialists; 2 patient representatives; the local YMCA; a pastor; a health educator, and, ex officio, the University of California Los Angeles (UCLA) principal investigator. They first met in June 2014 to discuss the aims, protocol, and instruments. This CAB was reconvened in March 2017 to discuss the study's preliminary results and to assist the investigators in interpreting the results and suggesting how to build on lessons learned.

Focus Groups and Key Stakeholder Interviews

Most of the focus group participants and key stakeholders who were interviewed were identified by members of the CAB. The investigators conducted 2 focus groups, one consisting of Spanish-speaking obesity patients, the second consisting of English-speaking obesity patients. Criteria for focus group participant selection were similar to the criteria expected in recruiting study participants to ensure comparability of demographic characteristics between focus group participants and study participants. The investigators also interviewed 6 community stakeholders, none of whom were patients from TCC but who were actively engaged members of the Long Beach community. The key stakeholder interviewees included a pastor, a community activist, and other community leaders knowledgeable about the dietary practices and physical activity habits of Long Beach residents and knowledgeable about the health promotion resources available in Long Beach.

The discussion guide that directed the focus group discussion and key stakeholder interviews asked 6 questions about the interlocutors' attitudes toward food choices, 5 questions about their attitudes toward physical activity, and 8 questions about strategies to improve both the quality of their food choices and their daily level of physical activity. The discussion guide preamble stated the participants'/interviewees' answers would help the investigators design a better weight-loss intervention for TCC patients who were overweight or obese and wanted to lose some of their excess weight. The questions for the discussion guide were based on the investigators' previous experience with promotora-led, home-based dietary intervention with Latinos. The focus group discussion lasted 1 hour and took place in a community setting operated by the Long Beach Health Department. The key stakeholder interviews lasted approximately 35 minutes. All participants were rewarded with a $20 gift card following their participation. Seven of the 8 focus group participants were women; 4 of the 6 key stakeholder interviewees were women.

A Spanish-speaking bilingual, bicultural Latina graduate student well experienced in the collection of qualitative data facilitated the focus groups. The notetaker was also a bilingual female graduate student. Although the discussions were audio-recorded they were not transcribed. The content analyses used the notetaker's notes as the source of the data. When questions arose concerning the meaning of the notetaker's notes, the content analysts consulted the audio record. The analysis protocol used to sort the themes was based on the protocol described in Krueger and Casey.⁶⁶

Results suggested that the original design of the MyP community health worker–based home environment–focused lifestyle change intervention was sound but needed minor modifications. Participants in the Spanish-speaking focus group were particularly receptive to the idea of a community health worker coming into the home to advise residents about visual cues, home equipment, and family routines that could, if adopted, increase residents' adherence to federal nutrition and physical activity recommendations. While participants in the English-speaking focus group were more ambivalent about a community health worker coming into the home, when it came to specific behavioral strategies they were equally supportive. All key stakeholder interviewees supported the idea of having a community health worker making home visits to advise residents on ways to increase their adherence to recommended guidelines, some of them enthusiastically so. The overall impression from all of these data, then, was support for the general concept of the original MyP intervention design, but specific cautions about employing certain intervention strategies, such as (1) insisting on starting every day with a healthy breakfast, (2) serving only noncaloric beverages to guests, (3) having household members do gardening every week (if a garden was available), and (4) insisting that household members needed to sleep 7 to 8 hours a night.

The MyP approach resonated with focus group members as a particularly appropriate vehicle for communicating nutritional priorities to the mostly low-income, Latino TCC patient population because it requires minimal literacy and numeracy skills. The CC approach, by contrast, requires the ability to read food labels (literacy) and the ability to track and add up daily calories consumed (numeracy). Moreover, the MyP emphasis on fruits, vegetables, legumes, and nuts accorded well with immigrants raised on the traditional Mesoamerican diet of maize, beans (eg, black, pinto), and squash (eg, pumpkin, acorn squash).⁶⁷ Using bilingual, bicultural community health workers as the change agents rather than masters-level health educators was also consistent with community health practices in low-income Latino communities.^68,69

Study Design

The investigators conducted a parallel group, a randomized controlled comparative effectiveness trial comparing MyP with CC partly to help reconcile 2 conflicting messages from government-sanctioned sources about practical strategies for losing excess body weight. More specifically, should individuals engage in portion control and restricting calories from all foods (as originally recommended by the DPP¹²) or should they eat more F&V (as recommended by MyP consumer messages⁷⁰), even as they try to reduce overall daily calorie intake to match what their bodies need for daily metabolic needs? If the study was well implemented (ie, high fidelity and low study attrition) and if results confirmed the hypothesis that increased consumption of fiber-rich plant foods (as recommended by MyP/DASH) facilitated weight loss measured at 12-month follow-up, this randomized controlled trial would permit confident causal inferences about the weight-control benefit of encouraging low-income Americans to eat more fruits, vegetables, legumes, whole grains, and nuts.

Forming the Study Cohort

African American and Latino adults in the United States have the highest age-adjusted rates of obesity relative to other major ethnic groups.¹ The investigators therefore partnered with TCC, whose baseline population of adult patients was 76% Latino and 13% African American. Eligibility criteria included (1) body mass index of between 27 and 40, (2) ability to communicate either in English or Spanish, (3) aged 18 years or older, (4) willingness to change diet and exercise patterns, (5) willingness to accept randomization to either intervention group, and (6) ability to give informed consent. Exclusion criteria included (1) pregnancy, (2) major cardiac event or stroke-related medical procedure in the past 6 months, (3) prior or planned bariatric surgery, (4) use of prescription medication for weight loss in the past 6 months, (5) chronic use of medications likely to cause weight gain or weight loss (eg, antidepressants, mood stabilizers), (6) glucose control diabetes medications, (7) corticosteroids, (8) antiseizure medications, (9) β-blockers, (10) current cigarette smoking, (11) problem alcohol use, (12) psychiatric hospitalization in the past year, (13) plans to move from the area in the next 12 months, (14) unstable angina, and (15) blood pressure (BP) >160/100 mm. The most common reason for these exclusions is that patients with these conditions would have difficulty adhering to intervention recommendations.^71,72 Patients with uncomplicated type 2 diabetes could participate in the trial but only after being permitted to do so by their primary care provider. We included this last proviso at the behest of physicians who argued that patients newly diagnosed with diabetes could benefit from participation in a behavioral weight loss program and should not be barred from participation if they had not yet experienced complications from their disease.

To achieve satisfactory statistical power to detect the expected experimental difference in satiety, we relied on past literature involving use of a fruit-and-vegetable approach to facilitate weight loss. With an effect size ([mean_baseline − mean_follow-up]/mean SD) of 0.52 ([53.5 − 46.7]/13.2 = 0.52), the estimated per-condition sample size needed to detect an effect at 12-month follow-up was n = 72 {Cohen, 1992 #6905}. To have the power necessary to evaluate differences in body fat assessment at 12-month follow-up, we relied on the 3 studies cited above, which yielded per-condition sample size estimates of n = 103 to n = 135. For the proposed 2-arm study and allowing for 20% attrition at 12 months, we set the prudent sample size target at N = 300. We halted accrual at N = 261 because of slower-than-expected accrual. Accrual was slower than anticipated despite the planned contingency of slightly enlarging BMI-contingent eligibility from 30 > BMI < 40 to 27 > BMI < 40. We made this change in BMI eligibility retroactively to include potential recruits who had originally been told that their BMI was too low but whose BMI was >27.

All participants were recruited in TCC waiting rooms. Bilingual male and female research assistants approached 2086 adult patients, regardless of perceived corpulence, as they waited for their appointment with their health care provider (see CONSORT flow diagram, Figure 2). The order in which waiting room patients were selected was determined by computer-generated random numbers to ensure representativeness of the TCC patient population. Overall, 44.7% (932 of 2086) declined to be screened for eligibility for the trial, and among the 364 who appeared to be eligible, 28.3% (103 of 364) declined to participate in the trial. The most common reason for declining to participate was lack of interest (45%). Another 24% were found ineligible based on self-report anthropometric and BP information. Subsequent assessment-using objective anthropometric measures or use of a sphygmomanometer led to an additional 7% being found ineligible. Of potential recruits, 7% were unable to complete the screening, generally because registration desk staff announced that the patient's primary care provider was now available to meet with him or her. The remainder were ineligible for other reasons, such as not speaking either English or Spanish.

Figure 2

CONSORT Flow Diagram Showing Reasons for Study Attrition.

Study Setting

Although initial contact with the patient was in the clinic waiting room, most of the health education sessions occurred offsite. Two of the sessions took place in the patient's home because those sessions focused on how to make the home environment more supportive of healthier lifestyle choices. One group education session took place in the grocery store because that session focused on strategies to make typical food shopping more supportive of healthy food choices. Most of the coaching sessions took place by phone at times convenient to the study participant. Group cooking sessions took place at TCC or at community sites close to TCC.

Interventions/Choice of Comparators

It has been established that overweight patients are highly interested in receiving advice from their primary care physicians about effective lifestyle change approaches to losing excess weight.⁷³ The investigators chose to compare 2 government-recommended lifestyle change approaches to healthy weight loss with somewhat conflicting recommendations.

Follow-up

In this prospective 1-year trial, follow-up assessment occurred at 6 and 12 months after the enrollment date for most measures. At each assessment, the questionnaire and anthropometric measures took 60 minutes to complete; the addition of program evaluation questions at follow-up increased the assessment time to 80 minutes. Ideally, the 20-minute Food Frequency Questionnaire (FFQ) assessment took place concurrently with the questionnaire and anthropometric assessment, but in practice participants preferred to complete the FFQ by phone with a trained nutritionist. Food frequency questionnaire assessments were limited to baseline and 12-month follow-up. Despite its desirability, collection of food frequency data at the 6-month follow-up was thought to pose an excess burden on study participants and was considered less important than 12-month follow-up for documenting intervention-related dietary change. For most participants, exposure to all study intervention activities had ceased by the 6-month follow-up assessment. They were expected to continue adhering to their respective lifestyle change prescriptions and knew that their level of adherence would be assessed at 12-month follow-up. To accommodate unexpected interruptions in some participants' lives, make-up intervention sessions were permitted for several weeks after the 6-month assessment.

Study Outcomes

The conventional primary outcome in previous trials of clinic-based weight-loss interventions has been body weight.^16,17,80,81 Successfully reduced body weight achieved at 12-month follow-up has not been enough to sustain a healthier body weight for 4 years or more,⁸² in part because the calorie restriction approach has been accompanied by increased hunger relative to a fruit- and vegetable-supplemented approach.²² For dietary changes to be sustained for a lifetime, not only does the excess weight need to be lost but the successful weight loss regimen also needs to leave the patient feeling satisfied after each day's meals.^83,84 Hence, the investigators chose to include the hunger scale used in previous research²² as well as 2 additional questions about meal satisfaction and a feeling of fullness⁸⁵ as primary end points. We evaluated the choice of terms for assessing these facets of the satiety construct using cognitive interviewing techniques to ensure study participants correctly comprehended the meaning ascribed to these terms by the investigators. These terms were also vetted by focus group participants, the patient representatives, and members of the CAB.

The published intervention trial that comes closest to the study proposed here was a comparison between a standard low-fat energy restriction (only) approach compared with a low-fat energy restriction approach accompanied by encouragement to consume more F&V. A key finding in this study was that the F&V intervention yielded greater 1-year weight loss but significantly less daily feeling of hunger than the more conventional low-fat energy restriction approach.²² This trial used “How hungry did you feel today?” The trial also used a Visual Analogue Scale (VAS). The VAS consisted of a 100-mm line anchored at either extreme by “not at all hungry” and “extremely hungry.” Participants placed a hash mark on the line that represented the level of hunger they remembered having experienced after the preceding day's last meal. We scored each VAS by measuring the distance from the left end of the line to the participant's hash mark.^86,87 In this trial, the 3 satiety items were prefaced by “Take a moment to remember the last meal you ate yesterday.” The wording of the hunger item was “Thinking about yesterday, how hungry did you feel during the day?” The VAS scale was anchored by “Not at all hungry” on the left and “Extremely hungry” on the right. The wording of the meal satisfaction item was “Thinking about the last meal you ate, how satisfied were you after you ate that meal?” The VAS scale was anchored by “Very satisfied” on the left and “Very unsatisfied” on the right. For analysis purposes, this scale was reverse-scored, so high scores connoted satisfaction. Finally, the fullness item was “Thinking about the last meal you ate, how full did you feel after you ate that meal?” The VAS scale was anchored by “Completely full” on the left and “Not at all full” on the right. For analysis purposes, this scale was reverse-scored, so high scores connoted fullness.

We originally assumed that these 3 measures of satiety would be sufficiently similar to justify including them in a single scale, to avoid the problem of inflation of type I error associated with multiple hypothesis testing, but the internal consistency (Cronbach α = .43) was unacceptably low. However, all 3 measures have been used in nutrition research to represent the satiety construct to good effect,^22,85,88 so we retained all 3. To correct for the inflation of type I error in multiple comparisons,⁸⁹ we used the Bonferroni correction to set the nominal critical P value to .0167 instead of .05.

While the hypothesis was that the MyP diet, with its doubling of F&V, would yield greater satiety and reduced feeling of daily hunger than the DPP-like diet, a confounding contributor to feeling hunger is meal skipping, both voluntary and involuntary. The lifestyle change coaches were trained to encourage breakfast eating in both conditions and to discourage meal skipping. Patients dependent on government food assistance may also experience periods of involuntary hunger. Two questions about food insecurity were asked of all participants and used as covariates to help control for the hunger-generating effects of periodic meal skipping. In this study, food insecurity means a household-level economic and social condition of limited or uncertain access to adequate food.⁹⁰ The specific questions were the following: “In the last 12 months, did you ever eat less than you felt you should because there wasn't enough money to buy food?” “In the last 12 months, were you ever hungry but didn't eat because you couldn't afford enough food?” The answer options were “yes,” “no,” “refused,” or “don't know.”

Primary Patient Medical Outcome Included 2 Indicators of Body Fat Composition: Weight and Waist Circumference

Anthropometric measures of body fatness are conventionally used to assess the impact of clinic-based weight-loss interventions.^16,17 Weight (in kg) was measured at each assessment in the clinic setting. Weight in light indoor clothes without shoes was recorded by trained, certified staff using a high-quality digital scale (Tanita). Duplicate measurements were made to ensure reliability. Weight was measured in pounds for ease of interpretation by the participants and subsequently converted to kgs for data analysis. Scales were calibrated weekly using Troemer standardized weights (Thorofare, NJ). The weight at screening/baseline determined eligibility (27.0 ≤ BMI ≤ 40.4). The difference between body weight obtained at screening/baseline and 12-month follow-up was the primary patient medical outcome.

Although easily measured by patients, body weight is an imperfect gauge of metabolically significant body fat because it can be influenced by exercise-induced hypertrophy of lean body tissue.⁹¹ Waist circumference is arguably a better reflection of abdominal fat, which is a more consistent risk factor for metabolic disease than subcutaneous fat.^92-95 Participants' waist circumference was therefore measured at each assessment. Waist circumference (in cm) was measured by trained staff using an anthropometric measuring tape (Gulick anthropometric tape) at a horizontal plane around the abdomen just above the uppermost lateral border of the right iliac crest (ie, the top of the hip bone).⁹⁶ Obesity cut points of 88 cm (women) and 102 cm (men)⁹⁶ were considered cut points separating those at risk of obesity-related disease from those not at risk.

Data Collection and Sources

Participants were typically phoned a week before the due date to remind them about their 6- or 12-month follow-up assessments. If they were unreachable by phone, a postcard was sent to them, asking them to call one of the research assistants. After securing permission from the IRB and participants for HIPAA-consistent access to their medical records, clinic records yielded information about patients who had moved away. More specifically, our strategy was to work with clinic front office staff to inquire about information relevant to patient retention (eg, patients' last clinic visit, updated contact information such as telephone number or home address). From prior research, it was known that maintaining regular contact with study participants optimizes retention. The nature of the 2 interventions being compared here—namely 11 contacts over 6 months—provides a sufficient frequency of contact that retention is likely to be high. Reminder calls and birthday cards were sent periodically to remind participants they were considered continuing participants in the study until completion of their 12-month follow-up assessment, even if they had been unable to complete all proposed health education sessions during the intervention period or if they did not complete any sessions at all. Patients were considered adherent to the program if they were reachable for their scheduled health education session within 5 contact attempts made by their community health educator. If more than 5 contact attempts were made for a session then patients were labeled as hard-to-reach. To ensure good participation at the baseline, 6-month, and 12-month assessments, the investigators employed additional strategies to facilitate patient retention. Some were providing $20, $30, and $50 incentives, respectively⁵³; having full-time staff members available during all clinic business hours; and making the clinic the primary workspace. If a patient came to the clinic for a medical appointment, staff was available to reach the patient and update contact information. Patients were offered the option of meeting at their preferred time to complete follow-up assessments (eg, after work, after school, on the weekends) and at a location of their preference in their community if unable to attend the clinic (eg, local coffee shops, their homes, community centers). All data were coded by a unique study identifier specific to each participant to ensure linkage of disparate sources of data, including participant questionnaires, research assistant records of anthropometry and BP measures, and food frequency questionnaires. Assessment research staff were blind to participants' experimental assignment at the time of assessment.

Analytical and Statistical Approaches

We conducted most of the cross-time analyses as random-intercept mixed-effects models because, in contrast to repeated measures analysis of variance, parameter estimates generated by mixed-effects modeling are robust in the presence of data missing at random.¹¹³ Cases with no missing follow-up data did not differ from study “dropouts” on either baseline demographic or outcome measures (all P > .15) except for the demographic age. The mean age of study dropouts was 37.6 (95% CI, 34.5-40.7) years compared with 43 (95% CI, 41.5-44.6) years for cases with no missing follow-up data. Age was not a significant correlate in any of the regression models showing significant changes over time and is probably not a confounding influence on observed experimental effects. For intent-to-treat analyses, we also analyzed missing primary and secondary outcome measures at follow-up by carrying forward previous values under the assumption that those participants who were missing at follow-up had not benefited from exposure to the intervention since the past assessment and therefore had experienced no (additional) improvement in their satiety or body fatness measures. As an alternative to the last-observation-carried-forward approach to imputing missing data, we used a regression approach assuming multivariate normality to impute missing data on the primary outcome measures and found a similar pattern of findings.¹¹⁴ For all regression analyses, covariates included age, sex, race/ethnicity, educational attainment, and participant marital status as standard demographic covariates used in comparable studies.

Conduct of the Study

A revised final study protocol was previously submitted. We conducted most of the trial as envisaged in the original protocol. Changes included expanding eligibility to include patients with a BMI between 27 and 30, not only those with a BMI between 30 and 40 as originally planned. This change accommodated patient interest in participating in the study and to facilitate accrual. Another change increased the incentive at 12-month follow-up from $40 to $50 to facilitate participant retention at the final assessment. A third change included access to patients' medical records to facilitate follow-up and provide information corroborative of weight status and health conditions in the event the patient was missing at follow-up. The informed consent form was revised to reflect these changes and approved by the IRB before we implemented the changes.

Baseline Characteristics

Table 2 includes participant baseline characteristics for the 261 participants who were fully enrolled in the trial. The table shows no statistically significant differences between experimental conditions on any demographic characteristics on the primary or secondary measures listed. The proportion of African American participants was only 8%, well below the estimated 13% expected based on publicly available TCC demographic statistics. The proportion of Latino participants was 86%, well above the estimated 78% expected. Initial estimates were that a third of participants would be men but, as is commonly observed in community-based weight loss interventions,¹¹⁵ men were underrepresented, with male participants composing just 5% of the study sample. We tested all major hypotheses for all participants as well as within subgroups defined by ethnicity (Latino participants only) and by gender (women only). Because women were such an overwhelming percentage (95%) of participants, results for women only were generally similar to results involving the full sample. Similarly, because Latinos were 86% of the total sample, results for Latinos rarely differed from results for the full sample. Unless otherwise highlighted, the findings of significance obtained for the sample as a whole were also obtained for the 95% of the sample who were women and the 86% of the sample who were Latinos.

Table 2

Descriptive Characteristics of the Baseline Sample (N = 261).

Intervention Exposure

Table 3 shows 44 study enrollees (17%) were exposed to none of the intervention sessions despite saying during eligibility screening that they were interested in participating. Of these enrollees, 51% were exposed to at least 6 sessions and 13% participated in all 11 sessions. Of the 3 types of sessions, the home visit education sessions were the most popular (93% participated), the group education sessions were the least popular (76% participated), and the phone coaching calls were intermediate (participants completed about half of the telephone coaching sessions). Whether they participated in a few sessions or many sessions, more than 92% said they were either somewhat or very satisfied with the weight-control program to which they were assigned. There was no difference in participants' high level of satisfaction with the program between the 2 experimental conditions. We assessed the variability in exposure to the planned intervention content as a moderating influence on the outcome and to gauge its threat to the internal validity of the study.

Table 3

Number of Patients by Number of Sessions Completed (11 Possible Sessions), Los Angeles Area Federally Qualified Health Center, 2015-2016.

Primary Outcomes

The primary patient-centered outcome was the satiety construct, represented by 3 measures of satiety measured on 100-mm visual analog scales: (1) How hungry did you feel (yesterday)? (2) How full did you feel after your last meal (yesterday)? (3) How satisfied were you after your last meal (yesterday)? We regressed these 3 indicators onto standard demographic variables and set the critical P value to .017 to correct for multiple comparisons using the conservative Bonferroni correction.⁸⁹ The demographic variables were those typically included in other clinical trials of behavioral weight-control programs and included sex, age, ethnicity, educational attainment, and whether the participant living with or without a partner. We reran the models including 2 food insecurity questions that asked about cutting back on food because of lack of money. Although these questions had a correlation of 0.57, we decided to include both as covariates rather than assuming they reflected the same underlying construct. There was a marginally insignificant reduction in reported hunger in the MyP condition (mean difference = −5.99; 95% CI, −11.64 to −0.34; P = .04) but unexpectedly a significant reduction in reported hunger within the CC condition (mean difference = −9.99; 95% CI, −15.73 to −4.25; P = .004) (see Figure 3). The addition of the 2 food insecurity covariates decreased the magnitude of the reduced hunger over time in both conditions but the reduction in hunger from baseline to 12-month follow-up remained significant for CC participants (mean difference = −9.57; 95% CI, −15.32 to −3.82). Figure 3 illustrates the monotonically declining levels of reported hunger experienced the previous day. Similar trends were observed for the sample excluding men and for the sample excluding non-Latinos.

Figure 3

Experimental Effect on Everyday Hunger as Measured on a 100-mm Visual Analogue Scale.

Comparisons between experimental conditions in estimated mean changes are also given in Table 4, showing no statistically significant difference between conditions.

Table 4

Comparison Between CC and MyPlate Conditions in Estimated Mean Change in Primary Patient-Centered Outcomes Over 12-Month Period in the Intention-to-Treat Population.

Figure 4 illustrates the significant increase in reported satiety as measured by meal satisfaction reported not only by MyP participants (difference = 16.58; 95% CI, 9.54-23.63) but also by the CC participants (difference = 13.79; 95% CI, 6.65-20.94). The MyP result was predicted; the CC result was not. Results were similar for the women and Latino subsamples. Comparisons between conditions in estimated mean changes are also given in Table 4, showing no difference between conditions.

Figure 4

Experimental Effect on Satiety (Meal Satisfaction) Using a 100-mm Visual Analogue Scale.

Figure 5 illustrates the significant increase in reported satiety as measured by “feeling full after last meal” reported not only by MyP participants (difference = 12.54; 95% CI, 6.05-19.04) but also by the CC participants (difference = 8.69; 95% CI, 2.10-15.28). Results were similar for the women and Latino subsamples. Comparisons between conditions in estimated mean changes are also given in Table 4, showing no difference between conditions.

Figure 5

Experimental Effect on Satiety (Feeling Full) Using a 100-mm Visual Analogue Scale.

Primary Medical Outcomes

With 2 primary medical outcome measures, we set the critical P value at .025 to correct for multiple comparisons, using the conservative Bonferroni correction.⁸⁹ Table 5 reports between-condition differences in within-condition weight difference scores observed from baseline through 6- and 12-month follow-up.

Table 5

Comparison Between CC and MyPlate Conditions in Changes Over Time in Estimated Mean Waist Circumference, Body Weight, and BMI Over 12-month Period in the Intention-to-Treat Population.

MyP participants experienced a reduction in body weight of a third of a kg (difference = −0.37; 95% CI, −1.30 to 0.56) from baseline to 12 months; the corresponding CC reduction was three-quarters of a kg (difference = −0.74; 95% CI, −1.72 to 0.23). These statistically insignificant results were replicated for the women-only and Latino-only subsamples, and are illustrated in Figure 6. Because of consistent evidence that acculturation to US dietary practices increases immigrants' risk of obesity and because 82% of baseline respondents were foreign-born, we explored the possibility that acculturation was a moderating influence on the outcomes. When participants were stratified by acculturation tertile, the middle tertile among CC participants only experienced a 2-kg decrease in body weight from baseline to 12-month follow-up (difference = −2.11; 95% CI, −3.78 to −0.44; P = .01). The body weight decline for this group at 6 months was similar to that observed at 12 months (difference = −2.00; 95% CI, − 3.64 to −0.37; P = .02) (see Figure 7).

Figure 6

Experimental Effect on Body Weight.

Figure 7

Experimental Effect on Body Weight Only for Participants in the Middle Tertile of Acculturation.

As illustrated in Figure 8, MyP participants enjoyed a nearly 2-cm reduction in their waist circumference (difference = −1.90; 95% CI, −3.29 to −0.50; P < .01). CC participants also experienced a significant reduction in their waist circumference from baseline to 12-month follow-up (difference = − 1.67; 95% CI, −3.11 to −0.23; P = .02). Ideally, all waist circumference measures would have been taken against the skin, not over clothing. In practice, most (73%) of the waist circumference measures were taken over clothing due to patient preference because most of the participants were female and often the research assistant taking the anthropometric measures was male. The additional length of measuring tape required to accommodate the clothing (mean increased waist circumference = 3.34 cm; 95% CI, 1.71-4.98 cm) could have introduced systematic error inasmuch as it was the participant's decision, not the investigator's decision, to opt for their waist circumference being measured over clothing. In other words, the 3.34-cm difference may have reflected other, unmeasured differences between the groups. The waist circumference analyses were therefore redone, subtracting 3.34 cm from the waist circumference measures taken over clothing for those participants whom the research assistants recorded as having the waist circumference measure taken over clothing. The resulting 12-month follow-up difference scores increased in absolute magnitude from 58% to 113% (MyP difference = −3.18 cm; 95% CI, −4.55 to −1.80; P < .001; CC difference = −2.70 cm; 95% CI, −4.13 to −0.71; P < .001). For sensitivity analysis purposes, differences of 1 cm, 2 cm, and 3 cm were also evaluated, with similar patterns of results intermediate between 0 correction and 3.34-cm correction of the waist circumference measures. These results were replicated in the women- and Latino-only subsamples. Table 5 reports between-condition differences in within-condition waist circumference difference scores observed from baseline through 6- and 12-month follow-up.

Figure 8

Experimental Effect on Waist Circumference Using Measures Unadjusted for Some Respondents Having Had Their Waist Measured Over Light Clothing Instead of Against the Skin.

While mixed-effects modeling generates parameter estimates that are robust in the presence of data missing at random, another way to address the potentially confounding issue of attrition-related selection bias is to conduct an intent-to-treat analysis. In the obesity field, a common strategy for imputing missing outcome data is to carry the last observation forward under the assumption that study dropouts are less likely to have lost weight in the interim than continuing participants.¹¹⁴ Making the assumption that dropouts had failed to reduce their waist circumference relative to their baseline weight systematically biases follow-up results in favor of the null hypothesis. Nonetheless, after imputing the missing waist circumference data this way, the cross-time MyP and CC results remained significant (P < .025) for the full-sample, women-only, and Latino-only analyses.

We observed a significant decline in systolic BP for MyP participants at 6-month follow-up, from 123 mm to 120 mm (difference = −3.08 mm; 95% CI, −5.61 to −0.54) but not for CC participants, whose systolic BP dropped only 1 mm, from 123 mm to 122 mm (difference = −1.07; 95% CI, −3.75 to 1.61) (see Figure 9). By 12 months, the decline in systolic BP was statistically insignificant for both conditions (MyP difference = −1.81 mm; 95% CI, −4.46 to 0.82; CC difference = −1.04; 95% CI, −3.77 to 1.70). We observed no decline in diastolic BP for participants in either condition.

Figure 9

Experimental Effect on Systolic BP Values Over Time Was Significant Only at 6 Months and Only for MyP Participants.

Intervention Check

Using the MyP icon (at www.choosemyplate.gov), the community health workers in the MyP intervention stressed the importance of filling half of one's plate with (minimally processed) F&V. CC participants were also encouraged to consume more F&V but only because of their low energy density. All participants answered questions about how much of their average plate they filled with F&V. The respondents were given 5 choices for indicating how much of their usual plate was filled with fruits, vegetables, or whole grains. The choices were (1) none, (2) a quarter plate, (3) a half plate, (4) a three-quarters plate, and (5) a whole plate. The marginal mean at 12-month follow-up for the MyP condition was 2.98 (approximately a half plate); the marginal mean at baseline for the MyP condition was 2.37 (closer to a quarter plate than to a half plate), for a difference of 0.61. MyP participants significantly increased the proportion of their plate they devoted to vegetables over time (difference = 0.61; 95% CI, 0.41-0.82), as did the CC participants (difference = 0.42; 95% CI, 0.22-0.63) (see Figure 10). In plate surface percentage terms, these represented 15.4% (from 32.9% to 48.4%) and 10.6% (from 34.1% to 44.7%) increases for the MyP and CC conditions, respectively. They reported a corresponding increase in plate space devoted to fruit as well (MyP difference = 0.62; 95% CI, 0.41-0.82 vs CC difference = 0.40; 95% CI, 0.20-0.60) (see Figure 11). In plate surface percentage terms, these represented 13.9% (from 30.5% to 44.4%) and 9.2% (from 31.8% to 41.0%) increases for the MyP and CC conditions, respectively. Analysis of the women-only and Latino-only subsamples yielded similar results.

Figure 10

Experimental Effect on the Proportion of the Typical Plate That the Participant Reports Devoting to Vegetables.

Figure 11

Experimental Effect on the Proportion of the Typical Plate That the Participant Reported Devoting to Minimally Processed Fruit (No Juices).

Internal Validity

Primary outcome results were moderated by exposure to the intervention. We documented participant exposure to intervention sessions. We used a prespecified categorical variable to represent an ordered classification of exposure to intervention sessions with 0 representing those participants who had participated in 0 sessions (17.3%), those who participated in 1 to 5 sessions (32.6%), and those who participated in 6 to 11 sessions (50.2%). The a priori expectation was that exposure to 0 sessions should be associated with the least change in satiety; exposure to more than half of all sessions should be associated with the greatest change in satiety; and exposure to some but not more than half of all sessions should be associated with an intermediate change in satiety. All 3 indexes of satiety were significantly related to intervention exposure, controlling for participant age, gender, educational attainment, ethnicity, and marital status. For example, participants' feeling of hunger during the previous day did not change significantly for either 0 sessions (−4.53; 95% CI, −9.53 to 0.47; P = .08) or 1 to 5 sessions (−2.28; 95% CI, −5.20 to 0.63; P = .12) but did change significantly for 6 to 11 sessions (−3.64; 95% CI, −5.83 to −1.44; P = .001). Similarly, participant meal satisfaction did not increase significantly for 0 sessions (4.57; 95% CI, −1.47 to 10.62; P = .14); did increase somewhat for 1 to 5 sessions (6.3; 95% CI, 2.74-9.87; P = .0005); and increased the most for 6 to 11 sessions (7.4; 95% CI, 4.71-10.11; P = .0000). Changes over time in waist circumference were also significantly related to intervention exposure, controlling for the usual covariates. Participants who had participated in 0 sessions experienced no decrease in waist circumference (−0.06 cm; 95% CI, −1.65 to 1.54; P = .94); those who participated in 1 to 5 sessions experienced some decrease in waist circumference (−0.68 cm; 95% CI, −1.48 to 0.11; P = .09); and those who participated in 6 to 11 sessions experienced the greatest decrease in waist circumference (−1.03 cm; 95% CI, −1.57 to −0.50; P = .0002).

Health Related QOL and Mental Health

In theory, the high-satiety approach of the MyP approach should lead over time to a lower sense of deprivation and hunger during active weight-control efforts than traditional calorie restriction approaches and therefore lead to enhanced health-related QOL and lower risk of depressiveness. Thus, we included the SF-12 health-related QOL scale and the Mental Health Index-5 mental health scale. Both were acceptably reliable (baseline SF-12 Cronbach α = .79; baseline MHI-5 Cronbach α = .76) Measures on both scales improved significantly for both MyP and CC participants by 12-month follow-up in all samples (MyP SF-12 difference = 0.17; 95% CI, 0.07-0.26; CC SF-12 difference = 0.22; 95% CI, 0.12-0.31; MyP MHI-5 difference = 0.31; 95% CI, 0.14-0.48; CC MHI-5 difference = 0.33; 95% CI, 0.16-0.51) (see Figures 12 and 13). There was a near-significant interaction effect for improvement in health-related QOL at 6-month follow-up favoring the CC condition (β = 4.50; 95% CI, −0.67 to 9.67; P < .09), but the QOL ratings for the 2 intervention conditions converged at the 12-month assessment. Mental health ratings steadily increased for both conditions, with no appreciable difference between the 2. Similar results were obtained for the women-only and Latino-only subsamples.

Figure 12

Experimental Effect on Mental Well-Being as Measured by the Mental Health Index-5 Items.

Figure 13

Experimental Effect on Health-Related QOL as Measured by the SF-12, Composed of 12 Items That Assessed Social and Physical Function as Well as Psychological Health.

Physical Activity

Advice to increase daily physical activity to at least 30 minutes of moderate to vigorous physical activity most days of the week was given to participants in both conditions. Participants in both conditions received a “gym in the bag” that included a 10-minute “Instant Recess” DVD featuring fun dance routines, resistance bands, a pedometer, and charts with which to monitor progress. None of these strategies for stimulating increased physical activity seemed to have had much impact, either on self-reported physical activity using the International Physical Activity Questionnaire— short version (see Figure 14) or on a proxy measure for physical fitness, namely research assistant– assessed heart rate (see Figure 15).

Figure 14

Experimental Effect on Daily Physical Activity as Assessed by Items From the International Physical Activity Questionnaire—Short Form.

Figure 15

Experimental Effect on Heart Rate, Measured as Beats Per Minute After at Least 5 Minutes of Sitting in a Chair, at Rest.

Acculturation

To examine the possible moderating impact of acculturation on change in waist circumference, especially in the Latino subsample, study participants completed 7 language preference questions used in previous research¹⁰⁸ that together yielded an acculturation scale with high reliability (Cronbach α = .94). We subjected these items to a principal components analysis with 1 factor summarizing the shared variance. We then used this factor to evaluate the impact of acculturation on study outcomes. For some analyses, we categorized this acculturation factor into tertiles. Figure 16 illustrates for MyP participants only the higher risk of central obesity faced by the most acculturated but also shows a significant decline in waist circumference over 12 months in the most acculturated (difference = −3.50 cm; 95% CI, −6.17 to −0.83) in contrast to no significant decline in the less acculturated MyP participants over 12 months. We observed no corresponding decline in CC participants.

Figure 16

Patient Acculturation Level Moderates Change in Waist Circumference Over Time.

TV Watching

At baseline participants in both conditions reported watching TV an average of fewer than 2 hours a day. Reducing time spent watching TV was particularly encouraged in the MyP condition but, contrary to prediction, TV watching declined in both the MyP condition (difference = − 0.88; 95% CI, −1.18 to −0.58) and the CC condition (difference = −0.45; 95% CI, −0.75 to −0.14). The reduction was equivalent to 37 minutes fewer watching TV per day in the MyP condition and 32 minutes fewer watching TV per day in the CC condition. The condition by assessment interaction was significant (P = .04) and is illustrated in Figure 17.

Figure 17

Experimental Effect on Amount of TV Watching Per Day.

Family Social Support for Healthy Eating

The scales were formed from the mean of the dummy values for the available items and generally varied between 2 and 4, with the higher numbers denoting higher levels of family support for healthy eating. We found the scale scores were found to be normally distributed and were therefore not subjected to transformation or dichotomization.

Both experimental conditions succeeded in boosting social support for healthy eating for the duration of the study, as depicted in Figure 18, and did not differ from each other. Mean social support for healthy eating ratings increased from 3.08 to 3.36 in the MyP condition and from 3.11 to 3.28 in the CC condition.

Figure 18

Experimental Effect on Family Social Support for Healthy Eating.

Family Social Support for Increased Physical Activity

The scales were formed from the mean of the dummy values for the available items and generally varied between 2 and 4, with the higher numbers denoting higher levels of family support for leisure time physical activity. Family social support for leisure time physical activity increased only in the MyP condition for the first 6 months, from 2.82 to 3.06, which remained elevated through 12-month follow-up (mean, 3.10). We observed an insignificant increase in social support in the CC condition, from 2.82 to 2.95 at 6-month follow-up, remaining insignificantly elevated at 2.95 at 12-month follow-up, as depicted in Figure 19. Family social support for leisure time physical activity did not differ between conditions.

Figure 19

Experimental Effect on Family Social Support for Exercise.

Food and Beverage Choices Paralleling Decline in Waist Circumference

The FFQ¹⁰⁶ was administered to participants at baseline and 12-month follow-up but not at 6-month follow-up. Hence, analyses including the food and beverage consumption data from the FFQs necessarily ignored the 6-month data. However, questions about sugary beverage consumption on the main questionnaire overlapped with questions on the FFQ. For these questions it was possible to model changes at 6 months and 12 months in consumption of sugary beverages.

Main Questionnaire Food and Beverage Choices

As Figure 20 illustrates, sugary beverage consumption dropped 50% in the MyP condition from an initial fourth-fifths drink per day to two-fifths drink per day and 50% in the CC condition from an initial one and one-sixth drinks per week to drinks per day between baseline and 6-month follow-up but then reverted to baseline in the MyP group by the 12-month follow-up. Conversely, as Figure 19 illustrates, plain water consumption increased in both conditions, increasing 14.9% (linear effect chi square[1] = 19.10; P < .0001) from an initial 4.24 times/day to 4.9 times/day in the MyP condition and 8.7% (linear effect chi square[1] = 6.65; P < .01) from an initial 4.41 times/day to 4.80 times/day in the CC condition through 12 months.

Figure 20

Experimental Effect on Daily Sugary Drink Intake.

Analyses of the FFQ data showed significant decreases in the proportion of food choices identified by the Block FFQ as sweet-tasting foods (MyP difference = −2.67; 95% CI, −4.51 to −0.82; CC difference = −2.25; 95% CI, −4.18 to −0.33). Relatedly, the FFQ data confirmed what we had observed using participant self-reported frequency of sugary beverage consumption on the main questionnaire, that the percentage of calories consumed as sugary beverages dropped significantly in both conditions (MyP difference = −17.87; 95% CI, −30.58 to −5.17; CC difference = −22.94; 95% CI, −36.21 to −9.68).

Water Intake

The survey questionnaire included the item “How often do you drink water on a typical day?” The answer options were (1) “I don't drink water,” (2) “I drink water once per day,” (3) “I drink water twice per day,” (4) “I drink water 3 to 4 times per day,” and (5) “I drink water 5 or more times per day.” At baseline, 55% said they drank water 5 or more times per day; the corresponding percentages at 6- and 12-month follow-up were 60% and 72%, respectively. Because of the negatively skewed distributions, these measures were dichotomized, with the dummy value of 0 representing those who drank water fewer than 5 times per day and the dummy value of 1 representing those who drank water 5 or more times per day. Figure 21 illustrates increasing water consumption in both conditions from baseline through 12-month follow-up. In the MyP condition, the probability of participants who drank water 5 or more times per day increased from 53% at baseline (95% CI, 44.6%-61.6%) to 72.5% at 12-month follow-up (95% CI, 63.9%-81.1%); in the CC condition, the probability of participants who drank water 5 or more times per day increased from 58% at baseline (95% CI, 49.7%-66.6%) to 71.2% at 12-month follow-up (95% CI, 62.2%-80.1%).

Figure 21

Experimental Effect on Daily Intake of Plain Water.

Total Gram Weight of Solid Food Consumed

One of the derived measures generated by the vendor of the Block FFQ was an estimate of the total gram weight of solid food consumed each day by each participant. As expected, participants in the CC condition reduced consumption of solid food, from 1125 g at baseline (95% CI, 1035-1213 g) to 1015 g at 12-month follow-up (95% CI, 908-1023 g). Unexpectedly, participants in the MyP condition also reduced consumption of solid food, from 1188 g at baseline (95% CI, 1100-1277 g) to 1044 g at 12-month follow-up (95% CI, 942-1147 g). These declines are illustrated in Figure 22.

Figure 22

Total Daily Gram Weight of Solid Food Consumed Declined Under the MyP Condition But Not Under the CC Condition From Baseline to 12-Month Follow-up.

Bean Intake

Curiously, despite MyP participants being encouraged to eat more beans, consumption of refried beans dropped significantly in both conditions (MyP difference = −5.55 g/day; 95% CI, −10.99 to −0.12 g/day; CC difference = −6.65 g/day; 95% CI, −12.34 to −0.96 g/day). Similarly, tortilla consumption dropped significantly in both conditions (MyP difference = −15.46 g/day; 95% CI, −24.23 to −6.70 g/day; CC difference = −10.11 g/day; 95% CI, −19.29 to −0.94 g/day).

FFQ Data on Sugary Beverage Intake in Relation to Consumption of Fiber From F&V

Figure 23 depicts box plots for both baseline and 12-month follow-up data showing an inverse association between categorical levels of daily sugary beverage intake and the ratio of F&V fiber (g) per kg of solid food weight. Relative to participants who drank no sugary beverages, participants who drank between 1 and 99 kcal of sugary beverages a day consumed 1.27 fewer g of F&V fiber per kg of food (95% CI, −2.26 to −0.28) and participants who drank ≥100 kcal of sugary beverage a day consumed 2.43 fewer g of F&V fiber per kg of food (95% CI, −3.53 to −1.33). Figure 21 shows the same data but in relation to change over time. Fruit and vegetable fiber intake was highest in the participants who consumed no sugary beverages and remained highest at 12-month follow-up but participants who consumed a low level of sugary beverage intake (1-99 kcal/day) increased their F&V fiber intake significantly over time (difference = 0.64 g/kg food; 95% CI, 0.28-1.00 g/kg) in contrast to participants who drank ≥100 kcal of sugary beverages per day, whose daily consumption of F&V fiber remained unchanged and well below the levels of the other 2 groups.

Figure 23

Relationship of SSB Intake and F&V Fiber Intake.

Inverse Associations Between Percentage of Calories From FFQ Sweet Food Choices (Added Sugars) and Fiber-Bearing Foods

Figure 24 summarizes the consistently inverse relationship between the proportion of calories from sweet-tasting food choices that are categorized by the Block FFQ as sweets or desserts (ie, added sugar) and the proportion of F&V g of fiber consumed per kg of solid food, termed here the F&V fiber ratio. The F&V fiber ratio was theoretically the most parsimonious way of assessing the impact of overall F&V intake on satiety because the accumulating literature on the gut microbiome has identified F&V fiber as critical determinants of satiety signaling.¹¹⁶ For ease of presentation we converted the percentage of calories from added sugar to a 3-way classification distinguishing participants who adhered to the AHA recommendation to consume fewer than 5% of calories of added sugar from participants who adhered to the USDA's recommendation to consume fewer than 10% of calories of added sugar and from participants whose added sugar intake exceeded both added sugar recommendations (≥10%). Both at baseline and at follow-up the participants who adhered to the AHA 5% added sugar recommendation had the highest F&V fiber ratio (baseline adjusted mean = 11.61; 95% CI, 11.04-12.19); adherents to the USDA 10% added sugar recommendation were intermediate (baseline adjusted mean = 10.55; 95% CI, 9.90-11.20); and participants whose added sugar intake exceeded both recommendations had the lowest F&V fiber ratio (baseline adjusted mean = 8.19; 95% CI, 7.61-8.77) (see Figure 25). This generic inverse association between percentage of calories from sweet foods and F&V fiber ratio was replicated in significant inverse associations observed between percent of calories from sweet foods and the following F&V: bananas, apples, pears, carrots, green salad, and tomatoes (all P < .04). The percentage of calories from sweet foods was also positively associated with salty snacks, tortillas, and, of course, sugary beverages (all P < .001). In contrast to the results for sugary beverage intake, for which the highest consumers experienced no intervention benefit in terms of increasing intake over time in F&V fiber, in the case of sweet-tasting solid food the high consumers showed the most intervention benefit (difference = 0.55 g/kg; 95% CI, 0.09-1.00).

Figure 24

SSB Intake Moderates F&V Fiber Intake.

Figure 25

Consumption of Sweet-Tasting Foods is Associated Cross-Sectionally With Significantly Inverse Proportions of F&V Fiber Intake per kg of Food Consumed at Both Baseline and 12-Month Follow-up.

The F&V fiber ratio, in turn, was inversely associated with feeling hunger the previous day (difference between <9 g fiber/kg of food and ≥12 g fiber/kg = −8.64; 95% CI, −16.62 to −0.65) (see Figure 26). The participants' F&V fiber ratio also moderated the change in their feeling full after meals from baseline to follow-up (see Figure 27). Feeling full as measured by a 100-mm visual analogue scale increased significantly from baseline to follow-up in study participants whose daily intake of F&V fiber was ≥9 g of fiber/kg of solid food (baseline to 12-month follow-up difference_{medium fiber ratio} = 11.66; 95% CI, 2.83-20.48; difference_{high fiber ratio} = 12.67; 95% CI, 3.71-21.64) but did not increase significantly in participants whose baseline F&V fiber intake was than 9 g of fiber/kg (difference_{low fiber ratio} = 6.48; 95% CI, −4.02 to 16.98).

Figure 26

Association Between the Ratio of F&V Grams of Fiber and Participant-Reported Level of Hunger Experienced During the Previous Day.

Figure 27

Association Between the Ratio of F&V g of Fiber and Participant-Reported Feeling Full After Yesterday's Last Meal.

The effect of increasing intake of water-rich foods on repeated hunger ratings over 1 year was also statistically significant, with a moderate estimated effect size (Cohen effect size = (53.5 − 46.7)/13.2 = 0.52).²²

Decisional Context

This study's inquiry was prompted by findings from 3 decades of research on calorie restriction approaches to treating obesity, calling attention to the consistent difficulty that most patients wanting to lose excess body fat have had in sustaining long term the weight loss achieved during the active intervention period with the calorie restriction approach.¹¹⁸ This study's central question was whether the new MyP approach could achieve at least equal weight loss success as the CC approach but without the increased hunger commonly associated with weight loss through nonketogenic calorie restriction approaches to desirable weight loss.^28,119 Low-carbohydrate ketogenic diets do appear to suppress feelings of hunger during active avoidance of carbohydrates, but increased hunger reappears immediately following restoration of any amount of carbohydrate consumption.¹²⁰ A confluence of findings from the bariatric literature,¹²¹ the gut microbiome literature,¹²² and literature on the use of F&V to promote satiety²² suggested that the new MyP approach introduced by the federal government in 2011 could yield at least equivalent weight loss with potentially less postmeal hunger and more satiation/satiety than the nonketogenic, conventional calorie restriction approach used in the Diabetes Prevention Trial.¹² If results were limited to comparing intervention effects on body weight, the 2 interventions would be judged a failure. If results include the slow but monotonically decreasing measure of waist circumference, then the results presented here suggest that both approaches yielded body fat reduction benefit measurable at 12-month follow-up, as hypothesized. The parallel increases in satiety in the 2 conditions were unexpected but not surprising in light of the spontaneous, compositional changes in the food choices made by participants in the CC condition. Instead of just cutting back on everything they had been eating before, the CC participants proportionately increased their F&V intake, much as the participants in the MyP condition were instructed to do.

To be consistent with previous literature and this study's dietary data,²² the most parsimonious explanation is that proportionately increased F&V intake—whether as part of a calorie restriction approach or a MyP/DASH approach—represents a gentle, user-friendly approach to reducing central adiposity in low-income patients, with satiety benefits enduring at least through 12 months of follow-up. The significant reduction in central obesity associated with exposure to both conditions was accompanied by increased satiety after meals even as participants actively engaged in weight-control efforts. Increased satiety-signaling during calorie restriction for weight loss was not expected based on past literature.⁸³ This reduction in central adiposity was also accompanied by increased mental health and health-related QOL, which are commonly observed short term during adherence to calorie restriction regimens.^38-40 More than 90% of participants enjoyed participating in the intervention and “definitely” would recommend it to their friends and relatives. Because the intervention failed to change participants' levels of self-reported physical activity, the observed benefits are more reasonably attributed to the observed changes in food choices during the 1-year study period. The principal dietary changes associated with the observed reduction in central adiposity were reductions in the percent of calories from added sugar, especially by reducing sugary beverage consumption, and by the proportionally increased consumption of minimally processed F&V, as reflected by F&V fiber intake (no juices). The percentage of calories from added sugar was strongly and inversely associated with the ratio of F&V fiber g relative to total g of solid food consumed. The F&V fiber ratio, in turn, was inversely associated with patient reports of everyday hunger experienced the previous day, consistent with the higher satiety signaling expected with increased intake of prebiotics such as F&V polysaccharides.¹²³

Study Results in Context

Study results confirmed recent findings¹²⁴ that CC is not required to achieve significant reduction in central adiposity comparable to that achieved by the traditional CC approach as long as one adheres to a DASH-style dietary pattern¹⁸ and limits consumption of inflammatory foods such as junk food,¹²⁵ highly processed foods with emulsifiers,¹²⁶ processed meats,¹²⁷ foods high in sodium,¹²⁸ and foods high in saturated fat.²¹ While statistically significant, the 12-month follow-up waist circumference reduction was smaller in magnitude than that reported in a previous low-income, predominantly African American clinic population.¹⁶ The seeming equivalence between intervention approaches in waist circumference reduction effectiveness seen in the results presented here could be different if the clinic population were predominantly Philadelphia African Americans rather than predominantly Long Beach Latinos. More dissemination and implementation research is needed to evaluate the relative waist circumference reduction effectiveness of these 2 approaches.

The satiety results for the CC control group, while unexpected, may reflect the decision made before trial onset to permit the CC community health workers to include strong and consistent encouragement to eat more F&V.²² While the classic calorie restriction approach treated all sources of calories as equivalent, the most popular commercial weight loss program, namely Weight Watchers, has popularized the notion that F&V were particularly helpful food choices during weight loss efforts because of their low-calorie density.²⁴ During the study design phase, our CAB objected to implementing the classic calorie restriction approach as being inconsistent with how calorie restriction approaches are implemented. We therefore permitted inclusion of a focus on encouraging CC participants to eat more F&V, even though doing so reduced the distinctiveness of the 2 weight-loss approaches. This change in CC intervention content may at least in part explain the surprising increase in satiety observed in CC participants at 12-month follow-up. The CAB insistence that the CC condition encourage patients toward more F&V consumption as a strategy to lose excess weight was borne out of the July 2017 DPP's release of its “Lifestyle Balance” program.¹²⁹ The latest version of the DPP explicitly and repeatedly encourages consumption of more F&V even as it continues to encourage consumption of fewer calories to achieve a healthier weight.

The use of community health workers instead of highly trained behavior change specialists can yield good weight loss results, as previously demonstrated by peer leaders in Weight Watchers–style interventions.¹³⁰ Throughout the study, all 4 community health workers were employed by TCC; 3 of them had years of prior experience working with TCC patients and were well prepared to make patient referrals to other TCC medical services, as needed. The high participant satisfaction with both of the intervention programs in this study suggests that the community health workers were successful in establishing rapport and providing culturally appropriate lifestyle change coaching. The high community support and patient-centeredness that characterized participants' experiences in both intervention conditions could be partially confounding influences in explaining why we observed similar intervention benefits in both trials but would not be parsimonious explanations for why proportionally higher intake of F&V would be associated with satiety ratings. More research is needed to disentangle the separate contributions of dietary change, community support, and the patient-centeredness of the interventions to increased satiety and reduced central adiposity.

The investigators were surprised at the dearth of literature on the use of behavioral economics¹³¹ in the home setting to create a home environment more supportive of healthier lifestyle choices. The most recent federal nutrition guidelines²¹ explicitly embrace a socioecological model approach to changing food choices, implicitly acknowledging the importance of the food environment for influencing population food choices. In the 1 major study that made changing the food in the home the focus of the intervention, desirable weight loss was achieved.⁵³ In the present study the community health workers were somewhat resistant to ask participants to complete the home environment audit. They were resistant in part because they believed it to be an intrusion on the participants' privacy. In theory, the participants were expected to be more receptive to changing physical features of their home environment than they were to changing their lifestyle practices. In practice, however, changing physical features of the home environment was challenging because there were typically multiple stakeholders in the household, not all of whom shared the same perceived benefits of changing home environmental cues to make them more supportive of healthier food choices.

Implementation of Study Results

TCC health care providers, according to TCC's senior administrators, were happy with the MyP intervention staff's minimally intrusive approach to apprise the providers of patient eligibility for participation in the trial and to obtain provider approval for the patient to be enrolled in the intervention. Using community health workers as change agents relieved the health care providers of managing the treatment of those patients who needed to lose excess body weight. Because the community health workers were actual TCC employees and not UCLA employees, they used their knowledge of TCC to obtain clinic resources that were not among those provided by the PCORI contract, such as referrals for mental health issues. Because TCC has a long-standing policy against using incentives to motivate patients to participate in health education offerings, no incentive was offered to induce participation in intervention sessions. Transportation vouchers were provided, however, to cover the costs associated with attending the group health education sessions at the clinic or at a local grocery store where nutrition education occurred. TCC provided childcare at the group health education sessions, allowing participants to devote their full attention to the health education. Participation in the phone coaching sessions and the group health education session was nonetheless low, suggesting that small per-session incentives (eg, $5 per session) might increase patient participation. The UCLA assessment staff used incentives to optimize participation in the assessments ($20 for baseline assessment, $30 for 6-month assessment, and $50 for 12-month assessment), resulting in 80% retention at 12-month follow-up (if the 9 participants not included in analyses because they became pregnant are included). If incentives worked so well for motivating participation in the assessments, they may also have motivated participation in the intervention sessions.

Generalizability

Because 86% of the participants were Latino, results may not be generalizable to African Americans, Whites, or other ethnic groups. Because 95% of the participants were women, results may not be generalizable to men. Because all participants were patients receiving medical care at a community clinic serving a low-income, urban population, results may not generalize to patients in other clinics in Los Angeles, in other cities, in rural populations, or to higher-income patients. These caveats notwithstanding, there may be benefits to clinics adopting some of the intervention strategies employed in this study, such as the use of home health education sessions, the use of community health workers as change agents, the use of cooking demonstrations, the use of home environmental audits, and the provision of a community resource guide to facilitate patient access to resources that may help them improve their food choices and increase their daily level of physical activity.

Subpopulation Considerations

The identifiable subgroups the investigators planned to examine closely in this study included men and African Americans. Unfortunately, instead of the projected 30% of participants being men, the observed proportion was only 5% (n = 12). Instead of the projected 13% of participants being African American, only 8% (n = 20) enrolled in the study. Hypothesis testing with samples of 20 or fewer is nearly futile.

These subgroups are nonetheless useful for hypothesis-generating purposes and represent future study populations in which to conduct MyP-type interventions culturally tailored to address their group's specific needs. Because 86% of the enrollees in this study were low-income Latinos, often with immigrant backgrounds (82% born outside the United States), potentially important lessons could emerge from analyses of the data regarding the impact of acculturation on receptivity and responsiveness to the MyP intervention. Indeed, exploratory subgroup analyses indicated that intervention impact was greatest on the moderately (second tertile) US-acculturated participants, whose food choices departed significantly more from federal nutrition guidelines at baseline relative to the minimally acculturated participants.

Study Limitations

We based most of the measures used in this study on self-report, which is typically subject to greater measurement error than biological measures. To optimize the patient-centeredness of trial procedures, we included no venipuncture to assess changes in diet or satiety hormones and conducted no maximal treadmill testing for assessing changes in fitness. The patient-centered decision to obtain waist circumference measures over clothing if patients objected to partial disrobing introduced measurement error but adjusting for that error did not meaningfully change the results, despite exhaustive sensitivity analysis testing. The failure of the hypothesis testing to confirm the expected experimental condition by time interaction on satiety was attributable to the CC condition yielding results that differed from results obtained in past calorie-restriction participants,¹⁶ possibly because the CC condition included strong encouragement to eat more F&V, because of contamination from inadvertent exposure of CC patients to the MyP condition by talking to other patients, or because of contamination from inadvertent exposure of the community health workers nested in the CC condition hearing about strategies being used by their community health worker colleagues nested in the MyP condition. Increased imprecision and possible selection bias might have resulted from the 33% of baseline participants not included in the 12-month follow-up analyses. The 261 baseline participants were selected at random from the participating clinic's patient population, but this small number may not fairly represent the millions of low-income patients living in California. From an ethical perspective, the investigators are pleased that participants in both conditions enjoyed similar satiety benefits but the reduction in feeling of hunger after eating in the calorie restriction condition is in contrast to a previous randomized controlled trial showing significantly greater hunger in the calorie restriction arm compared with the fruit- and vegetable-supplemented arm.²² The 12-month follow-up waist circumference outcomes of either intervention condition are impressive relative to secular trends toward increasing waist circumference over time in healthy middle-aged adults.¹³² Because follow-up of participants ended at 12 months, extrapolation of results to longer time intervals is problematic.

Eating is embedded in a web of daily influences ranging from variations in health status to variations in menstrual cycling, daily physical activity level, and income-related access to food, only some of which this study measured. Moreover, physiological functioning is not necessarily the primary determinant of food choices. Hedonic hunger and liking/wanting of highly palatable foods, regardless of feelings of fullness, could also influence eating duration and quantity of calories consumed.¹³³ The generalizability of this study's findings is necessarily constrained by the limited number of covariates that could be included without overburdening the study participant.

Future Research

Beneficial intervention features unique to each of the approaches studied here could be combined for greater impact. This, in fact, was accomplished implicitly in a recent weight-loss intervention targeting adults with serious mental illness.¹²⁴ The authors of that study built on the nutrition approach pioneered by the DASH trials^18,19 much as the investigators did with the MyP approach. However, they also included recommendations to restrict portion sizes and calorie-dense foods as commonly done in calorie restriction programs but explicitly rejected the calorie monitoring approach of traditional calorie restriction programs. The modest 1.5-kg weight-loss advantage observed in the intervention group compared with controls at 6-month follow-up became a 2.6-kg advantage at 1 year and a 3.2-kg advantage at 18-month follow-up, a pattern of continuing improvement rarely seen in the calorie restriction literature.⁸¹ Unsuspected inflammation-reducing benefits of calorie restriction have recently been identified in studies of the murine gut microbiota,¹³⁴ providing conceptual support for restricting consumption of proinflammatory foods (eg, processed foods with emulsifiers, processed meats, foods high in saturated fat),¹³⁵ even as patients are encouraged to eat more daily servings of minimally processed fruits, vegetables, whole grains, legumes, seeds, and nuts.²¹ In other words, there may be therapeutic benefits to including both caloric restriction (of proinflammatory foods) and the MyP recommendations (to eat more fiber-rich foods, to limit sodium-added, sugar-added, saturated fat-rich foods).