Predictors of race differences in weight loss: The PROPEL trial

Abstract

Objective:

Studies consistently show African Americans lose less weight in response to behavioral interventions, but the mechanisms leading to this result are understudied.

Methods:

Data were derived from the PROmoting Successful Weight Loss in Primary CarE in Louisiana (PROPEL) study, which was a cluster-randomized, two-arm trial conducted in primary care clinics. In the PROPEL trial, African Americans lost less weight compared to patients who belonged to other racial groups after 24 months. In the current study, counterfactual mediation analyses among 445 patients in the intervention arm of PROPEL were used to determine which variables mediated the relationship between race and weight loss. The mediators included treatment engagement, psychosocial, and lifestyle factors.

Results:

At 6 months, daily weighing mediated 33% (p=0.008) of the racial difference in weight loss. At 24 months, session attendance and daily weighing mediated 35% (p=0.027) and 66% (p=0.005) of the racial difference in weight loss, respectively. None of the psychosocial or lifestyle variables mediated the race-weight loss association.

Conclusions:

Strategies specifically targeting engagement, such as improving session attendance and self-weighing behaviors, among African Americans are needed to support more equitable weight losses over extended time periods.

Introduction

African American adults have higher rates of obesity compared to members of other racial (self-classified as White, Asian, American Indian or Alaska Native, Native Hawaiian or other Pacific Islander) and ethnic (self-classified as Hispanic/Latino) groups (1). Recent data indicate that while non-Hispanic Asian men have low rates of obesity (17.5%), there are few racial and ethnic differences in obesity prevalence between non-Hispanic African American (41.1%), non-Hispanic White (44.7%), and Hispanic men (45.7%). However, African American (56.9%) and Hispanic/Latino (43.7%) women have a higher prevalence compared to Asian (17.4%) and non-Hispanic White women (39.8%) (1). Obesity is associated with multiple comorbidities (2), such as cardiovascular disease, diabetes, and several cancers. These comorbidities are also more prevalent in African Americans and contribute to health disparities between African Americans and those of other races/ethnicities, potentially leading to excess health care costs and higher rates of premature mortality. Therefore, weight loss interventions may improve health and reduce health disparities for African American adults.

Behavioral interventions have been successful in producing weight loss in African American adults (3, 4, 5, 6). However, studies have consistently shown that African Americans lose less weight compared to non-Hispanic White Americans in the short (7, 8, 9) and long term (10, 11). It is important to move beyond describing racial and ethnic differences in weight loss and begin to identify the factors that are associated with weight loss that may differ between race and ethnic groups. Dietary patterns (5), physical activity (PA) behaviors (5), intervention attendance (6), caloric intake (6), and self-weighing (6) have been shown to be associated with weight loss across different racial and ethnic groups, yet these findings do not directly assess explanatory factors for differential weight loss between racial and ethnic groups. We are aware of only one study (8) that compared predictors of weight loss differences between race and ethnic groups. The study showed that non-Hispanic White women lost more weight compared to African American women; and that higher website logins and improved dietary behaviors associated with weight loss in non-Hispanic White women partially mediated the racial differences in weight loss. The study was conducted over a period of four months and only included women, and therefore, there is a need for more studies to explain racial differences in weight loss.

The PROmoting Successful Weight Loss in Primary CarE in Louisiana (PROPEL) study can assess factors associated with racial differences in weight loss over an extended time frame. PROPEL was a pragmatic trial conducted in 18 primary care clinics across the state of Louisiana and targeted an underserved, low-income population. The study compared a weight loss intervention delivered in clinics by trained health coaches to usual care. The primary outcome paper documented significant racial differences in weight loss across a 24-month intervention period (12). Therefore, the purpose of the current investigation is to determine the factors that mediated these racial differences in weight loss. Only data from the lifestyle intervention arm of the PROPEL study will be utilized because there were no racial differences in weight loss in the usual care condition at 24 months and assessments of all relevant mediators (e.g. engagement) were not collected on usual care participants (12). It is hypothesized that lower levels of engagement and weight loss related behaviors (e.g., PA, dietary adherence) will account for ethnic differences in weight loss.

Methods

Study design

Details on the trial design are provided elsewhere (13). Briefly, the PROPEL study was a cluster-randomized, two-arm trial conducted in primary care clinics. A total of 18 primary care clinics across Louisiana were randomized to one of two study arms: 1) the intensive lifestyle intervention (ILI) group or 2) a usual care group. Only the ILI group (n = 445, 73.9% African American) was analyzed herein. The nine ILI clinics were located in both urban and rural settings.

Participants

The key eligibility criteria included 1) 20–75 years old, 2) BMI of 30– 50 kg/m², and 3) patient at a participating clinic. Participants were excluded if they were 1) currently participating in a weight loss program, 2) currently taking weight loss medication, 3) had ever undergone bariatric surgery or planned to in the next two years, or 4) had lost >10 lbs of weight within the last six months. The complete list of inclusion and exclusion criteria can be found elsewhere (13). All participants provided written informed consent, and all study procedures were approved by the Pennington Biomedical Institutional Review Board.

Intervention

The ILI was embedded in the primary care clinics and consisted of weekly visits with trained health coaches (16 face-to-face and 6 via phone) during the first six months and at least monthly visits for the remaining 18 months, alternating between face-to-face and phone visits. Trained health coaches worked with participants to meet the predefined individual goal of 10% weight loss by coaching them to develop and adhere to personalized action plans focusing on changes in eating, diet, and physical activity behavior. The health coaches had higher education degrees in nutrition, physical activity, or behavioral medicine and received further training in the management of obesity and related comorbidities, fundamentals of health literacy, and patient communication and education before the start of the intervention. The ILI was based on empirically-proven weight loss interventions including Diabetes Prevention Program (DPP) (14), Look AHEAD (15), and CALERIE (16), and was consistent with the 2013 AHA/ACC/TOS Guidelines (17).

Each participant was provided with a BodyTrace© scale (BodyTrace Inc., Palo Alto, CA) and was encouraged to weigh daily. Self-monitoring of weight was implemented to monitor weight loss progress and promote intervention fidelity. The BodyTrace© scale transmitted weight data wirelessly and in real time to a computer tracking system. This system plotted the weight data onto a personalized weight graph, which was available via a website at any time, and allowed participants and health coaches to detect deviations from the intended weight loss progress quickly (18, 19, 20, 21). If deviations occurred, the personalized action plans were adjusted, utilizing additional components of the ‘toolbox’ approach (i.e., tailored behavioral, nutritional, and PA strategies) that has been shown to improve intervention efficacy in previous clinical trials (14, 15, 16). Primary care providers (PCP) in ILI clinics received a series of webinars on lifestyle weight management practices, lipid management, and ways to improve communication with patients with low health literacy and/or obesity throughout the intervention period.

Measurements

Trained technicians conducted the body weight measurement and were blinded to study group. Participants completed self-report measures. The current analysis is based on changes between baseline and 6 and 24 months of follow-up. The following formulas were used: change = (value at 6 or 24 months - the baseline value); percent = ((value at 6 or 24 months - the baseline value/baseline value)*100).

Predictors

Race

Participants were asked to self-identify their racial (American Indian or Alaskan Native, Asian, Black/African American, Hawaiian/Pacific Islander, White, Multi-race, Other) and ethnic group (Hispanic or non-Hispanic) at baseline. For the purposes of these analyses, participants were classified as “African American” (regardless of ethnic group), or “Other”. African American was coded as ‘1’ and Other was coded as ‘0’ in the analyses.

Outcome

Body weight

Body weight was measured to the nearest 0.1 kg using a digital scale (Seca Model 876). Each measurement was taken twice using standard procedures. If the two measurements differed by >0.5 kg, a third measurement was obtained, and the two measurements that were numerically closest together, whether < or > 0.5 kg, are averaged for analysis. Percent change in body weight from baseline was the primary outcome variable.

Mediators

Treatment sessions

As stated above, the program consisted of weekly (16 in-person; 6 phone) visits in the first six months, followed by visits held at least monthly (alternating in-person and phone) for the remaining 18 months. Participants were expected to attend each visit. The curriculum consisted of 43 total sessions. The percent of sessions received was used in the analyses.

Self-weighing

Self-weighing data were captured from the BodyTrace© electronic scale. As stated above, participants were encouraged to weigh themselves daily over the course of the intervention. The percent of days weighed was used in the analyses.

Eating attitudes and behaviors

The Eating Inventory is a 51-item questionnaire designed to measure different dimensions of eating attitudes and behaviors (22). Two of the three factor-analyzed subscales were administered to participants: Cognitive Restraint and Disinhibition. Change in Cognitive Restraint and Disinhibition scores were used in the analyses.

PA Levels

The International Physical Activity Questionnaire – Short Form (IPAQ-SF) is a self-report measure of PA (23). The IPAQ assesses physical activity performed over the previous 7 days. Self-reported minutes of PA per day were derived for all activity levels. Metabolic Equivalent of Task (MET) minutes of activity were derived by multiplying the number of minutes spent in moderate intensity activity by the MET equivalent. Change in MET minutes per week were used in the analyses.

Dietary intake

The dietary intake measure was composed of items from different scales, including the National Cancer Institute (NCI) fat screener (24), the NCI fruit and vegetable screener (25, 26), and the Brief Questionnaire to Assess Habitual Beverage Intake (BEVQ-15) (27). The National Cancer Institute (NCI) fat screener estimates the percentage of energy from fat by asking patients to report the frequency of consuming specific foods over the past 12 months.(24) A standard 7-item fruit and vegetable screener developed by the NCI and National 5 a Day Program grantees asks how often fruit and vegetables were consumed in the past month. Three questions related to the frequency of alcohol intake (beer, wine, hard liquor) were adapted from the Brief Questionnaire to Assess Habitual Beverage Intake (BEVQ-15). Change in the frequency of alcohol consumed per day as well as the frequency of fruits and vegetables consumed per day, were used in the analyses.

Weight-related quality of life (QoL)

The Impact of Weight on Quality of Life – Lite (IWQOL-Lite) is a 31-item questionnaire designed to measure obesity-specific aspects of QoL, which produces a total score and separate scores for physical function, self-esteem, sexual life, public distress, and work or daily activities (28). Change in the total score was used for analysis.

Health-related QoL (HRQOL)

The PROMIS-29 questionnaire, which is a 29-item questionnaire that includes questions covering health-related domains of physical function, anxiety, depression, fatigue, sleep disturbance, ability to participate in social roles and activities, pain interference and pain intensity, was used to measure HRQOL (29). Change in the total score was used for analysis.

Covariates

A self-report demographic and health history questionnaire was administered at baseline. Participants reported their age, sex, health insurance status, income, employment, education level, and food security status (30).

Statistical analyses

Frequencies were used for descriptive statistics. T-tests were used to assess baseline differences between groups (AA vs Other races). To explicate the racial differences in weight loss, we investigated the hypothesized mediating effect that a third variable might play by conducting three multivariable mixed linear regressions in SAS version 9.4 (SAS Institute Inc, Cary, NC).

$$\%WL={\alpha }_{1}Race+{\alpha }_{2}Sex+{\alpha }_{3}Age+{\alpha }_{4}Income+{\alpha }_{5}Edu+{\alpha }_{6}FoodSecurity+Zu+e$$ (Model 1)

$$\begin{gathered} Mediator={\beta }_{1}Race+{\beta }_{2}Sex+{\beta }_{3}Age+{\beta }_{4}Income+{\beta }_{5}Edu+{\beta }_{6}FoodSecurity+Zu \\ +e \end{gathered}$$ (Model 2)

$$\begin{gathered} \%WL={\gamma }_{1}Race+{\gamma }_{2}Mediator+{\gamma }_{3}Race\ast Mediator+{\gamma }_{4}Sex+{\gamma }_{5}Age+{\gamma }_{6}Income \\ +{\gamma }_{7}Edu+{\gamma }_{8}FoodSecurity+Zu+e \end{gathered}$$ (Model 3)

Factors including sex, age, income, education levels, and food security were controlled for in the models as covariates. Analyses were performed on data collected at months 6 and 24. All models included the cluster random effect of clinics explained by the term $Zu$, where $Z$ was the design matrix of clusters, and $u$ was assumed to be normally distributed with mean 0. The models assumed no correlation between clinics; therefore, variance component covariance structure was used for $u$.

The coefficient of race (${\alpha }_1$) in Model 1 estimates the total effect (TE) as the racial difference in percent weight loss ignoring the effect of the mediator. Model 2 tested for the relationship between mediator variable and race with ${\beta }_1$. The interaction between race and mediator variables was further considered in Model 3. Under the counterfactual framework, we estimated the controlled direct effect ((CDE), defined as the racial difference of % weight loss when the mediator is set to sample mean), natural direct effect ((NDE), defined as racial difference of % weight loss when the exposure is allowed to change from the level of African American to level it would take for Other and the mediator is set to the level of Other as reference), and natural indirect effect ((NIE), racial difference of % weight loss when the exposure is set to the level of Other and the mediator is allowed to change from the level for African American to the level it would take for Other). To evaluate the magnitude of mediation, the percentage of total effect mediated was computed as natural indirect effect/total effect*100. The total effect can be decomposed into natural direct effect and natural indirect effect. The methods described by Vansteelandt and Vanderweele (31, 32) were applied to quantify these effects’ standard error and 95% confidence intervals and also to test for significances. All P values were two-sided, and P < 0.05 was considered to indicate statistical significance (33).

Results

Only participants in the ILI arm (n = 445, 73.9% African American) were included in the analyses.

Demographics.

Participants in the ILI on average were middle aged, had Class II obesity, and the sample was majority female. Compared to African American patients, patients of Other (78.4% non-Hispanic White, 7.8% multi-race, 6.9% American Indian or Alaska Native, 6% other, and 0.9% Asian) race had significantly (p = 0.002) higher levels of disinhibition (tendency to eat in the presence of food). African American patients also endorsed significantly (p < 0.001) higher IWQOL-lite total scores compared to Other patients, indicating a lower negative impact of weight on quality of life. There was also a significant (p = 0.003) difference between racial groups on the PROMIS total score (Table 1).

Table 1.

Baseline characteristics

	AA	Other	p
Sample size (73.9% AA)	n = 329	n = 116
Weight (kg)	101.2 (15.6)	102.3 (18.2)	0.527
BMI (kg/m2)	37.0 (4.5)	37.8 (4.7)	0.130
Age, years	48.9 (12.3)	48.4 (13.5)	0.713
Gender (% female)	90.0	84.5	0.127
Annual Family Income (%)			0.135
Less than $10,000	20.3	16.8
$10,000 - $19,999	21.5	20.4
$20,000 - $39,999	25.9	23.9
$40,000 - $59,999	16.6	12.4
$60,000 and above	15.7	26.6
Marital status (%married)	37.7	46.6	0.342
Eating Inventory
Disinhibition	6.7 (3.5)	7.9 (4.0)	0.002
Restraint	9.8 (4.3)	9.1 (4.7)	0.098
IPAQ (MET min/week)	432.6 (800.2)	427.0 (794.2)	0.950
Dietary intake
Alcohol (times/day)	0.18 (0.39)	0.12 (0.25)	0.121
Fruits and vegetables (times/day)	2.4 (1.7)	2.3 (1.6)	0.702
Percent fat	32.1 (4.0)	31.6 (4.0)	0.378
IWQOL-lite (total score)	87.3 (15.1)	77.4 (18.7)	<0.001
Food security (%insecure)	28.0	29.3	0.782
PROMIS (total score)	350.1 (27.2)	359.6 (29.5)	0.003

Predictor variables.

There was a significant race difference in the percent of home weights taken. Other patients weighed at home at a higher percentage (73.5% vs. 67.1%; p = 0.031 at Month 6 and 53.6% vs. 44.2%; p = 0.003 at Month 24), compared to African Americans. There was a significant between group difference in session attendance. Other patients attended a greater percentage of intervention sessions at Month 24 (82.2% vs. 72.9% (4 session difference), p = 0.004) compared to African Americans, but not at Month 6 (80.3% vs. 76.2% (one session difference); p = 0.166). Levels of disinhibition were higher for Others than African Americans at Month 6 (5.7 vs. 4.9, p = 0.017) and 24 (6.3 vs. 5.3, p = 0.012). IWQOL-lite total score was lower for Others than African Americans at Month 6 (77.4 vs. 87.3, p < 0.001) and 24 (79.9 vs. 86.0, p < 0.004). Other participants had higher PROMIS scores compared to African Americans at Month 6 (359.6 vs. 350.1, p = 0.003). No other differences in predictor variables were found (Table 2).

Table 2.

Mediator values.

Month 6
Outcome	N	AA	Other	p
% of sessions	444	76.2 (1.6)	80.3 (2.7)	0.166
% home weights	444	67.1 (1.5)	73.5 (2.5)	0.031
Eating Inventory
Disinhibition	385	4.9 (0.19)	5.7 (0.31)	0.017
Restraint	385	16.0 (0.21)	16.0 (0.35)	0.946
IPAQ (MET min/wk)	418	433.3 (45.4)	427.0 (76.9)	0.945
Dietary intake
Alcohol (times/day)	383	0.18 (0.02)	0.12 (0.04)	0.122
Fruits and vegetables (times/day)	385	2.4 (0.10)	2.3 (0.16)	0.702
% fat	381	32.1 (0.24)	31.6 (0.39)	0.378
IWQOL-lite (total)	382	87.3 (0.96)	77.4 (1.6)	<0.001
Food security (% insecure)	363	20.8	19.4	0.774
PROMIS (total score)	385	350.1 (1.7)	359.6 (2.7)	0.003
Month 24
% of sessions	432	72.9 (1.7)	82.2 (2.7)	0.004
% home weights	432	44.2 (1.6)	53.6 (2.7)	0.003
Eating Inventory
Disinhibition	361	5.3 (0.2)	6.3 (0.35)	0.012
Restraint	361	14.5 (0.27)	14.2 (0.46)	0.667
IPAQ (MET min/wk)	407	425.9 (46.0)	431.0 (77.0)	0.954
Dietary intake
Alcohol (times/day)	359	0.17 (0.03)	0.12 (0.05)	0.323
Fruits and vegetables (times/day)	354	2.1 (0.09)	2.3 (0.16)	0.287
% fat	351	32.7 (0.26)	32.6 (0.44)	0.791
IWQOL-lite (total)	358	86.0 (1.06)	79.9 (1.80)	0.004
Food security (% insecure)	361	21.4	16.0	0.261
PROMIS (total score)	358	351.9 (1.77)	357.5 (3.0)	0.105

^*Values are expressed as Mean (Standard Error)

Weight loss outcomes.

African Americans lost significantly (ps < 0.01) less weight and percent weight compared to Others at 6 and 24 months (Table 3). The difference was approximately 2 – 2.5% body weight or kilograms in all cases.

Table 3.

Weight loss outcomes

	AA	Other	p
Month 6
% weight loss	−6.8 ± 0.33	−9.0 ± 0.55	0.001
Weight loss (kg)	−6.9 ± 0.35	−9.1 ± 0.58	0.001
Month 24
% weight loss	−4.5 ± 0.45	−7.0 ± 0.76	0.006
Weight loss (kg)	−4.5 ± 0.46	−7.2 ± 0.78	0.005

Mediation:

Table 4 shows the direct and indirect effect (through the designated mediator) of race on % weight loss at 6 months. Race was significantly associated with weight loss after adjustment for covariates with African Americans losing less weight than Other race participants (TE (SE) = 2.1 (0.66) %, CI: 0.83, 3.4). The disparity in daily weighing partially explained this interracial difference as it mediated 33% (p = 0.008) of the racial difference in weight loss. This implies that if daily weighing could be set equivalent between racial groups, the racial disparity in % weight loss at 6 months would be reduced by 33.3%. None of the psychosocial or lifestyle variables mediated the ethnicity-weight loss association.

Table 4.

Total, direct, and indirect effects and percent mediated of racial difference in % weight loss by mediators at 6-months

	Natural direct effect, difference (95% CI)	Natural indirect effect, difference (95% CI)	% mediated (95% CI)	p-value
% of sessions	2.0 ± 0.58 (0.81, 3.1)	0.16 ± 0.23 (−0.29, 0.61)	7.7 ± 9.5 (−11.0, 26.3)	0.209
% home weights	1.4 ± 0.54 (0.33, 2.5)	0.69 ± 0.31 (0.06, 1.3)	33.3 ± 11.7 (10.3, 56.2)	0.008
Eating Inventory
Disinhibition, change score	2.1 ± 0.62 (0.83, 3.3)	0.2 ± 0.14 (−0.07, 0.47)	9.0 ± 6.1 (−3.1, 21.0)	0.132
Restraint, change score	2.0 ± 0.61 (0.82, 3.2)	0.24 ± 0.17 (−0.09, 0.57)	10.4 ± 7.2 (−3.7, 24.6)	0.141
IPAQ (MET min/week), change score	2.0 ± 0.64 (0.7, 3.2)	0.06 ± 0.11 (−0.16, 0.28)	3.0 ± 5.3 (−7.4, 13.4)	0.584
Dietary intake
Alcohol (times/day), change score	2.1 ± 0.65 (0.87, 3.4)	0.07 ± 0.07 (−0.07, 0.21)	3.0 ± 3.1 (−3.1, 9.2)	0.328
Fruits and vegetables (times/day), change score	2.0 ± 0.63 (0.74, 3.2)	0.17 ± 0.14 (−0.1, 0.44)	7.9 ± 6.5 (−4.9, 20.8)	0.119
% fat, change score	1.8 ± 0.63 (0.58, 3.0)	0.01 ± 0.09 (−0.17, 0.19)	0.52 ± 4.6 (−8.6, 9.6)	0.924
IWQOL-lite (total), change score	2.2 ± 0.61 (0.97, 3.4)	0.08 ± 0.17 (−0.25, 0.41)	3.6 ± 7.2 (−10.5, 17.7)	0.609
PROMIS (total), change score	2.3 ± 0.62 (1.0, 3.5)	−0.09 ± 0.09 (−0.27, 0.09)	−4.2 ± 4.2 (−12.4, 4.0)	0.580
Total Effect	2.1 ± 0.66 (0.83, 3.4)

^*Values are expressed as Mean (Standard Error)

Table 5 shows the direct and indirect effect (through the designated mediator) of race on % weight loss at 24 months. Race was significantly associated with % weight loss after adjustment for covariates with African American losing less weight than Other (TE (SE) = 2.4 (0.91) %, CI: 0.56, 4.1). Disparity in session attendance and daily weighing partially explained this interracial difference. Session attendance and daily weighing mediated 35.4% (p = 0.027) and 66.2% (p = 0.005) of the racial difference in weight loss, respectively. This implied that if the session attendance and daily weighing could be set equivalent between races, the racial disparity in % weight loss would be reduced by 35.4% and 66.2%, respectively. None of the psychosocial or lifestyle variables mediated the ethnicity-weight loss association.

Table 5.

Total, direct, and indirect effects and percent mediated of racial difference in % weight loss by mediators at 24-months

	Natural direct effect, difference (95% CI)	Natural indirect effect, difference (95% CI)	% mediated (95% CI)	p-value
% of sessions	1.5 ± 0.94 (−0.37, 3.3)	0.81 ± 0.3 (0.21, 1.4)	35.4 ± 16.6 (2.8, 67.9)	0.027
% home weights	0.65 ± 0.86 (−1.04, 2.3)	1.3 ± 0.39 (0.51, 2.0)	66.2 ± 20.6 (25.8, 106.5)	0.005
Eating Inventory
Disinhibition, change score	2.1 ± 0.88 (0.38, 3.8)	0.21 ± 0.19 (−0.16, 0.58)	9.2 ± 7.9 (−6.3, 24.7)	0.260
Restraint, change score	2.2 ± 0.89 (0.5, 4.0)	0.23 ± 0.19 (−0.14, 0.6)	9.2 ± 7.6 (−5.7, 24.0)	0.198
IPAQ (MET min/week), change score	1.1 ± 0.93 (−0.75, 2.9)	0.0 ± 0.11 (−0.22, 0.22)	0.44 ± 4.9 (−9.2, 10.1)	0.935
Dietary intake
Alcohol (times/day), change score	2.2 ± 0.9 (0.41, 3.9)	0.05 ± 0.06 (−0.07, 0.17)	2.2 ± 2.7 (−3.2, 7.5)	0.766
Fruit and vegetables (times/day), change score	2.3 ± 0.93 (0.52, 4.2)	0.04 ± 0.18 (−0.31, 0.39)	1.8 ± 8.1 (−14.2, 17.7)	0.975
% fat, change score	1.3 ± 0.91 (−0.51, 3.1)	−0.02 ± 0.12 (−0.26, 0.22)	−1.8 ± 7.6 (−16.6, 13.1)	0.834
IWQOL (total), change score	2.4 ± 0.89 (0.61, 4.1)	0.29 ± 0.23 (−0.16, 0.74)	11.1 ± 9.3 (−7.2, 29.3)	0.119
PROMIS (total), change score	2.2 ± 0.9 (0.39, 3.9)	0.03 ± 0.1 (−0.17, 0.23)	1.5 ± 4.2 (−6.8, 9.7)	0.606
Total Effect	2.4 ± 0.91 (0.56, 4.1)

^*Values are expressed as Mean (Standard Error)

Discussion

The present study sought to understand the factors that may mediate the effect of race on weight loss in the PROPEL trial. The main study showed that African Americans lost significantly less weight over 6 and 24 months compared to Other patients (12). Engagement, psychosocial, and behavioral factors were investigated as factors that may explain the race differences in weight loss. The findings show that intervention engagement factors significantly mediated the race-weight loss results. In fact, 33% - 66% of the race-weight loss relationship was explained by engagement factors. Neither psychosocial nor other behavioral factors had a significant influence on the race-weight loss relationship.

The study findings point to the importance of intervention engagement in explaining race differences in weight loss. Our results are consistent with other studies showing that differences in treatment engagement mediate racial differences in weight loss. Blackman-Carr (8) showed that fewer study website logins in African American women explained 4-month race differences in weight loss compared to White women. In addition, our results are consistent with other studies demonstrating the importance of engagement in weight loss for African Americans. Engagement in the current study was defined as percent of sessions attended and percent of daily weighing. Although there was a relatively small difference in attendance between the groups (73% in African American and 82% in Other) at Month 24, it explained 35% of the racial difference in weight loss. Session attendance has been shown to be more strongly related to weight loss in African Americans compared to adults of other racial/ethnic groups (6). Self-weighing has also been shown to be an effective strategy to aid weight loss attempts within the context of weight management interventions (34). In the Look AHEAD trial, daily self-weighing was the strongest predictor of 1-year weight loss among African American men and women (along with use of meal replacements) (6). In PROPEL, self-weighing explained a larger proportion of the variance in weight loss differences at 24 months compared to 6 months and compared to session attendance at 24 months. Our data suggests that while the difference in self-weighing between the racial groups was only 6% - 9%, implementing strategies to promote parity in self-weighing may be effective in decreasing the racial weight loss disparity. Unfortunately, strategies to increase self-weighing among African Americans are not well understood. Qualitative research exploring self-weighing practice beliefs that can inform intervention design to promote self-weighing in African American populations would be particularly helpful. In sum, the consistency of findings suggests the variables influencing African Americans’ engagement are not being effectively addressed by current interventions.

The results of the study point to the need to address those factors that may be affecting engagement in African American participants enrolled in weight loss interventions. Racial differences across a variety of factors influencing weight loss, including weight loss goals (35), motivations for weight loss (e.g. body image, health improvement (36)), and cultural issues (e.g. cultural concordance with the counselor, cultural competence (37, 38)) may have impacted African American patients level of engagement with the intervention. It has been shown that African American men are more supportive of larger body sizes compared to men of other racial groups (39). Another possible influence on engagement is social determinants of health. African Americans experience more adverse social determinants, such as lower socioeconomic stability, lower quality education, reduced access to health care, and neighborhood and built environment barriers (40, 41). African Americans also experience poorer health across the spectrum of social determinants of health (42) and social determinants have been shown to influence treatment outcomes (43). While social determinants of health may not have been a factor in the current study given the lack of differences in participant characteristics (e.g. income, marital status, food security), these variables are likely relevant for other investigations of racial differences in weight loss. Nonetheless, future studies should be designed to determine which of these factors influence engagement and seek ways to overcome engagement barriers.

Lifestyle and psychosocial variables did not mediate the race effect on weight loss in the current study. Previous studies have shown these factors are important explanatory variables. Lifestyle behaviors, including eating behaviors and physical activity, have different associations across ethnic groups in studies showing ethnic differences in weight loss (8, 11). One reason for the difference in findings may be the measures used. For example, one study utilized objective measures of physical activity while PROPEL utilized a self-report measure (11). Also, PROPEL utilized a dietary screener as opposed to more detailed nutrition assessments. Thus, it is possible that lifestyle factors may mediate racial differences in weight loss, but studies must include measures sensitive enough to detect these differences.

Our study is one of the few long-term studies to assess racial differences in weight loss (5, 6). Long-term studies have tended to show that African Americans have slower weight regain even though there is less initial weight loss. This can result in non-statistically significant, but still lower, weight loss difference between the race groups over an extended follow-up period. One example of this is the Look AHEAD study (6). One-year follow up data showed that both African American men and women lost ~3 kg and ~2 kg less weight compared to non-Hispanic White American men and women, respectively. However, the 8-year follow up data show that African American men lost 1.5 kg less weight and African American women lost ~1 kg more weight than their non-Hispanic White American counterparts (6). Another study lasting 30 months showed weight loss differences of 0.5 kg and 1 kg for African American men and women compared to non-Hispanic White American men and women, respectively (5). PROPEL differs from these other studies in that it was a mostly underserved clinic- vs. academic institution-based intervention, and it was conducted in one state as opposed to being a multisite trial. Regardless of these methodological differences, more long-term data are needed to help clarify equivocal findings in the literature.

The findings of the current study should be interpreted in the context of its limitations and strengths. As with many studies of this nature, the original study was not designed to specifically assess racial differences in weight loss. As such, the sample sizes differ substantially between the racial groups by approximately three-fold. However, the sample sizes are large enough to conduct the mediation analyses. Another notable limitation is that some factors that may be influencing racial differences in weight loss, such as energy requirements (44), were not assessed. Unmeasured factors, including additional social determinants of health, may have played a role in explaining the observed racial differences. More research is required to determine the impact of social determinants of health on weight loss in different race/ethnic groups. There are several key strengths of the study. First, the study includes a large sample size, which provides greater confidence in the findings. It was also conducted in a primary care setting, which helps to generalize findings from previous research in this area. Finally, the predictors covered several important areas including intervention engagement, psychosocial factors, and behavioral factors, all known to influence weight loss.

The current study showed racial differences in weight loss across 24-months of treatment in that African American participants lost less weight than participants of other races. Lower levels of treatment engagement by African Americans explained a large fraction of the race differences, particularly self-weighing. The findings suggest that strategies specifically targeting engagement among African Americans are needed in order to support more equitable weight losses over extended time periods.

What is already known about this subject?

• African Americans have been shown to lose less weight in large weight loss trials. Few studies have assessed the potential mediators of differential weight loss in a largely low-income population.

• The few previous studies have shown that behavioral and engagement factors account for racial differences in weight loss, but these were largely short term studies.

What are the new findings in your manuscript?

• The mediational analyses showed engagement factors largely account for the racial difference in weight loss. Behavioral and psychosocial variables were not related, which is different from previous findings.

• This is the first study to show the importance of engagement factors in a primary care clinic study conducted with a largely low income population.

How might your results change the direction of research or the focus of clinical practice?

• Future research should focus on the factors that are influencing African American’s engagement in behavioral studies. These factors may include social determinants of health, beliefs, or mistrust.