Effects of Food Label Use on Diet Quality and Glycemic Control Among Latinos With Type 2 Diabetes in a Community Health Worker–Supported Intervention

Abstract

Objectives. To determine the impact of an intervention led by community health workers (CHWs) on food label use and to assess whether food label use and diet quality mediate the intervention’s impact on glycemic control.

Methods. From 2006 to 2010, 203 Latinos (intervention group, n = 100; control group, n = 103) in Hartford County, Connecticut, with type 2 diabetes were randomized to an intervention that included 17 CHW-led home-based sessions over a 12-month period in addition to the standard of care available in both study arms. Data on food label use, diet quality, covariates, and glycated hemoglobin (HbA1c) were collected at baseline and at 3, 6, 12, and 18 months. Data were analyzed via mixed effects and multilevel structural equation modeling.

Results. Food label use in the intervention (vs control) group was significantly higher at 3, 12, and 18 months (odds ratio = 2.99; 95% confidence interval = 1.69, 5.29). Food label use and diet quality were positive mediators of improved HbA1c levels.

Conclusions. Culturally tailored interventions led by CHWs could increase food label use. Also, CHW-delivered food label education may lead to better diet quality and improve glycemic control among Latinos with type 2 diabetes.

Type 2 diabetes is increasing in epidemic proportions in the United States,¹ where it is a leading cause of mortality,² renal failure,³ nontraumatic limb amputations,⁴ and blindness.⁴ Although proper diabetes self-care behaviors have proven benefits in improving glycemic control and reducing associated comorbidities,⁵ affected individuals frequently lack understanding of such behaviors, especially with respect to diet.⁶

A systematic review of studies conducted among diabetes patients identified a need for improved patient-initiated behaviors, including dietary modifications to improve type 2 diabetes self-management.⁷ Use of food labels may play an important role in sustaining such modifications. These labels, compulsory on prepackaged foods since the Nutrition Labeling and Education Act of 1990,⁸ contain food-specific information including serving size, nutrient content per serving, and percentage daily values,⁹ and as such they can help individuals make informed food choices.¹⁰ Evidence from studies focusing on Latinos with type 2 diabetes (Kollannoor-Samuel et al., unpublished data, 2016) and general population studies^11,12 suggests the positive impact of food label use on dietary choices. Similarly, positive associations between dietary components (foods with a low glycemic index,¹³ high-fiber foods,¹³ fruits and vegetables¹⁴) and glycemic control have been reported.

Latinos, the largest minority group in the United States,¹⁵ continue to suffer from a high prevalence of type 2 diabetes¹⁶ and related complications^16,17 as well as poor health care access.¹⁸ Latinos with type 2 diabetes often report poor diabetes health literacy¹⁹ and self-care behaviors,^20,21 including low diet quality and nonuse of food labels (Kollannoor-Samuel et al., unpublished data, 2016). There is limited empirical evidence of which we are aware documenting the effects of diabetes-specific education on food label use among Latinos with type 2 diabetes. Similarly, little is known about the impact of food label use or diet quality on glycemic control in this group. Culturally appropriate educational interventions have proven benefits with respect to type 2 diabetes management among members of racial/ethnic minority groups.^22,23 For example, in the Diabetes among Latinos Best Practices Trial (DIALBEST), we previously demonstrated improvements in glycated hemoglobin (HbA1c) levels among Latinos with type 2 diabetes who received health education from community health workers (CHWs).²³

Although various facets of diabetes care, including physical activity, weight management, and medication adherence, may have contributed to the improved HbA1c levels observed among DIALBEST participants, our primary aim in this study was to examine the mediating effects of 2 interrelated self-care behaviors, namely food label use and diet quality. We hypothesized that an educational intervention delivered by CHWs would have a positive impact on food label use among DIALBEST participants and that food label use and diet quality would mediate the HbA1c improvements attributed to the intervention.

METHODS

DIALBEST was a community-based randomized controlled trial that involved CHWs in the management of patients with poorly controlled type 2 diabetes. The main study design and primary outcome analysis (focusing on changes in HbA1c concentrations) have been described in detail elsewhere.²³ In brief, participants were Latinos attending a community-based ambulatory primary care clinic at Hartford Hospital in Hartford, Connecticut, that targets low-income individuals. To be eligible for the study, individuals were required to be of Latino race/ethnicity, to be aged 21 years or older, to reside in Hartford County, to have been diagnosed by a physician as having type 2 diabetes, and to have an HbA1c level of at least 7%. Individuals with significant diabetes-related complications, mental health problems, or other health conditions (e.g., cancer) that might limit their ability to engage in physical activity were excluded.

A total of 211 participants were enrolled from December 2006 to February 2010, and these individuals were randomly assigned in blocks to either the intervention (n = 105) or control (n = 106) group. In our analyses, we included only the 203 participants (103 from the control group and 100 from the intervention group) who completed the food label questionnaire at baseline (Figure 1). Power calculations were conducted via a set of expected changes in HbA1c values (from baseline starting values) over 4 follow-up time points gathered from a previous meta-analysis focusing on 28 diabetes education program interventions.²⁴ The baseline starting value was the same as the mean baseline HbA1c value (8.6%; 95% confidence interval [CI] = 8.2%, 8.9%) of patients previously enrolled in the Hartford Hospital Brownstone Clinic diabetes program. Accordingly, after we accounted for the 17% dropout rate, 104 individuals per group were required to detect significant (2-sided α = .05) differences between groups (80% power) across 5 time points (99% power) and to detect significant group by time interactions (81% power).

FIGURE 1—

Flow Diagram Showing the Study Design of the Diabetes Among Latinos Best Practices Trial; Hartford County, CT; 2006–2010

Standard of Care

All participants received the standard of care available in the clinic, which included monitoring of height, weight, and blood pressure at each visit; determination of HbA1c levels every 3 months; and an annual comprehensive metabolic panel. Other standard practices included referrals to dietitians, discounted diabetes medications, free glucometers, and prescriptions for glucose monitoring strips.

Intervention

The intervention was delivered by 2 bilingual (English and Spanish) CHWs who had previous experience working with type 2 diabetes patients. The CHWs received additional training from the DIALBEST health management team, which consisted of a medical doctor, 2 registered dietitians, 2 senior researchers, and a group of study coordinators. The study coordinators consistently monitored interviews and educational sessions to ensure adherence to the study objectives and protocols. Participants were compensated $10 at the end of each interview and blood draw.

The intervention group received 17 home visits over a 12-month period (weekly during the first month, biweekly during months 2 and 3, and monthly from months 4–12) from one of the 2 CHWs. Each 75- to 180-minute session was tailored to the participant’s level of literacy and conducted in the context of type 2 diabetes management. Sessions included information on nutrition and food labels, physical activity, type 2 diabetes complications, mental and cardiac health, and adherence to blood glucose monitoring, medications, and medical appointments.

Two of the 6 nutritional sessions provided food label education based on insights from previous randomized controlled²⁵ and qualitative²⁶ trials conducted among Latinos. In these sessions, CHWs initially offered information about carbohydrates, protein, sodium, cholesterol, saturated fat, trans fatty acid, and fiber. They also discussed the impact of various food items on blood sugar levels and lipid profiles.

After these initial lessons, CHWs accompanied participants to the grocery store of their choice to teach them how to use food labels to make wiser nutrition choices that fit their budgets. For example, they helped participants understand, via hands-on activities, the difference between whole and low-fat milk, the difference in the sodium content of canned and frozen foods, and the product content of healthier (monounsaturated or polyunsaturated) versus unhealthier (saturated and trans) fat; they also helped them do carbohydrate counting. Finally, they provided instruction on where to find and how to interpret the information included on food labels (e.g., serving size, servings per container, calories, percentage daily values).

Data Collection

Assessments were conducted at participants’ homes by trained bicultural/bilingual interviewers who were blinded to group assignments. Food label use, diet quality, medication history, body mass index (BMI), history of other chronic diseases, diabetes-related management knowledge, and HbA1c level were assessed at baseline and at 3, 6, 12, and 18 months.

At baseline, data on age, gender, monthly household income (< $500, $501–$1000, $1001–$1500, $1501–$2000, ≥ $2001; these categorical options were provided in the questionnaire), education (< eighth grade, attended high school or equivalent, at least some college/trade training), employment (yes or no), marital status (married/living together, single, separated/divorced, widowed), and health insurance coverage (yes or no) were collected. In addition, participants were asked whether or not they spoke English.

The first author assessed participants’ history of other nutrition-related diseases, such as hypertension, hyperlipidemia, and cardiovascular disease, via self-reports and medication histories. In addition, medication histories were used to obtain information on participants’ consumption of dysglycemic medications^27,28 (yes or no).

Potential Mediators

In the food label questionnaire, participants were asked whether they had ever seen the nutrition facts portion of a food label (after they had been shown a sample label); if they answered yes, they were then asked how often they used labels in selecting foods (always, sometimes, almost never, or never). Participants were classified as food label users if they responded always or sometimes and as nonusers if they responded almost never or never.

The Healthy Eating Index–2010 (HEI-2010), a diet scoring metric that describes the extent to which an individual’s dietary pattern matches the federal dietary guidelines, was used to measure participants’ overall diet quality.²⁹ Diet data for calculating HEI-2010 scores were obtained via 24-hour recall at each time point. These data were entered into the Nutrition Data System for Research (Nutrition Coordinating Center, University of Minnesota, Minneapolis) by trained graduate students blinded to intervention status.

We used the recommended scoring technique²⁹ for calculating total HEI-2010 scores. First, information from the Nutrition Data System for Research was used to calculate serving sizes of 9 “adequacy” components (total fruit, whole fruit, total vegetables, greens and beans, whole grains, dairy, total protein foods, seafood and plant proteins, and fatty acid ratio) and 3 “moderation” components (refined grains, sodium, and empty calories) consumed by each participant. Second, these components were adjusted for participants’ calorie intake, and recommended component-specific scores were calculated in SAS version 9.4 (SAS Institute, Cary, NC). Finally, the scores assigned for the 9 adequacy components and the 3 (reverse-scored) moderation components were summed. Higher scores reflect higher diet quality (maximum score = 100).

A survey comprising 12 items specific to type 2 diabetes was used to assess diabetes-related knowledge (Cronbach α = 0.74).³⁰ Details on the survey have been provided elsewhere (Kollannoor-Samuel et al., unpublished data, 2016). Each participant was assigned a score based on the percentage of correct responses (range = 0%–100%). BMI (defined as weight in kilograms divided by the square of height in meters) was calculated from body weight and height measurements calculated by trained personnel. We assessed physical activity using the American Diabetes Association guidelines,³¹ which recommend moderate physical activity (e.g., brisk walking) for 30 minutes a day at least 5 days a week; participants indicated whether or not they engaged in this level of activity.

The DIALBEST phlebotomist used the A1cNow INView device (Metrika Inc., Sunnyvale, CA) to measure participants’ HbA1c levels at their homes.³² The point-of-care INView instrument is approved by the US Food and Drug Administration and certified by the National Glycohemoglobin Standardization Program.

Statistical Analyses

We applied the intention to treat principle in all of our statistical analyses. Descriptive statistics were generated to characterize the study sample. We conducted the Mann–Whitney U test to assess baseline differences between the intervention and control groups with respect to continuous variables; medians and interquartile ranges (IQRs) are reported. Similarly, we conducted the χ² or Fisher exact test to analyze baseline between-group differences for categorical variables. The significance level was set at P < .05.

To test intervention effects on food label use, we conducted the χ² test at 3, 6, 12, and 18 months. In addition, mixed effects regression analyses were performed to test group and time effects after adjustment for baseline food label use. We used a multiple imputation method to replace missing data. A fully conditional model with linear and logistic regressions was used for imputing continuous and categorical variables, respectively. To ensure that we achieved at least 90% relative efficiency, we imputed four data sets. Estimates obtained from these data sets were combined into a single overall estimate.

The initial mixed model consisted of both fixed (group, time, group by time interaction, baseline food label use, quadratic and cubic time) and random (intercept, time, quadratic and cubic time) effects. Time effects included only postbaseline measurements. We used compound symmetry covariance structures to construct models, as is appropriate for unequally spaced follow-up times.³³ Because the intervention and control groups were balanced at baseline and after baseline, our mixed effects analyses were not adjusted for covariates. A final mixed effects model selected on the basis of Akaike’s information criterion included group status, time, baseline food label use, and random intercept and time values. Odds ratios (ORs) and 95% CIs are reported. Associations were considered significant if confidence intervals excluded 1.

Mediation analyses were conducted via a multilevel structural equation framework for a 2-level model with random intercepts and fixed slopes³⁴; Mplus version 7.4 (Muthen & Muthen, Los Angeles, CA) software was used in conducting these analyses. Repeated measurements (i.e., time; level 1) were assumed to be clustered within individuals (level 2). Because variables measured at level 2 (e.g., group status) were constant within individuals (i.e., time invariant), indirect effects of factors mediating the association between group status and HbA1c levels were assessed only at the between-individual level. Time-varying (level 1) variables, including mediators and HbA1c, were allowed to have both between- and within-individual components. Bayesian estimation was specified, and missing variables were replaced via the full information maximum likelihood method in Mplus. In addition to mediation via the food label use to diet quality path (i.e., group status to food label use to diet quality to HbA1c), mediations via diabetes-related knowledge (group status to diabetes-related knowledge to HbA1c), BMI (group status to BMI to HbA1c), and physical activity (group status to physical activity to HbA1c) paths were also tested.³⁵

The association of food label use with diet quality was adjusted for confounding by group status, age, education, and English-speaking status. These variables were significantly associated with both food label use and diet quality (P < .05), and led to a greater than 10% change in the food label use parameter estimate. Mediation analysis estimates are reported along with their respective 95% confidence intervals.³⁶ These confidence intervals are based on the percentile distribution (2.5th and 97.5th) of an estimate and are considered appropriate³⁶ for determining the significance of an estimate with an unknown (e.g., indirect effect) or nonnormal distribution.^37,38 Associations were considered significant if confidence intervals did not include 0.

RESULTS

The baseline characteristics of the intervention and control groups were comparable. The majority of respondents were middle aged (median age = 57.0 years; IQR = 48.0–65.0), and 73% were women. Approximately two thirds (64.5%) spoke Spanish only. More than half of the participants had less than an eighth-grade education (51.70%), and only 15.8% were employed. Most of the respondents were obese (median BMI = 32.8 kg/m²; IQR = 28.1–37.9).

Participants answered 50.0% (IQR = 41.7–66.7) of diabetes-related knowledge questions correctly. Approximately 59% of the participants used food labels in selecting foods. Participants’ median baseline HEI-2010 score (diet quality) and HbA1c level were 45.2 (IQR = 37.4–54.0) and 9.3% (IQR = 8.2–11.0), respectively (Table 1).

TABLE 1—

Baseline Characteristics of 203 Latino Participants With Type 2 Diabetes: Diabetes Among Latinos Best Practices Trial; Hartford County, CT; 2006–2010

Characteristic	Total (n = 203),a Mean (IQR) or No (%)	Control Group (n = 103),a Mean (IQR) or No (%)	Intervention Group (n = 100),a Mean (IQR) or No (%)	Pb
Age, y	57.0 (48.0–65.0)	59.0 (48.0–67.0)	55.0 (48.0–63.5)	.16
Gender				.78
Male	55 (27.1)	27 (26.2)	28 (28.0)
Female	148 (73.0)	76 (73.8)	72 (72.0)
Monthly household income, $				.53
0–500	14 (6.9)	7 (6.8)	7 (7.0)
501–1000	105 (51.7)	54 (52.4)	51 (51.0)
1001–1500	53 (26.1)	26 (25.2)	27 (27.0)
≥ 1501	15 (7.4)	7 (6.8)	8 (8.0)
Did not know/refused	16 (7.9)	9 (8.7)	7 (7.0)
Education				.73
< eighth grade	105 (51.7)	54 (52.4)	51 (51.0)
Attended high school/equivalent	79 (38.9)	41 (39.8)	38 (38.0)
Some college/trade training	19 (9.4)	8 (7.8)	11 (11.0)
Employed				.63
Yes	32 (15.8)	15 (14.5)	17 (17.0)
No	171 (84.2)	88 (85.4)	83 (83.0)
Marital status				.92
Married/living together	59 (29.1)	29 (28.2)	30 (30.0)
Single	59 (29.1)	31 (30.1)	28 (28.0)
Separated/divorced	56 (27.6)	27 (26.2)	29 (29.0)
Widowed	29 (14.3)	16 (15.5)	13 (13.0)
Health insurance coverage				.16
Yes	175 (86.2)	85 (82.5)	90 (90.0)
No	28 (13.8)	18 (17.5)	10 (10.0)
English speaking				.46
Yes	72 (35.5)	34 (33.0)	38 (38.0)
No	131 (64.5)	69 (67.0)	62 (62.0)
Body mass index, kg/m2	32.8 (28.1–37.9)	32.8 (26.9–38.0)	32.8 (29.2–37.8)	.55
Dysglycemic medication usec				.65
Yes	82 (40.4)	40 (38.8)	42 (42.0)
No	121 (59.6)	63 (61.2)	58 (58.0)
Other chronic nutritional diseases				.39
Yes	75 (37.0)	62 (60.2)	66 (66.0)
No	128 (63.0)	41 (39.8)	34 (34.0)
Food label use				.34
Yes	119 (58.6)	57 (55.3)	62 (62.0)
No	84 (41.4)	46 (44.7)	38 (38.0)
Healthy Eating Index-2010 score	45.2 (37.4–54.0)	46.4 (37.0–56.0)	44.7 (37.9–53.6)	.40
Diabetes-related knowledge score, %	50.0 (41.7–66.7)	50.0 (41.7–58.3)	50.0 (41.7–66.7)	.78
HbA1c, %	9.3 (8.2–11.0)	9.3 (8.1–11.0)	9.1 (8.4–11.1)	.99

Note. HbA1c = glycated hemoglobin; IQR = interquartile range.

^aPercentages may not sum to 100% owing to rounding.

^bP value for Mann–Whitney U Test (continuous variables) or χ² test (categorical variables).

^cDysglycemic medications include medications that can cause either hyperglycemia (thiazide diuretics, calcium channel blockers, nonselective beta blockers, thyroxine, atypical antidepressants, corticosteroids) or hypoglycemia (beta-2 agonists, fibrates, angiotensin converting enzymes).

Intervention Effects on Food Label Use

Food label use was higher among intervention group members than control group members at 3 months (66.3% vs 41.7%; P = .001), 12 months (68.4% vs 47.1%; P = .009), and 18 months (68.5% vs 45.5%; P = .006; Figure 2). Absolute percentage changes in food label use from baseline to 18 months among intervention and control group participants were 6.5% and −9.9%, respectively, resulting in an overall net improvement of 16.4% in the intervention group. Repeated measures analyses suggested that the overall odds of food label use were significantly higher among intervention group than control group members (OR = 2.99; 95% CI = 1.69, 5.29; P < .001).

FIGURE 2—

Food Label Use Among 203 Latinos With Type 2 Diabetes at Baseline and at 3, 6, 12, and 18 Months, by Group Status: Diabetes Among Latinos Best Practices Trial; Hartford County, CT; 2006–2010

Note. CHW = community health worker. Food label use between treatment groups were different at P = .34 (n = 203), P = .001 (n = 170), P = .26 (n = 161), P = .009 (n = 146) and P = .006 (n = 139) at 0, 3, 6, 12 and 18 months, respectively. P values were determined by χ² test.

Mediation Effects

The results of the mediation analyses are summarized in Table 2. At the between-individual level, the overall effect (−0.52%; 95% CI = −0.97%, −0.09%; P < .05) of the intervention on HbA1c levels was statistically significant. The direct effect was smaller than the overall effect and was not significant (−0.40%; 95% CI = −0.87, 0.02). Of the mediation paths tested (group status to food label use to diet quality to HbA1c, group status to diabetes-related knowledge to HbA1c, group status to BMI to HbA1c, group status to physical activity to HbA1c), only the food label use to diet quality path significantly mediated the reduction in mean HbA1c levels (−0.08%; 95% CI = −0.20, −0.02; P < .01) among the intervention participants relative to the control participants. Thus, approximately 15% ([−0.08/−0.52] = 0.150) of the total effect of the intervention on HbA1c levels was associated with the food label use to diet quality path (Table 2).

TABLE 2—

Results of Mediation Analyses Among Intervention and Control Group Participants: Diabetes Among Latinos Best Practices Trial; Hartford County, CT; 2006–2010

Mediatora	Intervention vs Control, OR (95% CI) or Mean Change in Mediator (95% CI)	Food Label Users vs Nonusers, OR (95% CI) or Mean Change in Mediator (95% CI)	Mean HbA1c Difference for Each Unit Increase in Mediatorf (95% CI)	Indirect Effect or Mediation Effectg (95% CI)
Food label use/diet quality	1.91 (1.26, 3.15)b	0.13 (0.05, 0.22)e	−0.99 (−1.84, −0.41)	−0.08 (0.20, 0.02)
Diabetes-related knowledge	0.45 (0.09, 0.85)c	. . .	−0.04 (−0.28, 0.19)	−0.01 (−0.15, 0.09)
Body mass index	1.04 (−1.19, 3.19)c	. . .	0.01 (−0.16, 0.32)	0.00 (−0.04, 0.06)
Physical activity	0.80 (0.52, 1.16)d	. . .	0.09 (−0.16, 0.32)	−0.01 (−0.13, 0.05)

Note. CI = confidence interval; HbA1c = glycated hemoglobin; OR = odds ratio.

^aThe following mediation paths were tested: group status to food label use to diet quality to HbA1c, group status to diabetes-related knowledge to HbA1c, group status to body mass index to HbA1c, and group status to physical activity to HbA1c. Mediation modeling was conducted at the between-individual level.

^bOR for food label use (95% CI) among the intervention participants relative to the control participants; results of simple logistic regression (with Bayesian estimation).

^cMean unit difference in mediator (95% CI) among the intervention participants relative to the control participants; results of simple linear regression (with Bayesian estimation).

^dOR for physical activity (95% CI) among the intervention participants relative to the control participants; results of simple logistic regression (with Bayesian estimation).

^eMean difference in diet quality among food label users relative to nonusers after adjustment for age, education, English-speaking status, and group status; results of multivariable linear regression (with Bayesian estimation).

^fMean difference in HbA1c level for each unit increase in mediator after adjustment for group status, baseline HbA1c, and other mediators (diet quality, diabetes-related knowledge, body mass index, or physical activity); results of multivariable linear regression (with Bayesian estimation).

^gMean difference in HbA1c level contributed by each mediator path among intervention participants relative to control participants.

At the within-individual level, HbA1c values varied significantly across time (−0.12%; 95% CI = −0.19, −0.05; P < .01) and with respect to diet quality (0.11%; 95% CI = 0.00, 0.21; P < .05).

DISCUSSION

Our study produced 2 novel findings. First, we demonstrated the impact of a CHW-led intervention on food label use, an important type 2 diabetes self-management tool. Second, we showed the mediating roles of food label use and diet quality in the association between an educational intervention and improvements in glycemic control.

In our study, the prevalence of food label use was significantly higher in the intervention group than the control group at all of the follow-ups other than 6 months. The overall improvement in intervention participants’ food label use supports the effectiveness of CHW-led education in modifying label use among Latinos with type 2 diabetes. Interestingly, this effect was significantly more pronounced in the 6 months after the completion of the intervention.

The increase in food label use attributed to the intervention is of clinical and public health relevance because increased label use can lead to better glycemic control, as suggested by our novel mediation findings. About 15% of the total effect of the intervention on HbA1c levels was contributed by the food label use to diet quality path, indicating the key role of food label use in consumption of diets of higher nutritional quality and, in turn, HbA1c control among disenfranchised individuals with type 2 diabetes. Thus, our study identified a modifiable mediation path through which better glycemic control can be achieved with the support of CHWs. Indeed, a simple nutritional education intervention delivered both at home and in grocery stores by the CHWs might have equipped our participants with the skills to effectively read food labels, choose a better diet, and achieve better glycemic control.

Our data confirmed the previously reported role of food label use in diet quality. However, most of these earlier studies were conducted among the general population, and either the Mediterranean diet score¹¹ or the HEI-2005 score¹² was used to measure diet quality. We used the HEI-2010, which is based on the current federal dietary guidelines, in our study. In addition, our patient population comprised Latinos with type 2 diabetes, a population that has been heavily understudied in terms of food label use and associated glucose control. Our study also confirmed the important role of diet quality in blood glucose control among patients with type 2 diabetes reported previously in multiethnic populations.^13,14

Our findings must be interpreted from a social and cultural perspective. Latinos often report a lack of certain cultural values (e.g., respect and warmth) on the part of health care providers.³⁹ One of the major challenges often faced by medical professionals in educating type 2 diabetes patients from diverse racial/ethnic backgrounds is a lack of understanding of these individuals’ cultural values and cooking or eating styles.⁴⁰ In addition, limited English proficiency may lead to communication barriers between health care providers and Latino patients.⁴¹ Approximately two thirds of our Latino participants did not speak English; in this context, medical information conveyed via bilingual CHWs might have led to higher levels of acceptance in our sample.⁴²

Limitations

Our Latino population was composed predominantly of low-income Puerto Ricans. Thus, the external validity of our findings with respect to other Latino subgroups remains to be determined. Also, because all of the participants lived in the same area, between-group contamination may have occurred, which may have biased our results toward the null. In addition, our study was based on a CHW home visitation model and included 17 sessions, which may affect the feasibility of implementing a similar intervention in another setting. Thus, hypotheses regarding the effectiveness of group session approaches in isolation as compared with combinations of short home visits and group sessions will require formal testing in future studies.

Conclusions

Our study highlights the positive impact of CHWs on the ability of Latinos with type 2 diabetes to improve their food label use. We found that food label use led to better diet quality, which in turn led to improvements in blood glucose control, and the benefits were sustained after the completion of the intervention.

Our findings are robust in that they are based on an experimental study that allows causal inferences to be drawn. In the context of the Affordable Care Act’s suggested use of CHWs in health care delivery,⁴³ our results contribute significantly to the literature on the benefits derived from CHW-led interventions targeting individuals with chronic diseases. The effect sizes we detected are moderate to strong, and our study provides definitive support for the inclusion of CHW-led food label education in the type 2 diabetes management protocols used in clinical and community settings.