Introduction

A high-quality plant-based diet with limited sweetened beverages during early life reduces the risk for gaining excess weight and developing diabetes and diabetes-related complications later in life.1–3 Dietary trends during the past decades in the USA, however, have shown an increase in the intake of high-energy and added sugar foods,4–6 possibly contributing to the increase in the incidence of obesity and early-onset type 2 diabetes (T2D).2 Since people with early onset T2D are more vulnerable to severe hyperglycemia and microvascular and macrovascular complications, compared with those who develop diabetes later in life, research that facilitates better dietary behaviors and improves diet quality is needed for young adults at risk for and/or with T2D.5–7

Food intake changes over time.8–11 During younger adulthood, there is a tendency to consume less fruit, vegetables, and dairy and more meat, potatoes, and sweets, which results in poor diet quality.9 11 12 Since diet quality is associated with the development of T2D, and early-onset T2D is increasing,1 13–15 it is important to know how young adults make dietary change over time, whether diabetes awareness or the age of diabetes awareness are associated with dietary quality, and what dietary changes are made when people learn of their diabetes. Existing studies are short-term, qualitative, have small samples, lower quality of evidence, start with middle-aged cohorts, or focus only on whether diet quality predicts future risk of developing diabetes and diabetes-related complications.8 15–17

Using the Coronary Artery Risk Development in Young Adults (CARDIA) prospective cohort, this study examined how American adults changed their diet over time and what longitudinal changes in diet quality and food intake were made once people learned of their diabetes. The hypotheses were: (1) lower diet quality in young adulthood is associated with incident diabetes later in life and (2) advancing age and diabetes awareness are associated with more favorable dietary changes in subsequent adulthood.

Research design and methodsStudy design

A nested case-control design referring to a case-control design in a cohort study was used.

Study population

CARDIA is a longitudinal, ongoing study that began in 1985–1986 with a cohort of 5115 healthy black and white women and men aged 18–30 years who were recruited from four US urban areas. The detailed study design has been described elsewhere.18 19 Briefly, participants were invited to participate in nine follow-up examinations from baseline with retention in the surviving cohort ranging from 91% in Y2 (1987–1988) to 71% in Y30 (2015–2016). By study design, the sample included an approximately similar number of participants by sex, self-identified race (black or white), age (18–24 vs 25–30 years), and education level (below high school vs above high school) at each center.18

For the current study, the CARDIA prospective cohort free of diabetes at baseline was used. Exclusion criteria included missing blood glucose data at baseline, having fewer than two CARDIA examinations, or having extreme values for energy intake (<800 or >8000 kcal/day for men and <600 or >6000 kcal/day for women). The final analysis included 4576 participants.

Diet assessment

Dietary intake was assessed during three CARDIA examinations (Y0, Y7, and Y20) using the CARDIA dietary history.19 20 Sample size for diet data at years (Y)0, Y7, and Y20 is 4576, 3761, and 3013, respectively. Trained interviewers asked participants open-ended questions about their food and beverage consumption over the previous month under 100 broad food categories. Questions included frequency of consumption, unit or serving size, and preparation method of food. Total energy and nutrient intake were calculated using the Nutrition Data System for Research (NDSR, University of Minnesota, Minneapolis, Minnesota, USA).21

At each examination, the recorded data were grouped into 166 food groups according to the NDSR food group query system and then summarized into 46 groups for creating a diet quality score, the A Priori Diet Quality Score (APDQS), a hypothesis-driven diet quality index based solely on foods or dietary patterns rather than individual nutrients.22 It is composed of 20 beneficially rated, 13 adversely rated, and 13 neutrally rated food groups, according to their presumed influence on cardiovascular diseases (CVDs).8 22 With beneficially rated food groups, participants in the highest quintile of a food category earned a score of 4, with scores of 3, 2, 1, and 0 for lower quintiles. Adversely rated food categories were scored using the inverse approach. A higher APDQS was driven primarily by increased consumption of nutrient-dense plant foods and decreased consumption of high-fat, red/processed meats, and unhealthy plant-based meals (eg, fried potatoes, salty snacks).8 23 Possible APDQS scores ranged from 0 to 132, and the obtained scores in the current study ranged from 35 to 95. A correlation between Alternative Healthy Eating Index and APDQS was 0.65.22

Ascertainment of incident diabetes, diabetes awareness, and group classification

Ascertainment of incident diabetes was determined as the first occurrence of any of the following criteria after baseline through Y30: fasting plasma glucose ≥126 mg/dL (7.0 mmol/mol) (measured at Y0, Y7, Y15, Y20, Y25, Y30), 2 hour postload glucose ≥200 mg/dL (11.1 mmol/mol) in the oral glucose tolerance test (measured at Y10, Y20, or Y25), glycated hemoglobin ≥6.5% (48 mmol/mol) at Y20 or Y25, or a self-antidiabetic medication use (per medication bottle brought to the clinic) at each data collection time point (Y0, Y2, Y7, Y10, Y15, Y20, Y25, Y30). The ascertainment (or diagnosis) of diabetes could result from a CARDIA biochemical test at an earlier examination or made by a physician outside of the scheduled CARDIA examination. About half of the CARDIA participants were first informed of diabetes presenting with abnormal test result(s) during a CARDIA examination while the remaining participants were diagnosed by independent healthcare providers not part of the CARDIA study. Once participants were informed of diabetes using the above criteria, glucose-lowering medication was prescribed by a physician outside of the CARDIA study and then participants were considered to have diabetes at all subsequent CARDIA examinations. Individuals who self-reported diabetes, but had normal glucose values and reported not currently taking glucose-lowering medication(s) at the corresponding CARDIA examination were considered to not have diabetes at that examination.

Participants who reported taking glucose-lowering medication or had hyperglycemia at a prior CARDIA examination were classified as being aware of their diabetes. Participants who had hyperglycemia for the first time at a given CARDIA examination and reported currently taking glucose-lowering medication or had elevated fasting glucose with self-reported diabetes in a survey item were considered to be aware of their diabetes at the corresponding examination. If a participant had hyperglycemia according to the CARDIA biochemical tests but reported not currently taking glucose-lowering medication and answered no to having diabetes in the self-reported survey item, the participant was considered as being unaware of the diabetes at that specific examination, but classified as being aware of diabetes in subsequent examinations.

We hypothesized that diabetes awareness would be associated with more favorable dietary changes in subsequent CARDIA examinations. Therefore, groups were determined by the first time of diabetes awareness relative to the year (Y0, Y7, and Y20) of diet assessment (figure 1), and participants were partitioned into either a control group (diabetes was never ascertained in any examination during the 30-year follow-up (Y0–Y30)) or one of three diabetes groups:

Early-onset group: participants aware of their diabetes before Y7 data collection.

Intermediate-onset group: participants aware of their diabetes after Y7 or before Y20 data collections.

Later-onset group: participants aware of their diabetes after Y20 data collection.

Figure 1

Group classification and milestone events. AHEI, Alternative Health Eating Index; APDQS, A Priori Diet Quality Score; CARDIA, Coronary Artery Risk Development in Young Adults; DM, diabetes mellitus; FBG, fasting blood glucose; HbA1C, hemoglobin A1C; HEI, Health Eating Index; MNT, medical nutrition therapy; N/A, not available; Y, years.

Assessment of covariates

Covariates (age, sex, race, education, smoking status and history, family history, and parental history of diabetes) were self-reported and collected at baseline. These covariates were chosen from the preliminary data analysis summarized in table 1 and the literature.8 23 24 Pack-years of smoking was computed by multiplying the number of packs of cigarettes smoked each day by the number of years the person had smoked. Height and weight were measured to the nearest 0.5 cm and 0.5 pounds, respectively, by trained professionals. Body mass index (BMI) was calculated by dividing the weight in kilograms by the height in meters squared (kg/m2). Physical activity level was obtained from a validated interviewer-administered questionnaire.25 Participants were asked about how often they engaged in 13 moderate-intensity or vigorous-intensity physical activities over the previous year. The overall physical activity score was computed by adding the weighted products of all activities (frequency in months×intensity of activity) and represented in exercise units (EU). According to a CARDIA substudy, 300 EU corresponds to 150 min/week of moderate or vigorous physical activity.26

Table 1

Participant characteristics at Y0, Y7, Y20 by four groups according to diabetes awareness status (n=4576)

Statistical analysis

To examine longitudinal changes of diet according to the aging process and diabetes awareness, a nested case-control design with repeated measures regressions stratifying groups by diabetes awareness and the timing of the CARDIA diet data collection was used. To investigate potential bias related to missing diet data, a sensitivity analysis of baseline characteristics was conducted, comparing those with all follow-up diet data with those missing diet data at either Y7 or Y20. Group differences on participants’ baseline characteristics were examined with χ2 tests for categorical data and analysis of variance tests for continuous data, respectively. The principal linear regression assumptions were checked for each independent variable, including the normality assumption by using the Kolmogorov-Smirnov test.

Hypothesis 1 was tested to examine the associations between diet quality in young adulthood and incident diabetes later in life. The mean APDQS and food intake (servings/day) in 12 collapsed food groups (5 beneficially rated, 3 neutrally rated, and 4 adversely rated food groups) at Y0, Y7, and Y20 were compared across the four case and control groups (online supplemental figure 1). For each of the case and control groups, adjusted means of APDQS with 95% CIs were estimated in linear regression analyses specific to each of Y0, Y7, and Y20 after adjusting for age, sex, race, maximal educational attainment, parental history of diabetes, physical activity, pack-years smoking, total energy intake, and BMI. To create a single net difference across the whole of CARDIA comparing people with diabetes versus control group when case groups were unaware of their diabetes, further assessment on APDQS and food intake as a repeated measures dependent variable using a Toeplitz variance-covariance structure in the model was performed after adjusting for age, sex, race (black and white), maximal educational attainment, parental history of diabetes, physical activity, pack-year of smoking, total energy intake, and BMI at corresponding examination year in order to remove the dietary changes occurring at the specific year and aging.

Hypothesis 2 was tested to examine the changes in diet as people get older (natural progress) and become aware of diabetes (intentional choice). Within the participants with diabetes, those who had diet assessments both before and after becoming aware of their diabetes status were included. Given this, the early-onset and intermediate-onset groups were only included. As a control group, individuals without diabetes over a 30-year of follow-up were included. Then, regressions parallel to those for hypothesis 1, but using changes in APDQS (Y7–Y0, Y20–Y7, and Y20–Y0) as the dependent variables were performed. Adjustments for the above variables at baseline and their changes for physical activity, pack-years of smoking, and BMI were made. This yielded five means and 95% CIs for the early-onset and intermediate-onset diabetes groups, plus three corresponding values for the control group. An overall comparison was performed using repeated measures regressions of the combined diabetes group versus the control group (net difference of differences), namely the combined mean of APDQS changes after—before awareness in case groups minus the corresponding APDQS mean change in the no diabetes group. Two-tailed tests of significance were set at p<0.05. All statistical analyses were performed using SAS V.9.4 (SAS Institute).

ResultsBaseline characteristics of participants

Table 1 shows participant characteristics at baseline. The mean age at baseline was 25.0±3.6 years with 55.1% female and 49.4% black race. The case (diabetes) groups were older, were more likely to include black participants, had less education, and had a higher prevalence of parental diabetes history. Over 30 years, 653 incident diabetes cases were identified.

Potential covariates at Y7 and Y20 are presented in table 1. Compared with the control group, diabetes groups had higher BMI, were less physically active, and had lower pack-year of smoking, generally graded across early onset, intermediate-onset, and later-onset groups. There was no association between energy intake and diabetes status at each Y7 and Y20.

Association between diet quality in young adulthood and incident diabetes

Lower diet quality in young adulthood was associated with incident diabetes later in life (table 2, online supplemental figure 2). However, no clear pattern was seen according to the timing of incident diabetes. For example, the mean APDQS at Y0 was 61.5 in the later-onset group, and the mean APDQS was slightly higher in the early- onset or intermediate-onset groups despite incident diabetes occurring earlier. The mean APDQS of all three case (diabetes) groups at Y0, however, was lower than a mean of 62.8 in the control group.

Table 2

Means (95% CIs) and differences of APDQS* at each diet assessment and the net differences of APDQS by diabetes awareness status

Diet quality and food intake by group over 20 years depending on diabetes awareness status

In general, the mean APDQS increased with advancing age and diabetes awareness in the CARDIA cohort. The greatest change in APDQS was observed between examinations when people first learned of their diabetes (online supplemental figure 2). The net difference of the adjusted mean APDQS, a combined mean across all diabetes groups in a repeated measures analysis, was 1.29 points lower than that of the control group when people were unaware of their diabetes. The improvement of APDQS in diabetes groups was greater than the change in the same time period in the control group after diabetes awareness. A net difference of mean APDQS was increased by 2.95 points once people learned their diabetes (table 2).

Food intake between case-control groups were examined depending on diabetes awareness. When participants were unaware of their diabetes, the combined diabetes groups had 0.84 less servings/day of beneficially rated food groups, and 0.45 servings/day of more neutrally rated foods, respectively, compared with the control group (online supplemental table 1). Most of these differences resulted from plant food choices. Diabetes groups had 1.27 more servings/day of adversely rated foods compared with the no diabetes group when they were unaware of their diabetes. Consuming more soft drinks, meat products, and less whole-fat dairy contributed to this difference (table 3, online supplemental table 1).

Table 3

Means (95% CIs) of food* intake (servings/day) in 12 collapsed food groups at Y0, Y7, Y20 and the net differences before diabetes awareness

Changes in diet quality and food intake over 20 years under the condition of diabetes awareness

While diet quality score increased with advancing age and diabetes awareness, this pattern was not seen consistently in the smallest group, early- onset group (online supplemental figure 2). The mean APDQS change from Y0 to Y20 in the early-onset group was an increase of 3.07 points resulting from a decrease of 4.03 points when the participant numbers decreased from 21 to 11 between Y20 and Y7. The greatest increase of mean APDQS change over 20 years occurred in the intermediate-onset group (11.01 points), which was bigger than an increase of 6.92 in the control group at the corresponding time (online supplemental table 2).

Changes in APDQS and food intake (servings/day) after becoming aware of diabetes are presented in tables 2 and 3, respectively. After people learned of their diabetes, the consumption of beneficially rated and neutrally rated food groups was increased by 2.08 servings/day and 1.82 servings/day, respectively. The consumption of adversely rated food groups decreased by 1.27 servings/day (figure 2, online supplemental table 2).

Figure 2

Three categories of food group subscores at Y0, Y7, Y20 according to diabetes awareness. (a) Beneficially rated food group_subtotal, (b) neutrally rated food group_subtotal, (c) adversely rated food group_subtotal. DM, diabetes mellitus.

The early-onset group showed a unique pattern between Y20 and Y7 with regard to the change in the adversely rated meat products (table 3, online supplemental figure 3C). There was an increase of 2.98 servings/day while control group showed a decrease at the corresponding time. Alcoholic beverage intake and whole-fat dairy intake became significantly lower at subsequent examinations once they learned of their diabetes for the first time, −0.62 serving/day and −1.42 serving/day, respectively (online supplemental table 2). The net differences of differences, mean changes across all diabetes groups minus a corresponding change in the control group, were modest, −0.27 to −0.38 servings/day, respectively. The diabetes groups drastically decreased soft drinks once they learned of their diabetes and this change maintained in early-onset group at the subsequent examinations while they increased diet soda and fruit juices when they first learned of their diabetes (table 3, online supplemental figure 3B and 3C).

Discussion

Using a unique investigation in tracking longitudinal dietary intake, this study revisited the importance of healthy dietary behaviors beginning in young adulthood to prevent or delay diabetes.8 15 27 While diet quality was improved with advancing age and diabetes awareness, especially when people learned their diabetes for the first time, the changes in food intake were not always concordant with evidence-based dietary recommendations. Therefore, individualized nutrition counseling and intervention need to be provided to people with diabetes at diagnosis and over time in order to promote longitudinal healthy dietary behaviors leading to diabetes remission and minimizing risks for developing diabetes complications.23 28

Forming well-balanced dietary habits should be emphasized beginning with young adulthood or even younger since it can continue throughout a lifetime and influence longitudinal health outcomes.8 29 In the current study, the diabetes groups compared with the control group consumed more adversely rated foods such as fried or salty plant foods, processed meat, or sweetened drinks in young adulthood or when they were free of diabetes. This dietary pattern may contribute to increasing their risk for diabetes along with other risk factors (ie, parental diabetes history, higher BMI, lower physical activity). Implementation of a realistic, practical and proactive nutritional intervention targeting at-risk young adults along with early detection using an appropriate nutritional screening tool is imperative.5 30

Over the years, the diagnostic criteria for diabetes and medical nutrition therapy (MNT) guidelines have been updated based on advances in research.5 31–34 Milestone research (ie, National Diabetes Prevention Program, Diabetes Control and Complication Trial) has reported the significance of diabetes prevention and management. However, the prevalence of T2D, especially early-onset T2D, is increasing, and the prevalence of optimal glycemic management continues to decline.35 36 Potential reasons may be related to translation time lags and food illiteracy.37 38 For instance, the diabetes group, especially the early-onset group, switched beverage choice from soft drinks to diet soda and fruit juices at Y7 when they first learned of their diabetes. This may be related to the people’s tendency to change self-care behaviors following disease diagnosis and symptom awareness without consideration of information source.17 39 40; many people at risk for or with diabetes tend to find healthy food substitutes when they learned of their health conditions. However, healthy eating does not mean to have ‘less tasteless’ food.17 40 Evidence-based individual counseling and education enable to provide support people to make better and enjoyable food selections without unpleasant feelings.5 Also, effective behavioral strategies with considerations of patient and disease characteristics are warranted to translate advances in nutritional sciences into MNT without time lags.6 38 41

This study raises questions about consuming dairy products. The APDQS rates whole-fat dairy products as adverse while low-fat dairy products are rated as beneficial. The case groups, however, consumed less whole-fat dairy products compared with the control group when they were unaware of their diabetes, while the difference disappeared after case groups learned of their diabetes. There are three possible explanations. First, dairy product consumption decreases with advancing age,17 42 43 and thus, the whole-fat dairy effect on developing diabetes may be diminished as participants’ age. In the current study, there was no significant net difference of differences on whole-fat dairy product consumption between case-control groups after participants learned of their diabetes and/or were getting older. Another possibility is a protective effect of black coffee, tea, and moderate alcohol consumption on diabetes development.7 44 Our data showed increases in coffee/tea in Y7 and Y20 across all groups. Also, the control group had modest alcohol consumption across all years while case groups decreased alcohol intake at Y7 and Y20.

Second, there is limited evidence about whether low-fat dairy products provide superior health benefits compared with whole-fat dairy products. A recent meta-analysis of randomized clinical trials evaluating the association between dairy product intake with the incidence of T2D concluded that intake of dairy products, especially low-fat dairy products, prevented diabetes via improving insulin resistance and reducing visceral fat and weight.45 However, the findings were not from experimental research that compared low-fat versus whole-fat dairy product consumption. The meta-analysis evaluated the dairy product effect on T2D with studies mainly focused on low-fat dairy products.

Lastly, the APDQS was created in 2007 and therefore reflects earlier advice when nutrition (eg, low-fat) and diet quantity (eg, energy intake) were emphasized.11 22 46 47 Over time, studies have shown the importance of healthy fat and high-density lipoprotein (HDL)-cholesterol and have initiated a re-evaluation of diets with full-fat dairy products.48 49 Although dairy fat may increase low-density lipoprotein-cholesterol, it also increases HDL-cholesterol and has uncertain protective effects on apolipoprotein C-III.48 Additionally, vitamin D (a fat-soluble vitamin) and calcium included in milk regardless of the level of fat content may play a protective role in diabetes development.49 That is, dairy fat may not be harmful to health as long as people are aware of the caloric intake and do not exceed their daily recommended total energy intake in order to prevent obesity.50 Further investigation is needed in this area.

This study extends prior CARDIA analyses that reported that changes in the APDQS were associated with subsequent diabetes risk as a binary outcome (event vs no event).2 8 The novel approach in the current study examined diabetes awareness with diet quality and food intake in a standardized fashion using serving size in a healthy cohort, which produced high-quality evidence supporting MNT.16 51 Using a nested case-control design with a relatively large sample size, the findings added high quality of evidence about how people make dietary change over time once they learned of their diabetes. Importantly, it shows a pattern of dietary changes in young adults with diabetes and provided a suggestion regarding the timing of nutrition education to be offered for young adults with early onset diabetes.52 APDQS temporarily increased and then decreased in the small early onset group, to a level that was even lower than the control group at Y20. Temporary improvement in diet quality may result from psychological distress (eg, being scared) due to diabetes diagnosis or transient efforts to make healthy lifestyles. Young adults, however, may be vulnerable of returning to long-term behavioral equilibrium (lower quality diet),40 53 and thus intensive and individualized intervention at appropriate timing is very important for this population. Additionally, the current study successfully identified certain food groups as a focus for future MNT to prevent or delay diabetes development and improve diabetes management while the difference in APDQS among groups seems very small. One possibility is that ‘slight fluctuations are sometimes not slight’. That is, younger individuals may provide compensatory physiological mechanisms to maintain euglycemia, but with small changes in dietary quality, along with BMI and age could affect diabetes onset, especially in those with high diabetes risk.54 Another possibility is that some specific advice, for example concerning whole-fat dairy, may be inaccurate. Further investigation using a rigorous research method is needed.

Our study has limitations. First, the CARDIA study was initiated by recruiting a biracial cohort in 1985–1986 to address racial disparities in CVDs,19 20 but greater diversity of people at risk for diabetes has occurred.4 55 Thus, research representing a current trend of population characteristics needs to be replicated. Second, few participants were classified as the early-onset group, thereby limiting the precision regarding estimates of subsequent changes made years later. To address this limitation, we calculated the net difference and the net difference of differences to provide an overall view of the differences between case and control groups over 20 years. Third, all changes in dietary quality and food intake were not observed due to unmeasured and unavailable variables. For instance, changes in dietary quality and food intake in the later-onset group were not included in the current study. Data from CARDIA Y35 including diet assessment will provide an answer soon. Fourth, we identified an unexpected result regarding alcohol intake. The case groups drank less alcohol than the control group and the difference was increased after diabetes awareness or with advancing age. Currently, there is a consensus about modest alcohol consumption (one or less drink per day) being beneficial for people with diabetes and prediabetes.7 This recommendation allows modest alcohol intake and avoids heavy alcohol consumption rather than encouraging the initiation of drinking.27 Message framing about alcohol consumption need to be carefully designed in order that the messages are delivered to the public as intended. Fifth, there is a concern about loss to follow-up, which is inevitable in longitudinal study design. However, in the sensitivity analysis of baseline characteristics comparing those with all follow-up diet data and those without follow-up diet data at either Y7 or Y20, there was no substantial difference with regard to the drop-out rate. Thus, the main estimate is unlikely to be considerably affected by selection bias due to loss to follow-up. Sixth, the CARDIA study did not collect autoimmune markers to determine type 1 diabetes vs T2D. Therefore, the type of diabetes could not be determined in this current investigation. Future research is needed to examine whether type of diabetes influences dietary behaviors.

Lastly, the present recommendation of macronutrient composition is a major change from previous dietary guidelines.22 23 33 Currently, the ADA recommends individualized meal planning based on current eating patterns, individual preferences, and metabolic goals.5 There is an increasing body of evidence suggesting that moderate consumption of carbohydrates (about 45% of total energy) and fat (36%~40% of total energy) may be beneficial to prevent and manage diabetes.5 7 While the reduction of energy intake was observed over time with no real difference across groups except Y0, CARDIA participants were not asked how they perceived their diet quality or whether they were aware of changing dietary guidelines. Further investigation about what motivates people to make dietary changes is warranted to identify a better way to deliver patient-centered MNT.