Unhealthy plant-based diet is associated with a higher cardiovascular disease risk in patients with prediabetes and diabetes: a large-scale population-based study

Abstract

Background

The role of plant-based dietary patterns in preventing cardiovascular disease (CVD) among individuals with prediabetes and diabetes remains unclear. We aimed to evaluate the associations of plant-based diet index (PDI), healthful PDI (hPDI), and unhealthful PDI (uPDI) with cardiovascular disease (CVD) risk and explore potential contributing factors among people with prediabetes and diabetes.

Methods

A total of 17,926 participants with prediabetes and 7798 with diabetes were enrolled from the UK Biobank between 2006 and 2010 and followed until the end of 2020. We calculated the PDI, hPDI, and uPDI based on 18 major food groups including plant-based foods and animal-based foods and applied Cox proportional hazard models to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for CVD risk related to PDI, hPDI, and uPDI. Decomposition analysis was performed to assess the role of dietary components, and mediation analysis was performed to assess the potential mediating role of serum biomarkers underlying these associations.

Results

A total of 2324 CVD events were documented among individuals with prediabetes, while 1461 events occurred among patients with diabetes. An inverse association was found between hPDI and CVD risk among individuals with prediabetes (HR T3 vs. T1 = 0.88, 95% CI = 0.79–0.98, P_trend = 0.025) but not those with diabetes. A positive association was found between uPDI and CVD risk among individuals with prediabetes (HR T3 vs. T1 = 1.17, 95% CI = 1.05–1.30, P_trend = 0.005) and those with diabetes (HR T3 vs. T1 = 1.14, 95% CI = 1.00–1.29, P_trend = 0.043). High-sugar-sweetened beverages (SSB) intake accounted for 35% of the hPDI-CVD association and 15% of the uPDI-CVD association among individuals with prediabetes, whereas low intake of whole grain accounted for 36% of the association among patients with diabetes. Elevated cystatin C levels explained the largest proportion of the association between uPDI and CVD risk among individuals with prediabetes (15%, 95% CI = 7–30%) and diabetes (44%, 95% CI = 9–86%).

Conclusions

Adherence to an unhealthy plant-based diet is associated with a higher CVD risk in people with prediabetes or diabetes, which may be partially attributed to low consumption of whole grains, high intake of SSB, and high blood cystatin C levels.

Background

International Diabetes Federation Diabetes Atlas reported a staggering 537 million adults living with diabetes in 2021 [1]. Prediabetes, a condition that significantly increases the risk of developing diabetes and its associated complications and mortality [2], is projected to affect more than 470 million people in 2030 [3]. Over the past two decades, both diabetes and prediabetes have emerged as a major medical concern worldwide [4].

Complications that have traditionally been associated with diabetes include macrovascular and microvascular conditions [5]. Cardiovascular disease (CVD) is the leading cause of morbidity and mortality in the population with diabetes [6]. Dietary patterns and nutrition management are important components of diabetes treatment [7] and play an important role in the primary and secondary prevention and resistance of diabetes [8]. A variety of dietary patterns emphasizing the consumption of nonstarchy vegetables, whole fruits, legumes, whole grains, nuts/seeds, and low-fat dairy products and minimizing consumption of meat, sugar-sweetened beverages, sweets, refined grains, and ultraprocessed foods are recommended [9, 10]. A high certainty of evidence was found for the beneficial effects of liquid meal replacement on reducing body weight and a low carbohydrate diet (< 26% of total energy) on HbA1c among persons with diabetes [11]. Epidemiological research suggests that risks of CVD may be reduced by following plant-based dietary patterns, especially the healthful version [12]. However, the role of plant-based dietary patterns in managing diabetes, particularly in preventing CVD among individuals with prediabetes and diabetes, remains uncertain. Studies have shown that improving the quality of plant-based diets is associated with lower CVD mortality, while adhering to unhealthy plant-based diets is associated with higher CVD mortality [13]. Additionally, greater adherence to an overall or healthful plant-based diet was associated with a lower risk of both prediabetes and type 2 diabetes (T2D) [14, 15]. However, no study has examined the association of a plant-based diet with CVD development among individuals with diabetes/prediabetes. Since the distinctive metabolic alterations observed in these participants, including impaired carbohydrate and fat metabolism, dyslipidemia, and a prothrombotic profile, the extrapolation of findings from healthy individuals to patients with diabetes remains uncertain. Furthermore, possible mechanistic pathways through which plant-based diets may influence cardiovascular health are uncertain and require further investigation [16].

To fill these gaps, this study examined the associations of various plant-based diets (overall, healthful, and unhealthful diets) with the risk of CVD in individuals with prediabetes and diabetes (type 1 and type 2 diabetes), using data from a large population-based prospective study. Additionally, we assessed the potential mediating role of dietary components and serum biomarkers underlying these associations.

Methods

Study population

The UK Biobank is a very large and detailed prospective study, recruiting over 500,000 participants aged 40 to 69 years between 2006 and 2010 [17]. Participants completed touchscreen questionnaires, took physical measurements, and provided biological samples at one of 22 assessment centers across Wales, Scotland, and England, including Edinburgh, Glasgow, Bristol, Cardiff, Birmingham, Oxford, etc. (https://www.ukbiobank.ac.uk/enable-your-research/about-our-data/baseline-assessment) [18].

This analysis focused on 60,540 participants with prediabetes and 31,422 patients with diabetes. Among them, we excluded those who had CVD or cancer at baseline and those without data of 24-h dietary recall [19]. A total of 17,926 individuals with prediabetes and 7798 with diabetes remained in the final analyses (Additional file 1: Fig. S1).

Estimation of PDI, hPDI, and uPDI

UK biobank used the Oxford WebQ (web-based dietary questionnaire), an online 24-h dietary recall that contains questions on the consumption of nearly 200 foods and drinks [20], to collect dietary data. The questionnaires have been validated using biomarkers [19]. Between 2009 and 2012, participants with accessible mail addresses were invited to complete five occasions dietary questionnaires. Participants who completed at least one of the five assessments were included in the analyses, and we calculated the average dietary intake using all available assessments to represent long-term dietary intake and reduce bias due to measurement error.

Plant-based diet indexes (PDIs), including an overall PDI, a healthful PDI (hPDI), and an unhealthful PDI (uPDI) were created to distinguish food sources and plant food quality [21]. Additional file 1: Table S1 shows detailed information on the classification of different food groups. We categorized 18 food groups into three categories: healthy plant foods, less healthy plant foods, and animal foods. Then we ranked the consumption of each food group into quintiles and assigned positive or reverse scores. For positive scores, the highest quintile of the food intake got 5 points and the lowest quintile got 1 point, while reverse scores had the opposite scoring rule. Healthy plant foods were assigned positive scores for PDI and hPDI but reverse scores for uPDI. Less healthy plant foods were assigned positive scores for PDI and uPDI but reverse scores for hPDI. Animal foods got reverse scores in all three patterns. Scores from 18 food groups were summed to obtain the final score for all three patterns, each ranging from 18 to 90. A higher score represented a higher adherence to the corresponding dietary pattern.

Definitions of prediabetes and diabetes

According to the 2021 diagnostic criteria from the American Diabetes Association guideline, participants were classified as having prediabetes based on a hemoglobin A1c (HbA1c) level ranging from 5.7% to 6.4% (39–47 mmol/mol) [22].

Patients with diabetes were ascertained using the UK Biobank diagnostic algorithms for diabetes [23]. Detailed information on diabetes ascertainment can be found at https://biobank.ndph.ox.ac.uk/showcase/label.cgi?id=2000. Individuals with diabetes were included if the participants reported they had diabetes or took hypoglycemic medicine at baseline assessment, covering type 1 and type 2 diabetes. Considering the date and etiology of hospital admissions, patients with diabetes were also identified using hospital inpatient records with the ICD-9 code 250 and ICD-10 code E10-14 and diagnosed before the baseline survey. The baseline type 1 diabetes was ascertained according to cumulative hospital inpatient records using ICD codes or self-reported T1D. Individuals with an HbA1c level above 6.5% (48 mmol/mol) who had not been previously diagnosed with diabetes were also included. Persons with gestational diabetes were not included. The detailed definition is shown in Additional file 1: Table S2.

Assessment of CVD

CVD incidence was the primary outcome of this study. Until 31 December 2020, the ICD-10 codes were used to define incident CVD events, including coronary heart disease (CHD), stroke, and atrial fibrillation (AF) events, using data from hospital inpatient records [17]. Hospital admission records were collected through linkages to the Health Episode Statistics for England, the Patient Episode Database for Wales, and the Scottish Morbidity Records for Scotland. Additional file 1: Table S2 shows the detailed definition of incident CVD, CHD, stroke, and AF cases. The person-year was calculated from the date of the first 24-h dietary recall to the occurrence of CVD, lost to follow-up, death, or the end date of follow-up, whichever occurred first.

Covariates

Baseline questionnaires collected variables including sociodemographic factors, medical histories, drug use, major demographic, and lifestyle factors. Detailed information included age, sex, race, centers, household income, Townsend deprivation index (TDI), smoking, alcohol drinking, physical activity, history of hypertension and hypercholesteremia, family history of CVD and diabetes, supplements use, and aspirin use. Blood samples were collected at baseline, and blood biomarkers were measured prior to the 24-h dietary recalls. Detailed information on these measurements is available online at https://biobank.ctsu.ox.ac.uk/showcase. BMI was calculated by dividing weight in kilograms by height in squared meters. TDI is a simple census-based index of material deprivation calculated on the basis of non-home ownership, non-car ownership, unemployment, and overcrowding in households [24]. Physical activity was expressed in metabolic equivalent of task (MET)-hours per week (MET-h per week) [25]. For participants with missing covariates, indicator variables were created if necessary.

Statistical analysis

After checking the proportional hazard assumption, Cox proportional hazards model was employed to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) of CVD incidence associated with PDIs. We built multivariable models (adjusted Cox models) to control known and potential confounders. Model 1 was adjusted for age and sex. Model 2 further included race, centers, BMI, household income, TDI, smoking, alcohol drinking, physical activity, history of hypertension and hypercholesterolemia at baseline, family history of CVD, family history of diabetes, vitamin supplementation, mineral supplementation, and aspirin use. Diabetes duration was also included in model 2 for patients with diabetes. Missing indicator categories were used for missing covariate data.

We conducted stratified analyses and tested potential interactions of following factors sex, age, BMI, household income, TDI, smoking, alcohol drinking, physical activity, history of hypertension, history of hypercholesterolemia, family history of CVD, family history of diabetes, vitamin supplementation, mineral supplementation, and aspirin use on the associations between PDIs and CVD risk. These potential interactions were tested by including the cross-product term of PDIs and each stratifying factor in the model.

To test the robustness of our findings, we performed sensitivity analyses. These included additional adjustments for the use of lipid-lowering drugs, glucose-lowering drugs, or sleeping patterns. We also excluded participants who had incident CVD within 2 years from the baseline, those with missing data on covariates, individuals with extreme energy intake (< 600 kcal or > 4200 kcal per day for men and < 500 kcal or > 3600 kcal per day for women [26]), those with type 1 diabetes, or individuals who completed only one 24-h dietary recall.

Decomposition analysis was conducted to evaluate the potential contributions of dietary components, while mediation analysis was used to assess the mediating role of serum biomarkers in these associations. Both analyses were performed using publicly available “mediate” SAS macros, following verification of the underlying assumptions [27, 28]. We investigated what proportion of the association between PDIs and risk of CVD, CHD, and stroke could be explained by various dietary components and blood biomarkers, including total cholesterol (TC), triglyceride (TG), low-density lipoprotein-cholesterol (LDL-C), high-density lipoprotein-cholesterol (HDL-C), creatinine, c-reactive protein (CRP), cystatin C, HbA1c, insulin-like growth factors-1 (IGF-1), and lipoprotein(a). As the serum concentration of CRP was a right-skewed distribution, we applied the log (base 10) transformation. The mediator was modeled with a multinomial Cox regression adjusted for age, sex, race, centers, BMI, household income, TDI, smoking, alcohol drinking, physical activity, history of hypertension and hypercholesterolemia at baseline, family history of CVD, family history of diabetes, vitamin supplementation, mineral supplementation, and aspirin use. Diabetes duration was only adjusted in the analysis among patients with diabetes.

Statistical analyses were conducted using SAS software version 9.4 (SAS Institute Inc). All tests were two-sided and, a P value < 0.05 was considered statistically significant.

Results

Population characteristics

The baseline characteristics of participants according to tertiles of PDI, hPDI, and uPDI are shown in Additional file 1: Table S3 for those with prediabetes and in Additional file 1: Table S4 for those with diabetes. Among participants with prediabetes, individuals with a higher PDI were predominantly white, non-smokers, older, more likely to use aspirin, and often had a family history of CVD; among those with diabetes, individuals with a higher PDI were less physically active, older, more inclined to use mineral supplements, and generally had a lower BMI.

PDIs and CVD risk among individuals with prediabetes and diabetes

During an average follow-up of 9.9 years (totaling 172,610 person-years), 2324 CVD events were documented among participants with prediabetes, including 1369 incident cases of CHD and 364 strokes. Over an average follow-up of 9.6 years (74,951 person-years) for patients with diabetes, we documented 1461 CVD events, including 994 CHD cases and 242 strokes. hPDI was found to be inversely associated with CVD risk among participants with prediabetes (HR comparing extreme tertiles: 0.88, 95% CI: 0.79–0.98; P_trend = 0.025), but not diabetes (HR comparing extreme tertiles: 1.00, 95% CI: 0.89–1.14; P_trend = 0.966). A positive association was observed between uPDI and CVD risk among participants with prediabetes (HR comparing extreme tertiles: 1.17, 95% CI: 1.05–1.30; P_trend = 0.005) and those with diabetes (HR comparing extreme tertiles: 1.14, 95% CI: 1.00–1.29; P_trend = 0.043) (Table 1). Further analysis of CVD subtypes revealed a positive association between uPDI and CHD risk, with an HR comparing extreme tertiles being 1.17 (95% CI: 1.01–1.34, P_trend = 0.038) (Table 2). No associations were found between PDI and the subsequent risk of CVD for prediabetes (HR comparing extreme tertiles: 1.05, 95% CI: 0.95–1.16; P_trend = 0.360) and diabetes (HR comparing extreme tertiles: 0.96, 95% CI: 0.85–1.09; P_trend = 0.548).

Table 1 Multivariable hazard ratios (95% CIs) for PDI, hPDI, uPDI, and CVD incidence in prediabetes and diabetes

Table 2 Multivariable hazard ratios (95% CIs) for PDI, hPDI, uPDI, and CHD and stroke incidence in prediabetes and diabetes

Subgroup and sensitivity analyses

In our stratified analyses according to potential CVD risk factors, we found that the associations between plant-based diets and CVD risk were generally similar across subgroups (Fig. 1 and Fig. 2). However, among participants with prediabetes, we observed inverse associations of hPDI with CVD risk only among younger individuals (HR per-SD: 0.89, 95% CI: 0.83–0.96; P_interaction < 0.001), smokers (HR per-SD: 0.85, 95% CI: 0.76–0.94; P _interaction = 0.049), without a family history of diabetes (HR per-SD: 0.89, 95% CI: 0.82–0.97; P_interaction = 0.033), or not taking vitamin supplement (HR per-SD: 0.94, 95% CI: 0.89–0.98; P_interaction = 0.030) (Fig. 1). In patients with diabetes, an inverse association between uPDI and CVD risk was observed among those with a lower TDI (HR per-SD: 1.13, 95% CI: 1.04–1.22; P_interaction = 0.026), as well as those without a history hypercholesterolemia (HR per-SD: 1.17, 95% CI: 1.06–1.29; P_interaction = 0.023) (Fig. 2).

Fig. 1

Multivariable hazard ratios (95% CIs) of plant-based diet index and CVD incidence among prediabetes from subgroup analyses. Forest plots show the multivariable HRs of CVD incidence among prediabetes associated with PDI, hPDI, and uPDI in subgroups. HRs were adjusted for age, sex, race, centers, BMI, household income, Townsend deprivation index, smoking, alcohol consumption, physical activity, history of hypertension, history of high cholesterol, family history of cardiovascular disease, family history of diabetes, vitamin supplement use, mineral supplement use, and aspirin use. Horizontal lines represent 95% CIs (*P_interaction < 0.05, **P_interaction < 0.01)

Fig. 2

Multivariable hazard ratios (95% CIs) of plant-based diet index and CVD incidence among diabetes from subgroup analyses. Forest plots show the multivariable HRs of CVD incidence among diabetes associated with PDI, hPDI, and uPDI in subgroups. HRs were adjusted for age, sex, race, centers, BMI, household income, Townsend deprivation index, smoking, alcohol consumption, physical activity, history of hypertension, history of high cholesterol, family history of cardiovascular disease, family history of diabetes, vitamin supplement use, mineral supplement use, aspirin use, and diabetes duration. Horizontal lines represent 95% CIs (*P_interaction < 0.05, **P_interaction < 0.01)

In sensitivity analyses, the associations of PDI, hPDI, and uPDI with CVD risk among individuals with prediabetes or diabetes remained largely unchanged after further adjusting lipid-lowering drugs, glucose-lowering drugs, sleeping pattern, or after excluding patients with incident CVD within 2 years, those with missing covariate data, those with type 1 diabetes, and those with extreme energy intake (Additional file 1: Tables S5-7). After excluding participants who completed only one 24-h dietary recall, results remained similar and the association of uPDI with CVD among prediabetes was weakened (HR comparing extreme tertiles: 1.14, 95% CI: 0.99–1.32; P_trend = 0.064).

Decomposition and mediation analysis

We further conducted a decomposition analysis to identify the dietary components potentially underlying the observed associations of hPDI and uPDI with the risk of CVD, CHD, and stroke, as well as to determine the proportions attributable to them (Additional file 1: Tables S8-9). High-sugar-sweetened beverages (SSB) accounted for 35% of the association between hPDI and CVD risk (95%CI = 7–79%, P = 0.005) and 15% of the association between uPDI and CVD risk (95%CI = 5–38%, P = 0.011), and animal fat accounted for 6% of the association between uPDI and CHD risk (95%CI = 1%-24%, P = 0.055) among individuals with prediabetes. Whole grain was the primary dietary component partially accounting for the association between uPDI and CVD risk in patients with diabetes, explaining 36% (95%CI = 6–82%, P = 0.028) of the association among all diabetes and 36% (95%CI = 5–86%, P = 0.036) among type 2 diabetes (Additional file 1: Table S8-10).

Meanwhile, a mediation analysis was performed to observe the potential mediation role of blood biomarkers. Blood biomarkers that contributed to the uPDI-CVD association included HDL-C, creatinine, cystatin C, and CRP (Table 3 and Additional file 1: Table S11-12). For both individuals with prediabetes and with diabetes, cystatin C accounted for the largest proportion of the association for CVD (prediabetes: 15%, 95% CI = 7–30%, P < 0.001; diabetes: 44%, 95% CI = 9–86%, P < 0.001), CHD (prediabetes: 13%, 95% CI = 5–32%, P < 0.001; diabetes: 50%, 95% CI = 3–97%, P < 0.001), and stroke (prediabetes: 16%, 95% CI = 5–40%, P = 0.001; diabetes: 15%, 95% CI = 3–48%, P = 0.005). For patients with diabetes, creatinine levels also significantly influenced the association for risk of CVD (20%, 95% CI = 5–53%, P < 0.001) and CHD (23%, 95% CI = 3–80%, P = 0.006) (Table 3), while IGF-1 accounted for 37% (0–100%) of the association (P = 0.0496). CRP marginally accounted for 2% (1–7%) of the association for CVD among prediabetic individuals and marginally accounted for 5% (1–23%) among diabetes patients. The results of the remaining indicators are shown in Additional file 1: Tables S11 and S12. Similarly, in patients with type 2 diabetes, IGF-1 accounted for a significant proportion of the association between the hPDI and CVD (24%; 95% CI, 0–98%; P = 0.048). Creatinine (23%; 95% CI, 5–65%; P < 0.001) and cystatin C (51%; 95% CI, 7–94%; P < 0.001) were significant mediators of the association between the uPDI and CVD (Additional file 1: Tables S13).

Table 3 Results of mediation analysis between an unhealthful plant-based diet and blood biomarker for risk of CVD, CHD, and stroke

Discussion

Dietary is increasingly being a crucial element of diabetes management. Our results show that greater adherence to a plant-based dietary pattern that emphasizes unhealthful plant foods, such as refined grains and SSB, was associated with a higher CVD risk among individuals with prediabetes and patients with diabetes. Healthful plant-based diet was associated with lower CVD risk among persons with prediabetes. Through decomposition analysis, we found that the positive association between the unhealthful plant-based dietary pattern and the CVD risk for participants with prediabetes and diabetes could be partly due to a lower intake of whole grains and a higher consumption of SSB and animal fat. In addition, blood biomarkers, such as creatinine, IGF-1, and cystatin C, could also partially account for the association between uPDI and CVD risk. Our research underscores the negative impact of uPDI on cardiovascular protection among individuals with prediabetes or diabetes and offers valuable guidance for creating dietary recommendations.

The existing research focused on the impact of PDI with the disease incidence based on the general population [12,13,14] or had a small cohort size or short follow-up times. Our study included 17,926 participants with prediabetes and 7798 with diabetes who were followed up for more than 9 years until the end of 2020. Based on the different physiological and metabolic states of patients with diabetes from healthy people, the study aimed to explore the impact of PDI, hPDI, and uPDI on the risk of CVD for individuals with prediabetes and diabetes. We did not find an association between PDI and CVD risk but found an inverse association between hPDI and reduced CVD risk among individuals with prediabetes (HR comparing extreme tertiles: 1.17, 95% CI: 1.05–1.30; P_trend = 0.005). Previous studies have reported beneficial associations of healthful plant-based diets with the risk of CVD [1,2,3] and T2D [14]. A population-based cohort study including 10,030 middle-aged community residents near Seoul showed that while PDI and uPDI were not significantly associated with T2D risk, the hPDI was inversely associated with the risk of T2D [14]. Similarly, a population-based, prospective study including 121,799 participants aged 40–69 years in the UK Biobank revealed that healthy plant-based dietary patterns significantly attenuated the risk of CVD for people at genetically higher risk of obesity [12]. This aligns with our findings that hPDI is associated with a lower risk of CVD among individuals with prediabetes. For individuals with diabetes, previous studies have shown the beneficial role of dietary fiber and whole grains in diabetes management, including improving glycemic control and reducing premature death [29]. A higher consumption of nuts was reported to be associated with lower CVD incidence and mortality among participants with diabetes [30]. Vegetables and fruits were associated with lower incident stroke in a Japanese cohort of T2D [31]. However, we showed hPDI was not related to CVD risk in persons with diabetes, which may be due to impaired glucose and lipid metabolism that is hard to reverse for this population. In contrast, uPDI, which consists of refined grains, sugar-sweetened beverages, and sweets and desserts, could aggravate the impaired glucose and lipid metabolism in patients with diabetes, leading to higher CVD risk.

Through decomposition analysis, we first identified dietary components contributing to the observed association between uPDI and CVD risk. We found that the contributing components were whole grain for patients with diabetes and animal fat and SSB for individuals with prediabetes. Whole grains are rich in dietary fiber which was beneficial for serum glucose, lipid, and body weight control and also anti-inflammation in diabetes management [29]. In our previous cohort study based on the Nurses’ Health Study and Health Professionals Follow-Up Study, monounsaturated fatty acids of animal origin were associated with a higher total mortality risk while those of plant origin were not among patients with diabetes [32]. Besides, a recent study in the same cohort also found a higher intake of SSB was associated with a 29% higher CVD mortality risk in participants with diabetes [33]. Then we also explored whether blood biochemical markers were important mediating factors underlying the association, addressing the gap in mechanism exploration present in existing studies and laying the foundation for further exploration. Our results of mediation analysis indicated that cystatin C, CRP, creatinine, and HDL were important mediators accounting for the association of uPDI with CVD risk among people with diabetes. Among a middle-aged population, subjects with high age, creatinine, and ejection fraction (ACEF) scores were significantly more likely to be diagnosed with CVD [34]. Previous reports suggest the accuracy of cystatin C in predicting CVD development in the elderly [35]. A systematic review and meta‐analysis of metabolically controlled and ad libitum trials showed that the Portfolio Diet significantly lowered LDL‐C by 17% and other CVD risk factors, including the alternate blood lipid targets of non‐high‐density lipoprotein cholesterol by 14% and apolipoprotein B by 15% and CRP by 32% [36]. A previous study has shown that an unhealthy diet (including the unhealthy plant diet) is associated with a higher inflammatory response [37]. Unhealthful plant-based diet, characterized by low whole grains content and high in SSB and animal fat, may activate the innate immune system through an increase in pro-inflammatory cytokines and a decrease in anti-inflammatory cytokines which may stimulate the formation of a pro-inflammatory state and can lead susceptible people to an increased prevalence of CVD [38]. Further mechanistic studies are required to elucidate the precise mechanism for these mediators.

Subgroup analysis showed that protective associations of hPDI were found among younger individuals (< 60 years) and those without hypertension. For patients with diabetes, the association of uPDI with CVD was evident among those without high cholesterol and not taking aspirin. These suggest the quality of PDI may play a more critical role for relatively healthy individuals. More mechanistic studies are needed to elucidate the interactions with baseline characteristics detected in our study.

The present study had a large sample and a long-term follow-up period, providing an opportunity to observe the different associations of PDI, uPDI, and hPDI with the risk of CVD in individuals with diabetes and prediabetes. The current study used mediation analysis and subgroup analysis to explore the potential mechanism of plant-based diet and CVD risk in people with diabetes and prediabetes disease. However, our study also has some limitations. Firstly, the study used multiple 24-h dietary assessments which were suitable for assessing the average intake of different food groups but measurement errors from recall bias were inevitable. Secondly, the consumption of some food components in the list of dietary patterns was assessed only once which may not capture potential changes in dietary habits, but we averaged all 24-h dietary assessments [19] as far as possible. Thirdly, although our model was adjusted for multiple established covariates, the possibility of residual confounding cannot be entirely ruled out. Despite our best efforts to minimize bias, confounding may still be present due to the complexity of multiple causal pathways assumed in the mediation and decomposition analyses. These analyses are particularly prone to confounding bias, and this should be considered when interpreting the results. Fourthly, the generalization of the findings to other ethnic groups might be limited as a previous study found no protective association of PDI with CVD and mortality among non-Hispanic black Americans [39]. Fifth, our study sample of T1D patients was small, and future research should explore whether the associations of PDI with CVD differ between patients with T1D and T2D. Finally, it is important to acknowledge that the temporal sequence in the mediation analysis involving blood biomarkers was not definitively established. As the blood biomarkers were measured at baseline, preceding the dietary recall, this may influence the interpretation of the mediation pathways. Further studies are needed to validate our mediation analysis findings, and the mechanisms by which uPDI mediates CVD risk require more in-depth investigation.

Conclusions

A higher adherence to an unhealthful plant-based diet was associated with an increased risk of CVD among individuals with prediabetes and diabetes. This association may be partly attributed to a lower intake of whole grains and a higher consumption of SSB and animal fat. Blood biomarkers such as creatinine, IGF-1, and cystatin C may also partially contribute to the association between uPDI and CVD risk. These findings highlight the harmful impact of uPDI, particularly for people with prediabetes or diabetes and provide important insights for the development of dietary guidelines aimed at CVD prevention in these patient groups.