Operational Differences in Plant-Based Diet Indices Affect the Ability to Detect Associations with Incident Hypertension in Middle-Aged US Adults

ABSTRACT

Background

Several distinct plant-based diet indices (PDIs) have been developed to characterize adherence to plant-based diets.

Objective

We contrasted 5 PDIs in a community-based cohort by assessing characteristics of the diet and evaluating whether these PDIs are associated with risk of incident hypertension.

Methods

Using FFQ data from adults (45–64 y, n = 8041) without hypertension at baseline in the Atherosclerosis Risk in Communities (ARIC) Study, we scored participants’ diets using the overall PDI (oPDI), healthy PDI (hPDI), less healthy (unhealthy) PDI (uPDI), provegetarian diet index, and PDI from the Rotterdam Study (PDI-Rotterdam). For the oPDI, provegetarian diet, and PDI-Rotterdam, higher intakes of all or selected plant foods received higher scores. For the hPDI, higher intakes of plant foods identified as healthful received higher scores. For the uPDI, higher intakes of less healthy plant foods received higher scores. All indices scored higher intakes of animal foods lower. We examined agreement between indices, and whether scores on these indices were associated with risk of hypertension using Cox proportional hazard models.

Results

The PDIs were moderately-to-strongly correlated and largely ranked subjects consistently, except for the uPDI. Over a median follow-up of 13 y, 6044 incident hypertension cases occurred. When adjusted for sociodemographic characteristics, other dietary factors, and health behaviors, the highest compared with the lowest quintile for adherence to the oPDI, hPDI, and provegetarian diet was associated with a 12–16% lower risk of hypertension (all P-trend <0.05). Highest adherence to the uPDI was associated with a 13% higher risk of hypertension, when clinical factors were further adjusted for (P-trend = 0.03). No significant association was observed with the PDI-Rotterdam. The oPDI, hPDI, and provegetarian diet moderately improved the prediction of hypertension.

Conclusions

In middle-aged US adults, despite moderate agreement in ranking subjects across PDIs, operational differences can affect the ability to detect diet–disease associations, such as hypertension.

Introduction

The 2015–2020 Dietary Guidelines for Americans recommend several eating patterns, such as the healthy Mediterranean-style diet, Dietary Approaches to Stop Hypertension (DASH) diet, and healthy vegetarian diet, for chronic disease prevention (1). Relative to a typical American diet, all 3 dietary patterns are rich in plant foods, by encouraging greater consumption of whole grains, fruits, vegetables, and plant proteins and lower consumption of red and processed meats. Recently, there has been a growing interest in whether plant-based dietary patterns, which consist predominantly of plant foods and are much lower in animal foods, have health benefits (2–4).

Observational studies have shown that vegetarian diets, which are a subset of plant-based diets with the exclusion of specific animal foods, are associated with lower risks of chronic disease (5–7). These studies defined vegetarian diets using consumption frequency of animal sources of food (eggs, dairy, fish/seafood) (8–11). Three plant-based diet indices (PDIs) have been developed recently, which take into account not only animal foods but also plant foods (12–15). An “overall plant-based diet index” (oPDI) was designed using data from 2 US cohorts: the Nurses’ Health Study and Health Professionals Follow-Up Study (12, 13); a “Provegetarian Diet Index” was developed using data from the European-based Prevención con Dieta Mediterránea (PREDIMED) trial (14); and a PDI (similar to the US oPDI) came from the Netherlands using data from the Rotterdam Study (PDI-Rotterdam) (15). These indices positively score intakes of plant foods and negatively score intakes of animal foods, allowing examination of whether a relatively higher intake of plant foods combined with a relatively lower intake of animal foods is associated with risk of chronic disease and related conditions. Studies have reported that higher adherence to these vegetarian diet indices and PDIs was associated with lower risks of insulin resistance, prediabetes, incident type 2 diabetes, coronary heart disease, all-cause mortality, and cardiovascular disease mortality (12–18).

To recognize that not all plant-based foods are considered healthy, additional indices have been developed to distinguish consumption of healthful from less healthful plant-based foods. A higher score on a “healthful” PDI (hPDI), representing diets that are higher in fruits, vegetables, whole grains, and plant protein, and lower in refined carbohydrates and animal foods, had a stronger inverse association with incident chronic kidney disease, type 2 diabetes, and coronary heart disease than the oPDI (12, 13, 17, 19). Conversely, higher adherence to a “less healthful” PDI (uPDI), consisting of higher intake of refined carbohydrates and lower intake of fruits, vegetables, plant proteins, and animal foods, was associated with an elevated risk of cardiometabolic conditions (12, 13, 19).

Considering the growing evidence and interest in plant-based diets, and differences in construction of PDIs, it is important to deconstruct and compare these indices to inform dietary guidance and the creation of future diet indices. Thus, we aimed to provide a comparative assessment of the PDIs by 1) examining similarities and differences in scoring with respect to food and nutritional characteristics, 2) assessing correlations and agreement in ranking of subjects, 3) evaluating whether greater adherence to the various indices was associated with incident hypertension in a general population, and 4) assessing which plant-based diet score improved the prediction of hypertension beyond established risk factors. We selected hypertension as an outcome because it is a strong risk factor for all of the outcomes reported in previous studies, and hypertension is modifiable by diet (20–22).

Methods

Data source

We scored all of the existing PDIs using dietary data from study participants in the Atherosclerosis Risk in Communities (ARIC) Study. The ARIC study is a prospective community-based cohort study of mostly black and white men and women (23). Middle-aged adults (45–64 y of age at baseline) from 4 US communities (Washington County, MD; Forsyth County, NC; Minneapolis, MN; and Jackson, MI) were recruited into the study and completed a baseline examination between 1987 and 1989 (visit 1). Follow-up visits were conducted in 1990–1992 (visit 2), 1993–1995 (visit 3), 1996–1998 (visit 4), 2011–2013 (visit 5), and 2016–2017 (visit 6). The Institutional Review Boards at all of the study sites approved the study protocol and participants provided informed consent.

Dietary assessment

Usual intake of foods and beverages was assessed at visit 1 and visit 3 using a modified version of the 66-item semiquantitative Willett FFQ. With visual aids (glasses and measuring cups), trained interviewers collected data on how often participants consumed foods and beverages of a defined serving size in the previous year. Participants had 9 options to select for frequency of consumption (“almost never” to “6 or more times a day”). Nutrient intakes were quantified by multiplying frequency of consumption and portion size by the nutrient composition of each food item. Reliability of the ARIC FFQ was assessed in a random subsample of participants from all study sites (n = 419) who completed the questionnaire at a follow-up visit (1990–1992) (24). Reliability coefficients for nutrient intakes ranged from 0.45 in black women to 0.63 in white men. Although a validation study was not conducted for the ARIC FFQ, the Willett FFQ reported validity coefficients ranging from 0.09 to 0.83 for plant foods, and from 0.33 to 0.77 for animal foods (25).

PDIs

Each of the PDIs was scored based on participants’ reported intake of foods and beverages in the FFQ (Supplemental Table 1) using previously published criteria (12–16, 19).

Briefly, food items on the FFQ were grouped into 1 of the food groups (Supplemental Table 2), with the number of food groups differing across the indices. The oPDI, hPDI, and uPDI had 17 food groups; the provegetarian diet index had 11 food groups; and the PDI-Rotterdam had 21 food groups. The PDI-Rotterdam had the highest number of food groups because it included alcohol-containing beverages and margarine, and scored 5 individual types of dairy (low-fat milk, full-fat milk, yogurt, milk-based desserts, and cheese).

Across all indices, the included groups were classified as plant-based or animal-based foods. For the hPDI and uPDI, plant foods were further categorized into healthy plant foods (i.e., whole grains, fruits, vegetables, nuts, legumes) and less healthy plant foods (i.e., refined grains, potatoes, fruit juices) (12, 13).

For all indices, energy-adjusted consumption of food groups was calculated using the residual method, and then participants were ranked into quintiles (26). However, differences existed across indices in the overall scoring approach. For the oPDI, provegetarian diet index, and PDI-Rotterdam, higher intakes of plant foods were positively scored, whereas higher intakes of animal foods were negatively scored (13–15). For the hPDI, only healthy plant foods were positively scored and all the other food groups were negatively scored. For the uPDI, only less healthy plant foods were positively scored and all other food groups were negatively scored.

Not all foods consumed were grouped and scored by each index. For instance, in the original design of the oPDI, hPDI, uPDI, provegetarian diet index, and PDI-Rotterdam, vegetable oil was positively scored, but the first 4 indices did not include margarine in the vegetable oil consumption. We did not include vegetable oil as a food group in these 4 indices because the ARIC FFQ only assessed margarine intake. However, we scored vegetable oil positively in the PDI-Rotterdam because the PDI-Rotterdam included margarine in the vegetable oil food group.

Assessment of incident hypertension

Incident hypertension was defined as either self-report of a doctor's diagnosis of hypertension, antihypertensive medication use, or elevated measured blood pressure. Doctor's diagnosis and antihypertensive medication use were ascertained at study visits and through annual telephone calls through 2016, in which participants were asked to report if a doctor had diagnosed them with high blood pressure or if they had taken antihypertensive medications in the past 2 wk. At study visits 1, 2, 3, 5, and 6 a certified technician measured participants’ blood pressure 3 times using a random-zero sphygmomanometer, and estimates from the last 2 measurements were averaged. At study visit 4, blood pressure was measured 2 times and estimates from the 2 measurements were averaged. Elevated measured blood pressure was defined as systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg.

Covariate assessment

At baseline, trained interviewers collected data on participants’ socio-demographic characteristics (age, sex, race/ethnicity, education), other dietary factors (alcohol consumption, total energy intake, margarine consumption, sodium consumption), health behaviors (smoking status, frequency and duration of physical activity), and clinical factors (lipid-lowering medication use, diabetes).

Trained study staff measured participants’ height and weight, which were used to derive BMI (in kg/m²). Normal weight was defined as <25, overweight was defined as 25 to <30, and obese was defined as ≥30. Total cholesterol and blood glucose (fasting and nonfasting) were measured at baseline using an enzymatic method and modified hexokinase/glucose-6-phosphate dehydrogenase, respectively (27). Serum creatinine measurements were used to estimate kidney function (estimated glomerular filtration rate) using the 2009 Chronic Kidney Disease Epidemiology Collaboration equation (28). Those with fasting glucose concentration ≥126 mg/dL, nonfasting blood glucose concentration ≥200 mg/dL, self-report of a doctor's diagnosis of diabetes, or diabetes medication use in the past 2 wk were considered to have diabetes.

Statistical analysis

We incorporated 2 dietary measurements using cumulative average diet from visit 1 and visit 3 (26). We used the mean of dietary intakes at baseline and visit 3 unless participants developed hypertension or were censored before visit 3. For such participants, we used their dietary intakes reported at baseline because their diets may have changed. We used cumulative average diet rather than a single baseline dietary assessment where possible to reduce measurement error, and to take into account that participants’ diets may have changed with hypertension.

To create the analytic sample from the overall ARIC cohort (n = 15,792), we excluded participants with implausibly low or high total energy intake (<500 kcal or >3500 kcal for women and <700 kcal or >4500 kcal for men, n = 98) and those missing hypertension classification at any one of the follow-up visits from visit 2 to 6 (n = 1727) (Supplemental Figure 1). Also excluded were those who reported that they were neither black nor white (n = 48), blacks in Washington County, Maryland (n = 33), or blacks in Minneapolis, Minnesota (n = 22) owing to small numbers. Necessarily excluded for analyses related to incident hypertension were those with hypertension at baseline (n = 5461) as well as those with evidence of heart disease, including myocardial infarction, a history of heart or arterial surgery, or cardiovascular procedures (coronary bypass, balloon angioplasty, and angioplasty of coronary arteries) (n = 362). The remaining analytic sample consisted of 8041 participants.

We compared baseline characteristics of the study participants by quintiles of all diet indices. To examine nutritional characteristics, we calculated food groups expressed as servings per day and nutrient density (macronutrients as a percentage of energy, and fiber and micronutrients as grams, milligrams, or micrograms per 1000 kcal). To examine food groups, we used the food groups defined from the oPDI. To visually depict differences across indices, we used radar plots to plot percentage differences between the highest and lowest quintile diet scores for intakes of 17 food groups, and nutrient density of macro- and micronutrients. In addition, we calculated the Healthy Eating Index-2015 to examine diet quality by quintiles of PDIs (29).

We calculated Spearman rank pairwise correlation coefficients between plant-based diet scores. Then, we examined the proportion of participants who were cross-classified into the same quintiles for all pairs of diet indices, using the same approach as a previous study (30). We determined the proportions of participants who were classified into the highest quintile, lowest quintile, and any quintiles (quintiles 1 through 5) to assess agreement between the PDIs.

To evaluate the associations between each index and risk of hypertension, we used 3 nested Cox proportional hazards models to estimate HRs and 95% CIs. The time metric was follow-up time from the date of baseline examination (visit 1) until the date of incident hypertension or censoring. In model 1, we adjusted for age, sex, a combined term of race and study center, and total energy intake. In model 2, we adjusted for the covariates in model 1, education (which was a proxy for socioeconomic status), smoking status, physical activity, alcohol consumption, margarine consumption, and sodium intake. For the association between the PDI-Rotterdam and hypertension, we did not adjust for alcohol and margarine intakes because these items were already incorporated into this index. In model 3, we adjusted for the covariates in model 2 and potential mediating variables (clinical factors). We tested for the proportional hazards assumption using the Schoenfeld residual and log(-log) plots, and did not find a violation of the assumption. We tested for linear trends by using the median score within each quintile. To compare these 5 plant-based diet scores against each other, we modeled the scores continuously using per-SD increase. Then, we modeled individual food groups within each diet index together to assess the association of one food group independently from the other food groups and also accounted for the covariates in model 3. Next, we calculated Harrell's C-statistics from the fully adjusted model (model 3) and assessed whether the addition of a plant-based diet score, total vegetable intake, or total plant food intake improved the prediction of incident hypertension. As sensitivity analyses, we 1) examined whether baseline characteristics differed for those with incomplete outcome data and those included in the analyses, 2) repeated the analyses by truncating follow-up time to visit 4 (31 December, 1998) owing to greater subsequent losses of follow-up (fewer participants attended study visits after visit 4), and 3) modeled quintiles of individual food groups within each diet index simultaneously. All analyses were conducted using Stata version 13.0 statistical software (StataCorp).

Results

Baseline characteristics

Those in the highest quintiles of the oPDI, hPDI, and provegetarian diet index were more likely to be female, white, a high school graduate, more physically active, and have lower fasting glucose at baseline; and less likely to be obese or a current smoker than those in the lowest quintiles (Table 1). However, they were more likely to use lipid-lowering medication. Those in the highest quintiles of the PDI-Rotterdam had similar characteristics as those in the highest quintiles of the aforementioned diet indices, except that they were more likely to be male.

TABLE 1.

Selected characteristics of the middle-aged adults free of hypertension at baseline and nutritional characteristics of the diet by quintiles of PDIs in the Atherosclerosis Risk in Communities Study¹

	oPDI		hPDI		uPDI		Provegetarian		PDI-Rotterdam
	Quintile 1	Quintile 5	Quintile 1	Quintile 5	Quintile 1	Quintile 5	Quintile 1	Quintile 5	Quintile 1	Quintile 5
Sample size	1818	1473	1792	1434	1639	1588	1987	1513	1711	1513
Median score	43	59	43	61	43	60	27	40	58	75
Female	42.9	61.0	47.8	62.3	62.7	45.2	46.8	60.0	57.7	48.8
Black	29.3	8.1	27.4	9.5	14.2	18.8	23.8	10.6	30.1	7.62
Age, y	53.1 ± 5.6	53.4 ± 5.6	52.4 ± 6.5	54.3 ± 6.6	54.0 ± 5.6	52.3 ± 5.5	52.7 ± 5.5	54.0 ± 5.7	53.0 ± 5.7	53.3 ± 5.6
High school graduate	74.9	88.6	75.8	88.7	85.4	78.9	78.8	87.3	76.6	85.2
BMI category, kg/m2
Normal weight (<25)	36.7	49.7	35.8	48.8	39.2	42.5	38.1	49.2	30.2	45.2
Overweight (25 to <30)	40.9	36.7	40.7	38.7	40.8	39.3	39.8	37.4	40.0	40.8
Obese (≥30)	22.4	13.6	23.5	12.5	19.9	18.2	22.1	13.4	29.8	13.9
Current smoker	34.9	21.5	30.4	23.1	27.2	28.8	34.3	18.8	26.4	28.2
Physical activity index	2.4 ± 0.8	2.6 ± 0.8	2.3 ± 0.8	2.7 ± 0.8	2.6 ± 0.8	2.4 ± 0.8	2.4 ± 0.8	2.7 ± 0.8	2.4 ± 0.8	2.6 ± 0.8
Alcohol, g/wk	61.2 ± 121	26.1 ± 50.5	35.8 ± 73.3	44.6 ± 90.6	38.2 ± 69.6	51.5 ± 116	53.9 ± 106	27.9 ± 59.6	37.1 ± 112	49.8 ± 85.6
Fasting glucose, mg/dL	104 ± 33.6	99.9 ± 24.1	104 ± 32.2	101 ± 25.3	104 ± 34.2	100 ± 19.4	103 ± 30.8	100 ± 23.9	108 ± 39.0	102 ± 25.6
Diabetes	6.3	4.6	7.0	5.9	7.6	3.1	5.3	5.2	10.6	5.4
Systolic BP, mm Hg	114 ± 12.1	111 ± 12.0	113 ± 12.2	112 ± 12.2	112 ± 12.2	113 ± 12.5	113 ± 11.9	112 ± 12.0	117 ± 12.1	112 ± 12.0
Diastolic BP, mm Hg	70.9 ± 8.9	68.0 ± 8.5	70.9 ± 8.6	68.1 ± 8.7	68.6 ± 8.7	70.1 ± 8.8	70.7 ± 8.8	68.3 ± 8.6	70.2 ± 8.5	68.8 ± 8.7
Lipid medication	1.3	3.3	1.0	3.3	1.7	1.4	1.0	3.4	1.2	2.8
eGFR, mL · min−1 · 1.73 m−2	105 ± 14.4	102 ± 12.1	105 ± 13.7	102 ± 12.0	103 ± 12.9	103 ± 13.5	105 ± 13.7	102 ± 12.2	104 ± 14.1	102 ± 12.5
Foods,2 servings/d
Whole grain	0.54 ± 0.79	1.15 ± 0.91	0.35 ± 0.50	1.44 ± 1.06	1.17 ± 0.96	0.52 ± 0.75	0.54 ± 0.67	1.19 ± 1.01	0.54 ± 0.70	1.14 ± 0.98
Fruit	1.08 ± 1.07	2.03 ± 1.17	0.87 ± 0.81	2.44 ± 1.47	1.96 ± 1.31	1.14 ± 1.08	0.97 ± 0.92	2.21 ± 1.34	1.26 ± 1.20	1.84 ± 1.25
Vegetables	1.29 ± 0.89	1.98 ± 0.98	1.05 ± 0.71	2.26 ± 1.04	2.09 ± 1.05	1.16 ± 0.67	0.88 ± 0.74	1.73 ± 0.90	1.28 ± 0.90	1.91 ± 0.98
Nuts	0.26 ± 0.42	0.56 ± 0.53	0.21 ± 0.34	0.63 ± 0.65	0.47 ± 0.53	0.32 ± 0.50	0.24 ± 0.40	0.61 ± 0.59	0.23 ± 0.34	0.60 ± 0.60
Legumes	0.48 ± 0.42	0.72 ± 0.44	0.41 ± 0.34	0.80 ± 0.52	0.75 ± 0.53	0.43 ± 0.34	0.40 ± 0.35	0.81 ± 0.47	0.46 ± 0.41	0.73 ± 0.53
Coffee and tea	2.15 ± 1.98	2.91 ± 2.04	1.85 ± 1.73	2.95 ± 2.09	3.10 ± 2.06	2.00 ± 1.91	2.58 ± 2.07	2.37 ± 1.96	1.90 ± 1.85	3.12 ± 2.09
Margarine	1.10 ± 0.90	1.18 ± 1.01	0.98 ± 0.87	1.30 ± 1.10	1.01 ± 0.87	1.13 ± 1.01	1.00 ± 0.90	1.22 ± 1.04	0.60 ± 0.87	1.54 ± 1.12
Refined grain	1.84 ± 1.27	2.12 ± 1.19	2.06 ± 1.23	1.76 ± 1.09	1.51 ± 0.97	2.41 ± 1.45	1.63 ± 1.10	2.24 ± 1.24	1.60 ± 1.10	2.36 ± 1.37
Potatoes	0.27 ± 0.32	0.31 ± 0.31	0.36 ± 0.36	0.18 ± 0.25	0.17 ± 0.22	0.44 ± 0.41	0.25 ± 0.27	0.30 ± 0.32	0.19 ± 0.25	0.40 ± 0.37
Fruit juices	0.37 ± 0.49	0.68 ± 0.54	0.51 ± 0.52	0.51 ± 0.56	0.42 ± 0.50	0.61 ± 0.65	0.46 ± 0.55	0.55 ± 0.52	1.26 ± 1.08	1.95 ± 1.41
SSBs	1.00 ± 1.14	1.19 ± 1.15	1.33 ± 1.13	0.70 ± 0.97	0.64 ± 0.85	1.72 ± 1.44	1.12 ± 1.19	0.95 ± 1.05	0.83 ± 0.84	1.70 ± 1.37
Sweets	1.03 ± 1.01	1.47 ± 1.30	1.26 ± 1.17	1.11 ± 1.19	0.74 ± 0.84	1.87 ± 1.48	1.15 ± 1.07	1.29 ± 1.25	0.48 ± 0.72	0.23 ± 0.58
Animal fat	0.56 ± 0.79	0.17 ± 0.42	0.46 ± 0.68	0.22 ± 0.56	0.47 ± 0.72	0.27 ± 0.58	0.56 ± 0.77	0.15 ± 0.40	0.37 ± 0.40	0.21 ± 0.34
Eggs	0.46 ± 0.44	0.15 ± 0.28	0.39 ± 0.41	0.20 ± 0.31	0.36 ± 0.38	0.24 ± 0.36	0.42 ± 0.42	0.16 ± 0.25	1.93 ± 1.36	1.52 ± 1.07
Dairy	1.91 ± 1.39	1.52 ± 0.99	1.47 ± 1.10	1.98 ± 1.33	1.92 ± 1.22	1.50 ± 1.17	1.86 ± 1.33	1.55 ± 1.04	0.32 ± 0.30	0.27 ± 0.31
Fish or seafood	0.30 ± 0.28	0.29 ± 0.38	0.27 ± 0.34	0.35 ± 0.34	0.41 ± 0.40	0.20 ± 0.22	0.29 ± 0.26	0.30 ± 0.38	1.59 ± 0.85	1.41 ± 0.80
Meat (total)	1.85 ± 0.89	1.18 ± 0.78	1.61 ± 0.95	1.31 ± 0.73	1.57 ± 0.93	1.46 ± 0.78	1.69 ± 0.85	1.23 ± 0.79	1.58 ± 0.34	1.46 ± 0.33
Poultry	0.37 ± 0.32	0.38 ± 0.29	0.32 ± 0.32	0.43 ± 0.33	0.45 ± 0.38	0.30 ± 0.24	0.35 ± 0.30	0.38 ± 0.29	1.19 ± 0.74	1.07 ± 0.74
Red and processed meat	1.48 ± 0.83	0.80 ± 0.69	1.29 ± 0.84	0.88 ± 0.66	1.12 ± 0.82	1.16 ± 0.73	1.34 ± 0.79	0.84 ± 0.70	0.39 ± 0.40	0.39 ± 0.41
Miscellaneous animal foods	0.57 ± 0.47	0.23 ± 0.32	0.53 ± 0.44	0.21 ± 0.27	0.38 ± 0.43	0.40 ± 0.41	0.44 ± 0.42	0.30 ± 0.39	0.54 ± 0.70	1.14 ± 0.98

¹Values are means ± SDs or percentages unless otherwise indicated. BP, blood pressure; eGFR, estimated glomerular filtration rate; hPDI, healthy plant-based diet index; oPDI, overall plant-based diet index; PDI, plant-based diet index; PDI-Rotterdam, plant-based diet index from the Rotterdam Study; SSB, sugar-sweetened beverage; uPDI, less healthy (unhealthy) plant-based diet index.

²One serving of beverages is defined as 236.6 mL.

Those in the highest quintile of the uPDI were more likely to be male, black, a current smoker, and drink a higher amount of alcohol than those in the lowest quintile. They were also less likely to be a high school graduate and to be obese, and were less physically active.

Food and nutritional characteristics

Those in the highest quintiles of all of the indices except for the uPDI had higher intakes of whole grains, fruits, vegetables, nuts, and legumes, and a lower total intake of meat (Supplemental Figure 2A). When we examined food group intakes for participants in the lowest quintiles, those in the lowest quintile of the uPDI had higher intakes of whole grains, fruits, vegetables, nuts, legumes, and coffee and tea (Supplemental Figure 2B). Those in the lowest quintile of the PDI-Rotterdam had the highest intakes of dairy compared with those in the lowest quintiles of other indices.

Those in the highest quintiles of all PDIs except for the uPDI consumed an average of 1.8–2.4 servings of fruits and 1.6–1.9 servings of vegetables per day compared with 0.9–1.2 servings of fruit and 0.8–1.2 servings of vegetables among those in the lowest quintiles (Table 1). Those in the highest quintile of the uPDI consumed an average of 1.1 servings of fruits and 1.2 servings of vegetables per day compared with 2.0 servings of fruit and 2.1 servings of vegetables among those in the lowest quintile.

In terms of nutritional characteristics, those in the highest quintile of the hPDI had higher intakes of protein and carbohydrates as a percentage of energy; fiber; potassium; magnesium; iron; vitamins A, C, E, and B-1; folate; and higher diet quality scores, and lower intakes of saturated fat, added sugar, and vitamin B-12 than those in the lowest quintile (Supplemental Table 3). Those in the highest quintiles of the oPDI, provegetarian diet index, and PDI-Rotterdam had similar characteristics, except that they had lower intakes of protein. When we examined the shape of the radar plots, micronutrient intakes (calcium; potassium; magnesium; iron; vitamins A, C, and B-2; and folate) were lower for those in the highest quintile of the PDI-Rotterdam than for those in the highest quintiles of the oPDI, hPDI, and provegetarian diet index (Supplemental Figure 3). In contrast, those in the highest quintile of the uPDI had higher total energy and carbohydrate intakes, and lower intakes of protein, saturated fat, fiber, calcium, potassium, magnesium, iron, vitamin A, vitamin C, thiamin, riboflavin, folate, and diet quality scores (Supplemental Table 3).

Correlations and concordance between diet indices

Spearman rank correlation coefficients ranged from −0.43 to 0.83 (all P values < 0.05, Table 2). We observed the strongest correlations between the oPDI and provegetarian diet index (ρ = 0.83) and the oPDI and PDI-Rotterdam (ρ = 0.75). Moderate correlations were observed between the provegetarian diet index and PDI-Rotterdam (ρ = 0.69), and the hPDI and provegetarian diet index (ρ = 0.66) (all P values < 0.05). Weak positive to moderate negative correlations were observed between the uPDI and all other indices: oPDI (ρ = 0.12), PDI-Rotterdam (ρ = 0.18), provegetarian diet index (ρ = −0.04), and hPDI (ρ = −0.43) (all P values < 0.05).

TABLE 2.

Spearman rank correlation coefficients (ρ) between all pairs of PDIs in middle-aged adults free of hypertension at baseline in the Atherosclerosis Risk in Communities Study¹

	oPDI	hPDI	uPDI	Provegetarian	PDI-Rotterdam
oPDI	—	0.53	0.12	0.83	0.75
hPDI		—	−0.43	0.66	0.44
uPDI			—	−0.04	0.18
Provegetarian				—	0.69

¹All were statistically significant (P < 0.05). hPDI, healthy plant-based diet index; oPDI, overall plant-based diet index; PDI, plant-based diet index; PDI-Rotterdam, plant-based diet index from the Rotterdam Study; uPDI, less healthy (unhealthy) plant-based diet index.

For all pairs of diet indices (except the uPDI), 30–50% of the study participants were ranked in the identical quintiles (Supplemental Table 4). Only 13–23% of the study participants were ranked in the identical quintiles by the uPDI as compared with the other indices, and agreement at the extreme quintiles (quintile 1 and quintile 5) was only 1–6%. Overall, there was a sizeable overlap in rankings for the oPDI, hPDI, provegetarian diet index, and PDI-Rotterdam, but less overlap with the uPDI.

Diet indices and risk of incident hypertension

During a median follow-up of 13 y (maximum of 30 y of follow-up), 6044 incident hypertension events occurred. Minimally adjusted (model 1) estimates showed that higher adherence to the oPDI, hPDI, and provegetarian diet was associated with a 12–19% lower risk of incident hypertension, whereas higher adherence to the uPDI was associated with a 10% higher risk of incident hypertension (all P values for trend < 0.05, Table 3).

TABLE 3.

Adjusted HRs and 95% CIs for incident hypertension according to quintiles of PDIs in middle-aged adults free of hypertension at baseline in the Atherosclerosis Risk in Communities Study¹

	HRs and 95% CIs
	Quintile 1	Quintile 2	Quintile 3	Quintile 4	Quintile 5	P-trend	Per SD increment
Model 12
oPDI	1 [Ref]	0.97 (0.89, 1.06)	0.93 (0.86, 1.01)	0.89 (0.81, 0.97)	0.88 (0.81, 0.97)	0.001	0.95 (0.92, 0.97)
hPDI	1 [Ref]	0.95 (0.87, 1.03)	0.95 (0.87, 1.03)	0.93 (0.85, 1.02)	0.81 (0.74, 0.89)	<0.001	0.94 (0.92, 0.97)
uPDI	1 [Ref]	1.02 (0.94, 1.11)	1.02 (0.93, 1.11)	1.04 (0.95, 1.14)	1.10 (1.01, 1.20)	0.04	1.03 (1.01, 1.05)
Provegetarian	1 [Ref]	0.97 (0.89, 1.05)	0.95 (0.87, 1.04)	0.92 (0.85, 1.00)	0.85 (0.78, 0.93)	<0.001	0.94 (0.91, 0.96)
PDI-Rotterdam study	1 [Ref]	0.99 (0.82, 1.08)	0.95 (0.87, 1.04)	0.97 (0.88, 1.06)	1.04 (0.95, 1.14)	0.68	0.99 (0.97, 1.02)
Model 23
oPDI	1 [Ref]	0.95 (0.87, 1.03)	0.89 (0.83, 0.97)	0.88 (0.80, 0.96)	0.88 (0.81, 0.96)	0.001	0.96 (0.94, 0.99)
hPDI	1 [Ref]	0.92 (0.85, 1.02)	0.95 (0.88, 1.02)	0.93 (0.86, 1.02)	0.84 (0.77, 0.92)	0.001	0.96 (0.93, 0.98)
uPDI	1 [Ref]	1.03 (0.95, 1.11)	1.02 (0.95, 1.12)	1.05 (0.97, 1.14)	1.06 (0.98, 1.16)	0.12	1.02 (0.98, 1.04)
Provegetarian	1 [Ref]	0.96 (0.89, 1.04)	0.94 (0.86, 1.02)	0.91 (0.84, 0.98)	0.86 (0.78, 0.91)	<0.001	0.95 (0.92, 0.97)
PDI-Rotterdam study4	1 [Ref]	0.99 (0.91, 1.07)	0.93 (0.86, 1.01)	0.99 (0.90, 1.07)	1.06 (0.96, 1.16)	0.69	1.00 (0.98, 1.03)
Model 35
oPDI	1 [Ref]	0.94 (0.87, 1.02)	0.89 (0.83, 0.97)	0.88 (0.81, 0.96)	0.90 (0.83, 0.98)	0.01	0.97 (0.94, 0.99)
hPDI	1 [Ref]	0.93 (0.86, 1.00)	0.96 (0.89, 1.04)	0.95 (0.87, 1.04)	0.87 (0.80, 0.95)	0.02	0.96 (0.93, 0.99)
uPDI	1 [Ref]	1.06 (0.99, 1.16)	1.07 (0.98, 1.16)	1.09 (1.01, 1.19)	1.13 (1.04, 1.23)	0.005	1.04 (1.01, 1.07)
Provegetarian	1 [Ref]	0.95 (0.88, 1.03)	0.92 (0.85, 1.01)	0.93 (0.86, 1.00)	0.87 (0.80, 0.94)	0.001	0.95 (0.92, 0.98)
PDI-Rotterdam study4	1 [Ref]	1.01 (0.93, 1.09)	0.96 (0.88, 1.04)	0.97 (0.89, 1.06)	1.05 (0.97, 1.15)	0.58	1.01 (0.98, 1.04)

¹hPDI, healthy plant-based diet index; oPDI, overall plant-based diet index; PDI, plant-based diet index; PDI-Rotterdam, plant-based diet index from the Rotterdam Study; uPDI, less healthy (unhealthy) plant-based diet index.

²Model 1 adjusted for age, sex, race–center, and total energy intake.

³Model 2 adjusted for the covariates in model 1 and education, smoking status, physical activity, alcohol consumption, margarine consumption, and sodium intake.

⁴Did not adjust for alcohol or margarine intake because alcohol and margarine were already accounted for in this index.

⁵Model 3 adjusted for the covariates in model 2 and baseline total cholesterol, lipid-lowering medication use, estimated glomerular filtration rate, diabetes, and BMI.

When we adjusted for sociodemographic characteristics, other dietary factors, and health behaviors, those with the highest compared with the lowest degree of adherence to the oPDI, hPDI, and provegetarian diet index had a 12%, 16%, and 14% lower risk of hypertension, respectively (all P values for trend < 0.05). When potential mediating variables were further adjusted for, the estimates did not change substantially (all P values for trend < 0.05). No significant association was observed for the uPDI in model 2 (P-trend = 0.12), but those with the highest degree of adherence to the uPDI had a 13% higher risk of hypertension in model 3 (P-trend = 0.005). No significant association was observed with the PDI-Rotterdam in model 1 (P-trend = 0.68), model 2 (P-trend = 0.69), or model 3 (P-trend = 0.58). Risk of hypertension was 3–5% lower with a 1-SD increase in the oPDI, hPDI, and provegetarian diet index, and 4% higher with a 1-SD increase in the uPDI in model 3. No association was observed with the PDI-Rotterdam in any of the models when we modeled this score continuously.

The C-statistic (95% CI) for the fully adjusted model was 0.603 (0.595, 0.611) (Table 4). Compared with the fully adjusted model, prediction of incident hypertension moderately improved with the addition of the oPDI, hPDI, and provegetarian diet index (all P values < 0.05). The addition of total vegetable intake moderately increased prediction of incident hypertension (P value = 0.044), whereas the addition of total plant food intake did not significantly improve the C-statistic (P value = 0.232).

TABLE 4.

Harrell's C-statistic and 95% CI for prediction of incident hypertension with the addition of a PDI to a fully adjusted model¹

	C-statistic (95% CI)	P value2
Fully adjusted model	0.603 (0.595, 0.611)	—
oPDI	0.605 (0.597, 0.613)	0.005
hPDI	0.605 (0.598, 0.612)	0.030
uPDI	0.604 (0.596, 0.612)	0.060
Provegetarian	0.605 (0.597, 0.613)	0.010
PDI-Rotterdam	0.604 (0.596, 0.611)	0.305
Total vegetable intake3	0.605 (0.597, 0.613)	0.044
Total plant food intake4	0.604 (0.595, 0.612)	0.232

¹The fully adjusted model included the following covariates: age, sex, race–center, total energy intake, education, smoking status, physical activity, alcohol consumption, margarine consumption, sodium intake, baseline total cholesterol, lipid-lowering medication use, estimated glomerular filtration rate, diabetes, and BMI. hPDI, healthy plant-based diet index; oPDI, overall plant-based diet index; PDI, plant-based diet index; PDI-Rotterdam, plant-based diet index from the Rotterdam Study; uPDI, less healthy (unhealthy) plant-based diet index.

² P for difference compared with the fully adjusted model.

³Total vegetable intake was adjusted using the residual approach.

⁴Total plant food intake was adjusted using the residual approach. Total plant food intake was derived by combining the intakes of whole grains, fruits, vegetables, nuts, legumes, coffee and tea, refined grain, potatoes, fruit juices, sugar-sweetened beverages, and sweets and desserts.

When we compared the baseline characteristics of those with incomplete outcome data and those included in the analysis, those who were not included were more likely to be black, have less than a high school education, and drink lower amounts of alcohol, and were less likely to exercise (all P values < 0.05). The associations for all diet indices and hypertension did not change when we repeated our analyses by following up participants through visit 4.

Individual food groups within PDIs

When individual food groups within each index were modeled together, higher vegetable intake (by 1 serving/d) was consistently associated with a 5% lower risk of hypertension, whereas higher intakes of sugar-sweetened beverages (SSBs) and, separately, vegetable oil/margarine (by 1 serving/d) were associated with a 4–5% higher risk of hypertension (Supplemental Table 5). When quintiles of individual food group intake were modeled simultaneously, the results were nearly identical, except that those in the highest compared with the lowest quintile of whole grain consumption had 11–13% lower risk of hypertension.

Discussion

To the best of our knowledge, this is the first study to calculate different types of PDIs and compare them in the same study population. We found that greater adherence to 3 of 5 PDIs (the oPDI, hPDI, and provegetarian diet index) was consistently associated with a lower risk of hypertension independent of sociodemographic characteristics, other dietary factors, and health behaviors. Greater adherence to a less healthy plant-based diet was associated with an elevated risk of hypertension in a fully adjusted model. We did not find a significant association with 1 PDI: the PDI-Rotterdam. The moderate to strong correlations between most PDIs suggest that they capture similar dietary patterns, but the results from the prospective analysis with incident hypertension highlight that the manner by which plant-based diets are operationalized can affect the ability to detect diet–disease associations.

Our analyses of correlations and concordances between PDIs suggest that the indices had several components in common with each other. Specifically, 30–50% of the study participants had identical rankings when we examined pairwise concordances between the oPDI, hPDI, and provegetarian diet index. Higher scores on these indices represented higher intakes of fruits, vegetables, whole grains, nuts, and legumes, and lower intakes of red and processed meats. The magnitude of the associations with risk of incident hypertension was similar for the hPDI (16% lower risk for quintile 5 than for quintile 1), provegetarian diet index (14%), and oPDI (12%). Further, all 3 indices moderately improved prediction of incident hypertension, whereas total plant food intake did not increase the C-statistic. Taken together, the findings support general dietary recommendations for higher intakes of plant-based foods (e.g., whole grains, fruits, vegetables, nuts, and legumes) and lower intakes of animal source foods (particularly red and processed meats), which were captured by the oPDI, hPDI, and provegetarian diet index.

Although not uniform, our results for the association between plant-based diets and hypertension are biologically plausible when considering the macronutrient and micronutrient composition of these dietary patterns. Those with higher adherence to the hPDI consumed a higher amount of protein (18–19% of total energy), fiber, and potassium than those in the lowest quintiles. Randomized controlled trials have reported that higher intakes of dietary protein and fiber are associated with lower systolic and diastolic blood pressure (31, 32). Similarly, higher potassium intake has been hypothesized to lower blood pressure through vasodilation and vascular homeostasis (21).

Finding no association between the PDI-Rotterdam and hypertension was surprising given that the PDI-Rotterdam had moderate to strong correlations with the oPDI and the provegetarian diet index, and higher adherence to the PDI-Rotterdam was associated with a lower risk of prediabetes and type 2 diabetes in the Rotterdam Study (15). However, the PDI-Rotterdam differed in scoring of alcohol, margarine, and dairy. As mentioned earlier, alcohol and margarine consumption were positively scored for the PDI-Rotterdam, but they were not scored for the other PDIs. Rather, we accounted for these 2 dietary components as covariates in the adjusted models for the other indices, consistent with the approaches from previous studies (12, 19). In the PDI-Rotterdam, margarine may have been scored positively because it had a healthier fatty acid composition in the Netherlands than in the United States (it was high in trans fat during the ARIC dietary assessments) (12). Alcohol intakes were higher for those in the highest quintile of the PDI-Rotterdam, the hPDI, and the uPDI, and lower for those in the highest quintiles of the oPDI and provegetarian diet index. Margarine intake was only slightly higher for those in the highest quintile of the PDI-Rotterdam compared with other indices. Thus, the scoring of alcohol or margarine intake does not appear to explain the inconsistent findings with respect to hypertension across indices.

The negative scoring of 5 dairy groups in the PDI-Rotterdam meant that overall dairy consumption was weighted more heavily than other food groups in the creation of the PDI-Rotterdam overall score, whereas dairy as a single component was weighted the same as other food groups in the oPDI and provegetarian diet index. Although it is not clear that this weighting led to the lack of association with incident hypertension, it is different from the approaches of other indices which reverse scored all animal products uniformly lower. This difference in scoring may have affected the results, because plant-rich dietary patterns that include dairy (e.g., DASH diet) have shown a strong inverse association with hypertension (20). A systematic review of 9 prospective studies reported that daily consumption of 200 g dairy was associated with a lower risk of incident hypertension (33). Future research on health impacts of plant-based diets that are low in dairy products is warranted.

We observed that the uPDI had the least overlap with the other diet indices and the uPDI was associated with a higher risk of hypertension. These findings are in agreement with previous studies which reported that greater degree of adherence to the uPDI was associated with an elevated risk of other related chronic disease outcomes, including chronic kidney disease, type 2 diabetes, and coronary heart disease (12, 13, 19). Our results add to this body of literature by showing that the uPDI may also be associated with hypertension, a strong risk factor for these cardiometabolic health outcomes. In addition, when individual food groups in each index were modeled simultaneously, higher intake of SSBs was consistently associated with a higher risk of hypertension. Prior studies have shown positive associations between SSBs and incident hypertension (34), and our results underscore that lowering the intake of SSBs (and added sugar) specifically in the context of an overall plant-based diet remains important.

We used data from a community-based cohort with a racially diverse study population. Our data have broader generalizability than several previous studies which were conducted in individuals with high socioeconomic status (health professionals, Spanish university students) or those at high risk of developing cardiovascular disease (12–14, 35). Our study also benefits from repeated dietary measurements and high-quality outcome ascertainment. Dietary assessment was conducted by trained interviewers, and incident hypertension was ascertained using a combination of self-reports and measured blood pressure.

Some limitations should be considered in the interpretation of the results. First, self-reported dietary intakes are subject to measurement error. Nevertheless, reproducibility of the FFQ was quantified in a reliability study of a random subsample of ARIC participants (24). Second, a validation study was not conducted for the ARIC FFQ. However, a similar FFQ was validated, and validity coefficients were reasonable for animal food intake (25). Third, baseline characteristics of those with incomplete outcome data and those included in the analytic sample were slightly different. However, when we repeated our analyses with earlier follow-up data (through visit 4), the results did not differ, underscoring the robustness of our findings. Fourth, participants were classified into quintiles of plant-based diet scores based on a distribution of intakes. All of the indices limit inference on the absolute amounts of animal foods or plant foods that are related to a lower risk of hypertension. Future studies should consider conducting multiple 24-h dietary recalls to quantify absolute amounts of food intakes to evaluate dietary patterns. Fifth, dietary intakes were measured in the late 1980s and the 1990s; thus, the dietary data are not necessarily representative of foods in the modern food supply, such as margarine intake, for which the composition of fatty acids has changed over time (12). However, the nutritional composition of foods such as fruits, vegetables, and whole grains would not be expected to be different between the 1990s and now. Sixth, although dietary intakes were assessed at 2 time points, we cannot rule out the fact that dietary habits may have changed over time. Lastly, even though potential confounders were rigorously assessed, there may be residual confounding due to unmeasured or incorrectly measured variables. For instance, an almost null association between the uPDI and incident hypertension in model 2 is likely due to residual confounding, particularly BMI, considering that those in the highest quintile of the uPDI were less likely to be obese. However, it is unclear if unhealthy plant-based diets are associated with body weight, because recent studies have shown mixed results (35, 36). Future studies which elucidate the pathways through which unhealthful plant-based diets are associated with cardiometabolic outcomes may be useful in interpreting our results.

In conclusion, the PDIs were similar in that they positively scored fruits, vegetables, whole grains, nuts, and legumes, but reverse scored all animal products. However, the diet indices differed in how they scored alcohol, margarine, and dairy, and how they weighted specific dietary components. We found that 3 PDIs (the oPDI, hPDI, and provegetarian diet index) were consistently associated with a lower risk of hypertension, and improved prediction of incident hypertension. However, the PDI-Rotterdam did not detect an association. The study provides evidence that differences in scoring need to be taken into account when evaluating adherence to a plant-based dietary pattern and health outcomes. Several factors, including differences in overall dietary patterns and usual preparation methods between populations for which the indices were developed, likely determined the scoring of specific food groups (i.e., alcohol, margarine, potatoes, and dairy). In future studies, it is important to provide justification associated with the scoring of key food groups to facilitate replication in different study populations.

Notes

The Atherosclerosis Risk in Communities Study was supported by National Heart, Lung, and Blood Institute (NHLBI), NIH, Department of Health and Human Services awards HHSN268201700001I, HHSN268201700002I, HHSN268201700003I, HHSN268201700004I, and HHSN268201700005I (to JC). HK was supported by the Department of International Health Tuition Scholarship, Bacon Chow Endowed Award, Harry D Kruse Fellowship, Harry J Prebluda Fellowship, and the Richard and Barbara Hall Student Award from the Program in Human Nutrition in the Department of International Health at the Johns Hopkins Bloomberg School of Public Health. CMR was supported by National Institute of Diabetes and Digestive and Kidney Diseases Mentored Research Scientist Development Award K01 DK107782 and NHLBI grant R21 HL143089.

Author disclosures: The authors report no conflicts of interest.

The funding agencies had no role in study design, data collection, analysis, drafting of the manuscript, and the decision to submit this manuscript for publication.

Supplemental Tables 1–5 and Supplemental Figures 1–3 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/jn/. In the Supplementary data, oPDI is labeled PDI.

Abbreviations used: ARIC, Atherosclerosis Risk in Communities; DASH, Dietary Approaches to Stop Hypertension; hPDI, healthy plant-based diet index; oPDI, overall plant-based diet index; PDI, plant-based diet index; PDI-Rotterdam, plant-based diet index from the Rotterdam Study; SSB, sugar-sweetened beverage; uPDI, less healthy (unhealthy) plant-based diet index.