Description of the Study Groups
For the study overview see the Study flow-chart Figure 1 and Table 1.
A total of 95 families (OM=25, VG=23,, VN=47) consisting of 187 adults, 65 children >3 years of age and 77 children <3 years of age were enrolled and examined in a cross-sectional setting. Adult vegans were on exclusive plant-based diets on average ≈for 7.4 years whereas vegetarians ≈12 years, and all children were on each respective diet from birth. Overall participants in the study were healthy. Twenty adult participants reported history of a thyroid disease (OM=3, VG=6, VN=11), of whom 10 were treated with thyroid hormone substitution. One participant had type 1 diabetes on insulin (VG=1), two participants had hypertension compensated on the treatment (VG=2, OM=1) and nine participants reported a history of hyperlipoproteinemia (VN=5, VG=1, OM=3) compensated in all without treatment. Three vegans reported a history of osteoporosis and there was a history of fractures in 87 subjects distributed evenly across groups (OM=27, VG=23, VN=27) that are described in detail in Suppl. Table 1. Self-reported prevalence of allergy was significantly lower in adult VN (13%) compared with adult OM (34%, p=0.003) and adult VG (27%, p=0.049). Parents reported lower allergy incidence in VN children < 3 years old compared with OM (2.5 vs 23%, p=0.018). Nine children had a history of atopy (OM=6, VG=2, VN=1) and five had a history of food intolerance (OM=4, VG=1). Psychomotor developmental delay was reported in three children (OM=1, VG=1, VN=1 together with autism and mild mental challenge). No thyroid disease or an autoimmune disease was reported in children across all groups and age strata. Other reported diseases in children < 3 years old included: valve insufficiency (OM=1), umbilical hernia (OM=1), neutropenia (OM=1), sideropenic anemia (OM=1); and in children >3 years old ADHD (VN=1); persistent foramen ovale (VN=2); persistent foramen ovale and epilepsy (VN=1).
Anthropometrics and clinical characteristics
Diet groups were compared separately across three age groups: (i) infants/toddlers (children < 3 years old), (ii) pre-schoolers/schoolers (children > 3 years old), and (iii) adults. Three different statistical techniques were employed: 1/ Kruskal-Wallis (KW) Test and Mann-Whitney U (MWU) Posthoc Test: These were used for raw (unadjusted) comparisons. 2/ Quantile Regression (QR): This method provided effect estimates adjusted for potential confounders such as age (log-transformed for children), sex, and breastfeeding (in children groups). It included cluster bootstrap to account for data dependence within families and modeled not only median values but also the 20th and 80th percentiles. 3/ Robust Mixed-Effects Models (rLME): These models provided adjusted effect estimates (mean values) and quantified the importance of family influence. rLMEs were also fitted for the merged dataset of all children to obtain a more robust estimation of the characteristics shared within families (Summarized in Tables 2-4 and Suppl. Tables 11-13). The previously observed raw median differences were considered 'supported' by the adjusted models when both the median from QR and the mean from rLME showed corresponding differences, and 'partially supported' when only one of these measures provided support.
We found no significant differences in anthropometric and growth characteristics in the children < 3 years old(Table 2), though the VN group in this age stratum tended to have lower median values of height, and weight, albeit normal BMI of values expressed as percentiles of population-based appropriate values per age. Four children (OM=1; VN=2; VG=1) below the third height percentile and two VN children below the weight percentile were identified. In vegans, we found significantly higher serum concentrations of active B12 compared to both OM and VG, further supported by adjusted models for VN vs. VG comparison. Similarly, vegans showed lower levels of homocysteine and methylmalonic acid compared with both OM and VG groups, further supported by the adjusted models. Vegan children had also significantly lower urinary iodine concentration (not further supported with adjusted models) and higher serum vitamin D concentration compared with omnivores (partially supported). Vegan and vegetarian children had higher levels of folate compared with the OM group, further supported with adjusted models. Of note, medians of the differing parameters felt in a normal range in all groups. Nevertheless, twenty children met the criteria for iodine deficiency (i.e., UIC < 100 µg/l; OM=1, VG=3, VN=16), whereby mean TSH=2,25 MU/L, i.e. in the normal range. Out of the sixteen vegan children, two children met the criteria for severe iodine deficiency (i.e., UIC <20 µg/l). Eight children with elevated active B12 levels (i.e. holotranscobalamin >169,2 pmol/l; VG=1, VN=7) were identified. On the contrary, vitamin B12 deficiency (i.e., holotranscobalamin <27,4 pmol/l) was identified in only two children (OM=1, VN=1). Sixteen children met the criteria for mild vitamin D depletion (i.e. 25(OH)D <75 nmol/l; OM=7; VG=2; VN=7).
Among the groups of children > 3 years old (Table 3) we observed no differences in anthropometric and growth characteristics. Three children (OM=1; VN=1; VG=1) below the third height percentile and two vegan children below the third weight percentile were identified. In vegan children, we found significantly lower serum concentrations of total and LDL cholesterol as well as higher P1NP and vitamin D status, though these differences were generally not supported with adjusted models. Vegans had lower urinary iodine concentrations, (not further supported with adjusted models), and higher folate levels which remained significant after adjustment for covariates included in the models. Serum B12 showed a trend towards significant differences in vegans having the highest levels while vegetarians having the lowest levels. Of note, medians of the differing parameters felt in a normal range in all groups. Nevertheless, four children had LDL cholesterol levels below the lower reference limit (OM=1, VG=1, VN=2), twenty-five children, mostly on an omnivorous diet, met criteria for mild vitamin D depletion (i.e. 25(OH)D < 75 nmol/l; OM=13, VG=7, VN=5). Eleven children with elevated B12 levels (i.e. hlotranscobalamin >169,2 pmol/l; VG=3, VN=8) were identified. On the contrary, vitamin B12 deficiency was not present in this children group.
In adults (Table 4) we found comparable anthropometric characteristics in all three groups. Vegans had lower diastolic blood pressure compared to the other groups. In omnivores, we observed significantly higher levels of total and LDL cholesterol than in both plant-based diet-adhering groups. We did not find any clear differences in serum concentrations of biogenic elements except for zinc, which was significantly different among all three groups with vegans being the lowest one (VN<VG<OM). Iron metabolism parameters were comparable among groups excluding the serum ferritin which was lower in both VN and VG when compared to OM. While vegans had significantly higher serum PTH and vitamin D concentrations and lower urea and creatinine concentrations when compared to the other groups, these values nevertheless fell in the normal reference range. Active B12 showed a trend towards significant differences in vegans having the highest levels while vegetarians having the lowest levels. Correspondingly, serum MMA values were highest in vegetarians and lowest in vegans. Folate levels were lowest in omnivores. Nearly all statistically significant differences among the groups identified remained significant after adjustment for non-dietary variables. Of note, we identified subjects with values out of the reference range: twelve subjects had low zinc (i.e., zinc < 9,8 mmol/l; OM=2, VG=1, VN=9), sixty-five subjects low iron stores (i.e. ferritin < 22 (μg/l); OM=12, VG=20, VN=33) and seventy-six low vitamin D status (i.e. 25(OH)D <75 nmol/l; OM=35, VG=25, VN=44). One vegan participant had vitamin B12 deficiency (i.e. vitamin B12 <27,4 pmol/l). On the contrary, we identified subjects above the upper reference limit in active vitamin B12 (OM=2, VG=1, VN=2), total and LDL cholesterol (VN=15, VG=13, OM=17), and PTH (VN=6).
Evaluation of nutritional risks in marginal subgroups
Potential nutritional deficiencies may not manifest in the majority population, but marginalized groups could still be at risk. Therefore, we used quantile regression models to analyze not only the central tendency (median values) but also the 20th and 80th percentiles of each clinical outcome.
In children, diet effects were mostly consistent across quantiles (see Supplementary Tables 2-3). An exception was found in urinary iodine levels, which significantly differed between VN and OM at the 20th percentile but not at another percentile. The adjusted VN-OM difference at the 80th percentile was also less than half compared to differences at the 20th and 50th percentiles in both age groups of children (see Suppl. Tables 2 and 3). Interestingly, vegan children included the 6 lowest but also the 3 highest urinary iodine levels. Altogether, it suggests that while urinary iodine levels may generally be lower in vegans compared to omnivores, vegan diets may be associated with wider spread towards higher iodine levels, hiding mean and median difference between diets.
In adults, the most interesting results were found for serum selenium, where the medians and 80th percentiles were similar across diet groups, but a significant and relatively large difference between VN vs OM in the 20th percentile (-0.17 µmol/l, p = 0.039) was found. VG vs OM showed a similar but opposing trend (-0.19 for the 20th percentile, but diff. -0.05 with p=0.01 for the median and the 80th percentile respectively). Moreover, although unadjusted serum selenium differences were modest between omnivores and plant-based diet (0.03 µmol/l), there was a larger unadjusted difference from OM levels lower quartiles (0.17 and 0.25 µmol/l in VN and VG respectively). This suggests that while serum selenium levels may not be universally higher in omnivores, part of vegans may have inadequate selenium stores.
For other characteristics, the most differences between diet groups were generally similar across percentiles (Supplementary Table 4). In some variables, there were significant differences between OM and VN (BMI, fat-free mass, serum zinc, and serum creatinine). However, the effects of diets were similar in these variables across quantiles, with differences in statistical significance potentially reflecting random variation or distributional characteristics, such as heavier tails for larger values causing inconsistent estimation of 80th percentiles.
Family clustering and covariates' importance
Besides diet, we anticipated that clinical characteristics are influenced by other factors, namely those clustering within families. rLME allows us to assess the relative importance of covariates and quantify the extent to which these characteristics cluster within families. To evaluate the importance of each variable in a model, we employed the Akaike Information Criterion (AIC), which estimates a variable's contribution to the model’s out-of-sample predictive accuracy. A larger decrease in AIC following the inclusion of a variable indicates a larger contribution to the model’s predictive capability. The results are summarized in Figure 2. When the AIC showed the large importance of family, we also reported an adjusted intraclass correlation coefficient (ICC), indicating how much of the total variability in the outcome is due to the grouping structure, i.e. family, after accounting for other variables.
In young children, age was the most important covariate for Ca, P, Fe, transferrin saturation, CTx, IGF-1, and particularly P1NP. Sex had negligible effects. Breastfeeding was the most important factor for ferritin levels, but also contributed to P1NP and IGF1. Diet was the most crucial for MCV, MMA, and folate. Anthropometric factors, mainly weight and height, were most importantly shaped by the birth weight. Surprisingly we did not find an effect of supplementation on relevant minerals or vitamins.
In children >3 years, age was crucial for HDL, creatinine, homocysteine, and IGF1. Sex had again negligible importance except for a modest contribution to calcium levels. Breastfeeding-related covariates had negligible importance, contrasting the situation in younger children. Birth weight shaped the actual weight and height. Iron supplementation was the strongest predictor for transferrin levels. Vitamin B12 supplementation was related to B12 levels and the diet was the most important factor for magnesium, MCV, and folate levels.
As family importance could be best inferred from larger data, with more observations per family, we decided to infer the importance of family clustering from the joint analysis of all children (merging both age groups together). Clustering within the family was found for most variables, but the most prominent was this factor for height (ICC = 57%), HDL cholesterol (59%), B12 (60%), PTH (59%), uric acid (56%), and particularly vitamin D (67%).
In adults, age was the main determinant of CTx but had little to no importance for other variables. In contrast to children, most variables were shaped by sex. In addition to anthropometrics, sex was crucial also for blood pressure, HDL, TG, Ca, and P levels, variables reflecting iron metabolism, hemoglobin, urea, and creatinine, uric acid, and homocystein. We did not identify the importance of vitamins and mineral supplementations. Diet was the main determinant of total and LDL cholesterol, PTH, but contributed to other variables as well. Family clustering was substantial for circulating selenium (ICC = 73%), zinc (41%), urinary iodine (58%), B12 (44%), and folate (35%), altogether likely reflecting family-specific diet habits.
Dietary intake
Dietary intake of the main macro- and micronutrients of interest is summarized in Figure 4 and Suppl. Table 5-7.
The differences among groups were negligible in children < 3 years old. In this age group, the diet composition was similar across all groups, only VN and VG children had significantly lower intake of saturated fats and cholesterol. We identified a tendency towards lower intake of selenium (p=0.057) and higher intake of fiber (p=0.074) in both groups preferring plant-based diets compared with the OM group.
Similarly, in the age stratum of preschool children (children > 3 years old), the total energy, carbohydrate, and fat intake was not different among the groups. Both groups adhering to plant-based diets (VN and VG) had a significantly higher intake of fiber and consumed less cholesterol (VN<<VG) compared with the OM group. The protein intake tended to be lower in VN and VG compared with OM but it reached statistical significance only in the VN group. Micronutrient intake was comparable among all groups except selenium, which intake was lower in both VN and VG.
Among adults, all groups had comparable total energy, sugar, protein, and fat intake. As expected, VN participants had a lower intake of saturated fats and cholesterol (VN < VG < OM) and a higher intake of fiber (VN > VG > OM) compared with the OM group. Carbohydrate intake was higher in VN only. Concerning micronutrients, the VN group had a higher intake of magnesium, zinc, and iron than both the VG and OM groups. Participants adhering to plant-based diets (VN and VG) had a lower intake of iodine and selenium than OM.
Supplementation habits
Micronutrients were supplemented by many study participants in the form of dietary supplements, but the exact dose is generally very complicated to quantify. The diet record may not reflect year-round supplementation and may underreport overall intake in irregularly supplementing persons. Therefore, we used a qualitative approach in surveying individual nutrient use among the study participants. The results are summarized in Suppl. Table 8. The groups differed significantly in supplementation habits, namely in the intake of B12; vitamin D, and n-3 fatty acids. A high proportion of vegans and vegetarians supplemented vitamin B12 across all age strata; omnivores did not supplement B12 at all. Vegans and vegetarians also supplemented n-3 fatty acids (VN>VG). For all groups, there was a significant proportion of individuals who supplemented vitamin D; although the number of supplementing omnivores is about half that of vegans or vegetarians.
Clinical variables as diet predictors
To explore how dietary patterns influence clinical characteristics, we employed elastic net logistic regression to determine whether the clinical characteristics could effectively discriminate between different diet groups. This approach provides insight into the extent to which diet shapes health outcomes, offering a predictive perspective on the role of diet in determining clinical profiles.
For each group, i.e. age-specific subgroups within children and adults, we began by fitting a baseline model incorporating basic subject characteristics (age, sex, and, for children, breastfeeding status) as predictors. Subsequently, we expanded our analysis with a more complex model that included also diverse clinical outcomes as predictors. The predictive capacity of clinical variables was estimated as the difference between the discriminative capacity of complex and baseline models, expressed as a difference between out-of-sample areas under ROC curves of both models (AUC_gain) (Table 5).
Generally, we were able to reliably discriminate between VN and OM in adults, with out-of-sample AUC 0.82 (95% CI: 0.69 to 0.92), whereas it was only 0.54 in the baseline model (not utilizing clinical characteristics), with a mean (?) AUC gain of 0.28 (0.08 to 0.49). The strongest predictors of VN diet are lower glycemia, total cholesterol, zinc, ferritin, and urea, and higher P1NP and folate.
In both children's age groups, the predictions were also relatively stronger when discriminating between VN and OM. Here the complete model performance was 0.74 (95% CI: 0.54 to 0.91) and 0.75 (95% CI: 0.45 to 0.97), respectively. However, these prediction models were shown unstable during bootstrap validation, providing more inconsistent performances over bootstrap resamples. The main predictors in these models were variables associated with vitamin supplementation, i.e. the strongest predictor for VN diet in children < 3 years old was low MMA serum concentration. When we omitted these variables (B12, homocystein, MMA, folate, vitamin D) from the set predictors, the performance of the complete models substantially decreased, i.e. AUCchildren<3yr=0.61 (95% CI: 0.37 to 0.82), AUCchildren>3yr=0.69 ((95% CI: 0.42;0.94) (Suppl. Table 10).