This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Formative Research, is properly cited. The complete bibliographic information, a link to the original publication on https://formative.jmir.org, as well as this copyright and license information must be included.
Dietary quality plays an essential role in the prevention and management of metabolic syndrome (MetS).
The aim of this pilot study is to organize personalized dietary advice in a real-life setting and to explore the effects on dietary intake, metabolic health, and perceived health.
We followed a one-group pretest-posttest design and included 37 individuals at risk of MetS, who indicated motivation to change dietary behavior. For a period of 16 weeks, participants received personalized advice (t=0 and t=8) and feedback (t=0, t=4, t=8, t=12 and t=16) on dietary quality and metabolic health (ie, waist circumference, BMI, blood pressure, lipid profile, fasting glucose levels, and C-peptide). Personalized advice was generated in a two-stage process. In stage 1, an automated algorithm generated advice per food group, integrating data on individual dietary quality (Dutch Healthy Diet Index; total score 8-80) and metabolic health parameters. Stage 2 included a telephone consultation with a trained dietitian to define a personal dietary behavior change strategy and to discuss individual preferences. Dietary quality and metabolic health markers were assessed at t=0, t=8, and t=16. Self-perceived health was evaluated on 7-point Likert scales at t=0 and t=16.
At the end of the study period, dietary quality was significantly improved compared with the baseline (Dutch Healthy Diet Index +4.3;
This exploratory study showed that personalized dietary advice resulted in positive effects on dietary behavior, metabolic health, and self-perceived health in motivated pre-MetS adults. The study was performed in a do-it-yourself setting, highlighting the potential of at-home health improvement through dietary changes.
ClinicalTrials.gov NCT04595669; https://clinicaltrials.gov/ct2/show/NCT04595669
Metabolic syndrome (MetS) is associated with a two-fold increased risk of cardiovascular diseases and a five-fold increased risk of type 2 diabetes [
Unhealthy dietary habits are a major risk factor for developing MetS and are probably even more relevant than sedentary lifestyles [
Personalized nutrition, that is, evidence-based dietary advice tailored toward an individual based on individual-specific information, is most likely an effective strategy to support dietary behavior change, resulting in measurable health benefits [
We distinguished two potential reasons for this effectiveness. First, each person receives advice that addresses the individual nutritional needs based on the person’s biology, thereby maximizing the individual health effect. In a study on healthy volunteers receiving placebo or anti-inflammatory dietary mix supplements, the inflammatory, oxidative, and metabolic responses were highly variable among individuals, suggesting different nutritional needs based on the person’s biology [
A second reason for personalized nutrition being effective is increased adherence to the advice when it is made personal. Each person receives only the information based on their characteristics, rather than generic information based on the characteristics of the population. Therefore, people are more likely to pay attention and feel more involved, especially when the information is tailored to the personal level of motivation [
Celis-Moralis et al reviewed the evidence on personalized interventions and concluded that there is a strong need for further development, testing, and implementation of digitally delivered, evidence-based, personalized interventions that incorporate effective behavior change techniques (eg, personal goal setting and feedback on performance) and are delivered digitally [
The primary aim of this pilot study is to organize personalized dietary advice in a real-life setting. We build upon the research described by Doets et al [
All participants provided informed consent for inclusion before they participated in the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of Tilburg University (file number NL61382.028.17).
An overview of the recruitment procedure is shown in
Flowchart of recruitment and screening procedure. HDL: high-density lipoprotein.
To improve health and behavioral changes through personalized advice, we included individuals at risk of MetS (second set of inclusion criteria). Therefore, individuals eligible for study participation were invited for additional screening to verify whether they were at risk of MetS, defined as an excessive waist circumference (≥88 cm for women and ≥102 cm for men) combined with elevated fasted triglycerides levels (≥1.7 mmol/L), reduced HDL level (<1.03 mmol/L for men and <1.29 mmol/L for women), high BP (systolic: ≥130 mm Hg or diastolic: ≥85 mm Hg), or elevated fasting glucose (>5.6 mmol/L).
On the basis of the study by Doets et al [
The study followed a one-group pretest-posttest design with a duration of 16 weeks (
Overview of study design: measurements, interventions, and planning.
Measurements and characteristics of the intervention | Timepoints (weeks), t | ||
|
|||
|
Diet quality (Dutch Healthy Diet Index) per food category and total score | 0, 8, 16 | |
|
Carotenoids in blood (biomarker fruit and vegetable intake) | 0, 8, 16 | |
|
Food purchase data at retailer (via customer card) | 4, 12 | |
|
|||
|
Self-perceived health questionnaire | 0, 16 | |
|
|||
|
Evaluation questionnaire | 16 | |
|
|||
|
Waist circumference | 0, 8, 16 | |
|
BMI | 0, 8, 16 | |
|
Blood pressure | 0, 8, 16 | |
|
Lipid profile (total cholesterol, HDLa, LDLb, and triglycerides) | 0, 8, 16 | |
|
Fasting glucose | 0, 8, 16 | |
|
C-peptide | 0, 8, 16 | |
|
|||
|
Stage 1: automated advice based on individual diet quality and metabolic health status | 0, 8 | |
|
Stage 2: telephone consultation with dietitian to define behavioral change strategy and discuss personal preferences | 0, 8 | |
|
|||
|
Diet quality discussed in telephone consultations with dietitian | 0, 8 | |
|
Alternatives for product purchases in email messages from dietitian | 4, 12 | |
|
Metabolic health via web-based platform | 0, 8, 16 | |
|
Integrated personal health score via web-based platform | 0, 16 |
aHDL: high-density lipoprotein.
bLDL: low-density lipoprotein.
During the 3 test days (t=0, t=8, and t=16 weeks), participants arrived in the morning in a fasted state to Wageningen University and Research, the Netherlands. Metabolic health parameters were assessed by trained research nurses using do-it-yourself devices, following standard operating procedures. Total cholesterol, HDL, low-density lipoprotein (LDL), and triglycerides were measured in finger-prick blood using the Mission Cholesterol 3-in-1 device (Acon Labs Inc). Glucose levels were assessed using a MediTouch 2 blood glucose meter (Medisana). BP was measured using a Medisana upper arm BP monitor. Both glucose and BP were measured twice for each participant, and the average result was used as input for feedback and personalized advice.
Dietary quality was assessed by using a web-based version of the Dutch Healthy Diet Index (DHDI;
Waist circumference was determined directly over the skin at the midpoint between the lower part of the last rib and the top of the hip. Body weight was recorded on a calibrated weighing scale to the nearest 0.1 kg. Finger-prick blood was blotted on dried blood spot (DBS) cards (Protein Saver TM 903R Cards, Whatman). To suppress the degradation of carotenoids in the DBS samples, the first two circles in the DBS cards were impregnated with a proprietary stabilizing solution supplied by Vitas AS. After air drying for several hours, the cards were stored in airtight resealable aluminum bags (Whatman) with a desiccant pouch (Reàl Marine A/S Stavanger) to remove any moisture from the DBS cards. C-peptide and carotenoids were assessed via high-performance liquid chromatography with UV detection and liquid chromatography-mass spectrometry (Vitas AS) [
At t=0 and t=8 weeks, the participants received personalized dietary advice. Personalized advice was generated in a two-stage process. During stage 1, the content of the advice was defined based on individual dietary habits (ie, DHDI and carotenoid levels as a biomarker of fruit and vegetable intake) and parameters of metabolic health. The results of these measurements were added to an automated personalized dietary advice system. First, the algorithm evaluated per food category (dairy, fats and oils, fish, fruit, nuts, sugar-containing beverages, vegetables, and wholegrain products) whether intake and nutrient status were sufficient based on predefined cut-off levels. If intake or nutrient status was insufficient, the food category was included in the advice. Second, the system evaluated the presence of metabolic abnormalities. If metabolic abnormalities were present, relevant food categories were included in the advice to emphasize the importance of adequate intake for a specific food category.
Stage 2 included a telephone consultation of 45-60 minutes, during which a trained dietitian discussed the system-generated advice with the participant following a standard protocol. During the consultation, a personal dietary behavior change strategy was defined by adapting the advice from stage 1 to individual preferences (eg, number of food groups to work on, selection of alternative products, and adjustment of portion sizes). In Table S1 of
Feedback on behavioral parameters was provided to participants by a dietitian as part of the individual telephone consultations at t=0 and t=8 weeks and via email at t=4 and t=12 weeks. The feedback by telephone addressed adherence to Dutch dietary guidelines based on the DHDI. The feedback by email addressed healthy alternatives for recent product choices and was based on purchase data registered on a supermarket customer card that participants were asked to share with the research team. Feedback on metabolic health parameters (ie, waist circumference, BMI, BP, glucose, cholesterol, C-peptide, and triglycerides) was directly communicated to the subjects via a web-based personal study portal at t=0, t=8, and t=16 weeks.
Furthermore, at t=0 and t=16 weeks, each participant received an integrated personal health score based on their metabolic health parameters.
The personal health score was produced using a so-called
Demographics and metabolic health parameters of study participants and the health space reference groups (N=85).
Variable | Study participants (n=34) | Healthy referencea (n=10) | MetSb referencea (n=41) | ||||
|
|||||||
|
Male | 9 (26) | 5 (50) | 19 (46) | |||
|
Female | 25 (74) | 5 (50) | 22 (54) | |||
Age (years), mean (SD) | 61 (8.2) | 57.6 (16.2) | 54 (21.0) | ||||
BMI (kg/m2), mean (SD) | 29.9 (4.18) | 21.3 (1.88) | 31.1 (5.68) | ||||
Waist circumference (cm), mean (SD) | 102 (11.4) | 83.2 (4.68) | 105 (10.8) | ||||
Total cholesterol (mmol/L), mean (SD) | 6.23 (0.78) | 5.32 (1.10) | 4.77 (1.02) | ||||
HDLc cholesterol (mmol/L), mean (SD) | 1.14 (0.27) | 1.49 (0.37) | 1.01 (0.13) | ||||
LDLd cholesterol (mmol/L), mean (SD) | 4.34 (0.74) | 3.12 (1.02) | 2.73 (0.99) | ||||
Triglycerides (mmol/L), mean (SD) | 1.67 (0.85) | 1.57 (0.52) | 2.25 (0.95) | ||||
Glucose (mmol/L), mean (SD) | 5.61 (0.65) | 5.33 (0.70) | 7.07 (3.08) | ||||
C-peptide (nmol/L), mean (SD) | 0.52 (0.33) | 0.75 (0.40) | 1.46 (0.90) | ||||
Systolic blood pressure (mm Hg), mean (SD) | 135 (18.0) | 138 (18.5) | 128 (21.5) | ||||
Diastolic blood pressure (mm Hg), mean (SD) | 78.6 (9.54) | 77.7 (14.0) | 67.4 (18.6) |
aData for the reference groups were obtained from the National Health and Nutrition Examination Survey 2003-2004 (CDC 2003) [
bMetS: metabolic syndrome.
cHDL: high-density lipoprotein.
dLDL: low-density lipoprotein.
This aggregated rank is based on the features included in the trained model, where the highest rank corresponds to the healthiest values of these features. The aggregated rank of this collection of features was calculated using the
A multivariate mixed-effects regression model was subsequently fitted to the data from the two selected reference groups with good classification performance with an accuracy of 99% and a Cohen κ coefficient of 0.94. The model includes triglycerides, LDL cholesterol, HDL cholesterol, glucose, and C-peptide as fixed effects and sex as a random effect. The random effect was included to allow for sex differences in the final model coefficients. Table S2 in
At baseline and at the end of the study, participants reported self-perceived health, self-perceived healthiness of the diet, and satisfaction with the diet using a 7-point Likert scale ranging from 1=very unhealthy to 7=very healthy and 1=very unsatisfied to 7=very satisfied.
At the end of the study, participants filled out an evaluation questionnaire on personal experiences regarding advice, feedback, and the digital platform (statements on 7-point Likert scales, ranging from 1=completely disagree to 7=completely agree).
Data on DHDI scores and metabolic health were analyzed using linear mixed models with
In addition, Pearson correlation coefficients were calculated between the Δ of the single dietary behavior variables and the single metabolic health variables and between the Δ of dietary behavior and metabolic health variables. Only significant correlations that could be visually confirmed in the scatterplots were regarded as reliable (Figure S1 in
For the analyses of the individual food categories, only participants that actually incorporated the specific food category in their dietary behavior change strategy were included.
Statistical significance was set at
A total of 37 individuals were enrolled in this study. During the study period, 3 subjects dropped out: 1 participant no longer met the inclusion criteria, and the other 2 experienced too many difficulties in using the web-based platform. The baseline characteristics of the 34 participants who completed the intervention as well as the reference populations used for modeling the health score are summarized in
Most subjects (33/34, 97%) were provided with advice on multiple food categories in their individual dietary behavior change strategies. One participant chose to focus on only one food category. Advice was provided most frequently on vegetables (31/34, 91%), followed by oils and fat (21/34, 62%), nuts (20/34, 59%), wholegrain products (19/34, 56%), dairy (14/34, 41%), fish (12/34, 35%), fruit (9/34, 27%), and sugar-containing beverages (3/34, 9%). The mean DHDI scores over time per food category are shown in
Dutch Healthy Diet Index per food category (score 1-10) and total score (score 8-80) and total carotenoids (µmol/L) at baseline, 8 weeks, and 16 weeks.
Food categorya | DHDIb score, mean (SD) | ||||
|
t=0 weeks | t=8 weeks | t=16 weeks |
|
|
Vegetable intake (n=31) | 6.6 (2.9) | 7.5 (3.0) | 7.1 (3.3) | .53 | |
Fruit intake (n=9) | 5.8 (3.3) | 8.2 (2.0) | 8.4 (2.0) | .70 | |
Intake of oils and fats (n=21) | 3.7 (3.9) | 3.6 (4.1) | 3.5 (3.8) | .85 | |
Fish intake (n=12) | 6.6 (3.2) | 7.5 (2.6) | 8.6 (1.9) | .18 | |
Intake of wholegrain products (n=19) | 6.3 (2.5)c | 7.7 (2.8)c | 7.9 (2.8)c | .009 | |
Dairy intake (n=14) | 3.1 (2.6) | 3.8 (3.1) | 4.1 (2.7) | .84 | |
Nut intake (n=20) | 6.2 (3.3)c | 7.0 (2.6)c | 8.4 (2.5)c | .009 | |
Intake of sugar-containing beverages (n=3) | 1.9 (1.8) | 5.5 (5.0) | 6.6 (2.8) | —d | |
Total DHDI (sum of all food categories; n=34) | 52.9 (13.1)c | 56.5 (11.3)c | 57.2 (11.5)c | <.001 | |
Carotenoid levels in blood (µmol/L; t=0, n=36; t=8, n=34; t=16, n=33) | 1.21 (0.43) | 1.39 (0.46) | 1.42 (0.56) | .66 |
aOnly participants who included the specific food category in their individual dietary behavior change strategy are included in the analysis.
bDHDI: Dutch Healthy Diet Index.
cNo significant difference following the post hoc analysis.
dNot available (as the sample size was not sufficient to obtain reliable statistical output).
After 16 weeks of intervention, triglycerides, HDL cholesterol, LDL cholesterol, BMI, waist circumference, C-peptide, and HOMA-IR were all significantly improved (
No significant correlation was found between the changes in the total DHDI and health scores (ρ=0.12;
Metabolic health parameters assessed at t=0, 8, and 16 weeks.
Parameter | t=0 weeks, mean (SD) | t=8 weeks, mean (SD) | t=16 weeks, mean (SD) | |
Glucose (mmol/L) | 5.61 (0.67)a | 5.63 (0.64)a | 5.84 (0.63)a | .04 |
C-peptide (nmol/L) | 0.52 (0.32)a | 0.52 (0.23)a | 0.43 (0.16)a | .01 |
HOMA-IRb | 7.45 (5.27)a | 7.42 (3.95)a | 6.31 (2.57)a | .049 |
Triglycerides (mmol/L) | 1.67 (0.86)a | 1.43 (0.60)a | 1.39 (0.55)a | .02 |
Total cholesterol (mmol/L) | 6.23 (0.78) | 5.91 (0.84)a | 5.90 (0.86)a | 0.01 |
HDLc cholesterol (mmol/L) | 1.14 (0.28)a | 1.09 (0.28)a | 1.44 (0.36)a | <.001 |
LDLd cholesterol (mmol/L) | 4.34 (0.74)a | 4.18 (0.79)a | 3.87 (0.78)a | <.001 |
Systolic blood pressure (mm Hg) | 135 (18.2) | 133 (13.7) | 132 (17.1) | .70 |
Diastolic blood pressure (mm Hg) | 78.6 (9.60) | 80.3 (8.93) | 79.6 (9.77) | .30 |
BMI (kg/m2) | 29.9 (3.94)a | 29.4 (3.60)a | 29.2 (3.66)a | <.001 |
Waist circumference (cm) | 102 (11.5)a | 100 (9.43)a | 99.4 (8.86)a | .01 |
Health score (arbitrary units) | 1.30 (0.31) | 1.23 (0.30) | 1.57 (0.32) | <.001 |
aNo significant difference following the post hoc analysis.
bHOMA-IR: homeostatic model assessment–insulin resistance; calculated based on glucose and C-peptide [
cHDL: high-density lipoprotein.
dLDL: low-density lipoprotein.
Association between the Δ dietary scores and Δ health scores calculated between week 0 and week 16 of the study.
The mean scores for self-perceived health, self-perceived healthiness of the diet, and satisfaction with the diet, as reported by the participants at baseline and end, are shown in
Self-perceived health, self-perceived healthiness of the diet, and satisfaction with the diet as reported at t=0 and 16 weeks.
Self-perceived health items | t=0 weeks, mean (SD) | t=16 weeks, mean (SD) | |||||
Self-perceived healtha | 4.68 (1.07) | 5.35 (1.10) | .005 | ||||
Self-perceived healthiness of dieta | 4.50 (1.05) | 5.56 (0.96) | <.001 | ||||
Satisfaction with dietb | 4.35 (1.39) | 5.29 (1.14) | .001 | ||||
|
|||||||
|
The personalized advice helped me to improve my diet | —d | 5.7 (1.5) | — | |||
|
The feedback helped me to improve my diet | — | 5.4 (1.4) | — | |||
|
If possible I would continue taking part in this program | — | 4.7 (2.0) | — | |||
|
I would recommend people in my surroundings to obtain personalized advice like in this study | — | 5.0 (1.8) | — | |||
|
I would be willing to pay for this program | — | 2.5 (1.7) | — |
a7-point Likert scale, ranging from “very unhealthy” to “very healthy.”
b7-point Likert scale, ranging from “very unsatisfied” to “very satisfied.”
c7-point Likert scale, ranging from “completely disagree” to “completely agree.”
dConsumer experiences were only assessed at the end of the study (t=16).
With the aim of exploring the combined behavioral and metabolic health effects in relation to personalized nutrition, we have shown that personalized dietary advice delivered through an automated advice system and discussed by a dietitian with the participant has a significant positive effect on dietary behavior, with a concurring beneficial impact on metabolic health in consumers at risk of MetS. Moreover, the perception of health and healthiness of and satisfaction with the diet improved.
Most earlier studies have shown the impact of personalized dietary advice either on dietary intake or on metabolic health parameters. In this study, we build upon the research described by Doets et al [
Compared with Doets et al [
Interestingly, our results revealed no correlation between the effect on dietary behavior and metabolic health, although both variables showed a significant improvement.
In our study, there was a large variation in the personalized advice between participants, as the advice was tailored to individual metabolic health status as well as dietary quality. Most participants in our study sample incorporated improved intake of vegetables, oils and fats, nuts, and wholegrain products in their behavior change strategy (n≥19). Among these, participants seemed to comply with the advice for wholegrain products and nuts, especially as these two food groups significantly improved over time. In contrast, no changes were observed in vegetables, oils, and fats. These results suggest that it is easier for motivated participants to replace refined products with wholegrain products or to include nuts in their dietary patterns as compared with increasing vegetable intake or changing the type of fat for the preparation of meals or for bread spread. Previous systematic reviews have shown significant pooled effects of dietary advice interventions on increased intake of fruits, vegetables, total fiber, and total fat [
Although our results indicated an improvement over time for fruit, fish, dairy, and intake of sugar-containing beverages, these effects turned out to be nonsignificant as only a small number of individuals included these food groups in their behavior change strategy (n≤14).
Several studies have reviewed available behavior change techniques that are effective for dietary behavior change [
We hypothesized that by optimizing the quality of the diet in terms of adherence to Dutch dietary guidelines on specific food groups, we were able to improve the metabolic health of our participants. Our analysis indeed showed significant improvements in metabolic health; however, whether these effects were due to improvements in diet quality could not be substantiated. Previous reviews have shown that the restriction of total energy intake, carbohydrate, or fat is a successful strategy to improve metabolic health status. Furthermore, enriching the diet with monounsaturated fatty acids (nuts and olive oil) or omega-3 fatty acids (fish) has been proven effective, especially in improving lipid profiles [
Although the absolute health score is also determined by subtle changes in triglycerides, glucose, C-peptide, and LDL cholesterol, it seems that the relatively strong Δ HDL cholesterol is the main driver for the change in the health score. It is known from the literature that HDL levels are affected by the increased consumption of fish and unsaturated fatty acids and decreased consumption of saturated fatty acids [
Self-perceived health summarizes the objective and subjective aspects of health within the perceptual framework of an individual. Some studies suggest that although the criteria for judging health status may vary between individuals, it is a valid indicator of overall health status and use of health services [
Participants may have become more aware of their dietary behavior throughout the study, which may have influenced their answers to the DHDI questionnaire, causing a learning bias [
No control group was included, which is a general challenge in studies investigating the efficacy of personalized nutrition. Therefore, it is not possible to separate the effect of diet from the potential effect of general health improvement as a behavioral consequence of taking part in the study. A semiplacebo control may be reached by comparing personalized advice with generic advice [
Although we could confirm the assumption that personalized dietary advice is effective in improving both overall dietary behavior (total DHDI score) and overall metabolic health (health score), interestingly, there was no significant correlation. It should be noted that the pilot study only included 34 individuals, all of whom received personalized dietary advice. In addition, there are some limitations to the DHDI score, in which each food category is weighted equally in the total score. An adjusted total score in which the food categories relevant for MetS would outweigh the other food groups could possibly reveal a significant effect.
In addition to the positive effects of improved dietary quality, previous research has also demonstrated the beneficial effects of moderate- to high-intensity physical activity training on lipid profile, BP, and C-reactive protein [
Contrary to our expectations, these data illustrate that positive effects at the population level are not necessarily indicative of associations between diet and health. We can thus conclude that personalized dietary advice works for dietary behavior and health, but the data did not allow us to conclude that metabolic health was improved as a consequence of dietary improvement. A larger sample size with a more equal distribution of men and women and the addition of a control group to the study design are warranted to further investigate and understand the association between diet and health at the individual level. Furthermore, a follow-up after a longer period (eg, 6 months or 1 year) would allow to determine whether initiated behavior changes are maintained over time.
In this exploratory pilot study in individuals at risk for MetS and motivated to change behavior, personalized dietary advice was indicative of positive effects on self-perceived health, dietary behavior, and metabolic health. The lack of association between diet and health improvement is reflective of the individual nature of diet-health relations and underlines the need for an integrated analysis focusing on individual improvements. The study was performed in a do-it-yourself setting, highlighting the potential of evidence-based at-home improvement of health through dietary changes. Follow-up studies are needed to confirm these effects and evaluate the maintenance of dietary behavioral changes.
An overview of the telephone consultation in stage 2 of the personalized advice with the trained dietitian (example of one food group for one participant) and standardized coefficients of the features in the health space model.
blood pressure
dried blood spot
Dutch Healthy Diet Index
high-density lipoprotein
low-density lipoprotein
metabolic syndrome
The authors gratefully acknowledge all participants in the study. Furthermore, they thank Meeke Ummels for her contribution to the study execution and Quinten Ducarmon and Angelique Speulman-Saat for their support in data collection.
This research was supported by the Dutch Personalized Nutrition and Health program funded by the Dutch Top Sector Agri & Food (TKI-AF-15262). The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
ELD, FPMH, MT, and AB designed the study; MT, AB, and TVDB designed the automated advice system; SVDH coordinated the study execution; TVDB, WJVDB, and SVDH analyzed the data; SVDH, FPMH, ELD, and WJVDB prepared the manuscript; MT, AB, and TVDB reviewed the manuscript.
None declared.