https://orcid.org/0000-0001-7738-7686
Figures
Abstract
Food-based dietary guidelines are helpful for governments and health agencies to encourage healthy eating at the population level. In order to assess adherence to such guidelines, index scores have been developed, the version in the Netherlands being the Dutch Healthy Diet-index (2015) (DHD2015-index), which reflect adherence to the 2015 Dutch dietary guidelines. Because a higher diet quality, i.e. a higher adherence to the dietary guidelines, is associated with better health outcomes, a higher DHD2015-index score would also mean better outcomes on measures of health, such as all-cause mortality. The present study aimed to elucidate this by investigating the association between DHD2015-index score and mortality in the Dutch population using data from 97 999 participants in the Lifelines cohort study. For the analyses, Cox Proportional Hazards regression was used, whilst accounting for age, sex, physiological measurements, exercise, and biochemical and lifestyle variables. There was a strong negative association between DHD2015-index score and mortality. Hazard ratios for DHD2015-index scores below 60 were approximately 1.2x larger than the mean. Every 10 unit increase in DHD2015-index scores between 60 and 90 led to a 0.1 reduction in hazard ratio, and every 10 unit increase between 90 and the highest DHD2015-index scores led to a reduction in hazard ratios of 0.05. The hazard ratio for the lowest quartile of DHD2015-index scores was 1.14 (95% CI = 1.04–1.26), whereas that for the highest quartile was 0.88 (95% CI = 0.84–0.92). Our results show a clear inverse relationship between DHD2015-index score and all-cause mortality.
Citation: Kirk D, Baart AM, McLean J, Feskens EJM (2024) Association between diet quality and all-cause mortality in a large Dutch cohort. PLoS ONE 19(8): e0302905. https://doi.org/10.1371/journal.pone.0302905
Editor: Harald Mischak, University of Glasgow, UNITED KINGDOM OF GREAT BRITAIN AND NORTHERN IRELAND
Received: January 24, 2024; Accepted: April 16, 2024; Published: August 23, 2024
Copyright: © 2024 Kirk et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The data is available exclusively through LifeLines, and the authors no longer have access to any of the data used in the study. Access to Lifelines can be requested at: https://www.lifelines-biobank.com/researchers/what-we-offer/data. An extensive description of the original dataset and the data that we used in our analysis is described in the methods and results and supplementary information, respectively. We provide key characteristics of the starting population and the sample population and describe every data processing step which led us to our final sample popoulation. We also submit two additional files, the R code for data processing (Data_processing.R) and the modelling procedure (Script.R). Both of these have been referenced in the paper. The authors received no special privileges in accessing the data that other researchers would not have.
Funding: The Lifelines initiative has been made possible by subsidy from the Dutch Ministry of Health, Welfare and Sport, the Dutch Ministry of Economic Affairs, the University Medical Center Groningen (UMCG), Groningen University and the Provinces in the North of the Netherlands (Drenthe, Friesland, Groningen). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Abbreviations: BP, blood pressure; DHD2015-index, Dutch healthy diet index 2015; HR, hazard ratio; LRT, likelihood ratio test
Introduction
Food-based dietary guidelines advise a population on how to eat for the purpose of health preservation [1]. Such guidelines exist in many countries, with the version in the Netherlands being the Dutch dietary guidelines in 2015 [2]. Given the impact of diet on health and the current and growing prevalence of nutrition-related chronic diseases in society [3], it is crucial to measure diet quality. This measure can be used to evaluate dietary interventions aimed at improving diet quality and health status. To this end, the Dutch Healthy Diet 2015-index (DHD2015-index) has been developed and validated [4].
Because of the intimate relationship between nutrition and health, it follows that, if DHD2015-index accurately reflects Dutch dietary guidelines adherence and higher adherence to the guidelines means a healthier diet, then a higher DHD2015-index score would also mean better outcomes on measures of health, such as all-cause mortality. Understanding the relationship between DHD2015-index score and outcomes of health would also be advantageous because it would mean that changes in dietary intake, reflected as a change in DHD2015-index score, following nutrition-based public health interventions could also approximate the subsequential impact on all-cause mortality. However, this is yet to have been thoroughly investigated. Previous studies have been done in either small sample sizes or as part of a comparison with other diet quality assessment measures [5,6], and the latter of these used quality-adjusted life years (QALYs), not all-cause mortality, as its final outcome in a linear regression.
The current study aims to investigate the association between diet quality, as quantified by the DHD2015-index, and all-cause mortality. To assess this, a sample from the Lifelines cohort, a large, multi-generational, prospective cohort study with over 150 000 individuals from the north of the Netherlands, was used. In addition to the primary research question, other covariates hypothesized or known to be relevant to all-cause mortality are also investigated for their relationship with survival probability.
Methods
Lifelines dataset
Lifelines is a multi-disciplinary prospective population-based cohort study examining in a unique three-generation design the health and health-related behaviours of 167,729 persons living in the North of the Netherlands [7]. It employs a broad range of investigative procedures in assessing the biomedical, socio-demographic, behavioural, physical and psychological factors which contribute to the health and disease of the general population, with a special focus on multi-morbidity and complex genetics. Participants for the Lifelines cohort study were recruited with the help of general practitioner’s practices across the northern three provinces of The Netherlands (Friesland, Groningen and Drenthe) from 2006 until 2013. Within these practices, all patients in the age range of 25–50 years were invited. Exclusion criteria included having a severe mental or physical illness, limited life expectancy (<5 years), and insufficient knowledge of the Dutch language to complete a Dutch questionnaire. Eligible participants received a first questionnaire and were invited to a Lifelines research facility for a comprehensive health assessment. During this visit, participants were also asked to indicate whether family members would be willing to participate in the study, and in the case of a positive response, family members were invited as well. Children were only allowed to participate if one of their parents was included in the study. In addition to this recruitment strategy, inhabitants of the three northern provinces could also register themselves via the Lifelines website. A more detailed description of the total study population of the Lifelines cohort study can be found elsewhere [8].
The data in the Lifelines database was accessed from 8th June 2022 until 11th November 2023. The data was processed and analyzed via access to the Lifelines Research Workspace, where the data was also stored. All participants signed informed consent upon participation, the research was conducted according to the principles of the Declaration of Helsinki the research code UMCG, and the Lifelines protocol was approved by the UMCG Medical Ethical Committee under number 2007/152.
Study population for the current study
Participants were considered for inclusion for the current study if they were over the age of 18, apparently healthy, and had food frequency questionnaire (FFQ) data available. We excluded all participants who had chronic or serious disease states and for whom the reported dietary intake data may not be reflective of their usual diet. This was achieved primarily on the basis of questionnaire responses but also through examination of outstanding values on boxplots for relevant biological variables (S1 Table in S1 File; S1 Fig in S1 File). Chronic or serious diseases states available on the basis of which participants were excluded were aneurysm or stroke, kidney disease, hepatitis, dementia, Parkinson’s disease, cancer at any point in life, type 2 diabetes, liver cirrhosis, heart attack, heart failure, and having had bypass surgery (N = 10 907). Additionally, participants with outliers in biochemical data which either signified serious or chronic disease or unreliable measurements were also excluded (N = 54; S1 Fig in S1 File, S2 Table in S1 File). Participants who responded “Yes” to the question “are you currently on a diet to lose weight?” (n = 7658) were removed. The number of participants excluded per variable can be seen in Fig 1 with additional information found in S3a and S3b Table in S1 File. The final dataset at the start of the analysis had N = 97 999 participants with 1659 events (deaths).
The Dutch healthy diet 2015-index
Dietary intake data was used to calculate the DHD2015-index [4], which is a measure of adherence to the Dutch dietary guidelines 2015 [2]. The DHD2015-index was derived from the Lifelines data base. A detailed description of the calculation of the DHD2015-index using Lifelines data has been described elsewhere [9].
In short, the DHD2015-index consists of fifteen components: vegetables, fruits, wholegrain products, legumes, nuts, dairy, fish, tea, fats and oils, coffee, red meat, processed meat, sweetened beverages and fruit juices, alcohol, salt and unhealthy choices. Recently, a component on unhealthy foods has been added [10]. The present sixteen components can be divided into adequacy, moderation, optimum, qualitative and ratio components. Adequacy components are derived from a guideline that recommends increasing intake (vegetables, fruits, legumes, nuts, fish, and tea). Moderation components are derived from guidelines that recommend limiting intake (red meat, processed meat, sweetened beverages and fruit juices, alcohol, salt, and unhealthy food choices). Dairy is an optimum component based on an optimal range of intakes, whereas coffee is a qualitative component based on the type of coffee. The fats and oils component is a ratio component and is based on the ratio of intake of healthy and unhealthy products in that food group. The wholegrain products component is considered as two types of components because two guidelines for grain products exist: an adequacy component for wholegrain intake and a ratio component to reflect the replacement of refined grain products by wholegrain products. All components are assigned a score based on intake of the specific food group. For all components, a minimum of 0 points and a maximum of 10 points can be allocated, resulting in a total score ranging from 0 to 160 points, with a higher score indicating better adherence to the guidelines.
Dietary intake in the Lifelines cohort study was assessed using the Flower-FFQ [11,12]. This FFQ consists of one main questionnaire (the heart FFQ) and three complementary questionnaires (the petal FFQs). The heart FFQ contains 110 food items used to estimate intakes of major food groups, energy, and macronutrients. The petal FFQs ask for detailed information on the types of food consumed within the food groups of the heart FFQ, as well as supplement intake, to estimate specific (micro)nutrients and food components. Using data from the Flower-FFQ, the DHD-2015-index was calculated [13]. For the current study, only data from the heart FFQ was used. From the Flower-FFQ, data on filtering of coffee and salt intake is not available, so these two components were not included in the DHD2015-index calculations. Additionally, From the heart-FFQ, regarding the wholegrain products component, only the adequacy component, and not the ratio component, with a maximum of 5 points can be assessed. This results in total scores ranging 0 to 135 points for the DHD2015-index in the current study.
Description of covariates
We included all variables for which there was a known or potential association with all-cause mortality or eating behavior (S1 Table in S1 File). This included established demographic, lifestyle, and anthropometric variables that have been consistently identified to impact all-cause mortality risk, such as age, BMI, education, smoking status, and physical activity [14–17]. Routinely measured biochemical plasma markers of health, such as lipid and HbA1c levels, among others, were also included [18,19]. Additionally, due to their previous associations with mortality, we included questionnaire responses assessing quality of life [20].
Data on sex, age, smoking, education status, physical activity, and disease status were obtained from questionnaires. Education status was categorized based on education attainment as follows: no education, primary education, lower vocational education, lower general secondary education (low); intermediate vocational education, higher general secondary education (moderate); higher vocational education, and university education (high) [21]. Smoking was categorized as current, former, and never smoker, although current smoker was excluded in favor of Total Number of Cigarettes Smoked per Day. Physical activity was assessed with the short questionnaire to assess health-enhancing physical activity (SQUASH), from which SQUASH moderate and intense activity scores were derived [22]. Anthropometric measurements, including height, weight, and hip and waist circumference, were conducted by well-trained staff at Lifelines research facilities. Biological variables, such as plasma biochemical profile, average blood pressure (BP), and average resting pulse rate were measured as described elsewhere [8]. A collection of quality-of-life variables was also included, which are categorized into questions based on self-rated health on the topics of emotional state, energy, pain state, physical health, social health, general health, and sleep [8].
Data pre-processing
Survival data were calculated using the difference between date of death and date of enrollment into the study. Participants without a date of death were assumed to be alive and their survival time was calculated as the latest possible date in the dataset subtracted by their enrollment date. All-cause mortality status was also coded accordingly. Only participants over the age of 18 were considered. The starting sample had a size N = 152 864. The number of missing data points that were present the variables either to be used in the final model or those used for excluding participants with chronic illness or serious disease can be seen in S3a Table in S1 File. The removal of these missing led to the exclusion of 26 550 participants. No differences in key characteristics were observed before and after the removal of missing data (S3b Table in S1 File). Data reduction was also performed to remove redundant variables (see supplementary section Data Reduction S4 Table in S1 File). Data used in the variables included in the model were the baseline data of the participants. Whilst Lifelines has longitudinal information, this information was not used in our analysis. The script for the data processing procedure can be seen in the R file “Data_preprocessing.R”.
Statistical analysis
The Cox proportion hazards (PH) model was conceived by David Cox to extend life tables, which had been used previously in medical statistics, to incorporate regression-like arguments and allow multivariate analysis of time until an event [23]. The assumptions and the way in which they were tested in the current study are described below. All statistical analyses were done in R version 4.2.2.
Influential observations
One of the assumptions of the Cox model is that the analysis is free of influential observations, which are data points whose inclusion significantly alters test statistics in a regression model [24]. Influential observations were identified using DFBETAS which quantifies the standard error-scaled change in the estimated regression coefficient upon removal of an observation [25]. DFBETAS plots suggested that 6 six observations from four participants for variables glucose, fasting LDL cholesterol, triglycerides, and HbA1c might have been influential, however since the change in the results before and after the removal of influential observations was insignificant, the data points were ultimately not removed. The interested reader can see the change in the Cox regression summary output in the R code (“Script.R”).
Linearity
Violation of the linear assumption between predictors and the outcome can lead to erroneous conclusions in the Cox PH model. Typically, Martingale residuals are used to test this assumption, however, this approach is influenced by the smoothing procedure used to produce its plots and is only appropriate when collinearity between the predictors is low, something which cannot be expected from our dataset [26,27]. Thus, instead, we employed a Poisson regression approach as recommended by Therneau [27] for visual assessment of the functional form of continuous, with shapes other than a straight-line suggesting nonlinearity. Penalized splines were used as a formal test for nonlinearity detection [28,29]. Penalized splines add a penalty term to the knots on the curve when minimizing the least squares estimation, which allows the use of more knots and a smoother fit (i.e., avoid overfitting) [30]. Differences in the β coefficients between knots on a spline suggest nonlinearity [28]. Penalized splines are easily implemented by adding ‘pspline’ terms from the Survival package to continuous variables, and evidence of nonlinearity can be quickly gleaned from the output [31].
The results of the Poisson regression and the linear and nonlinear interactions tested by penalized splines can be seen in S2 Fig in S1 File and S5 Table in S1 File, respectively. Taking the results together, it was decided that the variables age, total number of cigarettes smoked per day, hemoglobin, leukocyte, HDL cholesterol, diastolic BP, pulse, systolic BP, and DHD2015-index would be modeled with penalized splines to account for complex relationships with the output variable, whereas linearity would be assumed for the variables BMI, WHR, creatinine, glucose, LDL cholesterol, potassium, triglycerides, and SQUASH moderate and vigorous activity scores.
Proportional hazards
The assumption that hazards are proportional (i.e., do not change over time) is a crucial characteristic of the Cox model, necessitating that this assumption is verified for each variable in the model and violating offenders are dealt with accordingly [27,28,32]. A formal test of the β coefficient of the Schoenfeld residuals over time shows the statistical significance of the alternative hypothesis that hazards vary with time [33,34], and plots of the Schoenfeld residuals are used to assess if deviation from proportionality is meaningful.
Variables that violate this assumption can be dealt with by adding time-varying coefficients or time-varying covariates [35,36], stratifying by offending categorical variables (or discretization of continuous variables for the same purpose) [27,28], and partitioning of the time-axis [28,37]. It should also be noted that an incorrectly specified functional form for a given variable can cause PH violation, even in another variable [38]. Thus, it should arguably be a first step to experiment with different functional forms, ideally through the implementation of splines, for continuous variables when encountering nonproportionality.
A test on the time-dependency of the coefficient based on the weighted Schoenfeld residuals was done using the function `cox.zph`from the survival package [31] on each occasion that variables were transformed, such as with the addition of splines. This was done over sequential rounds of modelling in order to select an appropriate model whilst maintaining parsimony.
Eventually, penalized splines were added to all continuous variables except Creatine, Glucose and Potassium. The total degrees of freedom used in the model were acceptable given the number of events, and nonproportionality was not deemed problematic in any of the variables of the final model. The proportional hazards test and the Schoenfeld plots for the final model can be seen in S6 Table in S1 File and S3 Fig in S1 File, respectively.
Final model
The large sample size along with the representativeness of study population for the Dutch general population allowed us to include all of the variables that we considered relevant or potentially relevant for mortality into one model, along with penalized splines to correctly specify the functional form of continuous variables. This led to a final model with 48 variables and 109 degrees of freedom over 1659 events.
Results
The main characteristics of the study population after pre-processing can be seen in Table 1. The mean age of the participants was 43.2 years (SD = 12.5, IQR = 34–50) and there were slightly more females (57.0%) than males. Lifelines is a sample of citizens in the north of the Netherlands, and thus most (83.4%) participants identified as being white, with only 1.6% identifying as other, though there was also significant missing data on this aspect (15%). Most participants were moderately or highly educated (70.8%), almost half were never smokers (48.1%), around one-fifth were current smokers (20.7%), and the rest were former smokers (31.1%). Mean energy intake as assessed by food-frequency questionnaire (FFQ) was 2431.4 (SD = 688.8) and 1872.3 (SD = 489) kcal/day in males and females, respectively. Mean BMI was 26.3kg/m2 (SD = for males and 25.2 kg/m2 for females, however, a larger difference was seen in waist circumference (94.1cm for males and 85.1cm for females. The mean DHD2015-index score was 69.9 (SD = 14.3 overall, 66.9 [SD = 14.1] for males and 72.2 [14.0] for females (Fig 2. Finally, 1.7% (n = 1659) of the population died between the baseline examination and end of follow-up (2.1% in men versus 1.4% women). Fig 3 shows the survival curve. It should be noted that few participants were enrolled for more than 13 years, explaining the abrupt plateau after this timeframe.
Note that since fewer individuals participated in the study for longer than 10 years, the number at risk is small at longer years of participation.
Continuous variables are given as means (standard deviation).
A summary of the results of the Cox regression analyses (final model) is shown in S7 Table in S1 File. DHD2015-index score was strongly associated with all-cause mortality (p<0.001; Table 2). Fig 4. shows how the HR changes over different DHD2015-index scores. The HRs are 1.25 times higher for participants with DHD2015-index below 40 compared to those with the mean DHD2015-index score (69.9), which represented the baseline risk (i.e., HR = 1). An approximately linear decrease in HRs could be seen for scores between 60 and 90, where every 10 unit increase in DHD2015-index score led to a reduction of 0.1 in HR (Fig 4). For participants with scores between 90 and the highest scores (126), every 10 unit increase in DHD2015-index scores led to a reduction in HR of 0.07 (Fig 4). Participants in the lowest quartile had a HR of 1.14 (95% CI = 1.04–1.26) compared with those in the highest quartile, where the HR was 0.88 (95% CI = 0.84–0.92).
An asterisk in the final column represents p<0.05.
Likelihood ratio tests (LRT) were performed in order to estimate the importance of each variable to the model outcome. For each variable, the full model is compared to a model with every variable except one and the ratio of the likelihoods of both the full and reduced model is used to estimate whether the variable significantly contributes to the model [39]. These results are presented in Table 2. Along with DHD2015-index, other important variables in the model were age, number of cigarettes smoked per day, hemoglobin, HbA1c leukocyte levels, HDL cholesterol, diastolic and systolic BP, pulse rate, gender, living alone, and feeling tired during the previous 4 weeks. HRs for all-cause mortality increased as the values for age, number of cigarettes smoked per day, leukocyte and systolic blood pressure levels increased. In contrast, strong negative associations were found for HbA1c and hemoglobin. Variables with U-shaped relationships to mortality included HDL cholesterol, diastolic blood pressure, and pulse rate. Since these variables were modelled with splines and a simple interpretation of their HRs is not available, plots of these hazard ratios can be seen in S4 Fig in S1 File. For the categorical variables, females had a HR of 0.74 (95% CI = 0.61–0.89) compared to males, living alone had a HR of 0.80 (95% CI = 0.69–0.94) compared to not living alone, and “Never”, "Sometimes” or “Rarely” feeling tired in the previous 4 weeks was associated with a higher HR than those feeling tired less often, with those reporting feeling tired “All of the time”, “Most of the time” or “Often” having the lowest HRs (S4 Fig in S1 File).
Discussion
In a sample of almost 100 000 individuals from the north of the Netherlands, a strong inverse relationship was found between DHD2015-index scores and all-cause mortality [4]. HRs were approximately 1.2x higher than the mean for all DHD2015-index scores below 60 (Fig 4). suggesting that the effects of poor diet as measured by the DHD2015-index on mortality are comparable until this threshold. However, those with the lowest DHD2015-index scores tended to be younger and therefore had a lower risk of death since age was the most influential predictor of all-cause mortality, which may somewhat mask the true harmful effects expected from a low DHD2015-index score. The HR plot of DHD2015-index score was then characterized by two approximately linear lines between 60 and 90, where HR fell from ~1.1 to ~0.81 (a decrease in HR of 0.01 per 1 unit increase in DHD2015-index score), and between 90 and the highest DHD2015-index scores, where HR decreased from ~0.81 to ~0.54 (a decrease in HR of 0.007 per 1 unit increase in DHD2015-index score, This would suggest slightly diminishing returns once DHD2015-index scores reach around 90. Nonetheless, these results suggest a strong relationship whereby all-cause mortality is significantly reduced as DHD2015-index score increases.
The health benefits of key dietary components of the DHD2015-index are well established. High fruit and vegetable consumption consistently correlates with a reduced risk of cardiovascular disease and all-cause mortality [40]. These food groups serve as a rich source of essential micronutrients like vitamins A, C, and E, alongside other health-promoting nutrients like fibre and phytochemicals. The mechanisms underlying these beneficial effects are multifaceted. Dietary fibre and vegetable intake lowers low-density lipoprotein (LDL) cholesterol [41] and blood pressure [42], both established biomarkers for cardiovascular disease [43,44]. Furthermore, dietary fibre intake has also been linked to improved vascular [42] and immune function [45] Additionally, the antioxidants in fruits and vegetables may act as scavengers of reactive oxygen species (ROS), thereby mitigating cellular oxidation and lowering oxidative stress [46]. High consumption levels of these foods (vegetable and fruit) in addition to wholegrains and legumes are well-known dietary intervention targets for weight loss and maintenance [47,48], a separate, well-documented causal variable for cardiovascular disease [49], type 2 diabetes [50] and all-cause mortality [49].
After age, which had the largest effect on all-cause mortality, number of cigarettes smoked per day, gender, systolic and diastolic BP, and heart rate were also significantly related to all-cause mortality, as reported elsewhere [51–58]. Important plasma predictors were plasma leukocyte concentration, which had the highest Chi-Square of the other variables after age on the LRT, hemoglobin, and HDL cholesterol, all of which have established relationships with all-cause mortality [59–70]. Important self-reported variables included living with others (HR = 0.80) and “How much time feeling tired during the past 4 weeks”. Loneliness has been suggested to negatively influence metabolic health and is associated with poorer lifestyle habits [71–73]. Somewhat counterintuitively, “Never” feeling tired in the past 4 weeks had the highest HR, with “Always” having the lowest. The most likely explanation for this is that these levels were also associated with age, with the level “Never” having the oldest average age before decreasing at each level until “Always”, which had the youngest average age (S5 Fig in S1 File). This most likely reveals differences in how younger versus older people report their subjective tiredness, similar to how older people, despite generally having more health problems, more positively evaluate their own health compared to younger people [74].
Limitations
Some important limitations of the present study should also be noted. Pairwise differences were hardly present in the categorical variables, a reason for which could be the variability of results depending on the selection of the baseline factor. The reference is selected automatically by the R program and is the first level of the factor by default, however, this can be problematic in our case since participants may avoid entering extreme results on categorical data collected using a Likert scale with a middle point [75–78]. Thus, it is possible that the statistical significance of some of the factor levels would be different had another baseline reference been selected. This was not investigated thoroughly since this was not our primary goal, however, no meaningful findings were seen when the factor level was changed for some variables.
It has been argued that the use of splines can lead to overfitting of the data, which may limit the generalizability of our results [29]. Firstly, the Lifelines dataset is considered nationally representative of the Dutch population [79], and our final dataset was large with a sample size of almost 100 000. Despite this, we took steps to limit the possibility of overfitting, including limiting the degrees of freedom with respect to the number of events in the model [28,80,81], limiting the degrees of freedom on splined variables and assuming linearity for certain variables where nonlinearity was insignificant or not meaningful (see supplementary materials for details on the modeling procedure). However, whilst overfitting the data is undesirable, so too is underfitting; given that splines do not make assumptions about the nature of the data, they represent an appropriate way to model complex relationships to a high degree of accuracy, and thus provide a more accurate model. Finally, whilst Cox regression is a well-established approach for answering our research question, our approach was not compared against other alternatives such as modern machine learning techniques [82].
Conclusion
Adherence to the Dutch dietary guidelines as reflected by the DHD2015-index, is a strong independent predictor of all-cause mortality in the healthy Dutch population. Future work could investigate whether public health nutritional interventions that lead to an increase in diet quality, reflected as an increase in DHD2015-index scores, also reduce mortality.
Supporting information
S1 File. Supporting information for the manuscript “association between diet quality and all-cause mortality in a large Dutch cohort”.
https://doi.org/10.1371/journal.pone.0302905.s001
(DOCX)
Acknowledgments
The authors are grateful for the team at Lifelines for their excellent support and patience. The data used in this study is only available through the Lifelines service and is not publicly available.
References
- 1. Herforth A, Arimond M, Álvarez-Sánchez C, Coates J, Christianson K, Muehlhoff E. A Global Review of Food-Based Dietary Guidelines. Adv Nutr [Internet]. 2019 Jul 1 [cited 2023 Dec 18];10(4):590–605. Available from: https://pubmed.ncbi.nlm.nih.gov/31041447/. pmid:31041447
- 2. Kromhout D, Spaaij CJK, De Goede J, Weggemans RM, Brug J, Geleijnse JM, et al. The 2015 Dutch food-based dietary guidelines. Eur J Clin Nutr [Internet]. 2016 Aug 1 [cited 2023 Dec 18];70(8):869–78. Available from: https://pubmed.ncbi.nlm.nih.gov/27049034/. pmid:27049034
- 3. Chooi YC, Ding C, Magkos F. The epidemiology of obesity. Metabolism [Internet]. 2019 Mar 1 [cited 2023 Dec 18]; 92:6–10. Available from: https://pubmed.ncbi.nlm.nih.gov/30253139/. pmid:30253139
- 4. Looman M, Feskens EJM, De Rijk M, Meijboom S, Biesbroek S, Temme EHM, et al. Development and evaluation of the Dutch Healthy Diet index 2015. Public Health Nutr [Internet]. 2017 Sep 1 [cited 2023 Dec 18];20(13):2289–99. Available from: https://pubmed.ncbi.nlm.nih.gov/28625202/. pmid:28625202
- 5. Van Lee L, Geelen A, Kiefte-De Jong JC, Witteman JCM, Hofman A, Vonk N, et al. Adherence to the Dutch dietary guidelines is inversely associated with 20-year mortality in a large prospective cohort study. European Journal of Clinical Nutrition 2016 70:2 [Internet]. 2015 Oct 21 [cited 2023 Dec 18];70(2):262–8. Available from: https://www.nature.com/articles/ejcn2015163.
- 6. Fransen HP, Beulens JWJ, May AM, Struijk EA, Boer JMA, de Wit GA, et al. Dietary patterns in relation to quality-adjusted life years in the EPIC-NL cohort. Prev Med (Baltim) [Internet]. 2015 Aug 1 [cited 2023 Dec 18]; 77:119–24. Available from: https://pubmed.ncbi.nlm.nih.gov/26007298/. pmid:26007298
- 7. Sijtsma A, Rienks J, Van Der Harst P, Navis G, Rosmalen JGM, Dotinga A. Cohort Profile Update: Lifelines, a three-generation cohort study and biobank. Int J Epidemiol [Internet]. 2022 Oct 13 [cited 2024 Mar 3];51(5):e295–302. Available from: pmid:34897450
- 8. Scholtens S, Smidt N, Swertz MA, Bakker SJL, Dotinga A, Vonk JM, et al. Cohort Profile: LifeLines, a three-generation cohort study and biobank. Int J Epidemiol [Internet]. 2015 Aug 1 [cited 2023 Dec 18];44(4):1172–80. Available from: pmid:25502107
- 9. Baart AM, Brouwer-Brolsma EM, de Jong HBT, de Vries JHM, Feskens EJM. Assessment of the Dutch Healthy Diet index 2015 in the Lifelines cohort study at baseline. European Journal of Clinical Nutrition 2023 [Internet]. 2023 Nov 28 [cited 2023 Dec 18];1–11. Available from: https://www.nature.com/articles/s41430-023-01372-x. pmid:38017142
- 10. De Rijk MG, Slotegraaf AI, Brouwer-Brolsma EM, Perenboom CWM, Feskens EJM, De Vries JHM. Development and evaluation of a diet quality screener to assess adherence to the Dutch food-based dietary guidelines. Br J Nutr [Internet]. 2021 Oct 28 [cited 2024 Mar 23];128(8):1615–25. Available from: https://pubmed.ncbi.nlm.nih.gov/34776025/.
- 11. Brouwer-Brolsma EM, Perenboom C, Sluik D, Van De Wiel A, Geelen A, Feskens EJM, et al. Development and external validation of the “Flower-FFQ”: a FFQ designed for the Lifelines Cohort Study. Public Health Nutr [Internet]. 2022 Feb 14 [cited 2024 Mar 6];25(2):225–36. Available from: https://pubmed.ncbi.nlm.nih.gov/33988111/. pmid:33988111
- 12. Baart AM, Brouwer-brolsma EM, Perenboom CWM, de Vries JHM, Feskens EJM. Dietary Intake in the Lifelines Cohort Study: Baseline Results from the Flower Food Frequency Questionnaire among 59,982 Participants. Nutrients [Internet]. 2021 Jan 1 [cited 2024 Mar 6];14(1). Available from: https://pubmed.ncbi.nlm.nih.gov/35010922/.
- 13. Baart AM, Brouwer-Brolsma EM, de Jong HBT, de Vries JHM, Feskens EJM. Assessment of the Dutch Healthy Diet index 2015 in the Lifelines cohort study at baseline. Eur J Clin Nutr [Internet]. 2023 [cited 2024 Mar 6]; Available from: https://pubmed.ncbi.nlm.nih.gov/38017142/. pmid:38017142
- 14. Gellert C, Schöttker B, Brenner H. Smoking and all-cause mortality in older people: systematic review and meta-analysis. Arch Intern Med [Internet]. 2012 Jun 11 [cited 2024 Mar 23];172(11):837–44. Available from: https://pubmed.ncbi.nlm.nih.gov/22688992/. pmid:22688992
- 15. Löllgen H, Böckenhoff A, Knapp G. Physical activity and all-cause mortality: an updated meta-analysis with different intensity categories. Int J Sports Med [Internet]. 2009 Mar [cited 2024 Mar 23];30(3):213–24. Available from: https://pubmed.ncbi.nlm.nih.gov/19199202/. pmid:19199202
- 16. Mao L, Yin R, Cai J, Niu M, Xu L, Sui W, et al. The Relationship Between Successful Aging and All-Cause Mortality Risk in Older Adults: A Systematic Review and Meta-Analysis of Cohort Studies. Front Med (Lausanne) [Internet]. 2022 Feb 9 [cited 2024 Mar 23];8. Available from: https://pubmed.ncbi.nlm.nih.gov/35223877/. pmid:35223877
- 17. Visaria A, Setoguchi S. Body mass index and all-cause mortality in a 21st century U.S. population: A National Health Interview Survey analysis. PLoS One [Internet]. 2023 Jul 1 [cited 2024 Mar 23];18(7): e0287218. Available from: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0287218. pmid:37405977
- 18. Li S, Zhang W, Liu H. Association between lipid levels and all-cause and cause-specific mortality in critically ill patients. Scientific Reports 2023 13:1 [Internet]. 2023 Mar 29 [cited 2024 Mar 23];13(1):1–9. Available from: https://www.nature.com/articles/s41598-023-32209-z.
- 19. Arnold LW, Wang Z. The HbA1c and all-cause mortality relationship in patients with type 2 diabetes is J-shaped: a meta-analysis of observational studies. Rev Diabet Stud [Internet]. 2014 [cited 2024 Mar 23];11(2):138–52. Available from: https://pubmed.ncbi.nlm.nih.gov/25396402/. pmid:25396402
- 20. Martín-María N, Miret M, Caballero FF, Rico-Uribe LA, Steptoe A, Chatterji S, et al. The Impact of Subjective Well-being on Mortality: A Meta-Analysis of Longitudinal Studies in the General Population. Psychosom Med [Internet]. 2017 [cited 2024 Mar 23];79(5):565–75. Available from: https://pubmed.ncbi.nlm.nih.gov/28033196/. pmid:28033196
- 21. Vart P, Gansevoort RT, Coresh J, Reijneveld SA, Bültmann U. Socioeconomic measures and CKD in the United States and The Netherlands. Clin J Am Soc Nephrol [Internet]. 2013 Oct 7 [cited 2023 Dec 18];8(10):1685–93. Available from: https://pubmed.ncbi.nlm.nih.gov/23813554/. pmid:23813554
- 22. De Hollander EL, Zwart L, De Vries SI, Wendel-Vos W. The SQUASH was a more valid tool than the OBiN for categorizing adults according to the Dutch physical activity and the combined guideline. J Clin Epidemiol [Internet]. 2012 Jan [cited 2023 Dec 18];65(1):73–81. Available from: https://pubmed.ncbi.nlm.nih.gov/21840174/ pmid:21840174
- 23. Cox DR. Regression Models and Life-Tables. Journal of the Royal Statistical Society: Series B (Methodological) [Internet]. 1972 Jan 1 [cited 2023 Dec 18];34(2):187–202. Available from: https://onlinelibrary.wiley.com/doi/full/10.1111/j.2517-6161.1972.tb00899.x.
- 24. Cook RD. Influential Observations in Linear Regression. J Am Stat Assoc. 1979 Mar;74(365):169.
- 25. Belsley D a, Kuh E, Welsch RE. Identifying influential data and sources of collinearity [Internet]. 1980 [cited 2024 Mar 1]. 6–84 p. Available from: https://onlinelibrary.wiley.com/doi/book/10.1002/0471725153.
- 26. Lin DY, Wei LJ, Ying Z. Checking the Cox Model with Cumulative Sums of Martingale-Based Residuals. Biometrika. 1993 Sep;80(3):557.
- 27.
Therneau TM, Grambsch PM. Modeling Survival Data: Extending the Cox Model [Internet]. New York, NY: Springer New York; 2000 [cited 2023 Dec 18]. (Statistics for Biology and Health). Available from: http://link.springer.com/10.1007/978-1-4757-3294-8.
- 28.
Harrell FE. Regression Modeling Strategies [Internet]. Cham: Springer International Publishing; 2015 [cited 2023 Dec 18]. (Springer Series in Statistics). Available from: https://link.springer.com/10.1007/978-3-319-19425-7.
- 29. Perperoglou A, Sauerbrei W, Abrahamowicz M, Schmid M. A review of spline function procedures in R. BMC Med Res Methodol [Internet]. 2019 Mar 6 [cited 2023 Dec 18];19(1):1–16. Available from: https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/s12874-019-0666-3.
- 30.
Eilers PHC, Marx BD. Flexible smoothing with B-splines and penalties. https://doi.org/101214/ss/1038425655 [Internet]. 1996 May 1 [cited 2024 Mar 1];11(2):89–121. Available from: https://projecteuclid.org/journals/statistical-science/volume-11/issue-2/Flexible-smoothing-with-B-splines-and-penalties/10.1214/ss/1038425655.full.
- 31. Therneau T. A Package for Survival Analysis in R [Internet]. 2024 [cited 2024 Mar 1]. Available from: https://cran.r-project.org/web/packages/survival/vignettes/survival.pdf.
- 32. Hess KR. Graphical methods for assessing violations of the proportional hazards assumption in Cox regression. Stat Med [Internet]. 1995 [cited 2024 Mar 3];14(15):1707–23. Available from: https://pubmed.ncbi.nlm.nih.gov/7481205/. pmid:7481205
- 33. Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika [Internet]. 1994 Sep 1 [cited 2024 Mar 3];81(3):515–26. Available from: https://dx.doi.org/10.1093/biomet/81.3.515.
- 34. Schoenfeld D. Partial residuals for the proportional hazards regression model. Biometrika [Internet]. 1982 Apr 1 [cited 2024 Mar 3];69(1):239–41. Available from: https://dx.doi.org/10.1093/biomet/69.1.239.
- 35. Zhang Z, Reinikainen J, Adeleke KA, Pieterse ME, Groothuis-Oudshoorn CGM. Time-varying covariates and coefficients in Cox regression models. Ann Transl Med [Internet]. 2018 Apr [cited 2024 Mar 3];6(7):121–121. Available from: /pmc/articles/PMC6015946/. pmid:29955581
- 36. Therneau T, Crowson C, Atkinson E. Using Time Dependent Covariates and Time Dependent Coecients in the Cox Model. 2023.
- 37. Hess KR. Assessing time-by-covariate interactions in proportional hazards regression models using cubic spline functions. Stat Med [Internet]. 1994 [cited 2024 Mar 3];13(10):1045–62. Available from: https://pubmed.ncbi.nlm.nih.gov/8073200/. pmid:8073200
- 38. Keele L. Proportionally Difficult: Testing for Nonproportional Hazards in Cox Models. Political Analysis [Internet]. 2010 Jan 25 [cited 2024 Mar 3];18(2):189–205. Available from: https://www.cambridge.org/core/journals/political-analysis/article/abs/proportionally-difficult-testing-for-nonproportional-hazards-in-cox-models/A42DA31E4F09413718AF5425C6527797.
- 39. Vuong QH. Likelihood Ratio Tests for Model Selection and Non-Nested Hypotheses. Econometrica. 1989 Mar;57(2):307.
- 40. Wang X, Ouyang Y, Liu J, Zhu M, Zhao G, Bao W, et al. Fruit and vegetable consumption and mortality from all causes, cardiovascular disease, and cancer: systematic review and dose-response meta-analysis of prospective cohort studies. The BMJ [Internet]. 2014 Jul 29 [cited 2024 Mar 23];349. Available from: /pmc/articles/PMC4115152/. pmid:25073782
- 41. Soliman GA. Dietary Fiber, Atherosclerosis, and Cardiovascular Disease. Nutrients [Internet]. 2019 May 1 [cited 2024 Mar 23];11(5). Available from: /pmc/articles/PMC6566984/. pmid:31126110
- 42. Reynolds AN, Akerman A, Kumar S, Diep Pham HT, Coffey S, Mann J. Dietary fibre in hypertension and cardiovascular disease management: systematic review and meta-analyses. BMC Med [Internet]. 2022 Dec 1 [cited 2024 Mar 23];20(1):1–9. Available from: https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-022-02328-x.
- 43. Tardif JC, Heinonen T, Orloff D, Libby P. Vascular Biomarkers and Surrogates in Cardiovascular Disease. Circulation [Internet]. 2006 Jun 27 [cited 2024 Mar 23];113(25):2936–42. Available from: https://www.ahajournals.org/doi/abs/10.1161/circulationaha.105.598987. pmid:16801474
- 44. Kirk D, Costeira R, Visconti A, Khan Mirzaei M, Deng L, Valdes AM, et al. Bacteriophages, gut bacteria, and microbial pathways interplay in cardiometabolic health. Cell Rep [Internet]. 2024 Feb 27 [cited 2024 Mar 23];43(2). Available from: https://pubmed.ncbi.nlm.nih.gov/38300802/. pmid:38300802
- 45. PD SCJ F. The immune-enhancing effects of dietary fibres and prebiotics. Br J Nutr [Internet]. 2002 May 1 [cited 2024 Mar 23];87 Suppl 2(6):221–30. Available from: https://pubmed.ncbi.nlm.nih.gov/12088522/. pmid:12088522
- 46. Liu Z, Ren Z, Zhang J, Chuang CC, Kandaswamy E, Zhou T, et al. Role of ROS and Nutritional Antioxidants in Human Diseases. Front Physiol [Internet]. 2018 May 17 [cited 2024 Mar 23];9(MAY). Available from: https://pubmed.ncbi.nlm.nih.gov/29867535/. pmid:29867535
- 47. Mytton OT, Nnoaham K, Eyles H, Scarborough P, Ni Mhurchu C. Systematic review and meta-analysis of the effect of increased vegetable and fruit consumption on body weight and energy intake. BMC Public Health [Internet]. 2014 Aug 28 [cited 2024 Mar 23];14(1):1–11. Available from: https://bmcpublichealth.biomedcentral.com/articles/10.1186/1471-2458-14-886. pmid:25168465
- 48. Whigham LD, Valentine AR, Johnson LK, Zhang Z, Atkinson RL, Tanumihardjo SA. Increased vegetable and fruit consumption during weight loss effort correlates with increased weight and fat loss. Nutrition & Diabetes 2012 2:10 [Internet]. 2012 Oct 1 [cited 2024 Mar 23];2(10):e48–e48. Available from: https://www.nature.com/articles/nutd201222. pmid:23449500
- 49. Strelitz J, Lawlor ER, Wu Y, Estlin A, Nandakumar G, Ahern AL, et al. Association between weight change and incidence of cardiovascular disease events and mortality among adults with type 2 diabetes: a systematic review of observational studies and behavioural intervention trials. Diabetologia [Internet]. 2022 Mar 1 [cited 2024 Mar 23];65(3):424–39. Available from: https://link.springer.com/article/10.1007/s00125-021-05605-1. pmid:34859263
- 50. Chen Y, He D, Yang T, Zhou H, Xiang S, Shen L, et al. Relationship between body composition indicators and risk of type 2 diabetes mellitus in Chinese adults. BMC Public Health [Internet]. 2020 Apr 6 [cited 2024 Mar 23];20(1):1–6. Available from: https://bmcpublichealth.biomedcentral.com/articles/10.1186/s12889-020-08552-5.
- 51. Brunström M, Carlberg B. Association of Blood Pressure Lowering With Mortality and Cardiovascular Disease Across Blood Pressure Levels: A Systematic Review and Meta-analysis. JAMA Intern Med [Internet]. 2018 Jan 1 [cited 2023 Dec 18];178(1):28–36. Available from: https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2663255. pmid:29131895
- 52. Hansen TW, Jeppesen J, Rasmussen S, Ibsen H, Torp-Pedersen C. Ambulatory Blood Pressure and Mortality. Hypertension [Internet]. 2005 Apr 1 [cited 2023 Dec 18];45(4):499–504. Available from: https://www.ahajournals.org/doi/abs/10.1161/01.hyp.0000160402.39597.3b.
- 53. van den Hoogen PCW, Feskens EJM, Nagelkerke NJD, Menotti A, Nissinen A, Kromhout D. The relation between blood pressure and mortality due to coronary heart disease among men in different parts of the world. Seven Countries Study Research Group. N Engl J Med [Internet]. 2000 Jan 6 [cited 2023 Dec 18];342(1):1–8. Available from: https://pubmed.ncbi.nlm.nih.gov/10620642/. pmid:10620642
- 54. Jensen MT, Suadicani P, Hein HO, Gyntelberg F. Elevated resting heart rate, physical fitness and all-cause mortality: a 16-year follow-up in the Copenhagen Male Study. Heart [Internet]. 2013 Jun [cited 2023 Dec 18];99(12):882–7. Available from: https://pubmed.ncbi.nlm.nih.gov/23595657/. pmid:23595657
- 55. Münzel T, Hahad O, Gori T, Hollmann S, Arnold N, Prochaska JH, et al. Heart rate, mortality, and the relation with clinical and subclinical cardiovascular diseases: results from the Gutenberg Health Study. Clinical Research in Cardiology [Internet]. 2019 Dec 1 [cited 2023 Dec 18];108(12):1313–23. Available from: https://link.springer.com/article/10.1007/s00392-019-01466-2. pmid:30953178
- 56. Zhang D, Shen X, Qi X. Resting heart rate and all-cause and cardiovascular mortality in the general population: a meta-analysis. CMAJ [Internet]. 2016 Feb 16 [cited 2023 Dec 18];188(3): E53–63. Available from: https://pubmed.ncbi.nlm.nih.gov/26598376/. pmid:26598376
- 57. Leon DA. Trends in European life expectancy: a salutary view. Int J Epidemiol [Internet]. 2011 Apr 1 [cited 2023 Dec 18];40(2):271–7. Available from: https://dx.doi.org/10.1093/ije/dyr061. pmid:21415000
- 58. MacKenbach JP, Slobbe L, Looman CWN, Van Der Heide A, Polder J, Garssen J. Sharp upturn of life expectancy in the Netherlands: Effect of more health care for the elderly? Eur J Epidemiol [Internet]. 2011 Dec 30 [cited 2023 Dec 18];26(12):903–14. Available from: https://link.springer.com/article/10.1007/s10654-011-9633-y. pmid:22127495
- 59. Abete I, Lu Y, Lassale C, Verschuren M, Van Der Schouw Y, Bueno-De-Mesquita B. White cell counts in relation to mortality in a general population of cohort study in the Netherlands: a mediating effect or not? BMJ Open [Internet]. 2019 Oct 1 [cited 2023 Dec 18];9(10):e030949. Available from: https://bmjopen.bmj.com/content/9/10/e030949. pmid:31666267
- 60. Asadollahi K, Beeching NJ, Gill G V. Leukocytosis as a predictor for non-infective mortality and morbidity. QJM: An International Journal of Medicine [Internet]. 2010 May 1 [cited 2023 Dec 18];103(5):285–92. Available from: https://dx.doi.org/10.1093/qjmed/hcp182. pmid:20056764
- 61. Kabat GC, Kim MY, Manson JAE, Lessin L, Lin J, Wassertheil-Smoller S, et al. White Blood Cell Count and Total and Cause-Specific Mortality in the Women’s Health Initiative. Am J Epidemiol [Internet]. 2017 Jul 1 [cited 2023 Dec 18];186(1):63–72. Available from: pmid:28369251
- 62. De Labry LO, Campion EW, Glynn RJ, Vokonas PS. White blood cell count as a predictor of mortality: Results over 18 years from the normative aging study. J Clin Epidemiol. 1990 Jan 1;43(2):153–7. pmid:2303845
- 63. Ruggiero C, Metter EJ, Cherubini A, Maggio M, Sen R, Najjar SS, et al. White blood cell count and mortality in the Baltimore Longitudinal Study of Aging. J Am Coll Cardiol [Internet]. 2007 May 8 [cited 2023 Dec 18];49(18):1841–50. Available from: https://pubmed.ncbi.nlm.nih.gov/17481443/. pmid:17481443
- 64. Shankar A, Mitchell P, Rochtchina E, Wang JJ. The association between circulating white blood cell count, triglyceride level and cardiovascular and all-cause mortality: population-based cohort study. Atherosclerosis [Internet]. 2007 May [cited 2023 Dec 18];192(1):177–83. Available from: https://pubmed.ncbi.nlm.nih.gov/16730736/. pmid:16730736
- 65. Ali KM, Wonnerth A, Huber K, Wojta J. Cardiovascular disease risk reduction by raising HDL cholesterol—current therapies and future opportunities. Br J Pharmacol [Internet]. 2012 Nov [cited 2023 Dec 18];167(6):1177–94. Available from: https://pubmed.ncbi.nlm.nih.gov/22725625/. pmid:22725625
- 66. Briel M, Ferreira-Gonzalez I, You JJ, Karanicolas PJ, Akl EA, Wu P, et al. Association between change in high density lipoprotein cholesterol and cardiovascular disease morbidity and mortality: systematic review and meta-regression analysis. BMJ [Internet]. 2009 Feb 28 [cited 2023 Dec 18];338(7693). Available from: https://pubmed.ncbi.nlm.nih.gov/19221140/. pmid:19221140
- 67. Jacobs DR, Mebane IL, Bangdiwala SI, Criqui MH, Tyroler HA. High density lipoprotein cholesterol as a predictor of cardiovascular disease mortality in men and women: the follow-up study of the Lipid Research Clinics Prevalence Study. Am J Epidemiol [Internet]. 1990 [cited 2023 Dec 18];131(1):32–47. Available from: https://pubmed.ncbi.nlm.nih.gov/2293751/. pmid:2293751
- 68. Yi SW, Park SJ, Yi JJ, Ohrr H, Kim H. High-density lipoprotein cholesterol and all-cause mortality by sex and age: a prospective cohort study among 15.8 million adults. Int J Epidemiol [Internet]. 2021 Jul 9 [cited 2023 Dec 18];50(3):902–13. Available from: pmid:33313654
- 69. Li X, Guan B, Wang Y, Tse G, Zou F, Khalid BW, et al. Association between high-density lipoprotein cholesterol and all-cause mortality in the general population of northern China. Sci Rep [Internet]. 2019 Dec 1 [cited 2023 Dec 18];9(1). Available from: https://pubmed.ncbi.nlm.nih.gov/31594968/. pmid:31594968
- 70. MacMahon S, Duffy S, Rodgers A, Tominaga S, Chambless L, De Backer G, et al. Blood cholesterol and vascular mortality by age, sex, and blood pressure: a meta-analysis of individual data from 61 prospective studies with 55,000 vascular deaths. Lancet [Internet]. 2007 Dec 1 [cited 2023 Dec 18];370(9602):1829–39. Available from: https://pubmed.ncbi.nlm.nih.gov/18061058/. pmid:18061058
- 71. Cacioppo JT, Hawkley LC, Crawford LE, Ernst JM, Burleson MH, Kowalewski RB, et al. Loneliness and health: potential mechanisms. Psychosom Med [Internet]. 2002 [cited 2023 Dec 18];64(3):407–17. Available from: https://pubmed.ncbi.nlm.nih.gov/12021415/. pmid:12021415
- 72. Luo Y, Hawkley LC, Waite LJ, Cacioppo JT. Loneliness, health, and mortality in old age: a national longitudinal study. Soc Sci Med [Internet]. 2012 Mar [cited 2023 Dec 18];74(6):907–14. Available from: https://pubmed.ncbi.nlm.nih.gov/22326307/. pmid:22326307
- 73. Steptoe A, Shankar A, Demakakos P, Wardle J. Social isolation, loneliness, and all-cause mortality in older men and women. Proc Natl Acad Sci U S A [Internet]. 2013 Apr 9 [cited 2023 Dec 18];110(15):5797–801. Available from: https://pubmed.ncbi.nlm.nih.gov/23530191/. pmid:23530191
- 74. Jylhä M. What is self-rated health and why does it predict mortality? Towards a unified conceptual model. Soc Sci Med [Internet]. 2009 Aug [cited 2023 Dec 18];69(3):307–16. Available from: https://pubmed.ncbi.nlm.nih.gov/19520474/. pmid:19520474
- 75.
Albaum G. The Likert Scale Revisited. https://doi.org/101177/147078539703900202 [Internet]. 1997 Mar 1 [cited 2023 Dec 18];39(2):1–21. Available from: https://journals.sagepub.com/doi/10.1177/147078539703900202.
- 76. Chomeya R. Quality of Psychology Test Between Likert Scale 5 and 6 Points. Journal of Social Sciences [Internet]. 2010 Sep 4 [cited 2023 Dec 18];6(3):399–403. Available from: https://thescipub.com/abstract/jssp.2010.399.403.
- 77. Hasson D, Arnetz B. Validation and Findings Comparing VAS vs. Likert Scales for Psychosocial Measurements. Int Electron J Health Educ. 2005.
- 78. Chyung SY Yonnie , Roberts K, Swanson I, Hankinson A. Evidence-Based Survey Design: The Use of a Midpoint on the Likert Scale. Performance Improvement [Internet]. 2017 Nov 1 [cited 2023 Dec 18];56(10):15–23. Available from: https://onlinelibrary.wiley.com/doi/full/10.1002/pfi.21727.
- 79. Klijs B, Scholtens S, Mandemakers JJ, Snieder H, Stolk RP, Smidt N. Representativeness of the LifeLines Cohort Study. PLoS One [Internet]. 2015 Sep 2 [cited 2023 Dec 18];10(9). Available from: https://pubmed.ncbi.nlm.nih.gov/26333164/. pmid:26333164
- 80. Ogundimu EO, Altman DG, Collins GS. Adequate sample size for developing prediction models is not simply related to events per variable. J Clin Epidemiol [Internet]. 2016 Aug 1 [cited 2023 Dec 18]; 76:175–82. Available from: https://pubmed.ncbi.nlm.nih.gov/26964707/. pmid:26964707
- 81. Vittinghoff E, McCulloch CE. Relaxing the Rule of Ten Events per Variable in Logistic and Cox Regression. Am J Epidemiol [Internet]. 2007 Mar 15 [cited 2023 Dec 18];165(6):710–8. Available from: pmid:17182981
- 82. Kirk D, Kok E, Tufano M, Tekinerdogan B, Feskens EJM, Camps G. Machine Learning in Nutrition Research. Advances in Nutrition [Internet]. 2022 Nov 1 [cited 2024 Mar 6];13(6):2573. Available from: /pmc/articles/PMC9776646/. pmid:36166846