Can nutritional scores improve the WHO calibrated non-laboratory risk prediction model for cardiovascular disease? Golestan Cohort Study

Abstract

Purpose

Evaluation of the added value of Dietary Approaches to Stop Hypertension (DASH) and Mediterranean diet scores on the prediction model of the World Health Organization (WHO) to predict 10-year cardiovascular disease (CVD) mortality using the Golestan Cohort Study data.

Methods

A total of 44,648 participants (25,268 women and 18,531 men) were included in the final analysis. To assess the external validity of the non-laboratory risk model of WHO, the Area Under the Curve (AUC) and calibration plot methods were used. The multivariate Cox proportional hazards regression analysis was used to evaluate the association of 10-year CVD mortality risk with DASH and Mediterranean scores and their components. The added value of each significant variables was evaluated by the concordance C-statistic and integrated discrimination improvement (IDI). Statistical significance was defined as p-value < 0.05.

Results

DASH and Mediterranean diet scores were not significant predictors of 10-year CVD mortality in both genders (p > 0.05). However, sodium and total vegetable in both genders and added sugar in women were significant predictors for 10-year stroke mortality (p < 0.05). Sodium intake in women and monounsaturated fatty acid (MUFA) to saturated fatty acid (SFA) ratio in men had significant associations with 10-year mortality of myocardial infarction/coronary heart disease (MI/CHD). Calculation of IDI showed that none of the evaluated nutritional indices/variables could significantly improve the WHO model performance and predictive ability.

Conclusion

Inclusion of DASH and Mediterranean diet scores and their components did not improve WHO risk prediction model performance and predictive ability to predict 10-year CVD mortality.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40200-024-01463-x.

Introduction

Cardiovascular disease (CVD) is among the primary factors contributing to mortality. According to the World Health Organization (WHO), approximately 17.9 million individuals lose their lives annually due to CVD [1]. CVD in Iran is the foremost cause of mortality accounting for 46% of all deaths and 20–23% of total burden of diseases [2].

Prediction models have emerged as useful tools in the prevention of CVD. These models can estimate prognosis, offering objective information on specific outcome probabilities. Their proper utilization can contribute to informed decision-making and enhanced patient outcomes [3, 4]. To facilitate the expansion of efforts in CVD prevention and control, the WHO has developed tools and guidance. These resources are designed to address the specific challenges and limitations faced by low- and mid-income countries, considering their unique circumstances. By tailoring the tools and guidance to the needs of these countries, the WHO aims to enhance CVD prevention and control efforts, ultimately improving health outcomes in these populations [5].

The dietary research has newly focused on dietary scores/indices/patterns as dietary components are interrelated and consumed in combination to provide a more comprehensive understanding of their impact on health [6, 7]. Recently, there has been a growing utilization of evaluative nutritional indices in identifying individuals and prioritizing preventive interventions for those who are most susceptible to CVD-related health risks. Meanwhile, Dietary Approaches to Stop Hypertension (DASH) score and Mediterranean diet score have been more studied in relation to the risk of CVD outcomes [8–11]. There are several worthy evidence on the protective association of mentioned dietary scores with reduced risk of CVD outcomes and has made them as most popular dietary scores in CVD-related dietary researches [12–14]. In most prospective cohort studies, diet is assessed using standardized self-reported dietary assessment methods. To enhance model performance, many prediction models incorporate laboratory measurements as part of their functions. However, using lab-based prediction models can be expensive and resource-intensive. As a result, non-laboratory prediction models have been developed as an alternative. These models can provide accurate risk prediction without the need for laboratory-based tests, making them more cost-effective and accessible for various healthcare settings [15]. Still, there has been a lack of examination regarding the added values provided by the nutritional indices when used in conjunction with non-laboratory prediction models.

To the best of our knowledge, there is no previous study to evaluate the integration of dietary scores to improve the predictive value of existing prediction models for CVD risk. The current study aimed to examine the added value of DASH and Mediterranean diet scores and their components on the prediction model of WHO in predicting mortality outcomes using the Golestan Cohort Study (GCS) data.

Materials and methods

The golestan cohort study

The design of the GCS was previously described [16]. The GCS, as a part of a series of studies conducted to the investigation of upper gastrointestinal cancer’s etiology in Northern Iran, was launched in January 2004 in Golestan Province located in northeastern Iran. After reaching to the accrual goal of subjects’ number in June 2008, the enrollment was closed. The follow-up process has been done actively for all participants every twelve months. A total of 49,173 adults between the ages of 36 and 81 were recruited from Gonbad City and 326 rural villages, forming a cohort consisting of 20% urban and 80% rural participants. To include apparently healthy people without implausible risk factor levels who are targeted typically in efforts of primary prevention for CVD, exclusion criteria were applied to individuals with age less than 40 and more than 80 years, energy intake less than 600 and more than 4200 kcal, pre-existing cancer, heart disease, stroke, and rheumatic heart disease diagnoses, incomplete responses on the food frequency questionnaire (FFQ) or the general questionnaire (containing information on history of diabetes and hypertension, smoking, and anthropometric data), body mass index (BMI) values less than 16 and more than 60 kg/m², systolic blood pressure (SBP) more than 270 and diastolic blood pressure less than 60 mmHg, undefined reason of death, and follow-up duration of less than 1 year. After applying these criteria, a total of 44,648 participants, comprising 25,268 women and 18,531 men, were included in the final analysis (as depicted in Fig. 1). Ethical approval for the GCS was obtained from the Institutional Review Boards of the Digestive Disease Research Center (DDRC) at Tehran University of Medical Sciences, the US National Cancer Institute (NCI), and the World Health Organization International Agency for Research on Cancer (IARC) [16].

Fig. 1.

The flowchart of the study population

Lifestyle and demographic factors

Trained general physicians and nutritionists conducted interviews with all participants, during which they gathered information on lifestyle and demographic factors using an evaluated reliable questionnaire [16, 17]. Anthropometric measurements, including weight, height, and BMI were recorded according to the guidelines provided by the WHO [16].

Dietary assessment

Participants’ dietary information was collected using a validated FFQ specifically designed for this particular population [18]. The data on the usual portion sizes, consumption frequencies, and servings per day for each food item was gathered during the enrollment period. To calculate the daily intake of each food item, the consumption frequency was multiplied by the typical portion size and the number of servings per day. The participants reported how often they consumed a specific serving of each food item daily, weekly, or monthly for the previous year. The analysis process involved converting the reported consumption frequencies of each food item into daily intakes and converting portion sizes to grams. Total energy intake was then computed by summing up energy intakes of individual food items. The FFQ from the GCS included 158 food items and assessed the usual frequency and portion sizes of dietary intake over the past 12 months. The reliability and validity of the questionnaire were confirmed [18]. Dietary information in GCS were collected by trained personnel using a validated FFQ and photos of portion sizes through detailed face-to-face interviews to ensure that data collection is done with minimal bias. Data collected at recruitment included information on portion sizes, consumption frequencies, and servings consumed for each food item. Consumption frequencies were converted to daily intakes, and portion sizes were converted to grams using household measurement tools [19, 20]. Daily dietary intake was evaluated using the Nutritionist V software and the Iranian Food Composition Table [21].

Follow-up and cause of death ascertainment

Death from CVD was considered as the final endpoint of the study. The follow-up strategies of GCS have been described in more details in a separate publication [16]. Follow-ups were conducted at regular intervals of 12 months. The researchers either contacted the participants by phone or visited their homes to assess the vital status of the participants. In case of any reported deaths, the study group confirmed the information by conducting clinician visits and utilizing a validated verbal autopsy questionnaire [22]. Two external internists independently reviewed all available information, including the verbal autopsy questionnaire, and medical records to determine the cause of death. In situations where there was disagreement between the two external internists regarding the cause of death based on the available information, a third, more experienced internist was involved. This process helped to ensure that the cause of death determination was as accurate and reliable as possible, even in cases where initial assessments by the two specialists differed [16]. For the analysis, only subjects with confirmed deaths were included. To evaluate the calibration of non-laboratory WHO model in the present study, CVD mortality was defined as death from myocardial infarction (MI)/coronary heart disease (CHD) (ICD10 codes I21-I25), or death from stroke (ICD10 codes I60-I69).

The WHO prediction model to estimate CVD risk

The WHO has created new models specifically designed to estimate the risk of CVD in individuals aged 40–80 years across 21 regions based on the Global Burden of Disease [23]. These models include both laboratory-based and non-laboratory-based approaches. The non-laboratory-based risk model, referred to as the WHO model, is particularly suitable for regions with limited resources where blood-based biomarkers like lipid levels are not widely accessible to all individuals. In the context of Iran, the WHO model for East Asia has been recommended for predicting the CVD risk of individuals.

Construction of DASH diet score

To assess the score of DASH-style diet among participants, we developed DASH score based on the foods and nutrients recommended or discouraged in the DASH diet [24]. We focused on eight components: a high intake of fruits, vegetables, nuts and legumes, whole grains, sodium, dairy products, sugar-sweetened beverages and sweets, and red and processed meats. It is important to note that the present study used total dairy intake instead of low-fat dairy consumption because the FFQ used in GCS did not provide information on the type and amount of fat in dairy products. To calculate the DASH score, we categorized participants into quintiles based on their energy-adjusted intakes of desired foods and nutrients. For fruits, vegetables, dairy products, nuts, and legumes, individuals in the highest quintile received a score of 5, indicating a high intake, while those in the lowest quintile received a score of 1, reflecting a low intake. In contrast, for the consumption of red and processed meats, sugar-sweetened beverages and sweets, and sodium, we reversed the scoring system. For this purpose, participants in the highest quintile of consumption received 1 point, while those with the lowest consumption quintile obtained a score of 5. By summing up the scores for each of the eight components, we derived the DASH score for each participant. The lowest possible DASH score (8) was indicating poor adherence to the DASH-style diet, while the highest possible score (40) representing higher adherence [25].

Construction of mediterranean diet score

In the present study, calculation of Mediterranean diet score was done in line with the method introduced by Trichopoulou et al. [26] with a focus on the consumption of eight components including: fruits and nuts, fish, vegetables, legumes, whole grains, the ratio of monounsaturated fatty acids (MUFA) to saturated fatty acids (SFAs), meats (red meat, poultry, and processed meats), and dairy. For each of these items, participants were categorized into quintiles based on their consumption levels. A value of 5 was assigned to the highest quintile, indicating higher consumption, and a value of 1 was assigned to the lowest quintile, indicating lower consumption. A reverse scoring approach was applied for meat and dairy products. For these food groups, a value of 5 was assigned to the lowest quintile of consumption and a value of 1 was assigned to the highest quintile. The final Mediterranean diet score was obtained by summing up these component values. The score ranges from 8 (indicating low adherence to the Mediterranean diet) to 40 (indicating high adherence to the Mediterranean diet).

Statistical analysis

We fitted a Cox proportional hazards regression model to investigate the relationship between CVD mortality risk and dietary scores, including DASH, and Mediterranean diet scores. For each individual, the “time-to-event” was calculated as the interval between the time of entry to the GCS and the time of the following events whichever had happened earlier: death from MI/CHD/Stroke, death from other reasons, or the latest follow-up. Participants who completed the study without experienced the desired events, who did not complete 10-year follow-up for any reasons that were not related to the events of interest, or who died from reasons other than defined CVD outcomes have been considered as censored at their last follow-up time. The analysis was based on a 10-year follow-up period.

The risk of CVD mortality within time ‘t’ was calculated using the general equation:

,

where S0(t) represents the baseline survival function at follow-up time ‘t,’ βi is the estimated regression coefficient for the i-th variable, Xi is the value of the i-th variable, and ‘p’ denotes the number of variables. Age was centered at 60 years, SBP at 120mmHg, and BMI at 25 kg/m2. Sex-specific baseline survival for 10-year stroke and MI/CHD mortality in GCS and WHO study was reported in Supplementary Table 1.

Baseline characteristics of the participants were reported as mean (standard deviation) for continuous variables and as number (%) for categorical variables. For covariates with a skewed distribution, the median (interquartile range: IQR) was provided. The baseline characteristics between men and women were compared using convenient statistical tests, such as the student’s t-test for normally distributed continuous variables, the Mann-Whitney U test for non-normal variables, and the Chi-squared test for categorical variables. To account for our large sample size, statistically significant differences were also assessed using standardized differences unaffected by the sample size. The approach in standardized difference is comparing of difference in means of variables between two groups. This difference is calculated in units of pooled standard deviation of desired variables. Unlike t-test, sample size cannot influence on the standardized difference values. It has been suggested that a standardized difference of 10 per cent (0.1) indicates considerable imbalance in the baseline covariates. Given the low percentage of missing values (< 2%), the missing data were considered to be negligible and complete case analysis was applied.

To assess the external validity of the risk equation, the Area Under the Curve (AUC) and calibration plot methods were used. The AUC was employed to determine the discrimination of the predictor models, with cutoffs defined by Hosmer et al. [27] as follows: AUCs of 0.5–0.7, 0.70–0.80, 0.80–0.90, and ≥ 0.90 indicating poor, acceptable, excellent, and outstanding discrimination, respectively. The bootstrapping with 200 replications was utilized to estimate the uncertainty interval. Additionally, the observed-to-expected ratio (O/E) was calculated for the estimation of calibration, where a ratio < 1 indicated overestimation and a ratio > 1 indicated underestimation of the risk. Recalibration of the risk assessment tool for the GCS characteristics was done by adjusting the intercept of the model while maintaining the same predictors and regression coefficients from the original model [28]. The clinical performance of the validated model was evaluated using the same scoring points defined by The WHO CVD Risk Chart Working Group [23]. The non-laboratory risk score was calculated by summing the risk points over the defined variables for both men and women. The current study also examined whether the addition of DASH and Mediterranean diet scores would improve discrimination or calibration of model. The Harrell’s C index (95% CI) was used to assess the discriminative power of the models. We performed statistical analysis using STATA version 14 (Stata Corp LP, College Station, Texas). In this study, we constructed multivariate Cox models, which were evaluated in terms of model performance, discrimination and predictive ability, and general calibration. The concordance C-statistic and integrated discrimination improvement (IDI) were also calculated [29]. Statistical significance was defined as p-value < 0.05.

Ethical considerations

This project was approved by the research council of National Nutrition & Food Technology Research Institute (NNFTRI), Iran (Ethical code: IR.SBMU.NNFTRI.REC.1399.070).

Results

Baseline characteristics

The flowchart of the study population is shown in Fig. 1. The study population consisted of 18,531 men and 25,268 women at the study baseline with a mean (SD) age of 52.53 (9.17) and 51.34 (8.28) years, respectively. The baseline characteristics of men and women are shown in Table 1. Considering standardized difference, there were significant differences between men and women in case of age, BMI, SBP, and smoking status. The mean age of men was higher than women and they had a higher frequency of being a current smoker, whereas women had a higher level of BMI and SBP. During the median follow-up of 13.16 years (IQR: 12.88–14.25), event-specific incidence densities and ratios among participants was 14.5768 (13.6226–15.5979) for stroke mortality and 19.7257 (18.6104–20.9079) for MI/CHD mortality per 10,000 person-year (Table 2).

Table 1.

Baseline characteristics of participants

Variables	The study groups			P-value	Standardized difference
	Total population (n = 43,799)	Men (n = 18,531)	Women (n = 25,268)
Age (year)	51.85 ± 8.69	52.53 ± 9.17	51.34 ± 8.28	< 0.001a	0.137
BMI (kg/m2)	26.68 ± 5.36	25.10 ± 4.54	27.84 ± 5.62	< 0.001	0.535
SBP (mmHg)	127.23 ± 23.82	125.73 ± 22.71	128.33 ± 24.54	< 0.001	0.110
Current smokers	7371 (16.8)	7029 (37.9)	342 (1.4)	< 0.001b	1.037
Follow-up duration, median (IQR)	13.16 (12.88;14.25)	13.09 (12.09;14.22)	13.90 (12.96;14.26)	< 0.001c	-
DASH Score	23.94 ± 3.48	23.90 ± 3.48	23.97 ± 3.49	0.027	0.021
Mediterranean Score	23.93 ± 4.25	23.90 ± 4.25	23.95 ± 4.26	0.180	0.013

^a Data analysis was done by Independent Sample T test; ^b Data analysis was done by Chi-Square test; ^c Data analysis was done by Mann-Whitney U test. BMI: Body mass index, SBP: Systolic blood pressure, IQR: Interquartile range, DASH: Dietary Approaches to Stop Hypertension

Table 2.

Event-specific incidence densities and ratios among participants per 10,000 person-year in the Golestan Cohort Study, 2004–2021, Iran

Variables	Stroke mortality (CI 95%)	MI/CHD mortality (CI 95%)
Gender
Male	12.1991 (11.0737; 13.4389)	25.7125 (23.7570; 27.8289)
Female	14.5768 (13.6226; 15.5979)	15.4721 (14.1978; 16.8607)
Total	14.5768 (13.6226; 15.5979)	19.7257 (18.6104; 20.9079)

CI: confidence interval; MI: Myocardial infarction; CHD: Coronary heart diseases

Model performance

Risk assessment tool for sex-specific non-laboratory risk models for stroke and MI/CHD mortality developed in the WHO models is presented in supplementary Table 1. In the present study, the WHO risk model showed good discrimination for 10-year stroke mortality with AUC (95% CI) of 0.8233 (0.8019–0.8448) in men and 0.8064 (0.7809–0.8319) in women (Table 3). Also, calibration of model (O/E ratio) was acceptable in men (0.9590) and women (1.1403) (Supplementary files 1). Considering 10-year MI/CHD mortality outcome, calibration was poor in both genders (0.5226 for men and 1.5817 for women). However, discrimination was acceptable in both men and women (Table 3). Therefore, “recalibration in the large” with adjusting the GCS intercept was done to improve the calibration of original model for MI/CHD outcome (Table 3). Although the AUC showed similar discrimination compared to the original model, the O/E ratio was improved for the recalibrated model in both genders (0.9890 for men and 0.9736 for women) (Supplementary files 1).

Table 3.

Model performance for 10-year follow-up: Golestan Cohort Study

Stroke mortality
	Observed to predicted ratio	C-concordance (95% CI) *
Men
Original follow-up 10 years	0.9590	0.8233 (0.8019;0.8448)
Women
Original follow-up 10 years	1.1403	0.8064 (0.7809;0.8319)
MI/CHD mortality
	Observed to predicted ratio	C-concordance (95% CI) *
Men
Original (95% CI) *	0.5226	0.7311 (0.7093;0.7529)
Recalibrated in the large (95% CI) *	0.9890	0.7311 (0.7109;0.7512)
Women
Original (95% CI) *	1.5817	0.7508 (0.7274; 0.7741)
Recalibrated in the large (95% CI) *	0.9736	0.7508 (0.7288;0.7728)

* With (n = 200) Bootstrapping CI: confidence interval; MI: Myocardial infarction; CHD: Coronary heart diseases

Added value of nutritional scores and their components

In the present study, we examined the additional value of the two well-known nutritional scores (DASH and Mediterranean diet scores) and their components on the predictive power of sex-specific calibrated non-laboratory WHO prediction risk model for 10-year stroke and MI/CHD mortality. In the multivariate Cox regression model, DASH diet score was not significant predictor of 10-year mortality of stroke (p = 0.082 and 0.333 in men and women, respectively). Also, Mediterranean diet score were not significant predictor of 10-year stroke mortality in men (p = 0.055) and women (p = 0.799). Non-significant results were also found for DASH diet score in predicting 10-year mortality of MI/CHD in men (p = 0.982) and women (p = 0.702). Regarding Mediterranean diet score as a predictor of 10-year MI/CHD mortality, there was also non-significant association in men (p = 0.323) and women (p = 0.555) (Table 4). However, considering the dietary components of these scores, added sugar in women (p < 0.001), and sodium in men (p = 0.006) and women (p = 0.034), as well as total vegetable in men (p = 0.024) and women (p = 0.001), were significant predictors for 10-year stroke mortality. Considering 10-year mortality of MI/CHD, sodium in women (p = 0.002) and MUFA to SFA ratio in men (p = 0.017) had significant association with the outcome variable (Table 5). Finally, calculation of IDI showed that none of the mentioned nutritional scores and variables could significantly improve the performance and prediction ability of the WHO model (Table 6).

Table 4.

Multivariate Cox regression with 95% confidence interval for 10-year risk of stroke and MI/CHD in associations with calibrated non-laboratory risk model of WHO and dietary scores

Dietary scores	Men	Women
	Hazard ratio (95% confidence interval)	Hazard ratio (95% confidence interval)
Stroke mortality
Total DASH Score	0.9722 (0.9419; 1.0035), p = 0.082	0.9842 (0.9530; 1.0164), p = 0.333
Total Mediterranean Score	0.9750 (0.9502; 1.0005), p = 0.055	0.9965 (0.9706; 1.0232), p = 0.799
MI/CHD mortality
Total DASH Score	1.0003 (0.9748; 1.0264), p = 0.982	1.0054 (0.9780; 1.0335), p = 0.702
Total Mediterranean Score	1.0106 (0.9896; 1.0321), p = 0.323	0.9932 (0.9708; 1.0160), p = 0.555

MI: Myocardial infarction; CHD: Coronary heart disease; DII: Dietary inflammatory index, DASH: Dietary Approaches to Stop Hypertension

Table 5.

Multivariate Cox regression with 95% confidence interval for 10-year risk of sex-specific stroke and MI/CHD mortality in associations with calibrated revised non-laboratory risk model of WHO and components of dietary scores

Components of dietary scores	Men	Women
	Hazard ratio (95% confidence interval)	Hazard ratio (95% confidence interval)
Stroke mortality
Sodium a	0.9999 (0.9998; 0.9999), p = 0.006	0.9999 (0.9998; 0.9999), p = 0.034
Dairy a	0.9996 (0.9989; 1.0003), p = 0.268	0.9992 (0.9983; 1.0001), p = 0.084
Nuts and Beans a	0.9916 (0.9819; 1.0013), p = 0.089	0.9965 (0.9873; 1.0059), p = 0.465
Total vegetable a	0.9985 (0.9972; 0.9998), p = 0.024	0.9977 (0.9963; 0.9991), p = 0.001
Total fruits a	0.9996 (0.9987; 1.0004), p = 0.299	0.9997 (0.9986; 1.0008), p = 0.579
Added sugar a	1.0003 (0.9992; 1.0015), p = 0.579	1.0008 (1.0004; 1.0013), p = 0.000
Meat and processed meat a	0.9979 (0.9913; 1.0047), p = 0.556	0.9998 (0.9916; 1.0081), p = 0.963
Whole grain a	0.9993 (0.9986; 1.0001), p = 0.080	1.0003 (0.9995; 1.0011), p = 0.512
Dairy b	0.9996 (0.9989; 1.0003), p = 0.268	0.9992 (0.9983; 1.0001), p = 0.084
Fruit and Nuts b	0.9996 (0.9988; 1.0004), p = 0.293	0.9997 (0.9986; 1.0007), p = 0.550
Fish b	1.0029 (0.9968; 1.0091), p = 0.347	0.9942 (0.9836; 1.0049), p = 0.291
Beans b	0.9911 (0.9804; 1.0020), p = 0.110	0.9984 (0.9886; 1.0084), p = 0.756
Whole grain b	0.9993 (0.9986; 1.0001), p = 0.080	1.0003 (0.9994; 1.0011), p = 0.512
Total vegetable b	0.9985 (0.9972; 0.9998), p = 0.024	0.9977 (0.9963; 0.9991), p = 0.001
MUFA to SFA ratio b	0.8804 (0.5193; 1.4927), p = 0.636	0.6824 (0.4024; 1.1573), p = 0.156
Meat b	1.0013 (0.9998; 1.0028), p = 0.087	0.9997 (0.9979; 1.0016), p = 0.773
MI/CHD mortality
Sodium a	0.9999 (0.9999; 1.0001), p = 0.962	0.9999 (0.9998; 0.9999), p = 0.002
Dairy a	0.9996 (0.9991; 1.0002), p = 0.236	0.9994 (0.9986; 1.0002), p = 0.148
Nuts and Beans a	1.0007 (0.9941; 1.0074), p = 0.831	1.0039 (0.9971; 1.0108), p = 0.264
Total vegetable a	1.0000 (0.9990; 1.0009), p = 0.979	0.9996 (0.9985; 1.0007), p = 0.452
Total fruits a	0.9996 (0.9989; 1.0002), p = 0.200	0.9997 (0.9988; 1.0006), p = 0.543
Added sugar a	0.9993 (0.9979; 1.0007), p = 0.301	0.9979 (0.9959; 1.000), p = 0.052
Meat and processed meat a	1.0010 (0.9967; 1.0053), p = 0.637	0.9982 (0.9909; 1.0056), p = 0.635
Whole grain a	1.0004 (0.9999; 1.0010), p = 0.135	0.9996 (0.9989; 1.0004), p = 0.331
Dairy b	0.9996 (0.9991; 1.0002), p = 0.236	0.9994 (0.9986; 1.0002), p = 0.148
Fruit and Nuts b	0.9996 (0.9989; 1.0002), p = 0.206	0.9997 (0.9988; 1.0006), p = 0.508
Fish b	1.0016 (0.9966; 1.0067), p = 0.525	0.9959 (0.9873; 1.0046), p = 0.357
Beans b	1.0007 (0.9932; 1.0082), p = 0.849	1.0062 (0.9997; 1.0127), p = 0.060
Whole grain b	1.0004 (0.9999; 1.0010), p = 0.135	0.9996 (0.9989; 1.0004), p = 0.331
Total vegetable b	1.0000 (0.9990; 1.0009), p = 0.979	0.9996 (0.9985; 1.0007), p = 0.452
MUFA to SFA ratio b	1.5806 (1.0842; 2.3042), p = 0.017	1.0414 (0.6880; 1.5763), p = 0.848
Meat b	1.0008 (0.9995; 1.0021), p = 0.221	1.0006 (0.9991; 1.0020), p = 0.441

MI: Myocardial infarction; CHD: Coronary heart disease; DII: Dietary inflammatory index, DASH: Dietary Approaches to Stop Hypertension, MUFA: Monounsaturated fatty acid, SFA: Saturated fatty acid. ^a components of DASH diet score ^b components of Mediterranean diet score

Table 6.

C-statistics and IDI for the evaluation of improved predictive ability of calibrated model by adding the desired nutritional variables

Components of evaluated dietary scores	Men		Women
	IDI	C-concordance	IDI	C-statistics
MI/CHD mortality
MUFA to SFA ratio a	0.0003 (-0.0005; 0.0011), p = 0.475	0.7296 (0.7062; 0.7531), p < 0.001	-	-
Sodium b	-	-	0.0004 (-0.0012; 0.0020), p = 0.596	0.7533 (0.7479; 0.7587), p < 0.001
Stroke mortality
Sodium b	0.0002 (-0.0012; 0.0017), p = 0.751	0.8254 (0.8035; 0.8474), p < 0.001	0.0003 (-0.0005; 0.0009), p = 0.485	0.8079 (0.7759; 0.8399), p < 0.001
Added Sugar b	-	-	0.0004 (-0.0042; 0.0051), p = 0.847	0.8062 (0.7717; 0.8407), p < 0.001
Total vegetable b	0.0003 (-0.0002; 0.0008), p = 0.266	0.8067 (0.7819; 0.8314), p < 0.001	0.0007 (-0.0003; 0.0016), P = 0.162	0.7816 (0.7528; 0.8103), p < 0.001
Total Vegetable a	0.0003 (-0.0014; 0.0019), p = 0.730	0.8067 (0.7795; 0.8339), p < 0.001	0.0007 (-0.0005; 0.0019), p = 0.256	0.7816 (0.7499; 0.8131), p < 0.001

DASH: Dietary Approaches to Stop Hypertension, IDI: Integrated discrimination index/improvement, MUFA: Monounsaturated fatty acid, SFA: Saturated fatty acid, MI: Myocardial infarction, CHD: Coronary heart disease. ^a components of Mediterranean diet score ^b components of DASH diet score

Discussion

In the current study conducted on the largest cohort study in the Middle East and North Africa, we examined for the first time the added value of two common nutritional indices and their components on the predictive power and discrimination of calibrated sex-specific non-laboratory WHO model for 10-year stroke and MI/CHD mortality risk in Iran. As for 10-year mortality of stroke and MI/CHD in both genders in a multivariate cox regression analysis were adjusted for WHO risk score, DASH and Mediterranean scores were not significant association with the desired outcomes. However, considering the dietary components of these scores, sodium and total vegetable in both genders and added sugar in women were meaningful predictors for 10-year stroke mortality. Also, sodium in women and MUFA to SFA ratio in men had considerable associations with 10-year mortality of MI/CHD. However, none of the above mentioned meaningful nutritional variables had an added value on the calibrated sex-specific non-laboratory WHO risk model for the prediction of 10-year stroke and MI/CHD mortality.

Identification of high-risk individuals, improvement of disease awareness, and establishment of preventive strategies for CVD are some of the advantages of application of CVD risk prediction models [30]. Regarding the increasing incidence of CVD in worldwide, several types of risk prediction models and risk scores for stroke, CHD, or total CVD have been proposed for different types of participants, until now [30]. Meanwhile, office-based or non-laboratory prediction models with similar discriminative ability are more popular than laboratory-based models specially among the regions with low or middle-income rate [31].

Several studies conducted on the large sample sizes of populations have been reported the association of dietary indices and their components with the risk of CVD-related outcomes [12, 32]. Wang et al., in their study on 28,905 participants from the National Health and Nutrition Examination Survey (NHANES) with a median follow-up of 6.3 years found that a higher Mediterranean diet score was considerably associated with a lower risk of all-cause mortality in general population among the people with different glucose regulation states [33]. Also, they found that in diabetic patients a higher adherence to the DASH and the Mediterranean diet was associated with a reduced risk of all-cause mortality. Karam et al., in their systematic review and network meta-analysis of 40 randomized controlled trials found that Mediterranean diet can probably reduce the risk of mortality and nonfatal MI in people at increased CVD risk [34]. Also, they reported that Mediterranean dietary patterns are likely to reduce stroke risk. On the other hand, Mente et al., used data from the Prospective Urban Rural Epidemiology (PURE) study (among 147,642 people from 21 countries) to develop a globally applicable healthy diet score based on six foods which was associated with health outcomes (including major cardiovascular events and all-cause mortality). They reported the association of a diet contains high vegetables, fruit, legumes, nuts, whole-fat dairy, and fish with lower CVD risk and mortality in all world regions, especially in low-income countries with lower consumption of these foods [35]. Also, Fan et al., conducted a study on peoples with metabolic syndrome (N = 8301) who participated in the National Health and Nutrition Examination Survey (NHANES) from 2007 to 2018 [36]. They examined the adherence to the Mediterranean diet and the effect of its specific components on cardiovascular and all-cause mortality. They found that people with metabolic syndrome had considerably lower risk of all-cause and CVD mortality with higher intake of vegetables, nuts, legumes, and MUFA/SFA ratio. On the other hand, higher consumption of red or processed meat had a meaningful association with the higher risk of CVD mortality among people with metabolic syndrome [36].

In the current study, similar to the abovementioned reports, some dietary components of DASH and Mediterranean scores had meaningful associations with mortality risk of stroke and MI/CHD in multivariate cox regression analysis along with WHO risk score. Sex-specific non-laboratory 10-year risk prediction models of WHO for fatal and non-fatal CVD included information on BMI, age, SBP, and smoking status. First time, Panagiotakos et al., showed that dietary factors along with some important predictive factors of CVD risk such as smoking status, blood pressure, sex, and age could improve the risk estimation of CVD [37]. Then, Baik et al., examined the added value of some dietary factors on the predictive ability of developed risk models including an office-based model, a model developed by traditional CVD risk factors, and two diet-containing models [38]. They analyzed data from population-based prospective cohort studies of the Korean Genome Epidemiology Study including 9026 male and female aged 40–69 years and their outcome of interest was CVD incidence or death. Based on Akaike’s information criterion (AIC) they reported that the diet-containing models are better fits than others and C-statistics for the all models were acceptable and comparative. Also, they reported an improvement in model performance and predictability with the inclusion of dietary predictors, such as poultry and legumes intake, green tea or carbonated soft drinks to the traditional model (including smoking status, sex, age, SBP, diagnosis of diabetes mellitus or hypertension, and blood concentrations of high-density lipoprotein-cholesterol and total cholesterol), and the office-based model (constructed with BMI and predictors included in the traditional risk model) [38].

Improvement in risk prediction models is not an easy issue in practice [39]. For instance, despite the great clinical potential of ankle-brachial index (ABI) and C-reactive protein (hs-CRP) as atherosclerotic CVD risk enhancing factors, the added value of these factors (considering the discrimination, calibration, and reclassification) in CVD prediction risk models is not certain yet [40]. The use of nutritional variables has always been controversial due to the challenges related to the accuracy of their measurements in studies. But, the use of nutritional variables can be helpful in situations where access to some biochemical data is difficult. Sometimes nutritional factors or indices can be a proxy of some biochemical variables. For instance, it has been suggested that dietary diversity score [41], plant-based diet [42, 43] or high-quality dietary patterns [44] are associated with biochemical markers of blood antioxidant and oxidative stress.

Despite so limited previous evidence, in the current study we found that none of the investigated nutritional indices and their components had an added value in the prediction of 10-year risk of stroke and MI/CHD mortality. In a similar previous study, Georgousopoulou et al., [45] tried to evaluate added value of Mediterranean diet score (as an indicator of diet quality) in a 10-year risk prediction model of CVD among ATTICA study participants. They first calculated the HellenicSCORE as an indicator of CVD risk based on the following predictors: gender, age, smoking status, SBP, and total cholesterol. Their desired outcomes include fatal or non-fatal CVD incidence of several heart diseases according to the ICD-10 codes (i.e. stroke, different types of heart failure, MI, chronic arrhythmia, angina pectoris, and other forms of ischemia). They reported that, Mediterranean diet score had significant association with 10-year risk of CVD events and inclusion of the score to the 10-year risk prediction model of CVD could considerably improve the classification ability and accuracy of model by 56%. Georgousopoulou and their colleagues made this conclusion by the results of Harrell’s C and Net Reclassification Index (NRI) statistics. Prof. Kathleen F. Kerr in her special report has discussed about the unreliable results of NRI in assessing the added value and interpretation of reclassification by a new predictor [46]. So, the judgment about the reported findings on the reclassification ability of Mediterranean diet score in Georgousopoulou et al. study should be done with caution. Main models and their predictors used for the risk prediction of CVD outcomes, type of outcome (such as mortality or first incidence, the exact type of disease regarding ICD-10 codes), characteristics of studied populations at the baseline (such as health status, age, gender proportion, CVD incidence rate, follow-up duration, etc.), and type of nutritional scores were used in model improvement analysis are important factors lead to controversial results among different studies.

In the present study, some components of DASH and Mediterranean diet scores (including added sugar, MUFA to SFA ratio, sodium, and vegetable intake) were statistically meaningful predictors for 10-year mortality of stroke or MI/CHD in the study population. There are several considerable evidence on the importance of reducing added sugar [47, 48] and sodium (salt) [49] as well as increasing vegetable consumption [50, 51] in daily dietary program to reduce risk of CVD in healthy people. These associations have a message that these components have an important role in the prediction of CVD risk. So, assessment of dietary intake must be independently considered in the estimation of CVD risk in healthy populations. Also, the issue of examining changes in participants’ food intake and its evaluation during follow-up periods is the next important point. Evaluation of changes in dietary patterns or food intakes in CVD patients and using these factors as predictors or model improving factors in the prediction of CVD mortality in these patients can be more informative for modifying dietary patterns and even improving dietary behaviors of CVD patients.

Strengths and limitations

To the best of our knowledge, in the present study for the first time we evaluated the external validity of non-laboratory CVD risk prediction model of WHO and added value of DASH and Mediterranean diet scores and their components on the discrimination and predictability of model in the largest cohort study in the Middle East and North Africa. The dietary information in GCS were collected by trained personnel via detailed interviews using a validated FFQ and photos of portion sizes to ensure that data collection is done with minimal bias. However, there are some limitations which are listed as follows: First, in the present study we only had access to data on CVD mortality. Nutritional variables may not be able to precisely predict long-term mortality events in healthy individuals. So, evaluation of the first occurrence of the CVD incidence, as desired outcome, may be more associated with the improvement of the relevant models by nutritional predictors. On the other hand, in the present study, the nutritional data has been only evaluated at the baseline of the cohort study. Access to the nutritional data at follow-up intervals and evaluation of changes in participants’ nutritional habits and dietary patterns in relation to the desired outcomes may provide more valuable data for examination of risk prediction models with nutritional predictors.

Conclusion

In the present study, the inclusion of dietary predictors, such as DASH and Mediterranean diet scores and their components to the non-laboratory based WHO prediction risk model had not improve model performance and predictive ability to the prediction of 10-year risk of stroke and MI/CHD mortality. Investigation of the first CVD incidence instead of mortality, as desired outcome, investigation of models for patients with CVD to examine the mortality or recurrence of CVD events such as stroke and MI, and assessment of nutritional data at follow-up intervals and evaluation of changes in nutritional habits or dietary patterns in relation to the CVD outcomes may provide more valuable data in the field of prediction models and their improvement by dietary factors.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Funding

This project was approved and financially supported by the National Nutrition & Food Technology Research Institute (NNFTRI), Shahid Beheshti University of Medical Sciences, Tehran, Iran (Ethical code: IR.SBMU.NNFTRI.REC.1399.070). The Golestan Cohort Study was supported by Tehran University of Medical Sciences (grant number: 81/15), Cancer Research UK (grant number: C20/A5860), the National Cancer Institute/National Institutes of Health, Intramural Research Program, and various collaborative research agreements with International Agency for Research on Cancer. The funding sources had no role in study design, data collection, data analysis, data interpretation or writing of the report.

Data availability

The data will be shared on request to the corresponding authors.

Declarations

Conflict of interest

Not applicable.