Development and validation of dietary and lifestyle insulinemic indices among Iranian adult population

Insulin is one of the most important hormones maintaining metabolic homeostasis. Insulin affects different tissues in the body, including adipose tissue, liver, muscle, brain, bone [1], kidneys [2], and vascular system [3, 4]. Insulin-related disorders, including hyperinsulinemia (HI) and insulin resistance (IR), has been known as an early indicator and central place in metabolic dysfunction [5, 6]. If these initial disorders are left uncontrolled, they will contribute to the majority of chronic diseases such as type 2 diabetes (T2D), metabolic syndrome (MetS) [7].

Information on IR and HI statistics worldwide and in Iran is limited; however, considering the prevalence of T2D and MetS, as the two diseases most closely related to IR and HI, clarifies the current situation. According to the latest estimates, the prevalence of MetS is about 14–32% worldwide [8] and 30.8% in Iran [8]. The prevalence of diabetes in Iran in 2021 was 9.1% among adults, and it is projected to reach 10.6% by 2045. It is also estimated that the proportion of undiagnosed T2D in the Iranian adult population is about 36% [9]. This high rate of chronic diseases related to IR and HI globally, especially in Iran, requires special attention to improve the prevention and control of these disorders in the early stages. Lifestyle is the most important modifiable risk factor affecting insulin-related disorders development [10].

Diet is well-established to be an essential part of lifestyle [10]. Although specific dietary factors have been shown to affect insulin secretion and action [11‐13], dietary patterns or dietary indices, including several dietary factors and cover the complex interactions between nutrients and foods, can provide a better estimate about the relationship between diet and risk of chronic disease [14, 15]. Other lifestyle factors associated with HI and IR include body mass index (BMI) and physical activity (PA) [16‐18]. Given that each of these lifestyle factors is a great predictor of metabolic disorders per se, combining diet with them is likely to be more predictive of HI and IR risk than these factors alone.

Recently, Tabung et al. has developed four indices to assess the insulin potential of diet and lifestyle, including empirical dietary and lifestyle indices for HI and IR [19]. To create these indices, the concentration of connecting peptide (C-Peptide) and triglycerides (TGs) / high-density lipoprotein cholesterol (HDL-C) ratio were used as biomarkers for predicting HI and IR respectively. However, this study had some limitations. Questionnaires for medical history information were self-administered, reducing reporting accuracy [20, 21]. The study mentioned above also used blood data of participants in the Nurses’ Health Study (NHS), all of whom are women and highly educated; so, the generalizability of its results to the general population may be limited. Latterly two insulinemic indices were proposed by Mazidi et al. [22]. Similar to Tabung et al., Mazidi et al. have used C-Peptide and TG/HDL-C as biomarkers to develop the indices; however, they derived nutrient patterns, which are limited in covering interactions between different food components, such as those observed in the dietary pattern. Also, in Mazidi et al. study, the dietary intakes have assessed using a single 24-h recall, which may not be appropriate for determining the usual diet or irregularly consumed foods.

Based on the available literature on this topic, there is no study on developing a valid index to predict insulin-related disorders in the Iranian population based on their dietary habits, patterns, and lifestyle. Therefore, this study aims to develop and validate insulin potential indices of diet and lifestyle in the Iranian population.

Components of the four insulinemic dietary and lifestyle indices, including DIH, DIR, LIH, and LIR, with their specific weights and R² values, were reported in Table 1. Based on the relationship of components with serum insulin and HOMA-IR levels, as response biomarkers for each index, the components were divided into two categories, including (1) food groups that directly related to mentioned biomarkers and (2) food groups that showed an inverse relationship to these biomarkers. The first group components directly related to serum insulin and HOMA-IR levels were given positive weights, while components with inverse relationship with the biomarkers were assigned negative weights. Detailed information about food groups with their items used in developed indices were provided in Additional file 1: Table S1.

Table 2 presents the participants’ baseline characteristics across quartiles of the DIH and DIR. Serum insulin, HOMA-IR, WC, TGs, TG to HDL-C ratio, VAI, and percentage of men were significantly increased across quartiles of DIH and DIR (P for trend =  < 0.001). In contrast, participants’ age and HDL-C concentration were decreased across quartiles of DIH and DIR (P for trend =  < 0.05).

Pearson and Spearman’s correlations coefficients of the insulinemic dietary and lifestyle indices with serum insulin and HOMA-IR based on examinations 3 and 4 of TLGS were shown in Table 3. In examination 3 (baseline), a significant moderate correlation was observed between all developed indices (DIH, DIR, LIH, and LIR) and serum insulin and HOMA-IR levels. In contrast, in examination 4, this significant correlation was only observed for LIH with serum insulin and LIR with HOMA-IR. In Table 3, we also reported the results on the correlation of previously designed similar insulin-related indices, including insulin index, insulin load, empirical dietary indices for HI and IR, empirical lifestyle indices for HI and IR with serum insulin levels, and HOMA-IR. Although the Spearman correlation coefficients showed that three of four above mentioned indices, including empirical lifestyle index for HI, insulin index, and insulin load, were significantly correlated with serum insulin in both examinations, Pearson correlation was significant only for the including empirical lifestyle index for HI score with serum insulin. Furthermore, a weaker correlation was found between HOMA-IR and empirical dietary index for insulin resistance in examination 3 and HOMA-IR and empirical lifestyle index for insulin resistance in examinations 3 and 4.

The Relative values and 95% CIs of serum insulin concentration and HOMA-IR across quartiles of the DIH, DIR, LIH, and LIR based on the two examinations of TLGS (3 and 4) were shown in Table 4. RVs of serum insulin concentration and HOMA-IR level followed statistically significant upward trends across quartiles of most indices in both examinations. However, a significant trend for biomarkers RVs across DIH quartiles was observed only in examination 3.

In examination 3, participants in the highest quartile of DIH and LIH had 22% and 59% higher serum insulin concentrations than those in the first quartile (95% CI 1.15–1.29, and 1.52–1.67, P for trend < 0.001), respectively. Also, based on multivariable-adjusted models, the HOMA-IR index was 28% and 67% higher among individuals in the highest vs. lowest quartile of DIR and LIR (95% CI 1.20–1.36, and 1.59–1.75, P for trend < 0.001), respectively. In examination 4, the trend of the RV of serum insulin level across quartiles of DIH score was not significant; however, an upward trend was observed in mean serum insulin level and HOMA-IR index across quartiles of the DIR, LIH, and LIR. Based on multivariable-adjusted models, the RV of mean serum insulin has significantly increased across quartile of LIH (RV = 1.45; 95% CI 1.36–1.55, P_trend < 0.001). Also, in fully adjusted model, the RV of HOMA-IR index level have significantly increased across quartiles of DIR (RV = 1.28; 95% CI 1.13–1.42, P_trend = 0.019) and LIR (RV = 1.51; 95% CI 1.41–1.61, P_trend < 0.001).

Using the sensitivity analysis, the trend of the RV and 95% CIs of serum insulin across quartiles of DIH and LIH and the RV and 95% CIs of HOMA-IR across quartiles of DIR and LIR has been shown based on three sub-groups, including (1) excluding the smoked subjects, (2) excluding participants with a family history of T2D, and (3) including type 2 diabetic patients at baseline in Table 5. Our results showed that the RV of serum insulin concentration across quartiles has significantly increased across quartiles DIH and LIH based on age and sex adjusted and as well as multivariable-adjusted models. Also, the RV of HOMA-IR index levels has significantly increased across quartiles DIR and LIR based on both models. Furthermore, the stratified analysis showed that the mean serum insulin concentration and HOMA-IR of active-obese participants with the highest scores of DIH and DIR were 23% and 30% higher than those with the lowest scores. Similarly, non-active obese subjects in the highest quartile of DIH and DIR had 28% and 36% greater HOMA-IR values than those in the lowest quartile (Fig. 1).

In the present study, we developed and validated four indices, including DIH and DIR, consisting of a dietary pattern predictive of HI and IR, as well as LIH and LIR encompassing both diet and lifestyle factors that are significantly related to insulin disorders. Considering the importance of PA and obesity in HI and IR [16, 18], we performed a subgroup analysis based on the physically active obese group and physically inactive obese individuals, separately. Our analysis showed that in the physically active obese subgroup, subjects in the highest quartile of DIH and DIR compared to those in the lowest one had 22 and 31% higher serum insulin and HOMA-IR, respectively. Also, in the physically inactive obese subgroup, serum insulin concentration and HOMA-IR levels of individuals in the highest quartile of DIH and DIR were 26% and 36% higher than those in the lowest quartile, respectively. These analyses showed that individuals with higher DIH and DIR scores had higher serum insulin concentration and HOMA-IR levels, independent of BMI and PA.

In recent years, Tabung et al. [19] and Mazidi et al. [22], in two separate studies, introduced several indices related to insulin homeostasis status, using serum C-Peptide concentration and TGs to HDL-C ratio as biomarkers of HI and IR, respectively. Similar to our study, Tabung et al. derived four dietary and lifestyle indices with different food groups, BMI, and PA, which indicate the insulinemic potential of diet and lifestyle; however, Mazidi’s study used a pattern of nutrients to develop insulinemic dietary indices. Using nutrients instead of food groups in Mazidi’s study may not accurately estimate the diet-disease relationship due to ignoring the complex interactions of different substances in whole foods [29]. However, it seems that the most important concern about these indices introduced in Mazidi et al. and Tabung et al.’s studies is that they are all developed in the US population, which may limit the generalizability of their results for other populations. In this study, of all food groups that were entered in the stepwise regression analysis, only those with statistically significant (positive or negative) relationships (P < 0.1) with serum insulin level and HOMA-IR were included in the final patterns. In our study, intakes of red meat, broth, high- and low-fat dairy products, starchy vegetables, snacks, and higher PA level showed an inverse relationship with serum insulin level and HOMA-IR, while consumption of refined grains, sweetened beverages, pickle, doogh, lemon juice, fish and BMI directly associated with the biomarkers as mentioned earlier. Our findings are in line with the results of most of the studies that have suggested the reduced risk of HI and IR in individuals with higher intakes of starchy vegetables, low-fat dairy products, and higher PA levels, as well as lower intake of refined grains and sweetened beverages and obesity [30‐35].

In Iranian food culture, fish is mainly eaten fried form, which this cooking method can increase the intake of trans fats and weaken the beneficial properties of this food item. Such an effect is also noticed in other studies [19]. In contrast, red meat is mainly eaten unprocessed form along with legumes, vegetables, which may mitigate its potentially harmful effects based on results of previous studies [36]. Based on the extracted patterns in this study, higher consumption of doogh (a beverage made from yogurt, which has relatively high salt content), lemon juice, and pickles directly correlated with serum insulin and HOMA-IR levels. Although doogh is a low-calorie beverage, its high salt content can worsen IR through several mechanisms [37]. Moreover, dietary constituents like doogh, lemon juice, and pickles are often eaten with high-carbohydrate, high-calorie meals, meaning that consuming more indirectly indicates higher energy and carbohydrate intake, which raises IR risk.

In our study, lifestyle indices (LIH and LIR) were 2 to 2.5 times more correlated with biomarker levels than dietary indices (DIH and DIR). This higher correlation indicates the higher potential of lifestyle indices to predict the risk of HI and IR. Dietary indices showed a relatively low but significant correlation with serum insulin and HOMA-IR level. The low correlation of these dietary indices with the biomarkers does not mean that the developed indices are not valid; because the serum insulin and glucose level’ specialty in healthy adults, more influenced by genetic, endogenous factors (like regulatory systems), and other lifestyle factors (such as PA and BMI) compared with a usual diet [38], so the low correlation of some nutritional factors with serum insulin and HOMA-IR level was expectable. Similar findings have been observed in other studies [19, 39], e.g., in the Fung et al. study, a dietary pattern predicting serum C-peptide showed a relatively low correlation (r = 0.23) with the levels of this biomarker; however, the association of this pattern with the risk of colon cancer was still significant [39]. Also, in the Tabung et al. study, the reported correlation coefficients between developed indices and biomarkers were close to ours, and their indices had good predictive power for the risk of HI and IR [19].

Various analyses confirmed the robustness and validity of all four developed indices in our study. In the third examination of the TLGS study (baseline), the relative serum insulin concentration and HOMA-IR levels increased across the quartiles of DIH, DIR, LIH, and LIR with a significant trend. Similarly, these significant findings were observed in the fourth examination for all indices mentioned above except DIH. In general, dietary indicators are less capable than lifestyle indicators for predicting the response biomarkers. In the validation study, the adjusted RV of fasting insulin in the highest vs. lowest quartile of DIH was 4% higher; however, it showed no significant trend. In contrast, HOMA-IR showed a significant positive trend across quartiles of DIR. Some possible reasons may explain why DIH has shown a weaker predictive ability than DIR. First, DIH covers a smaller range of dietary components than DIR, as DIR, additional to similar items with DIH, includes snacks and low-fat dairy. Second, fasting insulin concentration (as the response variable for DIH) is less affected by diet compared with HOMA-IR (as the response variable for DIR) because FBS as a component of HOMA-IR is more affected by food intakes [40]. Finally, in our study, about 97% of individuals had a positive DIH score (compared to DIR with 75% positive and 25% negative). This may have undermined the DIH’s ability to discriminate between individuals with different serum insulin levels and, consequently, its predictive ability in our study.

Furthermore, we tested the sensitivity of the indices in three different sub-groups by excluding subjects with a family history of T2D, excluding smoked participants, and including diabetic patients in the analysis. Finally, based on each sub-group analysis, the relative values of serum insulin concentration and HOMA-IR significantly increased across quartiles of all developed indices.

We examined the correlation of the developed indices in the unweighted form and the similar indices developed in previous studies with the serum insulin and HOMA-IR levels. Our results showed that even unweighted forms of DIH and DIR correlate with serum insulin concentration and HOMA-IR more than the empirical dietary index for HI, insulin index, and insulin load. Weighted DIH and DIR showed a lower correlation with biomarkers than unweighted forms. One possible reason could be that in unweighted form, all components were equally involved in the correlation between indices and biomarkers regardless of their dietary intakes, e.g., in the weighted form, the effect of some components such as pickles that were high in weight and could increase correlation was attenuated by low dietary intake, The opposite is true for sweetened beverages. Among lifestyle indices, the large difference in weight and R² of the BMI compared to other items covered their overall impact and significantly affected increasing correlation. When BMI was considered with the same weight as other factors, its effect is attenuated, which reduces the correlation of index with serum insulin and HOMA-IR. Further analysis showed that the DIH, DIR, LIH, and LIR were significantly correlated (r < 0.40) with their unweighted forms (Additional file 1: Table S2, S3).

The indices introduced in the present study can be used in other populations, especially those with similar habits to the Iranian population, such as people in the Caucasus and the Middle East. The main benefit of our results is that researchers who do not have access to insulin biomarkers data can easily estimate the insulinemic potential of diet and lifestyle using only dietary and lifestyle data. More research and studies are still needed on the derived patterns and the food groups (to determine the optimal intakes). If the results obtained in this study are repeated in future studies, clinical applications for these findings can be considered. Physicians and dietitians can assess the diet and lifestyle of patients at risk of developing insulin-related disorders using the indices proposed in this study.

Our study had several strengths and limitations. This work is the first study conducted to develop and validate dietary and lifestyle indices to predict the risk of HI and IR among the Iranian population. We used serum insulin concentration as a response variable to develop DIH and LIH indices, which is a direct indicator for the diagnosis of HI. HOMA-IR index was also used to detect insulin resistance for DIR and LIR development. This index has been widely used in studies among various populations and on several outcomes due to the high correlation with the euglycemic hyperinsulinemic clamp techniques as the gold standard for diagnosing IR [41]. Our study has some limitations. First, due to the lack of another study on measuring the serum insulin concentration with an acceptable sample size in the Iranian adult population, we had to validate the developed indices using participants’ data from examination 4 of the TLGS study. However, this limitation is somewhat negligible because the validation of the indices was done in the fourth examination of the TLGS study with new entries and different sample sizes, in which their data were collected independently of the third examination (baseline). Second, despite using a valid and reliable FFQ for dietary assessment, measurement errors were inevitable in the estimates of dietary intake using FFQ. However, we entered dietary intakes as energy-adjusted into analysis to reduce the impact of measurement errors to the extent possible. Third, we have no complete data of other lifestyle factors such as income, culture, customs, etc. data to determine better the lifestyle index related to HI and IR. Fourth, although we attempted to adjust for major confounders in the present study, the confounding effect of some unknown and unmeasured residual confounders cannot be excluded. In addition, although direct measurement of PA is not common in large population studies, the use of indirect measurement for PA is another limitation that cannot be ignored completely; however, we collected the PA data using a valid questionnaire in a face-to-face interview to enhance the assessment precision.

In the present study, we have designed and validated four insulinemic indices, including DIH, DIR, LIH, and LIR, to determine the insulinemic potential of diet and lifestyle in the Iranian population, according to their demographic and dietary intakes. The findings of this study can be used for research and clinical purposes. These indicators may help identify individuals, groups, and even populations at risk for diabetes, metabolic syndrome, cardiovascular disease, etc. Researchers who do not have access to insulin biomarker data and researchers in other countries, especially in Southwest Asia, which has a food culture and lifestyle close to Iran, can use these indices in their studies. The results of this study need to be confirmed by large prospective studies and randomized clinical trials.