Background
The magnitude of the worldwide burden of diabetes continues to grow. It is estimated that 578 million people will be living with diabetes globally by the year 2030, approximately 90% of which will be type 2 diabetes [
1]. Abundant evidence has shown that adherence to a healthy diet (such as the Mediterranean diet) is crucial to the prevention of type 2 diabetes [
2]. However, these dietary patterns studied generally focused on conventional food groups such as fruits and vegetables [
3‐
5]. Recent studies show that higher intake of ultra-processed food (UPF) is associated with higher risk of type 2 diabetes [
6,
7]. However, UPF forms a highly heterogeneous food category, especially in terms of their nutritional composition, product types, and contribution to a habitual diet. It is therefore unclear whether previous results that identify total intake of UPF as a single risk factor for type 2 diabetes do apply to all underlying consumption patterns that fall under this “umbrella-term”.
Research on UPF has been facilitated by the development of the NOVA classification. The NOVA classification is a frequently used method to categorize food and drinks based on the nature, extent, and purpose of food processing. The NOVA classification comprises four categories, ranging from un-processed/minimally processed food to UPF [
8]. According to the NOVA classification, UPF is mostly formulated from food substances and industrial ingredients that undergo a series of chemical and physical manufacturing processes. The resulting food products are often pre-packed, contain little or no intact (un-processed) food, and are considered microbiologically safe, convenient, and palatable [
9].
Since the intake of UPF has substantially increased in most parts of the world over the past decades [
10], there is an increasing interest in the potential health impacts of UPF. Prospective cohort studies on the associations between UPF and health so far mostly focused on total intake of UPF. These prospective cohort studies found that higher intake of UPF was associated with higher risks of obesity [
11‐
13], cardiovascular diseases [
14‐
16], cancer [
17], mortality [
18‐
20], the metabolic syndrome [
21], and type 2 diabetes [
6,
7,
22,
23]. Associations established from these studies underline the fact that UPF is not neglectable when studying dietary effects on disease outcomes.
However, an often overlooked virtue of UPF is that it forms a highly heterogeneous food category. Food products considered as UPF are heterogeneous with respect to their nutritional composition, as well as their contribution to a habitual diet, and the context in which they are consumed [
24]. For example, according to the frequently used NOVA classification [
8], UPF includes pre-packaged bread, a staple food item which in many cultures is consumed with main meals; as well as cakes or fast food, which are consumed more occasionally. Therefore, results from previous studies analyzing UPF as one single food group may not apply to all underlying consumption patterns that fall within this food group. Scientific evidence so far may therefore not be sufficient to formulate evidence-based guidelines and health policies regarding UPF in the battle against type 2 diabetes.
In this study, we first aimed to assess the association between overall UPF intake and incident type 2 diabetes. More importantly, we aimed to identify underlying consumption patterns of UPF and to investigate how they were related to incident type 2 diabetes in a large cohort of Dutch adults.
Discussion
In this large population-based cohort study, the overall consumption of UPF was associated with a higher risk of type 2 diabetes, independent of overall diet quality and energy intake. We illustrated the importance of considering the heterogeneity of UPF when studying its health effects, as associations with incident type 2 diabetes varied across different patterns of UPF consumption. A positive association with incident type 2 diabetes was found for both warm savory snack and cold savory snack UPF consumption patterns, while a negative association was found for sweet snack UPF pattern. On the other hand, the absence of a clear association between diabetes risk and the traditional Dutch cuisine UPF pattern, which was high in main meal food items, suggests that not all types of UPF are necessarily detrimental to health.
Over the past few years, scientific interests and public awareness on UPF have risen substantially [
36,
37]. So far, four studies have investigated the association of UPF with type 2 diabetes [
6,
7,
22,
23]. Our results provide an independent confirmation of the association between UPF intake and incident type 2 diabetes in a different population setting. When comparing our results to those from the French NutriNet-Santé cohort and the UK Biobank cohort in which similar methods were used, the extent to which UPF contributed to the habitual diet differed considerably. The mean weight percentage of UPF in the diet was 35.9% in this Dutch cohort, versus 15.4% in the French cohort [
6] and 22.1% in the British cohort [
7]. Nevertheless, the reported hazard ratio of 1.15 in the previous French study and 1.12 in the UK Biobank study, regarding each 10 percent increment in the proportion of UPF in the diet, were comparable to our OR of 1.17 in our fourth model, adjusting for comparable potential confounding factors. In addition, it is noteworthy that in all three studies, associations were independent of the overall diet quality as well as total energy intake. This consolidates the potential role of UPF as an independent dietary factor in the development of type 2 diabetes. More importantly, it emphasizes that eating an otherwise healthy diet may not fully compensate for the detrimental effects of UPF.
Notwithstanding the high heterogeneity among different types of UPF, previous studies on the health consequences of UPF mainly focused on its overall intake. To our knowledge, the current study is the first that investigated the relation of overall intake and consumption patterns of UPF with incident type 2 diabetes in a large population-based sample. Our findings emphasize that it is crucial to consider various habitual UPF consumption patterns and their unique food groups when studying their health effects. In line with overall UPF intake, both the warm savory snack and the cold savory snack UPF patterns were associated with higher risks of type 2 diabetes. Results deviated for the traditional Dutch cuisine pattern and the sweet snack pattern, as the associations with type 2 diabetes were absent for the first, and inverse for the latter. The absence of an association for the traditional Dutch cuisine pattern illustrates that the detrimental effects of UPF may not be solely due to the degree of food processing. As UPF forms a highly heterogeneous food category, it is also important to consider their nutritional quality [
38]. For instance, a key food product in the traditional Dutch cuisine pattern was sliced bread. Despite mostly being ultra-processed, approximately 70% of the sliced bread consumed in The Netherlands is brown bread (made with a mixture of whole-wheat and white flour) or whole-wheat bread, and therefore often high in fiber and micronutrients. Higher intake of fiber and whole-wheat products was found to be associated with lower risk of type 2 diabetes [
39]. On the other hand, the UPF products identified in these two savory snack patterns are generally high in salt and fat and are often energy dense. It is conceivable that they may increase diabetes risk through metabolic disturbances, such as elevated blood pressure and lipid abnormality [
40,
41]. Therefore, a cautious interpretation of the health effects of UPF is warranted. More specifically, their effects on health may be determined by more than the level of food processing alone, which makes that not all types of UPF are necessarily detrimental to health.
Despite remaining statistically significant, our observation that estimates for the associations between UPF intake and type 2 diabetes were clearly attenuated when additionally adjusting for BMI, illustrating that BMI plays a role in the studied association. This role, however, may be two-fold, as BMI may be both a confounding and a mediating factor. Individuals with higher baseline BMI appeared to have higher total UPF intake, as well as a higher risk of type 2 diabetes, showing its confounding property. However, since previous prospective studies have illustrated that UPF is a risk factor for obesity [
11‐
13], higher intake of UPF may also increase type 2 diabetes risk through an increase in body weight, which illustrates the potential mediating role of BMI in the association studied. However, we also could not rule out the possibility of residual confounding, even in our analysis various involving covariates (including demographic, lifestyle, and socio-economic factors) were adjusted. Future studies, preferably in the form of randomized controlled trials, are required to help disentangle the role of BMI in the relationship between UPF and health.
Our finding that higher adherence to the sweet snack UPF pattern was associated with lower risk of incident type 2 diabetes was counterintuitive. Previous evidence indicates that the intake of dietary sugar from food and beverages was associated with weight gain and obesity, and may thus contribute to the risk of type 2 diabetes [
42,
43]. Nonetheless, a study in EPIC (European Prospective Investigation into Cancer and Nutrition) also found that non-consumers of cakes and cookies had a higher risk of type 2 diabetes [
44]. To assess for possible reverse causation, we performed a post hoc analysis to evaluate whether individuals’ baseline type 2 diabetes risk score was involved in this unexpected finding. As shown, a higher PROCAM diabetes risk score at baseline was associated with lower adherence to the sweet snack UPF pattern. Those with a high risk of type 2 diabetes could have been made aware of their situation through opportunistic screening by general practitioners, public health campaigns, or family history of the disease. Hence, awareness of high type 2 diabetes risk may have driven participants to avoid products that are high in sugar.
From a public health point of view, this can be perceived as a positive message, suggesting that public health initiatives to inform the public on the importance of a healthy diet in the prevention of chronic diseases, such as type 2 diabetes, did come across. In addition, the fact that the inverse association with baseline type 2 diabetes risk observed for the sweet snack pattern may be related to the layman’s term for type 2 diabetes, which is “sugar disease” in Dutch and several other languages. Although there is still some scientific uncertainty as to whether all types of sugar intake are associated with risk of type 2 diabetes [
45‐
48], limiting the consumption of energy dense, sugar-rich foods will be likely to benefit health, not only by reducing the risk of diabetes, but obesity and cardiovascular diseases as well [
48]. Furthermore, it is worth noticing that the adherence to both two savory UPF patterns was higher among individuals with higher diabetes risk scores at baseline, and both patterns were also associated with a higher risk of incident type 2 diabetes. Future research, preferably in the form of randomized controlled trials, is needed to confirm a detrimental effect on glucose homeostasis from both sugary and savory UPF items. A subsequent challenge would then be to create further awareness that it is not only sugary products, but also other kinds of UPF, which may be associated with higher diabetes risk. This could also bear relevance to prevention strategies not only by recommendations for health behaviors, but also by recommendations for product reformulation [
49].
Strengths of this study include the large sample size, which yields a strong statistical power. In addition, our study is the first that thoroughly investigated the habitual consumption patterns of UPF using PCA. The empirical consumption patterns identified reflected not only nutritional properties of UPF, but also its behavioral drivers, which provide a distinct added value over the nutritional information of the NOVA classification and strengthen the real-world robustness of the results of this study [
50]. Nevertheless, it should be noted that the four consumption patterns analyzed in total only explained 15.5% of the total variance of UPF intake, which inevitably left a certain proportion of the UPF consumption pattern information uncaptured. It is conceivable that this seemingly low explained variance is attributed to our large study sample size. Secondly, this may also be attributed to the fact that we did not apply massive food groupings in dietary pattern analysis (i.e., combining the intake of several food products into one, such as treating all sorts of cheeses as one single food product) [
50,
51], which in fact facilitates our study objective for the disentanglement of the consumption patterns of this highly heterogeneous food category. On the other hand, our 15.5% explained variance is comparable with previous studies using PCA and did offer us informative insights into the real-world eating patterns, especially considering our cohort setting [
51,
52]. We encourage future studies to further explore the UPF consumption patterns in a different population setting.
Furthermore, there are also several other limitations that should be noted. First of all, the FFQ used in this study was designed to assess the intake of major food groups, energy, and macronutrients. The aim to assess energy intake resulted in good coverage of energy dense food, including many kinds of UPF. However, since the FFQ was not designed to assess the intake of UPF, it included questions that covered the intake of food items with varying levels of processing, inevitably leading to some misclassification. Second, misclassification could also occur in the ascertainment of type 2 diabetes cases, since at T2 and T3 only self-reported data was available. However, as most cases were identified by objective laboratory measurements at T4, this limitation is not expected to influence our results. Third, the exact time of diagnosis of diabetes cases was not collected in the Lifelines study which unfortunately reduced the suitability of our data for survival analyses. Nevertheless, considering the low event rate and the relatively short follow-up time, logistic regression analysis may provide similar estimates for the effect sizes [
53,
54]. We therefore used logistic regression analysis instead. Furthermore, the use of self-reported questionnaires such as FFQ might lead to misreporting due to social desirability or recall bias. Finally, we illustrated that some reverse causation could be involved in the results of this study, despite our prospective design.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.