Introduction
Globally, 537 million adults were living with type 2 diabetes mellitus (T2D) in 2021, and the prevalence is projected to rise to over 780 million by 2045 [
1]. T2D, which accounts for the vast majority of diabetes cases, appears to be largely preventable by a balanced lifestyle. Besides obesity, a major modifiable risk factor, a suboptimal diet is considered a major contributor to the development of T2D. Evidence accumulated from both prospective observational studies and randomised controlled trials (RCTs) highlights the importance of single dietary factors in this context. Consumption of processed and unprocessed red meats as well as sugar-sweetened beverages have been observed being related to an increased T2D incidence, while whole grains, dairy products, nuts, green-leafy vegetables and coffee may reduce the risk of T2D [
2‐
5]. Furthermore, the quality of carbohydrates (fibre, glycaemic index) and fats (saturated versus unsaturated fatty acids) as well as individual nutrients (iron, magnesium) have been identified to affect the risk of T2D [
2,
6]. However, the role of total carbohydrate (CHO) and total fat intake, major macronutrients besides protein, appears to be less pronounced. For example, both prospective cohort studies as well as results from the Women’s Health Initiative RCT suggest that a general increase in CHO intake at the expense of fat is unlikely to affect T2D risk [
7]. Protein-rich foods show inconsistent associations with the risk of T2D, with red and processed meat being related to increased and dairy and whole grains to reduced risk, while poultry, fish, eggs, nuts, refined grain products and soya are not clearly related to risk [
6]. Given these inconsistencies, it remains unclear whether total protein intake resulting from these foods or the intake of protein from either animal-based or plant-based foods are related to T2D risk in humans.
Mechanisms that link the intake of dietary protein to the risk of T2D have to be related to one or both of the two known major pathophysiologic pathways of T2D: insulin resistance or impaired insulin secretion [
8]. In particular, insulin resistance is closely associated with body fat accumulation, and effects of protein-rich diets may be indirectly linked to lower insulin resistance by their potential beneficial effects on body weight (BW). However, reductions in BW can be achieved by energy-restricted diets with different macronutrient composition [
9,
10]. It is therefore important to clarify whether diets with different protein content affect insulin resistance, β-cell function and glycaemic status independent of energy intake and BW control (i.e. under isoenergetic settings). Dietary protein intake has been well established to augment postprandial insulin secretion, leading to enhanced glucose clearance from the blood by peripheral tissues [
11]. However, such acute postprandial effects are not synonymous with long-term effects of protein-rich diets on tissue insulin sensitivity or secretory capacity of β-cells. Whether protein intake affects important glycaemic traits (glycated haemoglobin A1c [HbA1c], fasting glucose, fasting insulin) over longer periods of time (weeks to months) has been investigated in several RCTs. Overall, data from RCTs do not provide consistent evidence that high-protein intake [mostly around 30 energy percentage (%E)] or the choice of animal versus plant protein substantially affects major glycaemic traits [
12‐
15].
Here, we present our findings from an umbrella review of systematic reviews (SRs) of prospective cohort studies and RCTs on protein intake (total as well as animal and plant protein) and the risk of T2D. We aimed to grade the overall certainty of evidence for such associations, considering the methodological quality of SRs, consistency of results, and biological plausibility. This work is part of a series of umbrella reviews on protein intake and health-related outcomes which are carried out as the basis of a new guideline on the effects of protein intake on health parameters by the German Nutrition Society [
16].
Discussion
In the present umbrella review, we identified eight SRs that evaluated an association between protein intake (total, animal and plant protein, respectively) and T2D risk, of which six also provided estimates from MAs. Overall, positive associations for total and animal protein and the risk of T2D were observed in the majority of SRs. However, the results of RCTs on important glycaemic traits do not provide strong support that total and animal protein intake adversely impacts the pathogenesis of T2D. In light of the lack of biological plausibility and the low outcome-specific certainty by NutriGrade for SRs, in particular for animal protein, the overall certainty of evidence for an increased T2D risk was judged to be possible for high total protein and insufficient for high animal protein intake. While some SRs support a lower T2D risk with higher plant protein intake, SRs findings were mainly inconsistent. This, together with the lack of a clear biological plausibility as indicated by a lack of effect on major glycaemic traits in RCTs, led us to conclude that there is insufficient evidence for such an inverse association. Given that the majority of SRs do not point out an association, the overall evidence that there is no association was considered possible.
Whether protein intake is associated with T2D risk has been evaluated in three umbrella reviews of SRs. Bellou et al. [
49] reviewed SRs that evaluated the association of various dietary and non-dietary factors with T2D, but identified only the SR by Alhazmi et al. [
26], which analysed the relation between protein intake and T2D risk. As discussed above, this SR included notably fewer individual studies than more recent SRs and its quality was rated low. Neuenschwander et al. [
6] conducted an umbrella review of dietary risk factors for T2D and concluded that there is evidence that total and animal protein intakes are related to increased T2D risk. The certainty of evidence was graded as moderate for both total and animal protein in this umbrella review, while we graded the evidence to be possible for total protein intake but insufficient for animal protein intake. Similar to our umbrella review, individual SRs were graded with AMSTAR and the certainty of evidence was evaluated with a modified version of NutriGrade. However, in contrast to our review, this umbrella review included a re-analysis (random effects MA) of the original studies identified in the SRs. Based on nine primary cohort studies, each 5% increase in energy from total protein intake was related to an RR of 1.09 (95% CI 1.04–1.13). The respective estimate for animal protein, based on eight primary studies, was 1.12 (95% CI 1.08–1.17). For plant protein no statistically significant association with T2D risk was observed (RR per 5%
E 0.87; 95% CI 0.74–1.01), and the certainty of evidence was rated as “low” [
6]. In contrast to our umbrella review, only one SR was selected for further evaluation for each dietary exposure, e.g. based on the largest number of individual studies or of T2D cases included. Thus, while the SRs by Shang et al. [
20] and Tian et al. [
21] were also identified in this umbrella review [
6], only the SR by Zhao et al. [
22] was selected. More recent SRs by Ye et al. [
25] and Fan et al. [
23] were not included in the umbrella review by Neuschwander et al. [
6], likely because of the later time of publication. Lv et al. have reviewed SRs on protein intake and multiple health outcomes, including T2D [
50]. The authors included 2 SRs (Zhao et al. [
22], Ye et al. [
25]). While more SR were identified [
13,
20,
23,
26], the authors included only those SRs which reported the highest number of primary studies. The two included SRs were graded to have “high” methodological quality using AMSTAR, similar to our AMSTAR 2 grading. Lv et al. reported “highly suggestive evidence” for a relationship of higher animal protein intake and higher T2D risk for the SR by Zhao et al. [
22], while the evidence was classified as “weak” for the SR by Ye et al. [
25]. Suggestive evidence was found for a positive relationship of total protein and T2D risk and for an inverse association of moderate plant protein intake. But again, this evidence classification refers to a single SR each, and the corresponding second SR was classified as either “weak” or “not significant”. The evidence classification by Lv et al. considered statistical significance and precision, number of cases, heterogeneity, evidence for small study effects, and evidence for excess significance bias. Lv et al. [
50] did not reveal the evidence from individual SRs to an overall certainty of evidence as we did in our UR. Our umbrella review and certainty of evidence evaluation therefore included considerably more SRs than previous umbrella reviews on the topic.
While SRs on total protein intake quite consistently observed higher T2D risk at higher protein intake, this association seems to be restricted to animal protein only, given that animal protein but not plant protein was found to be positively associated with T2D risk. Such a difference could be related to the relative abundance of different amino acids in animal versus plant protein. Animal protein provides higher amounts of branched-chain amino acids compared to plant proteins, and circulating leucine, isoleucine, and valine may be risk factors for T2D [
51]. Furthermore, circulating glycine was found to be associated with a higher risk of T2D; as it is abundant in animal protein, it has been considered as a potential mediator which links higher intake of red meat to T2D [
52]. In addition, higher circulating tyrosine levels appear to be causally related to reduced T2D risk [
53]; nevertheless, tyrosine is an abundant amino acid in both animal and plant foods and may unlikely explain the contrasting associations found for animal versus plant protein intake in cohort studies. However, despite some evidence that the amino acid composition of animal protein may be relevant for the pathogenesis of T2D, we downgraded the overall certainty of evidence for all protein exposures due to the lack of evidence that changes in total protein intake or the relative proportion of animal versus plant protein showed mid- to longer-term effects on major glycaemic traits in RCTs. The reasons for the discrepancy between RCTs on glycaemic traits and long-term cohort studies on T2D incidence remain unclear. Generally, protein intake in observational studies is estimated from food intake and associations observed may not necessarily reflect causal effects of the nutrient per se. It is rather possible that protein intake is a marker for specific protein-rich foods. In this context, associations of protein-rich animal foods with T2D risk are not homogenous [
2,
3,
6]; positive associations have largely been restricted to unprocessed red meat and processed meat. This makes it questionable that animal protein intake per se could have detrimental effects. Further, associations found for red meat may be explained by other food constituents than protein [
52]. Finally, residual confounding cannot be excluded as alternative explanation for observational study results.
There are several other limitations inherent to observational studies on dietary risk factors for T2D. Observational studies rely on self-reported dietary intake which is generally prone to misreporting. Cohort studies frequently rely on semi-quantitative assessment instruments like food frequency questionnaires, which are prone to measurement error and not designed to provide an accurate quantitative estimate of absolute protein intake. For example, correlations between protein intake estimated by questionnaires and urinary nitrogen excretion (the gold standard for validating self-report instruments) range from 0.07 to 0.57 [
54]. Given that measurement errors are unrelated to the disease status during follow-up in cohort studies, they would likely tend to lead to an underestimation of the true association. This is indicated by results of the Women’s Health Initiative which is included in several of the identified SRs: the association of total protein intake (as
E%) with the risk of T2D was markedly stronger when corrected for this measurement error using regression calibration [
33]. Furthermore, investigations on macronutrient composition generally reflect substitution effects under isoenergetic settings, but SRs provide usually only limited information whether associations included in MA refer to substitutions of protein for total CHO and/or total fat, if subgroups of CHO or fat were considered (e.g. low versus high quality CHO, saturated vs. unsaturated fatty acids), or if higher animal or plant protein intake reflect also a higher total protein intake or rather a substitution of one for the other—although the relevance of model choice to reflect specific macronutrient substitutions in relation to T2D risk has been well documented [
30,
55].
However, the limitation of current RCTs in relation to glycaemic traits might be also important. While an effect of protein intake on fasting insulin was shown, it is noteworthy that this may not reflect the effect to be expected in isoenegetic settings. The interventions of most RCTs included in the MA by Santesso et al. [
15] and Schwingshackl et al. [
12] involved energy-restricted weight-loss diets and several studies found different effects on BW. Thus, it remains unclear from the MA to which extent reductions in fasting insulin can be explained by the differences in weight-loss between the study arms. It is also noteworthy that high-protein interventions were applied in many of the included RCTs (often around 30
E% from protein), while there are apparently no MA of RCTs that investigate protein intake at the higher end of habitual intake (between 18 and 25
E%). Thus, the generalizability of findings on protein intake and biomarkers of glucose homeostasis from RCTs to real-world settings is limited. The PREVIEW RCT, which compared a high protein (25%
E) and low glycaemic index diet with a moderate protein (15%
E) and moderate glycaemic index diet among persons with prediabetes in a three-year weight maintenance intervention following an eight-week weight reduction, did not observe an effect on T2D incidence [
56]. However, incidence was overall low in the study, limiting the power for group comparisons. Interestingly, significantly fewer participants achieved normoglycaemia at three years in the high protein compared to the moderate protein group, although weight loss was comparable [
56].
The comprehensive and systematic literature search as well as the assessment of the methodological quality of the SRs with AMSTAR 2 and the rating of the outcome-specific certainty of evidence with NutriGrade are clear strengths of the current umbrella review. Next, all methodological steps of the review procedure have been defined a priori as described [
16]. Furthermore, we included evidence of all identified SRs in our overall certainty of evidence assessment. However, our procedure requires relatively high certainty of evidence in all or most individual SRs to result in high overall certainty of evidence. Our literature search revealed several SRs on the topic, but with different coverage of individual studies and with varying quality assessment. We applied NutriGrade instead of the GRADE approach (Grading of Recommendations, Assessment, Development and Evaluation) because an important novelty of NutriGrade (published in 2016) was the modified classification for MA of RCTs and cohort studies compared with the traditional GRADE approach (initially classifying RCTs with an initial high score and cohort studies with a low score) [
57]. We are aware that in the meantime the GRADE approach was amended (adjustments published in 2019, but after the guideline methodology was established in 2017) in a way that cohort studies can now also be assigned an initially high score, when risk of bias tools such as ROBINS-I are used [
58]. Furthermore, some reviews included individual cohort populations twice or included publications on endpoints other than T2D. Restricting the certainty of evidence assessment to the most recent or comprehensive SRs or to those which meet a pre-defined quality threshold may lead to a higher evidence grade. For example, the SR by Zhao et al. covers all individual cohorts identified also by any other SR except one, not considering duplicate publications from the same cohort population. This SR was rated “high” by AMSTAR 2 and “moderate” by NutriGrade. We also did not consider a re-analysis of original studies, although it is clear that none of the SRs included all relevant individual studies.
An umbrella review also has limitations. For example, more recent primary studies cannot be included in the evaluation and are therefore not taken into account. In addition, there is a dependency on the inclusion and exclusion criteria of the underlying SRs; for the umbrella review, the largest possible overlap with its own inclusion and exclusion criteria must be achieved. Accordingly, important results may not be taken into account because the inclusion criteria are not completely fulfilled. The quality of the umbrella review is largely dependent on the quality of the SRs and thus on the quality of the primary studies. Under certain circumstances, the primary studies included in the reviews differ considerably from one another, so that the informative value of the umbrella review is limited. In addition, it is dependent on the summary of findings at the level of the SRs, whereby outcome, subject structures and/or the interventions may be standardised or summarised. On the other hand, umbrella reviews have strengths. Umbrella reviews summarise the best possible evidence so that SRs with and without MA can be summarised. Umbrella reviews are considered as the highest level of evidence. The amount of available evidence is becoming more and more heterogenous, so umbrella reviews are a good way to summarise the available data [
59,
60].