Dietary Factors and All-Cause Mortality in Individuals With Type 2 Diabetes: A Systematic Review and Meta-analysis of Prospective Observational Studies

Diabetes is a major public health concern worldwide and is associated with several comorbidities and mortality. In 2021, it was estimated that ∼537 million people aged between 20 and 79 years lived with diabetes globally (1). The estimated number of deaths due to diabetes or its complications was 6.7 million people in 2021 (1). As hyperglycemia is associated with several comorbidities and subsequently with premature mortality, the glycemic management and secondary prevention of these comorbidities is of high public health interest.

For the primary prevention of type 2 diabetes, many modifiable risk factors, such as an unfavorable diet, are well known (2). Regarding diet, there are dietary recommendations available for individuals with type 2 diabetes, for example, from the International Diabetes Federation (IDF) or from the American Diabetes Association (ADA). The IDF clinical practice recommendations for the management of type 2 diabetes include a reduced daily caloric intake for overweight and obese individuals, the preference of high-fiber and low-glycemic–index foods, and the avoidance of sugar, sweets, and sweetened beverages (3). In contrast, the ADA Standards of Medical Care in Diabetes is more comprehensive and includes, for example, a Mediterranean diet, higher intakes of nonstarchy vegetables, fruits, and whole grains, as well as dairy products, polyunsaturated fatty acids (PUFA), and a limited intake of saturated and trans-fatty acids (4). However, these dietary guidelines are mostly not evidence-based, refer to surrogate markers, such as glycemic control or lipids, or are based on findings from the general population. Thus, specific recommendations for individuals with type 2 diabetes are scarce. So far, individual studies have investigated dietary factors, such as the Mediterranean diet (5), intake of fruit (6), vegetables (7), or protein (8), in association to all-cause mortality in individuals with type 2 diabetes. Recently, a few meta-analyses investigating a single dietary exposure in association to all-cause mortality among individuals with type 2 diabetes have been published (9,10). However, the impact of different dietary factors on mortality in type 2 diabetes has not been comprehensively summarized, and the risk of bias in primary studies and certainty of evidence needs to be assessed by using validated tools.

Therefore, the aim of the present systematic review and meta-analysis was to summarize the currently available evidence on dietary factors, including dietary patterns, food groups, foods, macro- and micronutrients, and secondary plant compounds, and the risk of all-cause mortality in individuals with type 2 diabetes derived from prospective observational studies and evaluate the certainty of evidence of these associations.

This report was planned, conducted, and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement (11) and registered in PROSPERO (registration number: CRD42018110669). A detailed protocol of the project has been published previously (12), and in this first report, we focus on all-cause mortality. There were no deviations from the protocol.

A detailed description of the inclusion and exclusion criteria is shown in the Supplemental Material. Briefly, we included studies if the following criteria were met: 1) participants with type 2 diabetes aged ≥18 years; 2) any dietary factors, including dietary patterns, food groups, foods, macro- and micronutrients, and secondary plant compounds, supplements, and biomarkers of dietary intake; 3) all-cause mortality; and 4) prospective observational studies published in a peer-reviewed journal.

The systematic literature search was performed in PubMed and Web of Science using predetermined search terms. No filters (e.g., publication language) were applied. In this context, we did not identify any relevant articles that were not in English. The full search strategy is presented in Supplementary Table 1. The systematic literature search was lastly updated on 7 June 2022.

The whole study selection process was conducted by two researchers independently. Any disagreements between the two researchers were resolved by consensus or by consultation of a third researcher. Reference lists of eligible studies and relevant reviews were screened to identify additional relevant studies.

If there were multiple publications of the same study population, we included the most recent and comprehensive report including the largest sample size and/or cases and longest duration of follow-up.

One reviewer extracted the data from all identified studies using a predefined data extraction form, and a second reviewer checked the data for accuracy. The data extracted can be found in the Supplementary Material. We successfully contacted authors for relevant missing data (13–16). We excluded studies from the meta-analyses if we were unable to obtain relevant data.

We evaluated the potential risk of bias using the Risk Of Bias In Non-randomised Studies–of Interventions (ROBINS-I) tool (17). In brief, the tool is divided into seven domains: 1) bias due to confounding, 2) bias in selection of participants into the study, 3) bias in measurement of the exposure, 4) bias due to misclassification of exposure during follow-up, 5) bias due to missing data, 6) bias in measurement of outcomes, and 7) bias in selection of reported results. A detailed description of the instrument and justification of the rating is presented in Supplementary Table 2. Two reviewers independently assessed the risk of bias, and any discrepancies were discussed to reach a consensus.

The certainty of evidence for each dietary exposure in association to all-cause mortality was evaluated using the Grades of Recommendation, Assessment, Development, and Evaluation (GRADE) approach (18). The GRADE approach takes into account the within-study risk of bias, inconsistency, indirectness and imprecision between the studies, publication bias, magnitude of the effect, and dose-response relationship. The GRADE approach classifies the certainty of evidence in one of four levels: high, moderate, low, and very low. High certainty of evidence indicates that there is a high confidence in the effect estimate and that further research probably will not change the effect estimate, whereas a moderate certainty of evidence indicates a moderate confidence in the effect estimate and that further studies may change the effect estimate. Furthermore, a low certainty of evidence indicates low confidence in the effect estimate and that further studies will likely change the effect estimate, and very low certainty of evidence indicates that there is very limited and uncertain meta-evidence available. Two reviewers independently rated the certainty of evidence. Any disagreements between the two investigators were resolved by consensus.

To calculate summary risk ratios (SRRs) and corresponding 95% CIs, we conducted meta-analyses using a random-effects model when two or more studies on the same exposure and all-cause mortality were available (19). We summarized studies on dietary intake and studies on biomarkers of dietary intake separately. A detailed description of the statistical analysis is provided in the Supplemental Material. Briefly, we conducted dose-response meta-analyses whenever possible; otherwise, high versus low meta-analyses were conducted if data for dose-response meta-analyses were missing and no additional data could be obtained. We conducted subgroup analyses and meta-regression to investigate possible sources of heterogeneity across studies, such as sex, age, study length, and duration of diabetes, if at least 10 studies were available. Moreover, potential publication bias was investigated visually using funnel plots and the Egger test if at least 10 studies were available. A P value <0.1 indicated potential publication bias. All data analyses were performed using Stata 14.2 (StataCorp, College Station, TX) statistical software.

We identified 28,251 articles after removing duplicates (Fig. 1). After title and abstract screening, 1,028 articles were read in full text. Of these, 90 studies were eligible for inclusion. The list of included and excluded articles with reasons are provided in Supplementary Table 3. In addition, we identified 17 articles through reference screening. In total, 107 articles were included in the present systematic review, of which 76 studies investigated dietary intake, and 31 studies investigated biomarkers of dietary intake. One study on dietary vitamin B₁₂ supplements was found.

The characteristics and the respective references of all included studies are presented in Supplementary Table 4. There were 42 studies conducted in the U.S., 37 studies in Europe, 24 studies in Asia, 1 study in Australia, and 3 studies were conducted internationally. The mean follow-up duration was 9.9 (minimum 1–maximum 34) years. The mean sample size was 5,879 (minimum 131–maximum 47,422). Of the 76 studies investigating dietary intake, 53 used a food frequency questionnaire or dietary history questionnaire, 12 studies used 24-h dietary recalls, 1 study used a 3-day dietary record, and 10 studies used questionnaires/instruments that were not specified.

There were 21 studies judged as being at moderate risk of bias, and 87 studies were judged as being at serious risk of bias (Supplementary Fig. 1). Overall, ∼80% of the studies were rated as being at serious risk of bias in the confounding domain (Supplementary Fig. 2), indicating that most of the studies did not adjust for the most relevant confounders: Age was considered in 98% of the studies, sex in 94%, education in 65%, smoking in 84%, physical activity in 69%, and diabetes duration/severity in 45% only. Of the studies investigating dietary intake, 59% considered total energy intake in the analysis. More than 50% of the studies were rated as being at serious risk of bias due to the selection of the participants.

Of the 107 articles, 72 studies were included in the meta-analyses, and 35 articles could not be included in a meta-analysis because there were <2 studies available for the same exposure or due to missing data for a dose-response meta-analysis.

The results of the meta-analyses of dietary patterns and all-cause mortality are presented in Fig. 2 and Supplementary Fig. 3. We identified 20 studies investigating a dietary pattern, including the Mediterranean diet (5,20–22), Alternate Healthy Eating Index (23,24), Dietary Approaches to Stop Hypertension (DASH) diet (14,22), Dietary Inflammatory Index (25–27), low-carbohydrate high-protein diet (28,29), low-carbohydrate score (16,30), or other dietary indices and behaviors (31–36). The certainty of evidence was rated as very low for all findings (Supplementary Table 5).

The results of the meta-analyses on foods or food groups and all-cause mortality are presented in Fig. 3 and Supplementary Fig. 4. We found an inverse association between fish consumption and all-cause mortality (SRR for an increase of one serving/week: 0.95; 95% CI 0.92, 0.99; τ² < 0.001; prediction interval [PI] 0.90, 1.01; I² = 0%; n = 6 studies, moderate certainty of evidence) (24,37–41). There was no indication for nonlinearity (Supplementary Fig. 5). Moreover, moderate certainty of evidence was found for the association of whole-grain intake and all-cause mortality (SRR for an increment of 20 g/day: 0.84; 95% CI 0.71, 0.99; τ² < 0.001; I² = 0%; n = 2 studies) (24,42). Intake of vegetables was inversely associated (SRR per 100 g/day: 0.88; 95% CI 0.82, 0.94; τ² < 0.001; I² = 0%; n = 2 studies, low certainty of evidence) (7,24), and egg consumption was positively associated with all-cause mortality (SRR per 10 g/day: 1.05; 95% CI 1.02, 1.07; τ² < 0.001; PI 0.98, 1.12; I² = 56%; n = 7 studies, low certainty of evidence) (24,38,39,43–46).

We further conducted meta-analyses on intake of nuts, dairy, meat, sugar and sweets, tea, soft drinks and juices, showing no associations and rated with low certainty of evidence (Fig. 3). The certainty of evidence of the associations between intake of fruit, cereal, and coffee and all-cause mortality was very low.

Eleven studies (21,47–56) were available for coffee consumption (high vs. low meta-analysis); thus, we performed subgroup analyses and meta-regression (Supplementary Table 6). No differences were found with respect to sex, geographic location, duration of follow-up, number of cases, dietary assessment method, risk of bias, and adjustment for education, total energy intake, smoking, physical activity, or diabetes duration. There was an indication for publication bias according to the funnel plot (Supplementary Fig. 6) and the Egger test (P = 0.067), indicating that there might be a lack of studies with a null effect or positive association.

The results of the meta-analyses on energy and macronutrient intake and all-cause mortality are presented in Fig. 4 and Supplementary Fig. 7. Intake of fiber (SRR per 5 g/day: 0.86; 95% CI 0.81, 0.91; τ² < 0.001; PI 0.88, 1.06; I² = 0%; n = 3 studies) (42,57,58) and n-3 PUFA intake were inversely associated with all-cause mortality (SRR per 0.1 g/day: 0.87; 95% CI 0.82, 0.92; τ² < 0.001; I² = 0%; n = 2 studies) (37,59), rated with moderate certainty of evidence. Intake of monounsaturated fatty acids (38,59) tended to be inversely and animal protein (24,60) tended to be positively associated with all-cause mortality, but the estimates were imprecisely estimated (moderate certainty of evidence).

Furthermore, dietary cholesterol was positively (SRR per 300 mg/day: 1.19; 95% CI 1.12, 1.25; τ² < 0.001; I² = 0%; n = 2 studies) (45,46) and intake of plant protein was inversely associated with all-cause mortality (SRR per 10 g/day: 0.91; 95% CI 0.86, 0.95; τ² < 0.001; I² = 42%, n = 3) (13,24,60), both rated as low certainty of evidence. Higher intake of saturated fatty acids pointed to an increased relative risk, but the estimate was imprecisely estimated, and the certainty of evidence was low. No association was found for carbohydrate intake.

Very low certainty of evidence was found for the intake of total energy, total PUFA, and protein.

The results of the substitution analyses are shown in Supplementary Fig. 8. Moderate certainty of evidence was observed for the replacement of 2% energy from carbohydrates by plant protein inversely associated with all-cause mortality (SRR 0.76; 95% CI 0.66, 0.87; I² = 0%; n = 2 studies) (30,61), and for the substitution of 2% energy from carbohydrates with saturated fatty acids positively associated with all-cause mortality (SRR 1.10; 95% CI 1.04, 1.16; I² = 0%; n = 2 studies) (30,62). No association was found for the substitution of carbohydrates with PUFA, monounsaturated fatty acids, and animal protein, graded with low to moderate certainty of evidence (Supplementary Table 5).

The results of the meta-analyses on the intake of micronutrients and secondary plant compounds and relative risk of all-cause mortality are presented in Fig. 5 and Supplementary Fig. 9. We found no association for caffeine consumption (53,63) and serum folate (64–66) (moderate and low certainty of evidence, respectively).

Very low certainty of evidence was found for all other associations: sodium intake, urinary sodium, serum 25(OH) vitamin D, serum calcium, serum phosphorus, plasma/serum phylloquinone, and serum vitamin B₁₂.

Alcohol intake was investigated in 13 studies (21,23,24,31,32,38,67–73). Seven studies could be used for high consumption versus no consumption (Supplementary Fig. 10A) and three studies for nonlinear dose-response meta-analysis (Supplementary Fig. 10B). The certainty of evidence was very low for alcohol intake and all-cause mortality.

Several dietary factors have been investigated only in one study, and, thus, a meta-analysis was not possible. Lower retinol levels were associated with higher relative risk of all-cause mortality (74). In agreement with dietary intake data, plasma docosahexaenoic acid and plasma n-3 PUFA were inversely associated with all-cause mortality, whereas no association was found for plasma monounsaturated fatty acids, saturated fatty acids, n-6 fatty acids, and linoleic acid (75). Moreover, inverse associations were found for selenium intake, green tea consumption (49), and intake of legumes (7) and all-cause mortality. Red blood cell folate (76) and choline intake from phosphatidylcholine (77) were positively associated with all-cause mortality.

No association was found for glycemic index, glycemic load, sugar or starch (57), and for a dietary pattern based on high intakes of “olive oil and vegetables,” “eggs and sweets,” and “pasta and meat” regarding all-cause mortality (5). Moreover, there was no association between α-tocopherol levels (78), serum ferritin (79), serum vitamin E (80), and all-cause mortality in individuals with type 2 diabetes.

In the present systematic review and meta-analysis, we provided a comprehensive overview of 107 studies investigating dietary factors and all-cause mortality in individuals with type 2 diabetes. We were able to conduct 45 meta-analyses. Higher intakes of fish, whole grain, fiber, and n-3 PUFA were inversely associated with the relative risk of all-cause mortality, rated as moderate certainty of evidence. Moreover, higher intakes of vegetable and plant protein may be inversely associated with all-cause mortality in individuals with type 2 diabetes, rated as low certainty of evidence. In contrast, higher intakes of eggs and cholesterol may be positively associated with all-cause mortality (low certainty of evidence). For other dietary factors, no association was found and/or the evidence was very uncertain.

A large body of evidence exists with regard to the prevention of type 2 diabetes (2). In addition, many meta-analyses on dietary factors and risk of all-cause mortality in the general population have been published. However, little is known on the secondary prevention of comorbidities and mortality in individuals with type 2 diabetes.

Our meta-analysis showed that higher intake of fish was associated with reduced relative risk of all-cause mortality in individuals with type 2 diabetes, which is comparable to findings of the general population (81). In addition, Jayedi et al. (10) also conducted a meta-analysis on fish consumption and mortality in individuals with type 2 diabetes and also found an inverse association as in the present meta-analysis. However, the authors included only three studies in the linear dose-response meta-analysis, and the SRR was somewhat lower (SRR 0.91 vs. 0.95).

Furthermore, we found that whole-grain and fiber intake were associated with reduced relative risk of all-cause mortality in individuals with type 2 diabetes, which is also comparable to meta-findings in the general population (82,83). Interestingly, the associations seem to be a stronger among individuals with type 2 diabetes compared with the general healthy population. Whole-grain products are rich in fiber, which has been shown to improve glycemic control, body weight control, and blood lipids in individuals with diabetes (84). Thus, fiber is an important dietary compound for diabetes management to prevent cardiovascular complications and mortality.

A meta-analysis of four prospective studies showed that an increment of 0.2 g of n-3 PUFA per day decreased the relative risk of all-cause mortality by 7% (81). With regard to our meta-analysis, the associations seem to be stronger in individuals with type 2 diabetes. As fish is an important source of n-3 PUFA, this result is in agreement with the meta-findings on fish consumption and mortality. Long-chain n-3 PUFAs have been hypothesized to have several beneficial effects on the vascular system, foremost antihypertensive, anti-inflammatory effects, and improved lipid profile (85), and, thus, it is possible that individuals with diabetes can especially benefit from n-3 PUFA intake.

In a meta-analysis based on 22 cohort studies, a dietary increment of 200 g of vegetables per day was associated with a 13% decreased risk of all-cause mortality in the general population (86). Compared with our meta-analysis, the risk estimate is quite similar, although we used a dose of 100 g/day. Moreover, in agreement with our findings, higher intake of plant protein has been shown to be inversely associated with all-cause mortality in the general population (87).

A meta-analysis of prospective cohort studies found a positive association of higher intake of eggs and cholesterol with all-cause mortality in the general population (88). One egg per day was associated with a relative risk of 1.07, indicating that there might be a weaker association in the general population compared with individuals with type 2 diabetes.

The Mediterranean diet is rich in many dietary factors mentioned above, and an inverse association was observed in the present meta-analysis. However, the certainty of evidence was very low, mainly due to serious risk of bias in the primary studies. For many other dietary factors, no association was found (e.g., meat, sugar and sweets, animal protein), or the association was graded with very low certainty of evidence (e.g., fruit and coffee consumption). Explanations for these findings could be the small number of studies for these meta-analyses, the high heterogeneity between studies, and the risk of bias of the studies.

The major strength of the present systematic review and meta-analysis is the comprehensive overview of the currently available evidence of dietary factors and all-cause mortality among individuals with type 2 diabetes. We assessed the risk of bias in all included studies and evaluated the certainty of evidence for all associations by applying validated tools. Moreover, we included only prospective observational studies, which reduces sources of bias, such as recall bias, and conducted linear and nonlinear dose-response analyses whenever possible.

Limitations of the present work include the serious risk of bias evaluated for most studies included in the meta-analyses. The main body of evidence in the current study came from studies at serious risk of bias, mainly due to uncontrolled confounding and selection of the participants into the study. More than half of the studies did not adjust for diabetes duration, and 41% of the studies investigating dietary factors did not consider total energy intake in the analysis. Moreover, as we included observational studies, residual confounding cannot be ruled out.

Dietary intake was assessed by self-reports in most studies. Thus, potential misclassification of the exposure cannot be ruled out. Biomarkers of dietary intake were used in 31 studies (29%), which may be a more objective measurement of dietary exposure. However, biomarkers are not available for all dietary factors. Moreover, since the exposure was measured only at baseline in most studies, changes in dietary behavior during follow-up could not be taken into account. However, large studies, such as the Nurses’ Health Study and Health Professional Follow-up Study, took dietary changes into account, and the risk estimates were similar to the overall findings of the meta-analyses.

Furthermore, in several studies, it was not clear whether only participants with type 2 diabetes were included or whether also participants with type 1 diabetes were included. However, as the proportion of type 1 diabetes is small compared with type 2 diabetes, we expect that the number of participants with type 1 diabetes is very low in the included studies.

Owing to the small number of studies in many meta-analyses, the effect estimates were mostly imprecisely estimated (wide 95% CIs), and, thus, drawing conclusions may be limited. Furthermore, we were not able to conduct subgroup analyses and investigate publication bias except for the association between coffee consumption and all-cause mortality. Moreover, we were not able to conduct stratified analyses for subgroups of individuals with diabetes, such as those with a different disease status, taking different treatments, or with comorbidities. Some individuals may respond differentially to dietary intake, and thus, associations might differ between subgroups of individuals with diabetes.

The main reason for downgrading of the certainty of evidence was the serious risk of bias evaluated for the main part of the studies, resulting in many associations with low or very low certainty of evidence. Thus, future studies should take into account important confounders, such as diabetes duration, diabetes treatment, socioeconomic status, and total energy intake, to strengthen the evidence. Furthermore, some studies were not primarily designed for our research question, and in most studies, prevalent cases of diabetes were included. So, future studies should be designed to include individuals with newly diagnosed type 2 diabetes to minimize selection bias. Moreover, future studies should use validated dietary assessment methods and consider changes of dietary behaviors during follow-up to increase the reliability of the exposure assessment and thus, increase the certainty of evidence.

This is the first systematic review and meta-analysis that comprehensively summarizes the currently available evidence on dietary factors and all-cause mortality in individuals with type 2 diabetes. We could show that higher intakes of fish, whole grain, fiber, and n-3 PUFA were inversely associated with all-cause mortality. Moreover, evidence suggests that higher intakes of vegetable and plant protein may be associated with all-cause mortality, whereas higher intakes of eggs and cholesterol may be positively associated with all-cause mortality in individuals with type 2 diabetes. More primary studies are needed to provide robust and comprehensive evidence on dietary factors and the progression of diabetes.

Funding. The German Diabetes Center (DDZ) is funded by the German Federal Ministry of Health and the Ministry of Science and Culture of the State North Rhine-Westphalia. This study was supported in part by a grant from the German Federal Ministry of Education and Research to the German Center for Diabetes Research (DZD).

The funders had no role in study design or data collection, analysis, and interpretation.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.Ba. conducted the statistical analyses. J.Ba., A.L., E.S., C.B., J.Be., L.S., M.N., and S.S. critically reviewed and approved submission of the final manuscript. J.Ba., A.L., E.S., J.Be., M.N., and S.S. conducted the systematic literature search. J.Ba., A.L., J.Be., M.N., and S.S. were involved in data acquisition. J.Ba., E.S., C.B., L.S., and S.S. conducted the assessment of risk of bias. J.Ba. and L.S. rated the certainty of evidence. J.Ba. and S.S. designed the study question and developed the search term of the systematic review and meta-analysis. J.Ba., and S.S. interpreted the results. J.Ba. and S.S. drafted the first version of the manuscript. S.S. assisted with the statistical analysis. J.Ba. and S.S. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented as an abstract at the 58th Annual Meeting of the European Association for the Study of Diabetes, Stockholm, Sweden, 19–23 September 2022. The abstract was published in Diabetologia 2022;65(Suppl. 1):1–469.