Discussion

The present study is the first to shed light on the possible association between dietary intake of fat-soluble vitamins and the risk of type 2 diabetes in a large cohort of Asians. Findings indicated that dietary intake of both vitamins K and E but not dietary intake of vitamin A or D were associated inversely with diabetes risk. These associations persisted even after controlling for traditional diabetes risk factors and other dietary variables and did not vary by different levels of a wide range of participants’ characteristics.

The antioxidant properties of vitamins A and E were the main interest of the research that investigated the impact of dietary and supplemental consumptions of these vitamins on the risk of diabetes⁽ Reference Montonen, Knekt and Järvinen ^{8 )}. Our findings indicated that high dietary intake of vitamin E, but not vitamin A, could benefit by reducing the risk of diabetes in our study subjects. The evidence from the scant previous literature showed inconsistent results; while dietary intake of both vitamin E (represented by α-tocopherol) and vitamin A (represented by β-carotene) were inversely associated with the risk of diabetes in Swedish men⁽ Reference Arnlöv, Zethelius and Risérus ^{6 )}, neither dietary intake of most of the carotenoids in Dutch⁽ Reference Sluijs, Cadier and Beulens ^{7 )} or Finnish⁽ Reference Montonen, Knekt and Järvinen ^{8 )}, nor dietary intake of vitamin E in Chinese⁽ Reference Zhou, Na and Shan ^{9 )} showed any association with the risk of diabetes. Moreover, while no association was found for dietary α- and β-carotenes, and only dietary β-cryptoxanthin was the most significant single carotenoid to inversely associate with the risk of diabetes in the Finnish study (relative risk 0·58, 95 % CI 0·44, 0·78)⁽ Reference Montonen, Knekt and Järvinen ^{8 )}, dietary consumptions of α- and β-carotenes (relative risk 0·78, 95 % CI 0·64, 0·95 and 0·85, 95 % CI 0·70, 1·03, respectively), but not β-cryptoxanthin, were inversely associated with the diabetes risk in the Dutch study⁽ Reference Sluijs, Cadier and Beulens ^{7 )}. Going further for discrepancies, another Finnish study reported an increased risk of diabetes with high dietary vitamins A and E consumptions among male smokers; however, the association disappeared after multivariable adjustment⁽ Reference Kataja-Tuomola, Kontto and Männistö ^{21 )}.

The exact mechanisms of why the reduced risk of diabetes is swinging within consumptions of dietary vitamins A and E in different populations are unclear. The sample size, participants’ characteristics and diagnostic criteria of diabetes are hypothesised justifications⁽ Reference Arnlöv, Zethelius and Risérus ^{6 –} Reference Zhou, Na and Shan ^{9 ,} Reference Kataja-Tuomola, Kontto and Männistö ^{21 )}. For example, the Sweden study assessed the dietary intake by dietary records among older men aged ≥70 years and diabetes status was retrieved from hospital registries⁽ Reference Arnlöv, Zethelius and Risérus ^{6 )}. The Finnish studies used FFQ for diet assessment among men and women aged 40–69 years⁽ Reference Montonen, Knekt and Järvinen ^{8 )} or male smokers aged 50–69 years⁽ Reference Kataja-Tuomola, Kontto and Männistö ^{21 )} and used the drug reimbursement register of the Finnish health insurance for diagnosing diabetes. The participants of the Chinese study aged 20–74 years and diabetes was diagnosed by an oral glucose tolerance test in some participants and by HbA1c levels in other participants⁽ Reference Zhou, Na and Shan ^{9 )}. In general, these results suggest other mechanisms besides the antioxidant effects of vitamins A and E are involved in reducing the risk of diabetes.

Ibarrola-Jurado et al. ⁽ Reference Ibarrola-Jurado, Salas-Salvadó and Martínez-González ^{10 )} followed 1069 diabetes-free Spanish individuals at high risk of CVD, who had ≥3 of the following risk factors: smoking, hypertension, dyslipidaemia, obesity or a family history of coronary artery disease for 5·5 years and reported that dietary consumption of vitamin K (phylloquinone) showed no association with fasting plasma glucose concentrations at baseline, but at the end of follow-up the risk of type 2 diabetes was reduced by 17 % with each additional daily dietary intake of 100 µg phylloquinone. Juanola-Falgarona et al. ⁽ Reference Juanola-Falgarona, Salas-Salvadó and Estruch ^{22 ,} Reference Reddi, Henderson and Meghji ^{23 )} confirmed these observations in their cross-sectional and longitudinal assessments among Spanish population. Following 38 094 Dutch men and women, aged 20–70 years for 10 years, yielded a multivariable hazard ratio for the risk of type 2 diabetes in the highest v. lowest quartiles of phylloquinone and menaquinones intake of 0·81 (95 % CI 0·66, 0·99) and 0·80 (95 % CI 0·62, 1·02), respectively⁽ Reference Beulens, van der A and Grobbee ^{11 )}. However, phylloquinone supplementation was not associated with the risk of diabetes in a systematic review that included studies investigating the relationship of intake and serum levels of vitamin K with risks of type 2 diabetes and CVD. However, this conclusion was based on the findings from only one clinical trial of vitamin K supplementation⁽ Reference Rees, Guraewal and Wong ^{12 )}. So far, there is no clear biological mechanism for the inverse association between dietary vitamin K intake and the risk of type 2 diabetes observed in our study and previous studies⁽ Reference Ibarrola-Jurado, Salas-Salvadó and Martínez-González ^{10 ,} Reference Beulens, van der A and Grobbee ^{11 ,} Reference Juanola-Falgarona, Salas-Salvadó and Estruch ^{22 )}. However, suggested pathways include the modulating effects of vitamin K on cytokines to decrease inflammation and improve insulin sensitivity⁽ Reference Reddi, Henderson and Meghji ^{23 )}, and the vitamin-K-dependent biosynthesis of osteocalcin, a bone-specific protein that has regulating effects on both the expression of insulin genes⁽ Reference Ferron, Hinoi and Karsenty ^{24 )} and Ca metabolism⁽ Reference Beulens, van der A and Grobbee ^{11 )} that can reduce insulin resistance.

The last fat-soluble vitamin to be discussed (vitamin D) has gained a huge attention in the research area in the past decade⁽ Reference Valdés-Ramos, Guadarrama-López and Martínez-Carrillo ^{5 ,} Reference Liu, Song and Ford ^{13 –} Reference Poolsup, Suksomboon and Plordplong ^{16 )}. The rapidly increasing evidence postulates the impact of vitamin D on diabetes risk through its capacity to stimulate the expression of insulin receptors and to unite with the pancreatic β cells’ receptors enhancing insulin sensitivity⁽ Reference Valdés-Ramos, Guadarrama-López and Martínez-Carrillo ^{5 )}. Systematic reviews showed that some but not all clinical studies that examined the effect of 10–5000 μg vitamin D supplements’ use for period of 2 months to 7 years have proven a modest favourable effect on impaired type 2 diabetes indicators such as fasting blood glucose and insulin sensitivity but not on glycated Hb⁽ Reference Valdés-Ramos, Guadarrama-López and Martínez-Carrillo ^{5 ,} Reference George, Pearson and Witham ^{15 ,} Reference Poolsup, Suksomboon and Plordplong ^{16 )}. However, the up-to-date studies that reported the association of dietary vitamin D and risk of type 2 diabetes showed inconsistent findings. Inverse associations were reported with the metabolic syndrome in the Women’s Health Study⁽ Reference Liu, Song and Ford ^{13 )} but not with the risk of incident diabetes in the EPIC-InterAct study⁽ Reference Abbas, Linseisen and Rohrmann ^{14 )}. In our analysis, dietary vitamin D had no obvious association with the risk of type 2 diabetes.

We ran several sensitivity analyses to explore whether the reported association will be modified by varying levels of hypothesised effect modifiers; however, interactions of these factors with dietary vitamin D towards the risk of diabetes were not significant. Of a special interest was the effect modification of smoking status on the association between antioxidants and risk of diabetes. In fact, smoking not only increases the oxidative stress but also reduces serum tocopherols and carotenoid levels⁽ Reference Gabriel, Liu and Crott ^{25 )}. Kataja-Tuomola et al. ⁽ Reference Kataja-Tuomola, Kontto and Männistö ^{21 )} found no beneficial effects of dietary antioxidants consumption, including tocopherols and carotenoids, on the risk of diabetes in middle-aged male smokers. However, similar to our findings, there was no interaction with smoking status, which was proven in Dutch⁽ Reference Sluijs, Cadier and Beulens ^{7 )} or Finnish studies⁽ Reference Montonen, Knekt and Järvinen ^{8 )}.

Our study, however, is not without limitations. The self-report of the outcome variable is one limitation; however, comparison with treatment and laboratory data showed very high sex-specific sensitivity and specificity for the self-reported physician-diagnosed diabetes in our participants⁽ Reference Eshak, Iso and Maruyama ^{4 ,} Reference Eshak, Iso and Yamagishi ^{19 )}. Second, the dietary records-based validation study indicated that the FFQ underestimated participants’ intake of fat-soluble vitamins by at least 20 %⁽ Reference Eshak, Iso and Yamagishi ^{19 )}. In this regard, we have to mention that the JACC FFQ did not include information about one important source of vitamin K in Japanese diet, natto (fermented soyabean). However, in general, the reproducibility of the FFQ was satisfactory and it was a good tool to sort individuals into the corresponding quartiles of dietary intake. Third, 64 % of participants did not report their diabetes status 5 years from the baseline and accordingly our findings were based on data from 36 % of eligible study participants. However, no significant differences in participants’ characteristics such as age, BMI, dietary intakes of fat-soluble vitamins and other variables were found between individuals with and without missing data of diabetes history at 5 years from the baseline. Fourth, we do not have data on supplementation with fat-soluble vitamins, except for vitamin E which was reported by 2·6 % of our study participants. Our baseline was in the late 80s, where Japanese population was unfamiliar with such supplementations; moreover, sensitivity analyses that excluded those supplement users did not alter the associations materially (data not shown). Last, we controlled for a wide range of hypothesised confounders; yet, the possibility of residual confounding remains.