2.1 Study population

The data of the current work was from eight centers of the REACTION (Risk Evaluation of cAncers in Chinese diabeTic Individuals) study.¹⁵ A total of 53 639 participants aged over 40 or older participated in the study between March and December 2012. A total of 3042 participants with preexisting kidney disease and other serious illnesses, 2231 participants using lipid-lowering drugs, 11 280 participants with missing important data, and 402 participants with calculated RC values ≤0 were excluded. Finally, 36 684 participants were enrolled.

2.2 Data collection

Trained investigators helped participants fill out detailed questionnaires including demographic data, past medical history, family history of diabetes, current medication, current smoking and drinking habits, current occupation, and physical activity. Information on physical activity was gathered using the short form of the International Physical Activity Questionnaire: high-level physical activity was classified as vigorous physical activity that lasted longer than 10 min each within the past 7 days, which would make people breath with difficulty, for example, playing basketball, swimming, and running; moderate-level physical activity was defined as a minimum of 10 min of physical activity in each of the past 7 days, which required people to breathe a little harder than normally, for instance jogging, playing table tennis, golf, or tai chi but no vigorous physical activity; low physical activity levels were determined as no physical activity or only some physical activity like walking, but could not meet the criteria for high- and medium-level physical activity.

Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times, each 5 min apart. The average of the three results was taken for statistical analysis. The height, weight, waist circumference (WC), and hip circumference were measured and recorded after the participants removed jackets and shoes.

The participants fasted for 8–10 h before the assay. Serum TG, low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), total cholesterol (TC), aspartate transferase (AST), alanine transferase (ALT), glutamine transferase (GGT), serum creatinine (SCr), and other biochemical indexes were measured on an autoanalyzer; hemoglobin A1c (HbA1c) was measured by high-performance liquid chromatography method; Fasting insulin was measured with chemiluminescent immunoassay. After fasting blood glucose extraction, participants without or with diabetes were tested for 75 g oral glucose tolerance or 100 g steamed-bread meal, respectively, post-load blood glucose (PBG) was extracted 2 h after the first blood sample was drawn.

2.3 Calculations

RC = TC−HDL-C−LDL-C.¹⁶ Body mass index (BMI) (kg/m²) = weight (Kg)/height² (m). Waist-to-hip ratio (WHR) = WC/HC. eGFR (mL/min per 1.73 m²) = 175 × (SCr in mg/dL) −1.154 × age−0.203 × (0.742 for women) × (1.212 if African American).

2.4 Definitions

RC was divided into four groups based on quartiles: <25% group (the control group), 25%–50% group, 50%–75% group, and ≥75% group. Diabetes was defined according to the American Diabetes Association's criteria¹⁷: FPG ≥7.0 mmol/L or PBG ≥11.1 mmol/L or HbA1c ≥6.5% or self-reported diabetes.

2.5 Statistical analysis

Statistical analysis was performed using SPSS 24.0 (IBM, Chicago, IL). Continuous variables were expressed as mean ± SD and categorical variables were presented numerically (proportionally). One-way analysis of variance (ANOVA) was used to test the difference between continuous variables and the chi-square test was used to analyze the categorical variables. The association between RC and diabetes was delineated by logistic regression analyses. Adjusted variables were examined using collinearity diagnosis according to the following criteria: (1) variance inflation factor >5; (2) condition index >30; and (3) variance proportions >50%. The adjusted variables had the following criteria: (1) in the current study, there was a significant difference between the diabetic group and the nondiabetic group, or it was significant for the occurrence and development of diabetes in clinical practice; and (2) there was no collinearity. Model 0 was unadjusted. Model 1 was adjusted for age, sex, and center. Model 2: Model 1 + myocardial infarction (MI), stroke, coronary heart disease (CHD), occupation, smoking habit, drinking habit, physical activity level, and family history of diabetes. Model 3 = Model 2 + ALT, AST, GGT, and eGFR. Model 4 = Model 3 + SBP, DBP, BMI, WHR, and insulin. Model 5 = Model 4 + HDL-C and LDL-C. Model 6 = Model 5 + TG. Moreover, according to the appropriate levels of LDL-C, HDL-C, and TG recommended by the 2016 edition of the guidelines for the management of dyslipidemia in Chinese adults,¹⁸ the association between RC and diabetes was also tested when LDL-C <3.4 mmol/L, HDL-C ≥1.0 mmol/L, and TG <1.7 mmol/L, respectively. In addition, the current study explored the relationship between RC and diabetes in BMI, BP, age, and eGFR subgroups. All statistical tests were two-sided and p < .05 was considered statistically significant.

3.1 Clinical characteristics of the study population

The present study included 36 684 participants (Table 1), with 11 321 (30.9%) males and 25 363 (69.1%) females, and their median age (Q1–Q3) was 57 (52, 64). Compared with participants without diabetes, participants with diabetes had older age, higher RC, LDL-C, TC, TG, BMI, WHR, ALT, AST, GGT, SBP, and DBP and lower HDL-C and eGFR. Meanwhile, males, frequent smoking, MI, stroke, CHD, manual labor, retirement or unemployment, and a family history of diabetes were more common in participants with diabetes than in participants without diabetes.

3.2 Association between RC and diabetes

Table 2 shows the association between RC and diabetes after potential confounders were adjusted and separated by sex. In the total participants, RC exhibited a significant adjusted odds ratio (OR) in models 0–6. After all potential confounders were adjusted in model 6, the third and fourth quartile of RC were still significantly associated with diabetes (Q3: OR: 1.132, 95% CI: 1.047–1.223, p = .002; Q4: OR:1.147, 95% CI: 1.049–1.254, p = .003). In the female participants, RC remained significantly associated with diabetes though all potential confounders were adjusted in model 6 (Q3: OR: 1.168, 95% CI: 1.057–1.291, p = .002; Q4: OR: 1.228, 95% CI: 1.097–1.975, p < .001). In the male participants, RC and diabetes were significantly associated in model 5 (Q3: OR: 1.133, 95% CI: 1.002–1.282, p = .046; Q4: OR: 1.144, 95% CI: 1.010–1.295, p = .034), however, after TG was adjusted in Model 6, the association between RC and diabetes no longer existed (Q4: OR: 0.985, 95% CI: 0.848–1.145, p = .846).

3.3 Association between RC and diabetes in participants with TG <1.7 mmol/L, LDL-C < 3.4 mmol/L, or HDL-C ≥1.0 mmol/L

In Table 3, RC and diabetes were still significantly associated when TG <1.7 mmol/L (Q4: OR: 1.155, 95% CI: 1.005–1.327, p = .042), LDL-C <3.4 mmol/L (Q2: OR: 1.103, 95% CI: 1.008–1.208, p = .034; Q3: OR: 1.151, 95% CI: 1.047–1.264, p = .003; Q4: OR: 1.130, 95% CI: 1.011–1.264, p = .032), or HDL-C ≥1.0 mmol/L (Q3: OR: 1.140, 95% CI: 1.045–1.243, p = .003; Q4: OR: 1.116, 95% CI: 1.007–1.237, p = .037).

3.4 Stratification analysis

The results of the stratification analysis were presented in Table 4. RC was significantly associated with diabetes regardless of whether participants were overweight or not (BMI <24 Kg/m²: Q3: OR: 1.148, 95% CI: 1.015–1.297, p = .028; Q4: OR: 1.175, 95% CI: 1.013–1.364, p = .033; BMI ≥24 Kg/m²: Q3: OR: 1.124, 95% CI: 1.016–1.243, p = .024; Q4: OR: 1.125, 95% CI: 1.006–1.259, p = .033). RC and diabetes were significantly correlated when SBP <140 mm Hg and DBP <90 mm Hg (Q3: OR: 1.126, 95% CI: 1.010–1.255, p = .032; Q4: OR: 1.179, 95% CI: 1.039–1.338, p = .011), but there was no correlation when SBP ≥140 mm Hg or DBP ≥90 mm Hg. For participants <55 years of age, RC was significantly associated with diabetes (Q3: OR: 1.238, 95% CI: 1.047–1.427, p = .003; Q4: OR: 1.322, 95% CI: 1.124–1.555, p = .001). However, there was no association between RC and diabetes among participants aged 55–65 years or ≥65 years. When 60 ≤ eGFR ≤90 ml/min per 1.73 m², RC was significantly related to diabetes (Q3: OR: 1.170, 95% CI: 1.044–1.311, p = .007; Q4: OR: 1.211, 95% CI: 1.067–1.374, p = .003), whereas RC and diabetes were not associated when eGFR ≥90 ml/min per 1.73 m² or eGFR <60 ml/min per 1.73 m².

4.1 Main findings

The present study's findings suggested a significant association between RC and diabetes in the general Chinese population. In addition, RC was still associated with diabetes even when TG, LDL-C, or HDL-C was at the appropriate level recommended by guidelines. Moreover, people with elevated RC were more likely to develop diabetes, especially females, those with normal blood pressure, those with an eGFR between 60 and 90 ml/min per 1.73 m², and those younger than 55 years of age. Hence, people with these characteristics should pay more attention to monitoring RC values and timely correction of risk factors to prevent the occurrence of diabetes. To the best of our knowledge, this is the first large sample, multicenter study to investigate the relationship between RC and diabetes in the general population.

4.2 Previous studies

Studies on the relationship between RC and diabetes are scarce and have some minor limitations. A single-center study of 15 464 Japanese found that RC could predict diabetes.¹⁹ However, their study did not measure PBG, so their definition of diabetes did not include PBG≥11.1, the definition of diabetes in the current multicenter study strictly followed the American Diabetes Association's criteria, making it more accurate and scientific, and the current study was multicenter and fully adjusted for confounders. Szili-Torok et al²⁰ analyzed 480 patients who had undergone kidney transplantation, and they found a significant association between RC and new-onset diabetes after kidney transplantation. Zhou et al²¹ found a positive correlation between RC and diabetes in 13 721 Chinese hypertensive patients. Nevertheless, the current study population was the large sample general Chinese population, including subjects with various chronic diseases and healthy subjects; hence, the findings of the current study were more generally applicable.

4.3 Potential mechanisms

The reason for the association between RC and diabetes remains unclear. Atherosclerosis might be a major contributor. Numerous investigations have shown a strong link between atherosclerosis and hyperglycemia.^22-24 Elevated RC levels aided in the atherosclerosis progression,^{10, 25-27} which causes hyperglycemia. RC could be penetrable to the arterial wall, be phagocytosed by macrophages, and result in the formation of foam cells. Furthermore, RC speeds up the creation of foam cells via upregulating scavenger receptor expressions.²⁶ Atherosclerosis lowers blood circulation to the pancreas and impairs pancreatic function, which reduces insulin secretion levels and causes hyperglycemia.²⁸ Atherosclerosis also causes liver dysfunction and reduces glycogen production in the liver, thus leading to hyperglycemia.²⁹

Insulin resistance (IR) could potentially account for the link between RC and diabetes. In a study of 86 nonobese Japanese diabetics, researchers found that the IR group had higher levels of RC than the insulin-sensitive group.³⁰ Funada et al discovered that RC could forecast IR independently.³¹ A recent study found that empagliflozin reduced RC in diabetic patients, which was highly related to a reduction in IR.³² Besides, RC could directly disrupt pancreas islet β-cells' function, resulting in decreased insulin secretion.³³

4.4 The association between RC and diabetes when TG, LDL-C, or HDL-C is within the appropriate range recommended by the guidelines

“Atherosclerotic lipid triad,” characterized by elevated levels of TG and LDL-C and reduced levels of HDL-C, is thought to predict diabetes risk.¹⁹ Nevertheless, in the current study, when TG was <1.7 mmol/L, LDL-C <3.4 mmol/L, or HDL-C ≥1.0 mmol/L, the association between RC and diabetes remained significant. Because many human cells could degrade TG, but none could degrade cholesterol, the content of cholesterol in the residue (RC) is the primary cause of atherosclerosis.²⁷ An Icelandic study has shown that the detrimental effects of a TG-elevating gene variant are mediated by atherogenic effects of the cholesterol component of RC.³⁴ On the other hand, increased RC concentrations have been reported to lead to atherosclerosis risk even in individuals with normal TG levels.³⁵ As our results show, even if TG levels are appropriate, elevated RC and the corresponding risk of diabetes could not be ignored.

Varbo et al found that elevated LDL-C could generate atherosclerosis, but not inflammation, and the inflammatory component of atherosclerosis was driven by elevated RC.²⁷ Despite lowering LDL-C to suggested levels, there is a significant adverse residual risk of atherosclerosis,³⁶ which could be explained by the link between RC and inflammation. In addition, because RC is ingested in an uncontrolled way through scavenger receptors, RC causes greater damage to β cells than LDL-C.²⁰

It has been demonstrated that there is no causality between low HDL-C and atherosclerotic incidents, and low HDL-C is merely a strong symbol of elevated RC.³⁷ After getting into the intima, HDL particles could penetrate the media and leave the arterial wall,^{13, 38} but RC might be trapped in the intima and generate the foam cells, causing regional injury and inflammation.^{13, 38} Moreover, TGRL promotes the remodeling of HDL-C into smaller, cholesterol-poor particles that might lack atherosclerotic protection regardless of HDL-C levels.³⁹ Combined with the results of the current study, we recommend that people whose TG, LDL-C, or HDL-C levels is at the appropriate level recommended by the guidelines should also be vigilant against the increased risk of atherosclerosis and diabetes associated with elevated RC.

4.5 Gender difference

The gender difference in older populations might be due in part to the gender-specific hormones. Our participants' median age (Q1–Q3) was 57 (52, 64), indicating that the women in the current study were either perimenopausal or postmenopausal. The atherogenic effect of RC is more pronounced in women at this stage.⁴⁰ Ossewaarde et al observed a 15% reduction in RC levels in a group of healthy Dutch postmenopausal women given estrogen + progestin combination therapy for 3 months.⁴¹ Hence, perimenopausal and postmenopausal women are more prone to develop RC abnormalities owing to hormonal changes, increasing their risk of atherosclerosis and diabetes.

4.6 Stratification analysis

Previous studies demonstrated that the causality between RC and low-level inflammation still existed even in participants who were neither overweight or obese,²⁷ suggesting that the relationship between elevated RC and low-grade inflammation is not attributable to weight gain. The present study also found that RC and diabetes remained strongly associated regardless of whether the participant was overweight or not. Zhou et al²¹ found a positive correlation between RC and diabetes in 13 721 hypertensive patients; however, RC and diabetes were associated only in persons with normal blood pressure in the current investigation. We speculate that this is because the number of hypertension patients in the current research was modest in comparison to Zhou's team's investigation. Further validation of the relationship between RC and diabetes in a larger sample of hypertensive patients is needed in the future. We found the correlation was just significant in participants younger than 55 years old, not in older participants. We hypothesize that older adults tend to maintain healthy lifestyle habits and possess better compliance, which might help to a prevention of dyslipidemia and diabetes. Prior literature has demonstrated a U-type relation between eGFR and all-cause mortality,⁴² highlighting the significance of both high and low eGFR. The current study found a correlation between RC and diabetes only in participants with 60 ≤ eGFR ≤90 ml/min per 1.73 m². We consider that the damage caused by low or high eGFR might cover up the pathogenicity of elevated RC.

4.7 Limitations

The current study benefited from a vast collection of multiple community-based samples, broadly representative of most Chinese people. Meanwhile, the current study also comprehensively adjusted for major risk factors and conducted a detailed stratification analysis. However, some limitations still remained. First, VLDL residues are a major component of RC during fasting, so the role of chylomicron residues may be underestimated. Second, the current study was cross-sectional, so we could we could make only correlational inferences, not causal ones.