This was a cross-sectional study. We included 495 patients, who were admitted at the National Metabolic Management Center of Jining First People’s Hospital Affiliated to Shandong First Medical University from May 2020 to April 2022. All participants included in this study were adults (age ≥18 years) and diagnosed with T2DM. The study protocol was approved by the Committee on Human Research of the Jining NO.1 People’s Hospital (2021 Ethical Approval No. 025). All participants provided written informed consent. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology reporting guideline and was conducted in accordance with the declaration of Helsinki.

Demographics, medical history, and medication use were collected using a standardized questionnaire through face-to-face interviews. Body height and weight were measured using height and weight meters (OMRON HNH-318; OMRON Corporation, Shenzhen, China). Blood pressure was measured using an electronic blood pressure meter (OMRON HBP-1100U; OMRON Corporation, Dalian, China) in a seated position after 5 min of rest.

The venous blood sample was drawn in the morning after overnight fasting for at least 8 h to measure the following laboratory indices. Triglyceride (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), alanine aminotransferase (ALT), aspartate aminotransferase (AST), and serum creatinine were assayed enzymatically by an automatic biochemistry analyzer (AU5831, Beckman Coulter, USA). Plasma FFA and adiponectin were assayed by an automatic biochemistry analyzer (BS-2000M, Mindray, China), with ACS-ACOD method for FFA and immunoturbidimetry method for adiponectin. Tumor necrosis factor-α (TNF-α) was assayed using chemiluminescence method by an automatic chemiluminescence immunoanalyzer (IMMULITE1000, Siemens, USA). Glycosylated hemoglobin (HbA1c) was assayed by a hemoglobin analyzer (Bio-Rad D-10) using high-pressure liquid chromatography.

A standardized steamed bread meal test was performed in the morning after overnight fasting and the discontinuation of medication that would affect the test. The standard steamed bread meal is made of 100 g flour, which contains carbohydrates approximately equivalent to 75 g glucose, and has similar experiment results as the oral glucose tolerance test [12, 13]. The venous blood sample was drawn to measure blood glucose, insulin, and C-peptide at fasting, 60 min, 120 min, and 180 min following the ingestion of 100-g flour. Insulin and C-peptide were assayed by an automatic electrochemiluminescence analyzer (Cobas e801, Roche Diagnostics, Mannheim, Germany). Blood glucose was assayed enzymatically by an automatic biochemistry analyzer (AU5831, Beckman Coulter, USA).

Body mass index (BMI) was calculated as weight (kg) divided by height (m) squared.

The estimated glomerular filtration rate (eGFR) was calculated by the Chronic Kidney Disease-Epidemiology Collaboration equation [14]. Homeostatic model assessment of insulin sensitivity (HOMA2S) and homeostatic model assessment of beta-cell function (HOMA2B) were calculated by the HOMA2 model using fasting blood glucose (FBG) (mmol/L) and fasting C-peptide (nmol/L) [15].

T2DM was defined as FBG ≥ 7.0 mmol/L, 2-hour blood glucose from the oral glucose tolerance test ≥11.1 mmol/L, HbA1c ≥ 6.5%, or self-reported physician-diagnosed diabetes [16]. Hypertension was defined as systolic blood pressure ≥140 mmHg, diastolic blood pressure ≥90 mmHg, any use of antihypertensive drugs, or self-reported history of hypertension [17]. Dyslipidemia was defined as TC ≥ 6.2 mmol/L, TG ≥ 2.3 mmol/L, HDL-C < 1.0 mmol/L, LDL-C ≥ 4.1 mmol/L, any use of lipid-lowering drugs, or self-reported history of dyslipidemia [18]. Chronic kidney disease was defined as having eGFR <60 mL/min/1.73 m2 [19]. Coronary heart disease and stroke were confirmed by self-reporting medical history.

Continuous variables with normal distribution were described as mean ± standard deviation. Those with skewed distribution were described as median (P25–P75). Categorical variables were described as numbers and percentages. The distribution of adiponectin among FFA groups was visualized using the box plot and the trend of adiponectin levels was analyzed using the univariable linear regression model.

To identify the factors associated with adiponectin, we first performed the univariable linear regression analysis to screen potential influence factors. Screened factors included age, sex, duration of diabetes, BMI, TC, TG, LDL-C, HDL-C, FFA, HbA1c, FBG, fasting insulin, ALT, AST, current insulin therapy, current oral hypoglycemic agents therapy, and current lipid-lowering therapy. Notably, adiponectin, duration of diabetes, TG, HDL-C, FFA, ALT, AST, and fasting insulin were log-transformed due to skewness distribution. Then, the factors with P < 0.10 in the univariable model were selected and included in the multivariable linear regression model to compare their contributions to adiponectin.

To assess the dose-response relationship between FFA and adiponectin, we performed the restricted cubic splines (RCS) analysis. We set three knots at 5th, 50th, 95th percentiles of FFA and set 0.7 mmol/L of FFA as a reference. To avoid reverse causality, we repeated the RCS analysis with adiponectin as the independent variable and FFA as the dependent variable. Poverall < 0.05 and Pnon-linear < 0.05 indicate a nonlinear association.

To determine whether adiponectin mediated FFA-related deleterious effects, we performed the mediation analysis. In the mediation model, we used FFA as the independent variable, adiponectin as the mediating variable, and TNF-α, HOMA2S, HOMA2B, HbA1c, dyslipidemia, hypertension, coronary heart disease, stroke, chronic kidney disease as the dependent variables. We estimated the mediating effect and the mediating effect percentage of adiponectin after adjustment for potential confounders. We further performed subgroup analyses in participants with different ages, sexes, HbA1c levels, and conditions of current insulin therapy.

All statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc, Cary, NC). A two-tailed P < 0.05 was considered statistically significant.