Study design and participants
This study retrospectively analyzed the clinical data of patients who received maintenance HD treatment at the Central Hospital of Longhua District, Shenzhen, from 2015 to 2022. Patients aged 18 or older with less than two years of HD and with no gender or ethnicity restrictions. We excluded patients who were pregnant, breastfeeding, suffering from rheumatic heart disease or congenital heart disease, and those who had undergone surgical or interventional treatment for valvular disease, post-parathyroidectomy, arrhythmias, multiple myeloma, amyloidosis, chronic liver disease, systemic lupus erythematosus, potential malignant tumors, severe infections or other conditions causing renal insufficiency, as well as patients with considerable clinical data deficiencies. The study ultimately enrolled 285 patients, of which 101 (35.4%) were diabetic, comprised 92 (91.1%) of diabetic nephropathy, and 9 (8.9%) of CKD combined with diabetes; the non-diabetic group (64.6%) comprised 184 cases, of which 99 (53.8%) were chronic nephritis, 35 (19.0%) were hypertensive nephropathy, and 50 (27.2%) were other renal diseases (Additional File: Figure S1).
Data collection and definitions
The data was collected from our hospital's medical records system, including the age, history of diabetes, gender, duration of renal failure (months), duration of hypertension (years), admission systolic and diastolic blood pressure, 24-hour urine volume (ml), body mass index (BMI, kg/m2), smoking history, and history of cardiovascular and cerebrovascular diseases. The diagnosis of diabetes was based on the World Health Organization's criteria (1999)[12]. All fasting serologic indices were collected from patients in the early morning before dialysis by our hospital's laboratory system, including hemoglobin (Hb, g/L), HbA1c (%), albumin (ALB, g/L), total cholesterol (TC, mmol/L), triglyceride (TG, mmol/L), low-density lipoprotein cholesterol (LDL-c, mmol/L), high-density lipoprotein cholesterol (HDL-c, mmol/L), serum creatinine (Scr, μmol/L) urea (UA, mmol/L), intact parathyroid hormone (iPTH, pmol/L), phosphorus (Pi, mmol/L), calcium (Ca, mmol/L). Calcium was corrected if ALB was less than 40g/L (corrected calcium, mmol/L = measured calcium - 0.02×(ALB-40)). Calcium and phosphorus production (mg/dl)=Ca(mmol/L)×Pi(mmol/L)×12.4. The eGFR was estimated according to the 2009 CKD-EPI equation[13].
The above information was collected and validated by two seasoned clinicians, and all data requests were approved and lodged by the Shenzhen Longhua District Central Hospital Ethics Committee.
CVC and mortality outcomes
We documented cardiac color Doppler ultrasound and chest plain CT in patients admitted to the hospital for the first time based on the case and imaging systems. Considering the differences in the sites of calcification, we classified the calcification into three regions: heart valve calcification, coronary artery calcification, and thoracic aorta calcification. The criteria for determining heart valve calcification were mainly based on cardiac Doppler ultrasound, which was determined by observing one or more echogenic solid refractions ≥1 mm in the aortic valve, mitral valve, or annulus. Calcification of the coronary arteries and the thoracic aorta was determined by multislice CT scanning, combined with the imaging system and the report of the radiologist, to determine the presence of thoracic aortic and coronary arterial calcification, the presence of calcification in any one place was considered to be detected by the CT.
Our study identifies mortality as the endpoint and tracks participants' survival from January 1, 2015, to December 31, 2022, using hospital records or family-provided death information. All natural death events will be included.
Statistical analysis
The study population was divided into diabetic and non-diabetic groups. Continuous variables were reported as mean ± SD (x̄±s). t-test was used to compare groups with a standard data distribution. Wilcoxon rank-sum test was used with non-normal data. Count data were reported as percentages (%) and compared using the chi-square or Cochran-Mantel-Haenszel (CMH) stratified chi-square test.
Secondly, we examined the connection between diabetes and calcification. Three logistic regression models were built sequentially with different confounders to test the results' robustness. Model 1 was unadjusted, whereas Models 2 was adjusted by age sex and Model 3 were ajusted by CHD/stroke, ALB , HbA1c, Scr , BUN, Ca, Pi, iPTH, TC, TG , LDL-C, HDL-C. Additionnally, subgroup analysis included gender (male, female), age (<50, ≥50), albumin (<35 g/L, ≥35 g/L), creatinine (<707 μmmol/L, ≥707 μmmol/L), 24-hour urine volume (<1000 ml, ≥1000 ml), Ca (<2 mmol/L, ≥2 mmol/L), Pi (<1.8 mmol/L, ≥1.8mmol/L), and LDL-C (<1.8mmol/L). Multiplicative interactions were used to assess differences between theme.
Thirdly, a multivariate-adjusted Cox regression model assessed how calcification and diabetes affect death risk. Participants were divided into four calcification site groups: 0, 1, 2, and 3. These groups were analyzed as continuous and graded variables (with the 0-cite group as the reference group). After that, subjects were divided into four groups based on calcifications and diabetes: non-DM & non-CA, non-DM & CA, DM & non-CA, and DM & CA. We used similar multivariate Cox regression models to compare survival rates in the non-DM and non-CA groups. Each analysis used longitudinal forest plots to compare risk and Kaplan-Meier (K-M) survival plots to compare survival by Log-rank test.
Finally, our study built a mortality risk model. According to statistical and clinical significance in Table 1, 14 clinical features were incorporated into LASSO regression for 100 bootstrap iterations, and six features with non-zero coefficients and a minimum lambda value were selected. Then, nomograms and risk-linkage plots showed how each variable predicted mortality risk. Additionally, we calculated each patient's total score and divided the study population into high- and low-risk groups using a total point threshold (cutoff value). K-M survival curves were used to compare these two groups' survival times to test the Nomogram model's identification and prediction abilities.
Data analysis and visualization were done in R (version 4.3.3, Windows 11) and Microsoft PowerPoint 2021 . Some critical packages were forestploter and ggplot2 for forest plots, ggrisk for risk-linkage maps, adjustedCurves for survival curves, glmnet for Lasso regression, corrplot for heat maps, DynNom for nomograms, and survminer for cutoff values. Statistical significance was set at p < 0.05.