Table of Contents
Our data source was a nationwide collaborative network that collected electronic health records of hospitalizations and outpatient care from 19 tertiary-level comprehensive medical centers across China between January 1, 2000 and May 26, 2022.twenty three This has also been described in previous publications.twenty four Briefly, in this collaborative network of hospitals, de-identified raw data from each center were collected, pooled, and standardized by trained medical personnel and professional engineers at Digital Health China Technologies Co., LTD (Beijing, China). All participating centers are required to have their laboratories pass the annual external quality assessment of the China National Clinical Laboratory Center. Further quality control protocols are implemented to ensure the quality of the data. Currently, data cleaning has been completed for nine data modules, including personal information, key vital signs at each visit, visit details, diagnosis information coded by the International Classification of Diseases, 10th Revision (ICD-10 code), surgical procedure information, drug prescriptions classified by Anatomical Therapeutic Chemical (ATC) codes, other prescription details, laboratory results, and endpoint events of interest. All data are securely stored and accessible at the National Kidney Disease Clinical Research Center, Guangzhou, China. At the time of accessing the dataset in July 2022, there were 7,084,405 participant records available for analysis.
The protocol of this study was approved by the Medical Ethics Committee of Nanfang Hospital of Southern Medical University (approval number: NFEC-2019-213), and individual informed consent was exempted.
We included patients with new T2D who were diagnosed with T2D within 6 months at the baseline visit. The baseline visit was defined as the first record of a patient being diagnosed with T2D or the first record of a patient in the database with physician-recorded diabetes duration of 6 months or less and already diagnosed with T2D. The index date was defined as the date of discharge if the baseline visit was an inpatient visit or the date of visit if the baseline visit was an outpatient clinic visit.
Patients who were diagnosed with T2D (ICD-10 code E11) within 6 months at baseline met the inclusion criteria. We excluded patients who (1) had a diagnosis of type 1 diabetes or other types of diabetes (ICD-10 code E10 or E13) in any record present in the database, (2) were pregnant at baseline, (3) had significant illness including malignancy, organ transplant, end-stage renal disease or liver disease at baseline, (4) were lost to follow-up after the index date, or (5) had a diagnosis of cardiovascular disease (CVD) at baseline. CVD was defined as CHD (ICD-10 codes I20-I25), HF (ICD-10 code I50), or cerebrovascular disease (ICD-10 codes I60-I69).
Exposure was early insulin therapy at baseline, defined as patients receiving subcutaneous insulin therapy within 6 months after diagnosis of type 2 diabetes. Treatment should continue for at least 2 weeks with or without other antihyperglycemic agents. Duration of insulin therapy was defined in line with previous studies.13,14,15,16 Short-term intensive insulin therapy induces remission of type 2 diabetes. Insulin therapy is identified using ATC code A10, regardless of insulin type. Codes for other antihyperglycemic agents are listed in Table S1 in the supplemental material.
Outcomes included the incidence of three cardiovascular outcomes: CHD (ICD-10 codes I20–I25), stroke (ischemic and hemorrhagic stroke, ICD-10 codes I60–I64), and hospitalization for HF (ICD-10 code I50). Detailed definitions of these outcomes are provided in Table S2 in the supplementary material.
Descriptive statistics were compiled for baseline characteristics of study participants. t Test (continuous variables, normal distribution), Mann-Whitney test you Chi-squared tests (continuous variables, skewed distribution) and chi-squared tests (categorical variables) were used to compare T2D patients who received early insulin therapy with those who did not. Multiple imputation by chained equations with predictive mean matching techniques was applied to impute continuous data.
Distributions of time to events were summarized with Kaplan-Meier curves. Cox proportional hazards models were used to compare the risk of cardiovascular outcomes between study participants who received and did not receive early insulin therapy. Associations between cardiovascular events of interest and early insulin therapy were estimated by calculating hazard ratios (HRs) and 95% confidence intervals (CIs). Follow-up time was calculated from the index date to the earliest of cardiovascular events, death, or May 26, 2022. The proportional hazards assumption was considered using Schoenfeld residuals, and as a result, data were censored at 140 months (11.7 years) from the index date in all analyses to allow compliance with the proportional hazards assumption. Models were adjusted for baseline demographics, comorbidities, and potential confounders, including clinical laboratory characteristics, and the presence or absence of use of statins, antiplatelet agents, and antihypertensive agents during the follow-up period. Baseline use of antihyperglycemic agents was also ascertained. These variables were ascertained from structured electronic records. In particular, use of statins, antiplatelet agents, and antihypertensive agents was identified from prescription data using the ATC codes shown in Table S1 in the supplemental material.
To evaluate the robustness of our results, we performed several prespecified sensitivity analyses. First, we performed analyses of each outcome in a propensity score-matched (PSM) population that was 1:1 matched according to patients' age, sex, BMI, baseline HbA1c level, baseline estimated glomerular filtration rate (eGFR) determined by the Diet in Renal Disease (MDRD) equation (revised for Chinese population), systolic blood pressure, serum high-density lipoprotein cholesterol (HDL-C), serum low-density lipoprotein cholesterol (LDL-C), and baseline use of metformin, statins, and antiplatelet agents.twenty five Baseline characteristics of the two groups in the PSM population were compared using standardized mean difference (SMD) to assess balance between groups: variables were considered well matched if SMD < 0.1.26 We then explored the association using inverse probability treatment weighting (IPTW) analysis. In IPTW analysis, we used the predicted probabilities from the propensity score model to determine the inverse probability weight for each participant and applied this weight to the multivariate Cox regression model. Third, to consider the potential effect of reverse causation, we repeated the analysis after excluding outcome events that occurred within 90 and 180 days from the index date in participants who received and did not receive early insulin therapy. We also further adjusted for the mean HbA1c level 6 months after treatment to examine whether the effect of early insulin therapy on CVD was attributable to better glycemic control. To further exclude potential confounding by the cardiovascular protective effects exerted by GLP-1 RAs and SGLT2is, we repeated the analysis excluding patients who had ever used GLP-1 RAs or SGLT2is in both groups.
Additionally, age (<60 vs. ≥60 years), sex, and BMI (<24 vs. 24–27.9 vs. 28 kg/m2), diagnosis of hypertension at baseline, baseline HbA1c level (> 9%) [75 mmol/mol] For ≤9% [75 mmol/mol]), baseline serum LDL-C level (≥2.6 mmol/L vs. <2.6 mmol/L), baseline metformin use, history of statin use, and history of antiplatelet drug use.
PA mean score of less than 0.05 was considered statistically significant in all analyses.http://www.r-project.org) was used for all statistical analyses.