We enrolled a cohort of AKI patients with type 2 diabetes who were exposed or not to metformin prior to admission in the database of Medical Information Mart for Intensive Care (MIMIC)-III (version 1.4). MIMIC-III is a real-world and publicly available clinical database contained more than 60,000 intensive care unit (ICU) admissions in Beth Israel Deaconess Medical Center between 2001 and 2012[9]. We were approved to use the database. All reporting followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines[10].
Study population
AKI patients with type 2 diabetes were eligible in our study. AKI was defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria on admission were considered eligible for study inclusion. KDIGO criteria are as follows[11]: increase in serum creatinine (SCr) to ≥1.5 times baseline must have occurred within the prior 7 days; or ≥0.3 mg/dl increase in SCr occurred within 48 h; or urine volume < 0.5 ml/kg/h for 6 h or more. The minimum of the SCr values available within the 7 days before admission was used as the baseline SCr[12, 13]. When the preadmission SCr was not available, the first SCr measured at admission was used as the baseline SCr[12, 14]. The diagnosis of type 2 diabetes was based on International Classification of Disease, Ninth Revision (ICD-9). We only included adult patients (age>16 years). For patients admitted to the ICU more than once, only the first ICU stay was considered.
Metformin Exposure
Preadmission metformin exposure was defined as a record of using metformin in “Medications on admission” in MIMIC-III.
Covariates
We included the following variables: demographic characteristics, marital status, insurance, and service unit, heart rate, mean arterial pressure (MAP), respiratory rate, SPO2, white blood cell (WBC) count, hemoglobin, platelet, creatinine, lactate, glucose, SOFA score, simplified acute physiology score (SAPS) II score, ventilator use, vasopressor use, renal replace treatment (RRT) use, and comorbidity disease included cardiovascular disease, liver disease, malignancy, neurological disease, chronic pulmonary disease, hypertension. Vasopressor included norepinephrine, epinephrine, phenylephrine, vasopressin, dopamine, dobutamine, and Isuprel. We also included marital status and insurance. These variables included those representing the health habits of patients who received preadmission metformin that may capture a healthy user effect[15].
Outcomes
The primary outcome was 30-day mortality. The secondary outcomes were neutrophil-to-lymphocyte ratio (NLR), and length of stay (LOS) in the hospital.
Statistical analysis
A descriptive analysis was performed for all participants. Categorical variables were expressed as numbers and percentages (%). Continuous variables were expressed as mean and standard deviation (SD) when normally distributed or median and interquartile range (IQR) when skewed. The chi-square tests (categorical variables) and One-Way ANOVA (normal distribution), Kruskal-Wallis (skewed distribution) test were used for comparison of categorical, normally, nonnormally distributed continuous variables, respectively.
To minimize the potential bias of treatment allocation and confounding, we generated a propensity score to estimate by logistic regression the likelihood that patients had preadmission metformin exposure[16]. A 1:1 nearest neighbor matching algorithm was applied using a caliper width of 0.01. The following variables were selected to generate the propensity score: age, sex, ethnicity, marital status, insurance, admission type, service unit, heart rate, MAP, respiratory rate, SPO2, WBC, creatinine, hemoglobin, platelet, ventilator use, vasopressor use, and SAPS II score. A standardized mean difference (SMD) was used to examine the PSM degree. A threshold of less than 0.1 was considered acceptable. On the PSM cohort, we used a 2-sided t test to compare preoperative NLR and LOS in hospital. We applied the Kaplan-Meier and log-rank analyses were used for 30-day survival curves.
Using the estimated propensity scores as weights, an inverse probabilities weighting (IPW) model was used to generate a weighted cohort[17]. A Cox proportional hazards regression was then performed to adjusted propensity score. We also used a univariable Cox proportional hazards regression model with the robust variance estimator to calculate the hazard ratio (HR) for mortality.
All analyses were performed with the statistical software packages R. version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria). EmpowerStats (X&Y Solutions, Inc., Boston, MA). The threshold of p <0.05 (two-sided) was considered statistically significant.
Sensitivity Analyses
To ascertain whether the results were sensitive to the matching method, we performed several additional sensitivity analyses using the full cohort. Multivariable logistic regression analyses were performed to assess the independent association after adjusted metformin prescription after hospital admission and adjusted for propensity score. Additionally, A previous study excludes patients with myocardial infarction during the previous month, who were contraindicated to metformin[18]. We exclude patients admitted to CCU for sensitivity analyses.