Source of data
The data sources is a large database called Medical Information Mart for Intensive Care III (MIMIC III) database. This free database contains information on various hospitalization categories from more than 40,000 patients who were older than 16 and admitted to Beth Israel Deaconess Medical Center’s intensive care units (ICU) between 2001 and 201221. To obtain data access, we successfully completed the Cooperative Institution Training Initiative exam and Cyber Training Program of the National Institutes of Health (NIH) which was to protect participants of human research. Due to patient’s anonymity and the absence of protected information in the database, the ethics committee decided to waive informed consent.
Inclusion and exclusion criteria
In this study, we comprised a total of 3,177 first time ICU admissions for AMI patients [using the International Classification of Diseases Version 9 (ICD-9) code]. There excluded 1,984 patients who were under 18 years of age, over 80 years of age or with an inexact age, 59 patients who didn’t have lactate or albumin at admission.
Data were collected from the MIMIC III database using Structured Query Language (SQL) with PostgreSQL (version 9.6). Demographic information (age, sex, ethnicity), Vital Signs (systolic blood pressure, diastolic blood pressure, mean blood pressure, heart rate, respiratory rate, body temperature), type of hospital admission (Elective, Emergency, Urgent), past medical history [AF, hypertension, diabetes, hyperlipidemia, COPD, AKI, ARDS, sepsis, CHF], laboratory indicators were measured multiple times within 24 h of admission, the first measurement was used [pH, SpO2, BE, AG, lactate, hemoglobin, platelet (PLT), WBC, albumin, BUN, Scr, glucose, sodium, potassium, PT, prothrombin time activity (PTA), oral medication on admission (aspirin, clopidogrel, beta blockers, diuretics, digitalis, statin), anti-diabetic therapy (the use of insulin and oral hypoglycaemic agents), cronary artery revascularization [(PCI, percutaneous transluminal coronary angioplasty (PTCA), CABG], SAPS II and SOFA were recorded. To minimize bias due to missing data, variables with more than 10% missing values were excluded from the final cohort. In fact, the missing values of indicators included in our study were less than 10%, such as body temperature, BE. We predicted the missing data using the multiple imputation method.
We obtained patient’s clinical outcome and the specific time of death through the records of the Social Security Death Index, and the endpoints were 14-day, 28-day and 90-day all-cause mortality after ICU admission.
Based on the tertiles of L/A ratio, the patients were divided into three groups : T1 group (L/A ratio<0.4063, n=379), T2 group (0.4063≤L/A ratio≤0.6667, n =379), and T3 group (L/A ratio>0.6667, n =376). Continuous variables that conform to the normal distribution are expressed as mean ± standard deviation. Continuous variables that are not normally distributed are expressed as median (interquartile range). Categorical variables are expressed as numbers (percentages). Independent sample T test or Mann-Whitney U test were used for continuous variables, while categorical variables were subjected to the Chi-square test. Then we established univariate and multivariate Cox proportional hazard models to test the correlation between the tertiles of L/A ratio and clinical outcomes ( the reference group was the first tertile). A variable of 0.05 was included in separate multivariate models of all-cause mortality at 14-day, 28-day and 90-day: Model 1, unadjusted; Model 2, involved age, sex, SBP and DBP; Model 3, involved variables in model 2 and hypertension, diabetes, hyperlipidemia, AF, COPD, CHF; Model 4 involved variables in model 3 and aspirin, clopidogrel, beta blockers, diuretics, digitalis and statin, insulin, oral hypoglycemic agents; Model 5 involved variables in model 4 and BUN, Scr, glucose, WBC, Hb, BE, and SpO2. At the same time, when L/A ratio was considered as a continuous variable, for more flexible modeling and visualizing the association between L/A ratio on admission and 14-day, 28-day and 90-day risk of mortality, we used the restricted cubic spline (RCS) curve. The cumulative incidence of 14-day, 28-day and 90-day all-cause mortality was calculated through performing Kaplan-Meier survival curves. Receiver operating characteristic curve (ROC) was used to compute the area under the curve (AUC) of the L/A ratio for predicting all-cause mortality. Meanwhile, we conducted subgroup analysis and presented it in the form of forest plot. A two-sided P <0.05 was considered statistically significant. All analyses were performed using R (R Foundation for Statistical Computing, Vienna, Austria).
Ethical approval and consent to participate
The data we used was extracted from a publicly available critical care database-Medical Information Mart for Intensive Care III (MIMIC III, Version 1.4). The privacy of patients in MIMIC III were protected by using anonymized personal identifier. To protect the privacy of the participants, their identification information was concealed. Therefore, we did not need the specific consent procedures from our institutional ethics committee.