2.1. Data Source.
We retrieved all data from an openly available critical care database named Medical Information Mart for Intensive Care III (MIMIC-III, version 1.4)[12], which included more than 60000 intensive care unit(ICU) stays and more than 50000 stays for adult patients. The data in MIMIC-III were collected from June 2001 to October 2012 in Beth Israel Deaconess Medical Center, including general information(patient demographics, birth and death, ICU admission and discharge information), vital signs, laboratory data, the balance of body fluid, reports, medication, nursing record, etc. Protecting Human Research Participants exam was passed to gain access to MIMIC-III database and our certificate number is 36571208.
2.2. Study Population.
All adult patients (≥18 years) admitted to CCU from MIMIC-III database were included, and only the first admission of each patient was included. Patients who meet the following criteria were excluded: (1) age under 18 years old; (2) length of CCU stay< 2 days; (3) missing data for neutrophil percentage and albumin ; (4) data missing values for individuals are greater than 5%. Finally, 2364 patients were included in this study (Figure 1).
2.3. Definition of NPAR and Outcomes.
NPAR was calculated as neutrophil percentage numerator divided by serum albumin concentration. Neutrophil percentage and serum albumin concentration were obtained by the first blood test after admission to CCU and measured at the same time. Primary outcome was in-hospital mortality, secondary outcomes were 30-day mortality, 365-day mortality, length of CCU stay, length of hospital stay, acute kidney injury, renal replacement therapy.
2.4. Data Extraction.
All data used in this study was extracted using SQL from MIMIC-III database. Demographics, vital signs, diagnoses of heart diseases, comorbidities and medical history, laboratory parameters, medication use, scoring systems and survival data were collected. Demographic data included age, gender and race. Vital signs included systolic blood pressure, diastolic blood pressure, mean blood pressure, heart rate, respiratory rate, temperature, oxygen saturation. Diagnoses of heart diseases included coronary artery disease, acute myocardial infarction, third-degree atrioventricular block, atrial fibrillation, congestive heart failure, ventricular arrhythmias (ventricular tachycardia; ventricular flutter; ventricular fibrillation), primary cardiomyopathy (hypertrophic obstructive cardiomyopathy and other primary cardiomyopathies), valve diseases(disorders of mitral, aortic, pulmonary, tricuspid valve; rheumatic diseases of valves, congenital diseases of valves), endocarditis, cardiogenic shock. Comorbidities and medical history included hypertension, diabetes, chronic liver disease, hypercholesterolemia, chronic lung disease, chronic kidney disease, malignancy, autoimmune diseases, sepsis, respiratory failure, prior myocardial infarction and prior stroke. Medication use included antiplatelet, oral anticoagulant, Beta-blockers, ACEI, ARB, statin and vasopressin. Laboratory parameters included neutrophil percentage, albumin, white blood cell, lymphocyte, platelet, hemoglobin, hematocrit, creatinine, blood nitrogen urea, sodium, potassium, glucose. All laboratory parameters were extracted within 24 hours after admission to CCU. Scoring systems included SOFA (sequential organ failure assessment score)[13] and SAPS II (simplified acute physiology score)[14].
Demographics were extracted from tables named “admissions” and “patients” of MIMIC-III database. Vital signs were extracted from table named “vitalsfirstday” of MIMIC-III database. Diagnoses of heart diseases, comorbidities and medical history were extracted from table named “diagnoses_icd” of MIMIC-III database. Laboratory parameters were extracted from table named “labevents” of MIMIC-III database. Medication use was extracted from table named “prescriptions” of MIMIC-III database. SOFA and SAPS II were extracted from table named “sofa” and “sapsii” of MIMIC-III database.
2.5. Statistical Analysis.
All CCU patients were divided into four groups based on NPAR quartiles. Normally distributed variables were described as mean ± standard deviation (SD) and non-normally distributed variables were described as median [interquartile range (IQR)]. Difference between groups was tested by Kruskal–Wallis test or one-way ANOVA analysis. Categorical variables were described as numbers (%) and difference between groups was tested using Chi-square test.
Binary logistic regression analysis was used to analyze the relationship between NPAR and clinical outcomes. Covariates were included in the regression model based on statistical evidence and clinical judgment. Subgroup analysis was used to assess the impact of NPAR on in-hospital mortality in different subgroups. Receiver-operating characteristic (ROC) curves were drawn and areas under the curves (AUC) of different parameters were compared using DeLong test. Log-rank test was used to compare the 30-day and 365-day survival rates of different groups, and Kaplan-Meier curves were plotted.
P<0.05 was considered statistically significant, all tests were two-sided. We used MedCalc version 15.2.2 and Stata v.11.2 for statistical analysis. GraphPad Prism 8 was used to draw Kaplan-Meier curves and ROC curves.