Sources of Data
The data for our study were collected from a publicly accessible critical care database named Multiparameter Intelligent Monitoring in Intensive Care Database III (MIMIC-III, version 1.4)[29]. MIMIC III is a large, single-centre database containing information of 61532 ICU admissions to Beth Israel Deaconess Medical Center (a teaching hospital of Harvard Medical School in Boston, Massachusetts) between 2001 and 2012. This database is approved by the institutional review boards of the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC). Because all the data is de-identified, no informed consent is required. In recent years, some high-quality papers using this database’s data have been published[30]. In this study, data presented were extracted by author Chen, who completed the National Institutes of Health (NIH) Web-based training course and the Protecting Human Research Participants examination (No. 36328122). Data extraction was performed using structure query language (SQL) with pgAdmin4 PostgreSQL.
Selection of Participants and Stratification Method
In our study, all patients admitted to ICU in this database were included. Exclusion criteria as follows: (I) age<18 years; (II) patients who did not complete serum phosphate measurement within the first day after admission to the ICU or had unclear records of the first serum phosphate measured value; (III) patients who received phosphate supplementation during the ICU stay; (IV) for patients with multiple records of ICU admission, excluding records other than the first ICU admission. Finally, the total number of 27131 patients were included. The detail was shown in Figure 1.
(Figure 1 should be placed here).
According to the normal range of serum phosphate as mentioned before, we divided patients into six groups based on the admission serum phosphate levels: G1: hypophosphatemia group (<2.5mg/L); G2: very-low-normal phosphate group (≥2.5mg/dL and <3.0mg/dL); G3: low-normal phosphate group (≥3.0mg/dL and <3.5mg/dL); G4: high-normal phosphate group (≥3.5mg/dL and <4.0mg/dL); G5: very-high-normal phosphate group (≥4.0mg/dL and <4.5mg/dL); G6: hyperphosphatemia group (≥4.5mg/dL). The data were also analyzed in terms of subgroups based on different ICU departments: CCU、CSRU、MICU、SICU、TSICU.
Extraction of variables
As mentioned before, using SQL with pgAdmin4 PostgreSQL, we extracted the following data: (I) age; (II) gender; (III) ethnicities; (IV) admission types; (V) ICU types; (VI) vital signs (within the first day at ICU admission): heart rate, mean arterial pressure, respiratory rate, temperature, oxygen saturation; (VII) severity at ICU admission: Sequential Organ Failure Assessment (SOFA) score, Simplified Acute Physiology Score II (SAPS II), Overall Anxiety Severity and Impairment Scale (OASIS) score; (VIII) comorbidities: congestive heart failure, coronary artery disease, hypertension, diabetes, chronic pulmonary disease, atrial fibrillation, liver disease, fluid and electrolyte disorder, malignant tumor, chronic kidney disease, (IX) laboratory results (within the first day at ICU admission): serum phosphate, serum sodium, serum potassium, serum calcium, serum creatinine, blood urea nitrogen, serum bicarbonate, serum chloride, white blood cell, hemoglobin, platelet; (X) interventions (within the first day at ICU admission): use of mechanical ventilation, renal replacement therapy, vasopressor. For some patients whose multiple laboratory measurements within the first day at ICU admission were available, the first measurement was used in our study. Because all variables included less than 3% of missing observations, missing values were imputed as the mean values.
Outcome Variables
We extracted the following outcomes variables: (I) all-cause 28-day mortality after ICU admission; (II) all-cause 90-day mortality after ICU admission; (III) all-cause mortality during ICU stay; (IV) length of ICU stay. Among the four, all-cause 28-day ICU mortality and 90-day ICU mortality were the primary endpoints, all-cause mortality during ICU stay and length of ICU stay were the secondary endpoints. Moreover, the secondary endpoints were extracted only for descriptive purposes. The data for the 28 or 90-day ICU mortality was confirmed by inspection of the death records in the database.
Statistical Analysis
Patient characteristics were described using descriptive statistics. Continuous variables were examined for normality using the Shapiro–Wilk test. According to the types and distributions of variables, normally distributed continuous variables are presented in the tables as the mean with standard deviation (SD). Skewed variables were presented as the median with interquartile ranges (IQR). Categorical variables are presented as a percentage. Analysis of variance (or the Kruskal-Wallis test) and Chi-square (or Fisher’s exact) tests were used for comparisons between groups.
Kaplan-Meier survival analysis for the cumulative rate of all-cause 28 or 90-day mortality after ICU admission was used to compare the death distributions of patients among six groups of serum phosphate at admission.
We used the restricted cubic spline functions to explore nonlinear relationships between the different levels of the serum phosphate at ICU admission as a continuous variable and our primary endpoints after ICU admission.
In order to evaluate independent associations between serum phosphate levels at ICU admission and the primary endpoints, we used univariate and multivariable Cox regression models. We used two different models to adjust potential confounders. Model 1: including age, gender, ethnicities, heart rate, mean arterial pressure, respiratory rate, temperature, oxygen saturation; Model 2: including the same as Model 1, furthermore, including admission types, ICU types, SOFA score, SAPS II, OASIS score, congestive heart failure, coronary artery disease, hypertension, diabetes, chronic pulmonary disease, atrial fibrillation, liver disease, fluid and electrolyte disorder, malignant tumor, chronic kidney disease, serum sodium, serum potassium, serum calcium, serum creatinine, blood urea nitrogen, serum bicarbonate, serum chloride, white blood cell, hemoglobin, platelet, use of mechanical ventilation on the first day, use of renal replacement therapy on the first day, use of vasopressor on the first day. We chose the G2 as the model’s reference group and calculated adjusted hazard ratios (HRs) for the other groups in comparison to the reference group. In addition, we used the same way to evaluate independent associations between serum phosphate levels at ICU admission and the primary endpoints at the different ICU departments.
Software Stata version 16.0 and R version 3.6.3 were used for statistical analyses. P-values of less than 0.05 was considered to indicate statistical significance in all analyses.