In this retrospective observational research, all included data were sourced from the Medical Information Mart for Intensive Care-IV (MIMIC-IV; version 2.2) database, which is a large, publicly open dataset administered and managed by the Laboratory for Computational Physiology of Massachusetts Institute of Technology. MIMIC-IV database contains the medical health records at the Beth Israel Deaconess Medical Center in the United States (12). After completing the online course and passing the exam, the author acquired access to the database (record ID: 57369428).

The cohort population included 3,557 patients from the MIMIC-IV with both CHD and CKD diagnoses. All patients were diagnosed per the International Classification of Diseases, 9th (ICD-9) and 10th (ICD-10) revisions. The criteria below were used to exclude ineligible patients: (1) several ICU admissions (with collected medical records pertaining only to the first admission), (2) ICU stay <24 h, (3) age <18 years at first admission, (4) no serum baseline levels of creatinine and phosphate on the first day of ICU admission, or (5) missing data exceeding 30%. Furthermore, Upon first admission to the ICU, the quartiles of serum phosphate levels were used to classify all patients into four distinct groups. Figure 1 presents the complete details concerning the inclusion of the study patients.

The MIMIC-IV database was searched to obtain patient information, such as demographics, laboratory test results, comorbidities, vital signs, and severity of illness scores, collected during the initial ICU admission using Structured Query Language (SQL) in PostgreSQL (version 16.0) and Navicat Premium (version 16.1) software. The extracted data were as follows: (1) demographics: gender, age, weight, height, and body mass index; (2) vital signs: diastolic blood pressure (DBP), mean blood pressure, systolic blood pressure (SBP), and heart rate; (3) laboratory tests: hemoglobin, blood urea nitrogen (BUN), platelet, hematocrit, estimated glomerular filtration rate [eGFR; calculated using the formula for Modification of Diet in Renal Disease (13)], white blood cell (WBC), bicarbonate, serum phosphate, red blood cell (RBC), serum chloride, serum sodium, serum potassium, anion gap, prothrombin time (PT), serum calcium, international normalized ratio (INR), serum creatinine, partial thromboplastin time (PTT), and urine output; (4) comorbidities: acute kidney injury (AKI), peripheral vascular disease, atrial fibrillation (AF), diabetes mellitus (DM), rheumatic disease, congestive heart failure (CHF), dyslipidemia, liver disease, hypertension, AMI, chronic pulmonary disease, respiratory failure, and mechanical ventilation, all of which were diagnosed utilizing the combination of the ICD-9 and ICD-10 codes; and (5) severity of illness scores: the sequential organ failure assessment (SOFA) score, the acute physiology score III (APS III), the simplified acute physiology score II (SAPS II), the Oxford acute severity of illness score (OASIS), and the logistic organ dysfunction system (LODS) score. The missing data was filled up using multiple imputations.

The primary outcomes comprised the length of ICU and hospital stays as well as vital status during the short-term follow-up times of 30 and 90 days. ICU and hospital all-cause mortality were the secondary outcomes.

The collected data were analyzed using descriptive statistics, with categorical variables presented as measures of quantity and frequency (%), and continuous variables presented as median with interquartile range or mean with standard deviation. Notably, the Shapiro–Wilk test was applied to establish the normality of continuous data. Subsequently, for non-normally distributed continuous data, we used the Wilcoxon rank-sum or Kruskal–Wallis tests. For those with normal distribution, we used t-tests or ANOVA. Fisher's exact tests or Pearson χ² tests were utilized to compare categorical data. The Cox proportional hazards models were employed in the calculation of 95% confidence intervals (CIs) and hazard ratios (HRs) for serum phosphate and the secondary outcomes, along with adjustments for certain variables. Confounding factors with a variance inflation factor of ≥5 were excluded to eliminate model overfitting caused by multicollinearity among the variables. The multivariate model approach considered only clinically relevant factors linked with prognosis, yielding three models: an unadjusted model; a partially adjusted model including gender and age; and a fully adjusted model incorporating gender, age, SBP, cerebrovascular disease, DM, mechanical ventilation, respiratory failure, BUN, urine output, glucose, PTT, LODS score, and OASIS. The quartile levels were used to evaluate trends of P-values. The occurrence rate of the primary outcomes among the groups categorized based on the distinct serum phosphate levels was determined via Kaplan–Meier (KM) survival analysis. Between-group variance was evaluated via log-rank tests. The associations of serum phosphate levels with HRs were depicted using restricted cubic spline (RCS) regression models with a setting of five knots. The specificity and sensitivity of serum phosphate levels were examined using receiver operating characteristic curves and area under the curves (AUCs). Additional subgroup analysis was performed to determine the predictive value of serum phosphate for the secondary endpoints. The subgroups that were assessed were as follows: gender, age (<65 and >65 years), AKI, AF, CHF, DM, hypertension, and AMI. Statistical analyses were performed using R software (version 4.3.1). A double-sided P-value of <0.05 has been accepted as statistically significant.

Table 1 presents the basic features of the patients categorized based on the quartiles of serum phosphate levels. A total of 3,557 patients [average age, 77.10 (69.0–84.2) years] were included, of which 1,152 (32.4%) were female. The mean serum phosphate level of the enrolled patients was 3.9 (3.3–4.7) mg/dl. Additionally, the hospital and ICU all-cause mortality rates were 14.6% and 10.0%, separately. Specifically, patients were divided based on the following quartiles: quartile 1 (Q1): <3.3 mg/dl, quartile 2 (Q2): 3.3–3.9 mg/dl, quartile 3 (Q3): 3.9–4.7 mg/dl, and quartile 4 (Q4): >4.7 mg/dl. The mean serum phosphate levels of these four groups were 2.9 (2.6–3.1), 3.6 (3.5–3.8), 4.2 (4.1–4.4), and 5.6 (5.1–6.5) mg/dl, respectively.

Compared with the groups with lower serum phosphate levels, those with increased concentrations were associated with younger age; higher prevalence rate of DM, CHF, liver disease, respiratory failure, mechanical ventilation, AKI, and AMI; higher severity of illness scores; higher levels of WBC, platelets, anion gap, BUN, serum calcium, glucose, serum potassium, INR, PT, PTT, and serum creatinine; and lower values of SBP, RBC, hematocrit, hemoglobin, eGFR, bicarbonate, urine output, serum chloride, and serum sodium. Moreover, the increasing serum phosphate levels from Q1 to Q4 were correlated with extended duration of stay in the ICU (2.4 days vs. 2.4 days vs. 2.6 days vs. 3.2 days) and hospital (8.1 days vs. 8.8 days vs. 9.0 days vs. 9.6 days) as well as higher all-cause mortality in the ICU (6.9% vs. 6.1% vs. 8.8% vs. 18.6%) and hospital (11.9% vs. 9.3% vs. 12.3% vs. 25.3%), with P < 0.001 for all.

The baseline characteristics of the patients who survived for 30 days (30-day survivor group) compared with those who did not survive for 30 days (30-day non-survivor group) are displayed in Table 2. The 30-day non-survivor group exhibited older age and elevated heart rate levels in comparison to the 30-day survivor group, along with a higher prevalence of AKI, AF, cerebrovascular disease, CHF, liver disease, mechanical ventilation, AMI, and respiratory failure (all P < 0.05). In the case of the laboratory test results, the 30-day non-survivor group exhibited higher levels of WBC, BUN, anion gap, glucose, serum potassium, PTT, PT, INR, and serum creatinine, and lower values of eGFR, serum bicarbonate, urine output, and serum chloride in comparison to the 30-day survivors (all P < 0.05). Nevertheless, gender, DBP, dyslipidemia, rheumatic disease, platelets, hematocrit, hemoglobin, serum calcium, serum sodium, and serum potassium weren't different significantly (all P > 0.05). Regarding the severity of illness scores, the 30-day non-survivors recorded higher APS III, LODS score, SOFA score, OASIS, and SAPS II than those of the 30-day survivors. Furthermore, 30-day non-survivors had remarkably higher serum phosphate levels in comparison to 30-day survivors [4.3 (3.5–5.3) mg/dl vs. 3.8 (3.3–4.5) mg/dl, P < 0.001]. In the 90-day survivor and non-survivor groups, similar differences were noticed (Supplementary Table S1).

Figure 2 depicts the KM survival curve analysis for the occurrence of primary outcomes among the quartile groups. Patient mortality risk was higher in those with elevated serum phosphate levels during the follow-up periods of 30 and 90 days (all log-rank P < 0.001).

Serum phosphate was shown to be substantially linked to death risk from all causes in the hospital, as determined by the Cox proportional hazards analysis. This correlation was observed in all three models when it was regarded as a continuous variable, i.e., the unadjusted [HR, 1.30 (95% CI: 1.24–1.36)], partially adjusted [HR, 1.36 (95% CI: 1.30–1.44)], and fully adjusted [HR, 1.10 (95% CI: 1.03–1.18)] models (all P < 0.05). Moreover, patients in the higher quartiles had an increasing tendency in serum phosphate levels when compared to those in the lowest quartile and were related to an elevated hospital mortality risk in the unadjusted [Q1 vs. Q4: HR, 2.37 (95% CI: 1.88–2.98)] and partially adjusted [Q1 vs. Q4: HR, 2.65 (95% CI: 2.11–3.34)] models (both P < 0.001; both P for trend <0.001), with serum phosphate levels as a nominal variable (Table 3 and Figure 3A). The analysis of the ICU death risk revealed similar results in the unadjusted [Q1 vs. Q4: HR, 2.96 (95% CI: 2.22–3.97)], partially adjusted [Q1 vs. Q4: HR, 3.22 (95% CI: 2.41–4.32)], and fully adjusted [Q1 vs. Q4: HR, 1.39 (95% CI: 1.03–1.89)] models (all P < 0.05; all P for trend <0.05) (Table 3 and Figure 3B).

As demonstrated in Figure 4, the HRs exhibited an upward trend when the serum phosphate levels surpassed the values of 4.37 and 4.45 mg/dl in the ICU and hospital, respectively. Conversely, the HRs of the serum phosphate values lower than these specified thresholds remained relatively stable. Furthermore, the RCS regression analysis demonstrated that the ICU and hospital death risk was strongly nonlinearly correlated with serum phosphate levels (both P for non-linearity <0.001). Lastly, the AUCs of the serum phosphate levels for predicting death risk in the ICU and hospital and during the follow-ups at 30 and 90 days were 0.637 (95% CI: 0.604–0.670), 0.611 (0.583–0.640), 0.611 (0.586–0.636), and 0.598 (0.576–0.620), respectively (Figure 5).

We further evaluated the risk stratification significance of the serum phosphate levels for mortality in the hospital and ICU according to the various subgroups of gender, age, AKI, AF, CHF, DM, hypertension, and AMI (Figure 6). In the case of hospital mortality, mortality differences were detected in only the gender subgroup [males: HR, 1.36 (95% CI: 1.28–1.45) vs. females: HR, 1.19 (95% CI: 1.09–1.31), P for interaction = 0.020] (Figure 6A). Further, in the case of ICU mortality showed differences in mortality risk within two subgroups, i.e., the gender subgroup [males: HR, 1.42 (95% CI: 1.32–1.52) vs. females: HR, 1.22 (95% CI: 1.10–1.36), P for interaction = 0.019], and AKI subgroup [patients with AKI: HR, 1.21 (95% CI: 1.12–1.31) vs. patients without AKI: HR, 1.38 (95% CI: 1.26–1.50), P for interaction = 0.033] (Figure 6B).