Association between serum calcium and prognosis in patients with acute ischemic stroke in ICU: analysis of the MIMIC-IV database

This retrospective investigation utilized the Medical Information Mart for Intensive Care (MIMIC)-IV database (v2.0), an iteration succeeding MIMIC-III. The database, aligned with the Health Insurance Portability and Accountability Act Safe Harbor provision, ensures deidentification. MIMIC-IV encompasses robust clinical data from 70,000 adult intensive care unit(ICU) patients at BIDMC between 2008 and 2019. Approval for employing the MIMIC-IV databases was granted by the Institutional Review Board of the Beth Israel Deaconess Medical Center and Massachusetts Institute of Technology. All patient data within the database is anonymized, obviating the need for informed consent. In adherence to the ethical standards articulated in the 1964 Declaration of Helsinki and its subsequent amendments, the study was conducted. Access to the database was secured following the completion of the National Institutes of Health Web-based training course and the Protecting Human Research Participants examination (No. 52784856).

Between 2008 and 2019, we identified individuals in the MIMIC-IV database meeting the following criteria: adults (aged ≥ 18 years) diagnosed with ischemic stroke, as indicated by ICD-9 codes 433/434/436/437.0/437.1, or ICD-10 codes I63/I65/I66 (Fig. 1). Only the initial ICU admission date was considered for patients with multiple ICU admissions. Exclusions were made for individuals with (i) incomplete serum iCa and tCa data; (ii) ICU stays of less than 24 hours; (iii) more than 10% missing individual data.

SQL (Structured Query Language) programming in Navicat Premium 15.0 software was used for data extraction. Patient characteristics were extracted as follows: (1) baseline demographic variables: age, sex, race. (2) vital signs (initial values on the 1^st day of the ICU): mean artery pressure (MAP), heart rate, percutaneous oxygen saturation (SpO2), temperature, and respiratory rate. (3) laboratory data (from the first record after ICU admission): tCa, iCa, potassium, phosphate, chloride, sodium, lactate, bicarbonate, creatinine, pH, white blood cell count (WBC), platelet, hemoglobin, magnesium, serum-glucose, BUN, and estimated glomerular filtration rate (eGFR; measured through the CKD-Epi formula). (4) comorbidities (processed into categorical variables for statistical analysis): hypertension, hyperlipidemia, coronary artery disease (CAD), diabetes, chronic pulmonary disease, congestive heart failure, liver disease, chronic kidney disease (CKD), malignancy, atrial fibrillation, rheumatic disease, heart failure, renal failure. (5) Severity scoring system (measured from the first record after the ICU admission: the Sequential Organ Failure Assessment (SOFA), acute physiology score III (APS III), the Glasgow Coma Scale (GCS), and systemic inflammatory response syndrome (SIRS). Additionally, treatment information data were acquired, including renal replacement treatment (RRT), mechanical ventilation, mechanical thrombectomy and thrombolytic drugs.

Serum Ca levels were assigned to 4 groups based on the quartiles (Q1-Q4) of their concentrations.

The primary outcome of this research was 30-d all-cause mortality, whereas the secondary outcomes encompassed 90-d and 365-d all-cause mortality.

Continuous data were presented as mean±standard deviation or median (interquartile range), while categorical data were expressed as numbers (percentages). To assess data normality, the Shapiro–Wilk test was employed. One-way ANOVA and Kruskal–Wallis H tests were conducted for continuous data with normal and skewed distribution, respectively, while Pearson’s Chi-square (χ2) test or Fisher’s exact test was utilized for categorical data.

Potential nonlinear associations between serum Ca levels and 30-d, 90-d, and 365-d mortality were examined using restricted cubic splines. Analyses were adjusted for multiple variables, with trimming of the highest and lowest 0.5% for serum Ca levels. Knots were positioned at the 5/25/75/95th percentiles for serum Ca measures. Likelihood ratio tests were conducted to test for nonlinearity.

The patients were assigned to 4 groups based on serum Ca levels (iCa<1.07 mmol/L, 1.07 mmol/L≤iCa<1.12 mmol/L, 1.12 mmol/L≤iCa<1.17 mmol/L, and iCa≥1.17 mmol/L, tCa<7.9 mg/dl, 7.9 mg/dL≤tCa<8.4 mg/dl, 8.4 mg/dL≤tCa<9.0 mg/dl, and tCa≥9.0 mg/dl). Log-rank tests and Kaplan–Meier methods estimated the absolute risk of events for each group. Univariate and multivariate Cox analyses identified associations between serum Ca quartiles and 30-d, 90-d, and 365-d mortality.

In the Cox regression models, Model I adjusted for gender, race, and age; Model II further adjusted for MAP, respiratory rate, heart rate, temperature, SpO2, SIRS, APS III, liver disease, malignancy, renal failure, hyperlipidemia, mechanical ventilation, RRT, mechanical thrombectomy, and thrombolytic drugs; Model III further adjusted for serum glucose, hemoglobin, platelet, WBC, creatinine, BUN, sodium, chloride, magnesium, bicarbonate, lactate, pH, phosphate, and eGFR based on Model II.

In total, 1773 IS patients admitted to the ICU were identified in the MIMIC-IV database based on the selection criteria (Fig. 1). Table 1 presents the demographic features of the subjects, categorized according to serum iCa quartiles. The median age of the participants was 71.4 (61.7-80.2) years, with 1020 (57.5%) subjects being male. The median admission serum tCa and iCa levels were 1.12 (1.07-1.17) mmol/L and 8.40 (7.90-9.00) mg/dL, respectively. Next, serum iCa levels were assigned to the Q1 group (iCa<1.07 mmol/L), Q2 group (1.07≤iCa<1.12 mmol/L), Q3 group (1.12≤ iCa<1.17 mmol/L), and Q4 group (1.17 mmol/L≤iCa). Similarly, serum tCa levels were categorized as follows: Q1 group (tCa<7.9 mg/dL), Q2 group (7.9≤tCa<8.4 mg/dL), Q3 group (8.4≤tCa<9.0 mg/dL), and Q4 group (9.0 mg/dL≤tCa). Within these groups, 428 patients were in Q1 group (iCa<1.07 mmol/L), 401 patients in Q2 group (1.07≤iCa<1.12 mmol/L), 436 patients in Q3 group (1.11≤iCa<1.17 mmol/L), and 508 patients in Q4 group (1.17 mmol/L≤iCa). Compared with those in Q2-4 groups, patients in Q1 group were more likely to exhibit higher MAP, respiratory rate, heart rate, temperature, serum-glucose, hemoglobin, platelet, WBC, creatinine, BUN, and phosphate. Additionally, they had a higher prevalence of comorbidities, including liver disease and CKD. Furthermore, this group was more inclined to receive interventions such as to receive mechanical ventilation, RRT, and thrombolytic drugs.

In Fig. 2, the results of multivariable-adjusted restricted cubic spline analyses revealed U-shaped associations between serum Ca levels (iCa) with 30-d and 90-d mortality. Nonlinear trends were observed for iCa with both 30-d and 90-d mortality (P<0.05). Notably, the lowest risk of mortality was identified at 1.16 mmol/L for iCa. Specifically, when iCa was less than 1.16 mmol/L, the risk of mortality reduced with increasing iCa concentration. Conversely, when iCa exceeded 1.16 mmol/L, the risk of mortality increased with iCa concentration. However, the relationship between iCa and 365-d mortality demonstrated linearity (P<0.05). Furthermore, the analyses highlighted significant linear relationships between tCa and 30-d, 90-d, and 365-d mortality (P>0.05).

Among the 1773 IS patients analyzed, 23.0% (407/1773) died during the first 30 days, 29.7% (527/1773) died during the first 90 days, and 36.5% (647/1773) succumbed over the 1-year follow-up period. Notably, the 30-d mortality rates were 29.2% for serum iCa <1.07 mmol/L, 26.7% for 1.07-1.12 mmol/L, 15.6% for 1.12-1.17 mmol/L, and 21.1% for ≥1.17 mmol/L. The 90-d mortality rates were 37.6% for serum iCa of <1.07 mmol/L, 32.2% for 1.07-1.12 mmol/L, 22.5% for 1.12-1.17 mmol/L, and 27.4% for ≥1.17 mmol/L. The 365-d mortality rates were 44.9% for serum iCa <1.07 mmol/L, 39.9% for 1.07-1.12 mmol/L, 30.3% for 1.12-1.17 mmol/L, and 32.1% for ≥1.17 mmol/L.

Figure 3 illustrates Kaplan–Meier curves depicting all-cause mortality across serum Ca quartiles. The curves for serum iCa quartiles exhibited significant differences (log-rank test: P<0.01 for 30-d, 90-d, and 365-d mortalities), with patients in the lowest serum iCa quartile displaying the highest cumulative incidence of mortality. In contrast, there was no obvious difference in the curves for serum tCa quartiles (log-rank test: P>0.05 for 30-d, 90-d, and 365-d mortalities).

Serum Ca was selected as the independent variable, while 30-d, 90-d and 365-d mortality as the dependent variables in the multiple regression analysis. Other variables served as covariates to enhance the model's stability, leading to the construction of 4 models (Table 2). In the non-adjusted models, the results indicated that the low serum iCa level quartile (Q1 or Q2) emerged as a significant predictor of 30-d, 90-d, and 365-d mortalities compared to the reference group (Q3). This observation persisted in Model I, even after adjusting for race, gender, and age. Model II, which further adjusted for covariates such as gender, age, race, MAP, respiration rate, heart rate, temperature, SpO2, SIRS, APSIII, liver disease, malignancy, renal failure, hyperlipidemia, mechanical ventilation, RRT, mechanical thrombectomy, and thrombolytic drugs, yielded similar results. The robustness of these findings continued in Model III, which further adjusted for serum-glucose, hemoglobin, platelet, WBC, creatinine, BUN, sodium, chloride, magnesium, bicarbonate, lactate, pH, phosphate, and eGFR on the basis of Model II. Specifically, the outcomes indicated that the high serum iCa level quartile (Q4) significantly predicted 30-d and 90-d mortalities, but not 365-d mortality, compared to the reference group (Q3), after adjusting for potential confounders in Model III. However, both low serum tCa level (Q1 or Q2 vs. Q3) and high serum tCa level (Q4 vs. Q3) were not associated with the incidence of 30-d, 90-d, and 365-d mortality in the four models (P>0.05).