This retrospective study used the single-center critical care MIMIC-III database (24), de-identified following the Health Insurance Portability and Accountability Act (HIPAA), and it was approved by the Massachusetts Institute of Technology (MIT) and the Beth Israel Deaconess Insurance Center (BIDMC). Each patient provided written informed consent under the principles of the Declaration of Helsinki.

The database contains clinical data, such as demographics, vital signs, laboratory test results, imaging reports, and mortality, associated with 46,520 distinct hospital admissions of adult patients (aged 16 years or older) at the critical care units of the BIDMC between 2001 and 2012 in Boston, Massachusetts. ICU nurses recorded the physiological data obtained from bedside monitors hourly. The present study used the International Classification of Diseases and Ninth Revision (ICD-9) codes for disease diagnosis. The requirement for individual patient consent was waived because the project did not impact clinical care, and all protected health information was de-identified. To have access to the database, users (scientific researchers) have to complete the recognized course: “Protecting Human Research Participants” in the National Institutes of Health’s website. One author, Dong-Jie, completed the course and accessed the database for data extraction.

According to the ICD, patients with ICH with an ICD-9 code of 431 were selected for this study. To ensure the accuracy of this study, we excluded patients aged <18 years and those missing important variables, including lactate and albumin data. For patients with multiple hospital and ICU admissions, only those with an ICU admission at the first hospital admission were selected. We used the first recorded value of more than one recorded clinical characteristic in the ICU.

Structured query language (SQL) with PostgreSQL (version 12.13) was used to extract baseline characteristics, vital signs, comorbidities, laboratory results, and other parameters during the first 24 h of ICU stay. The baseline characteristics included age, sex, weight, ethnicity, insurance, and length of ICU stay. Vital signs included weight, heart rate (HR), respiratory rate (RR), systolic blood pressure (SBP), diastolic blood pressure (DBP), mean blood pressure (MBP), temperature, and percutaneous oxygen saturation (SpO₂). Information on comorbidities, including congestive heart failure, cardiac arrhythmias, hypertension, diabetes, renal failure, liver disease, and metastatic cancer, was collected for sensitivity analysis based on the ICD-9 code. Laboratory variables were extracted within the first 24 h following ICU admission, including anion gap (AG), bicarbonate, creatinine, chloride, glucose, hematocrit, hemoglobin, platelets, potassium, activated partial thromboplastin time (PTT), international normalized ratio (INR), prothrombin time (PT), sodium, blood urea nitrogen (BUN), and white blood cell (WBC) count. The severity score of each patient was measured using the Oxford acute severity of illness score (OASIS), simplified acute physiology score II (SAPS II), systemic inflammatory response syndrome (SIRS) score, sequential organ failure assessment (SOFA) score, and acute physiology score III (APS III).

Patient follow-up was calculated from the day of hospital admission to the day of death. The clinical endpoints of interest were in-hospital and ICU mortalities. In-hospital mortality was defined as death during hospitalization, and ICU mortality was defined as death during ICU admission.

The normality of the data was examined for continuous variables using the Kolmogorov–Smirnov test. Because all continuous variables were not normally distributed, they were expressed as medians (interquartile ranges [IQRs]). Categorical variables were expressed as overall numbers and percentages. Fisher’s test and the chi-square test were used to compare categorical variables between groups (patients with in-hospital death vs. those without in-hospital death; patients with ICU death vs. patients without ICU death). The Wilcoxon rank-sum test was used to compare the continuous variables.

A univariate Cox proportional hazards model was used to investigate the association between LAR and clinical endpoints (risk of in-hospital and ICU mortality). Other potential confounding factors relevant to the clinical endpoints, including age, sex, heart rate, respiratory rate, SBP, DBP, creatinine, AG, WBC, BUN, bicarbonate, glucose, renal failure, and hypertension, were entered into a multivariate Cox proportional hazards model as covariates to further study the association of LAR with these clinical endpoints. As malignancy and insurance status are important causes of death in older individuals, and also considered the potential impact of surgery on outcomes of patients, we also used insurance, metastatic cancer and surgery as covariates in the sensitivity analysis. Thereafter, the maximum selection test method was utilized to determine the cut-off value of LAR for stratified analysis and the association of LAR with the risk of in-hospital mortality and ICU mortality. The Kaplan–Meier (K-M) curve, followed by the log-rank test, was used to visualize relationships. In the stratified analysis, we used two multivariate Cox proportional hazards models adjusted for potential confounders. Model I was adjusted for age, sex, HR, and RR. Model II was adjusted for age, sex, HR, RR, SBP, DBP, AG, creatinine levels, renal failure, and congestive heart failure.

Subgroup analyses of the association of LAR with in-hospital mortality and ICU mortality were further performed using comorbidities (congestive heart failure, cardiac arrhythmias, liver disease, renal failure, and hypertension) and severity scores (SOFA score and SAPS II) in patients with ICH. All statistical analyses were performed using the R software (version 4.1.3). Statistical significance was set at p < 0.05.

Among 46,520 patients in the MIMIC-III database, 237 patients with ICH were selected for this study according to the aforementioned inclusion and exclusion criteria, as shown in the flow chart of population selection (Figure 2). In our study cohort, the median age of the enrolled patients was 64.9 years, and 130 (55%) were male. The in-hospital and ICU mortality rates of the enrolled population were 37.5% (n = 89) and 30.8% (n = 73), respectively. Among the enrolled patients, 153 (65%) were white, and 20 (8.4%) were black. The median length of ICU stay was 4.2 (2.04–9.78) days. The lactate and albumin test values were measured for all patients in the first 24 h following ICU admission, with median values of 1.98 and 3.60 mg/dl, respectively. The lactate test results between the surviving and non-surviving groups were significantly different: lactate levels in the non-surviving group were significantly higher than those in the surviving group (Supplementary Figure S1). However, albumin levels in the surviving and non-surviving groups were essentially the same, with median values of 3.6 mg/dl. As expected, there was a significant difference in LAR levels between the non-surviving and surviving groups. These data will be useful for our subsequent studies. The baseline characteristics of the enrolled patients, classified according to survival status (alive or dead), are detailed in Table 1. Moreover, patients who died had a longer ICU stay and more comorbidities (diabetes and renal failure) than those who survived. LAR, AG, glucose, WBC, and lactate levels were significantly higher in the individuals who died than in those who survived (p < 0.05).

The univariate Cox proportional hazard analysis demonstrated that LAR was significantly poorly associated with increased risk of in-hospital mortality [unadjusted hazard ratio (HR), 1.79; 95% confidence interval (CI), 1.32–2.42; p < 0.001] and ICU mortality (unadjusted HR, 1.88; 95% CI, 1.38–2.55; p < 0.001) (Table 2). Similar results were obtained after adjusting for age, sex, HR, RR, SBP, DBP, creatinine, AG, WBC, BUN, bicarbonate, glucose, renal failure, and hypertension with covariates in the multivariate Cox proportional hazard model (all HRs > 1 and p < 0.05). Based on a sensitivity analysis using insurance coverage, metastatic cancer, and surgery as covariates, we showed that the association of LAR with in-hospital and ICU mortality changed only slightly, suggesting that there was no significant effect of insurance, malignancy, and surgery on the association (Table 3).

To further analyze the association of LAR with in-hospital and ICU mortality, the maximum selection test method was used to determine the cut-off value of LAR. After multiple grouping tests using the chi-square test, the optimal cut-off value corresponding to the maximum statistic was 0.963 (Supplementary Figures S2A,B). Based on the cut-off value, 273 patients with ICH were classified into a low LAR group (<0.963, n = 196) and a high LAR group (≥0.963, n = 41); demographics, vital signs, comorbidities, laboratory results, and scoring systems of these two groups are shown in Supplementary Table S1. Compared to those in the low LAR group, patients in the high LAR group had the higher risk of in-hospital mortality (56% vs. 34%, p = 0.007) and ICU mortality (49% vs. 27%, p = 0.006) and significantly high levels of AG, bicarbonate, creatinine, chloride, hematocrit, hemoglobin, platelet, potassium, PTT, INR, PT, sodium, BUN and WBC (p < 0.05). Patients in the high LAR group had a higher rate of comorbidities, such as congestive heart failure, diabetes, hypertension, renal failure, liver disease, and metastatic cancer, than patients in the low LAR group.

To further analyze the association of LAR with in-hospital and ICU mortality, multivariate Cox proportional hazards analyses were used to develop two adjusted models. Model I was adjusted for age, sex, HR, and RR, and Model II was adjusted for age, sex, HR, RR, SBP, DBP, AG, creatinine levels, renal failure, and congestive heart failure. As a continuous variable, LAR was significantly related to in-hospital mortality (Crude model: HR, 1.79; 95% CI, 1.32–2.42; p < 0.001; Model I: HR, 1.75; 95% CI, 1.27–2.41; p < 0.001; Model II: HR, 1.59; 95% CI, 1.12–2.27; p = 0.01) and ICU mortality (Crude model: HR, 1.88; 95% CI, 1.38–2.55; p < 0.001; Model I: HR, 1.83; 95% CI, 1.32–2.55; p < 0.001; Model II: HR, 1.65; 95% CI, 1.04–2.38; p = 0.008). Moreover, as a categorical variable, patients in the high LAR group had a remarkably increased risk of in-hospital mortality (Crude model: HR, 1.93; 95% CI, 1.20–3.11; p = 0.007; Model I: HR, 1.90; 95% CI, 1.15–3.12; p = 0.012; Model II: HR, 1.77; 95% CI, 1.02–3.06; p = 0.041) and ICU mortality (Crude model: HR, 2.07; 95% CI, 1.24–3.46; p = 0.006; Model I: HR, 2.06; 95% CI, 1.20–3.53; p = 0.009; Model II: HR, 1.88;, 95% CI, 1.04–3.40; p = 0.037) compared with those in the low LAR group. It is clear whether LAR as a categorical variable is correlated with a higher risk of in-hospital and ICU mortality than as a continuous variable (Table 4). The K-M curves for patients with in-hospital and ICU mortality are shown in Figures 3A,B; patients in the low LAR group had longer survival times and higher survival rates (log-rank p = 0.0048 and 0.0058, respectively) than those in the high LAR group.

A subgroup analysis was conducted to further evaluate the association of LAR with in-hospital and ICU mortality from various perspectives, including comorbidities and severity scores. Our results illustrated that only liver disease significantly interacted with the association of LAR with in-hospital mortality (value of p for interaction, 0.013) and ICU mortality (value of p for interaction, 0.020). Other comorbidities (congestive heart failure, cardiac arrhythmias, renal failure, hypertension, and diabetes) and severity scores (SOFA score and SAPS II score) had no significant interaction with the association of LAR with in-hospital mortality and ICU mortality (value of p for interaction >0.05 in all groups) (Table 5).