To obtain preliminary diagnostic criteria, we searched for clinical English articles published between January 1, 1990, and December 31, 2022, on PubMed using Mesh search (keywords + free words): critical illness and (immune system diseases or immunosuppression therapy or immune system or immunity innate or adaptive immunity or inflammation or cytokine release syndrome or monitoring, immunologic or immunomodulation or immunotherapy). One hundred and twenty-six documents were selected by reading the titles and abstracts, followed by a reading of the full text (Supplementary Figure S1), resulting in 37 documents and preliminary diagnostic criteria.

This diagnostic criterion was validated by clinical data from the Medical Information Mart for Intensive Care (MIMIC)-IV version 2.2, a publicly accessible ICU database that has detailed data on around 250,000 patients who were admitted in hospital and about 70,000 patients in the ICU between 2008 and 2019 (13). Researchers can filter and analyze various aspects of each patient’s medical records, such as demographic characteristics, vital signs, laboratory test results, and imaging examinations, using a unique code assigned to them upon admission. In order to gain access to this database, Author Zhou successfully completed online training courses and exams, obtaining certification number 53360516. As the data in this database is de-identified, informed consent from the patients was not required.

A comprehensive analysis was conducted by examining the records of all adult individuals (18 years or older) who were admitted to ICU. For patients who had several hospitalizations, we only took into account data from the first hospital admission and first ICU stay to assure accuracy. Three individuals were eliminated from the analysis due to data errors, where the recorded time of death preceded the time of ICU admission. Patients with missing values were also excluded. Furthermore, to ensure the accuracy of subsequent analyses, patients with outlier values, defined as values exceeding the 99th percentile or falling below the 1st percentile, were removed. A detailed description of the screening process can be found in Figure 1.

In the current study, we collected data on various patient parameters, including (1) Demographic features (age, gender, race); (2) Acute Physiology and Chronic Health Evaluation II (APACHE II), Charlson Comorbidity Index; (3) Acute etiology (sepsis, septic shock, acute pancreatitis, acute myocardial infarction etc.); (4) Chronic comorbidities (HIV infection, diabetes, malignant cancer, congestive heart failure, hypertension, cerebrovascular disease); (5) Some vital signs and laboratory indicators (temperature, heart rate, respiratory rate, mean blood pressure, glucose, platelet, urea nitrogen); (6) The use of mechanical ventilation (MV) and renal replacement therapy (RRT) during ICU stay; (7) the length of ICU stay, 30-day mortality, 180-day mortality and ICU mortality of all patients. Furthermore, we collected vital signs and laboratory indicators from the initial dataset that was gathered upon the patient’s admission to the intensive care unit. We classified septic shock as sepsis accompanied by a mean blood pressure (MBP) of less than 65 mm Hg or the use of vasopressors such as norepinephrine, epinephrine, phenylephrine, and others (14). Besides, some variables, such as white blood cell (WBC), lactate and c-reactive protein (CRP) were deleted due to missing values >20%.

Our primary end point was on determining the overall mortality rate within 30 days. Additionally, we assessed secondary outcomes – overall mortality rate within 180 days.

We performed four normality tests to determine which normal distribution exists for continuous variables (Supplementary Table S1). The data for continuous variables were represented as median (interquartile ranges, IQR) due to non-normal distribution, while those for categorical data were represented as frequencies. The Kruskal–Walli’s test was used for continuous variables, and the χ² test or Fisher’s exact test was used for categorical variables. The APACHE II score was employed as an assessment of the severity of the disease, and the findings were depicted in the form of line graphs.

The Kaplan–Meier (KM) method draws a cumulative incidence curve, indicating the occurrence of ICU and in hospital deaths in different groups of patients, and the differences in risk between the groups were compared using log-rank tests. The potential factors that could affect the results were taken into account and adjusted for in four different models. In the initial phase of our analysis, we conducted intergroup comparisons without taking into account any additional variables, which we labeled as Model 1. Then we made age, gender, and race adjustments (Model 2). Additionally, we made adjustments for several illness severity ratings (APACHE II and Charlson Comorbidity Index) as well as associated etiology and comorbidities, like diabetes and septic shock (Models 3). The remaining factors are then modified (Models 4).

Furthermore, one-factor COX regression was performed to determine the connection between immune dysfunction and ICU mortality. We also evaluated the multicollinearity between the variables using variance inflation factors (VIF). Supplementary Table S2 revealed that the VIF of each variable was less than 4, indicating that there was no multicollinearity between them. In addition, we performed calculations for the Concordance Index (C-index) to assess the performance of the model and make comparisons between multiple models.

The p-values with less than 0.05 were taken as the statistically significant (two-sides). Navicat Premium 16.0 software was used for data extraction, and analysis of statistical data was conducted with the R software (version 4.3) or Empowerstats (version 4.1).

The search yielded 21,658 relevant publications initially, followed by 6,657 publications after excluding non-English articles, non-human clinical trials or reviews, and publications not from the period 1990 to 2022. One hundred and twenty-six documents were selected by reading the titles and abstracts, followed by a reading of the full text, resulting in 36 documents and preliminary diagnostic criteria (Table 1).

First, all patients admitted to the ICU were determined as the critical illness. According to the criteria, we also separated the patients into two groups: immune dysfunction (ID) group and non-immune dysfunction (NID) group. Additionally, within the ID group, we identified three subgroups. Patients who met the second and third criteria were classified into the hyperinflammation (HI) subgroup, while those who met the fourth or fifth criteria were placed in the immunosuppression (IS) subgroup. Finally, patients who met all the aforementioned criteria were categorized into the combined hyperinflammation and immunosuppression (HI + IS) subgroup.

After a series of screenings, we were left with a final cohort of 43,965 patients for further analysis. A total of 9,939 (22.61%) of them were categorized as belonging to the NID group. In contrast, there were 34,026 patients (77.39%) who met the criteria for being the ID group and were included in our analysis accordingly. Meanwhile, the HI group has 18,035 (53.00%) patients, while the IS group has 2,797 (8.22%) and the HI + IS group has 13,194 (38.78%) individuals, respectively.

The detailed information for each variable can be found in Table 2. Most of the covariates have significant differences between them apart from race.

It is worth noting that the median age of patients in the ID group was slightly higher compared to the NID group, although the difference between the three subgroups was small. In terms of etiology, it was discovered that infections like sepsis accounted for a bigger percentage of patients in the ID group (54.57%), while that percentage reached as high as 77% in the HI + IS group. Additionally, with the exception of patients with hypertension, the proportion of patients in the ID group was always larger than the proportion of patients in the NID group, regardless of comorbidity or whether they were admitted to the ICU with mechanical ventilation or RRT.

Supplementary Figure S2A clearly shows the gap in number distribution between the two groups. The NID group consists primarily of patients with scores less than 20, whereas the ID had higher scores. Moving on, subgroup comparisons in Supplementary Figure S2B show that the majority of patients in all three subgroups have higher APACHE II scores than the NID. Notably, patients in the IS subgroup had significantly lower peak scores, whereas patients in the HI + IS subgroup have the highest peak scores.

During our analysis of the KM curve, we noticed significant discrepancies in the survival rates between the ID and the NID group at both the 30-day and 180-day marks (Figure 2 and Supplementary Figure S3). In both short-term and long-term mortality, we can see that patients in the ID always have a higher mortality rate than those in the NID. And with the gradual correction of covariates, it can be seen that the survival rate increases in both groups, but the survival rate of patients in the ID shows a substantial increase. Of these, the most significant reduction in patient mortality was seen when correcting for scores and associated comorbidities.

Similar results can be found in the survival curves of subgroups (Supplementary Figures S4, S5). When uncorrected for all confounders, the HI + IS group’s curve for short-term survival decreases the most and has a greater mortality rate (Supplementary Figure S4A). The HI + IS group’s curve was gradually brought closer to the HI curve as factors were corrected over time (Supplementary Figures S4B,C), and when all covariates were taken into account (Supplementary Figure S4D), patients in the HI were shown to have a greater death rate. In terms of long-term mortality (Supplementary Figure S5), adjusting for covariates causes all four curves to increase and come closer together. Interestingly, patients in the HI had a greater survival rate than those in the other two subgroups, while those in the IS had a higher mortality rate (Supplementary Figure S5D).

When COX analyses were performed (Table 3), after correcting for all covariates, patients within the ID have a significantly higher susceptibility to death and short-term mortality compared to patients within the NID. When subgroups were compared, it was discovered that patients in the HI (1.72 95%CI: 1.50–1.96) had the highest risk of dying, followed by those in the HI + IS (1.65 95%CI: 1.43–1.89), and those in the IS (1.58 95%CI: 1.32–1.88) had the lowest risk. The outcomes were reversed when it came to the risk of long-term mortality. Patients in the HI (1.31 95%CI: 1.20–1.43) had the lowest risk of mortality among the three groups, while those in the IS (1.62 95%CI: 1.45–1.81) had the highest risk.

In the subsequent C-index calculations of the models for each group, we observed a gradual increase in the C-index values as we continued to correct the covariates (Supplementary Table S3). When all the covariates were corrected, we noticed that the C-index for short-term immune dysfunction with non-immune dysfunction was the highest (0.822 se: 0.003), followed by long-term immune dysfunction (0.821 se: 0.003). Additionally, during the subgroup comparison, we found that the C-index of the model was 0.796 (se: 0.002) for short-term immune dysfunction and 0.797 (se: 0.002) for long-term immune dysfunction. These results demonstrate that the model exhibits a strong predictive performance.

After controlling for all confounders, we discovered that the C-index for short-term immunological dysfunction with non-immune dysfunction was the greatest, followed by long-term immune dysfunction. Furthermore, we discovered that the model’s C-index was 0.796 for short-term immunological dysfunction and 0.797 for long-term immune dysfunction during the subgroup comparison. These findings show that the model performs well in terms of prediction.