Systemic lupus erythematosus reduces survival of ICU patients mediated by renal dysfunction: retrospective study of critically ill patients

Abstract

Background

Systemic lupus erythematosus (SLE), an autoimmune disease, damages multiple organs. Studies showed higher all-cause mortality in SLE patients compared to the general population. However, it remains unclear whether the persistent immunodeficiency and organ damage exacerbate the prognosis of intensive care for patients with comorbid SLE.

Methods

A cohort of 50,920 critically ill patients from 2008 to 2019 in USA was obtained from the MIMIC IV database. Matching was employed to create control groups, mitigating the influence of known confounders. Wilcoxon test and Pearson’s Chi-squared Test were utilized to compare quantitative and categorical variables between groups, respectively. Log-rank test was used to compare survival differences. Univariate and multivariate Cox regression analyses were conducted to explore influencing factors. Mediation analysis was employed to investigate the mediating role of influencing factors.

Results

Patients with comorbid SLE showed notably lower 180-day survival than controls (HR = 1.485, P = 0.015). While hemoglobin, platelets, white blood cells, creatinine, urine output, SOFA performed significance in univariate Cox regression analysis, only creatinine and urine output remained significant in multivariate analysis. Mediation analysis revealed the significant mediating effect of renal function (represented by creatinine). SLE patients treated with glucocorticoids did not exhibit a significant decrease in survival compared to controls (HR = 1.482, P = 0.095), whereas those without glucocorticoids showed a significant decrease (HR = 1.660, P = 0.027).

Conclusion

SLE diminishes survival among critically ill patients by affecting renal function, while glucocorticoids can partially mitigate the decline in survival.

Supplementary Information

The online version contains supplementary material available at 10.1186/s41927-025-00600-0.

Key messages

What is already known on this topic?

What this study adds?

How this study might affect research, practice or policy

SLE is a life-threatening autoimmune disease that significantly increases all-cause mortality.

Immunologic derangement and multi-organ damage in SLE is highly heterogeneous and can be long-lasting.

• Patients with cormobid SLE showed a significant decline in survival within 180 days following the admission to intensive care unit.

• Renal dysfunction emerged as the primary mediating factor, but hematological abnormalities and other factors should also be considered.

• Glucocorticoid administration partially mitigated the decline in survival due to SLE.

• ICU patients with comorbid SLE deserve more attention, especially to their renal function.

• The use of glucocorticoids in patients with comorbid SLE during intensive care may be beneficial.

Supplementary Information

The online version contains supplementary material available at 10.1186/s41927-025-00600-0.

Background

Systemic lupus erythematosus (SLE) is a multifaceted autoimmune disorder with manifestations spanning across multiple organs including the hematological, renal, articular, cutaneous, and neurological systems. SLE is typified by an aberrant activation of both B and T lymphocytes, accompanied by a breakdown in immunological self-tolerance towards endogenous antigens. Defects in the production and elimination of nuclear antibodies, circulation and tissue deposition of immune complexes, activation of complement and cytokines culminate in a spectrum of intricate clinical presentations and organ damage.

SLE predominantly afflicts young women, with an estimated prevalence ranging from 30 to 150 per 100,000 individuals globally, and an annual incidence ranging from 2.2 to 23.1 per 100,000 individuals [1]. Notably, in spite of its relatively low incidence, SLE constitutes one of the primary causes of mortality among young females [2]. A meta-analysis has demonstrated a substantial 2.6-fold increase in all-cause standardized mortality among SLE patients compared to the general population [3]. Recent research, based on a European cohort over the past decade, has revealed that mortality peaks within the initial year following SLE diagnosis, although there’s less disparity in 5-year and 8-year survival rates, which reach 91% and 89%, respectively [4]. This mortality trend mirrors the dynamic nature of SLE, often diagnosed during its active phase. Furthermore, besides immediate threats posed by disease activity, persistent immune dysregulation and resultant organ dysfunction significantly influence long-term mortality outcomes.

Whether the persistent immunologic dysfunction and organ damage in SLE lead to poorer outcomes in ICU is an unresolved yet critical question. Therefore, this study employed a substantial cohort from the Multiparameter Critical Care Intelligent Monitoring (MIMIC) database to conduct a comprehensive analysis of how SLE affects mortality of critically ill patients, including organ dysfunction and pharmaceutical interventions.

Methods

Study design

This was a retrospective study of a large cohort of critically ill patients from the MIMIC IV database [5]. The MIMIC IV database provided critical care data for more than 50,000 patients admitted to the intensive care unit at Beth Israel Deaconess Medical Center (BIDMC). MIMIC IV was de-identified and patient identifiers were removed, so patient privacy was protected. Data was obtained through PhysioNet, and the relevant qualification certificate and Data Use Agreement have been obtained before data acquisition [6]. The data used in this study were obtained from publicly available databases; therefore, no clinical trial registration number is applicable.

Patients

The database encompassed information of all patients admitted to the emergency department or intensive care unit (ICU) between 2008 and 2019, with exceptions made for individuals under the age of 18 at their initial visit or those listed as requiring heightened protection. To avoid duplication, only data from the first ICU admission were included. Patient screening was conducted based on diagnostic codes, encompassing both ICD-9 and ICD-10 classifications, thereby ensuring comprehensive coverage. Screening of patients was not limited to the first diagnosis, meaning that patients with the disease was included regardless of whether the disease was active or not. According to the definition of nosocomial infections, those identified more than 48 h after admission are considered hospital-acquired. Thus, patients hospitalized for less than 48 h were excluded.

Matching of control group

In this study, matching was employed to establish corresponding control groups, aiming to mitigate potential confounding variables and enhance result accuracy. Recognizing substantial variations in gender and age distributions across distinct rheumatic disease populations, employing a single control group for multiple disease cohorts could directly influence outcome interpretation. To address this concern, matched control groups were utilized during survival analysis, mainly comprising two parts. Firstly, when assessing the impact of rheumatic disease on survival, control groups were matched to specific disease cohorts. Secondly, when investigating the influence of glucocorticoids (GC) on survival rates, separate matched control groups were formed for SLE patients undergoing glucocorticoids treatment and those not receiving glucocorticoids treatment.

Matching was performed utilizing the R package “MatchIt” [7]. For this study, the optimal matching method was selected, which called functions from the “optmatch” package [8]. The benefits of optimal pair matching encompass the absence of necessity to specify a matching order and a reduced likelihood of large within-pair distances compared to nearest neighbor matching. Age, gender, and race were selected as matching covariates to construct a control group at a ratio of 1:2.

Processing of missing data

In this study, missing values were predominantly below 5%, except for the respiration score and liver score within the Sequential Organ Failure Assessment (SOFA). Feature comparisons between groups utilized the original data directly, while Cox regression analysis and mediation analysis employed interpolated data. Missing data were imputed using Multivariate Imputation by Chained Equations (MICE), implemented via the R package “mice“ [9]. This package generates multiple estimates (replacement values) for multivariable missing data and assesses estimate quality through diagnostic graphs. The method is based on a complete conditional specification, wherein each incomplete variable is addressed by a separate model. The imputed data’s distribution was evaluated using density plots, with the distribution closest to the original data chosen for subsequent analysis.

Notably, the respiratory and liver scores exhibited a 20% missing rate, potentially introducing bias to the overall SOFA score. Although this presented a concern, given that the deletion proportion is consistent across the SLE group and the control group, we interpreted it as random deletion. Consequently, statistical inferences based on the available data for difference analysis were deemed acceptable.

Selection and construction of variables

In the first section, the rheumatic included in this study were determined based on published review [10]. In the later analysis, variables that reflect systemic damage due to SLE were selected. Hemoglobin, platelets, and white blood cell counts were chosen to evaluate blood system involvement, while creatinine and urine output were selected to assess renal involvement. Glasgow Coma Scale (GCS) was utilized to gauge consciousness [11]. Given that the skin and joint systems were not central to critical care considerations, they were not included in the analysis. To capture the clinical significance of these continuous variables, additional binary variables were introduced to indicate whether each parameter fell within abnormal ranges for the patient. Specific criteria for defining binary variables were outlined below:

1. Hemoglobin(binary): “1”: Hemoglobin < 12 mg/dL, “0”: Hemoglobin >= 12 mg/dL [12];

2. Platelets(binary): “1”: Platelets < 100 × 10⁹/L, “0”: Platelets >= 100 × 10⁹/L [13];

3. White blood cell(WBC)(binary): “1”: WBC < 4 × 10⁹/L, “0”: WBC >= 4 × 10⁹/L [14];

4. Creatinine(binary): “1”: Creatinine >1.5 mg/dL, “0”: Creatinine < = 1.5 mg/dL [15];

5. Urine output(binary): “1”: Urine output < 400 ml/24 h, “0”: Urine output >= 400 ml/24h [16];

6. GCS(binary): “1”: GCS < 9, “0”: GCS >= 9 [11];

7. SOFA(binary): “1”: SOFA >= 2, “0”: SOFA < 2 [17].

Statistical analysis

The primary analyses conducted in this study encompassed inter-group difference analysis, regression analysis, and mediation analysis. Given that most variables did not adhere to normality assumptions, Mann-Whitney/Wilcoxon test was employed to assess inter-group differences in continuous variables [18]. For comparisons of categorical data, Pearson’s Chi-squared Test and Fisher’s Exact Test were selected [19]. These tests are implemented through the R package “stats”. Log-rank test was utilized to compare survival curve disparities between the two groups, executed via the R package “survival“ [20, 21]. Furthermore, the Cox proportional hazards regression model was employed for survival data modeling, also implemented using the R package “survival“ [20, 22]. Mediating effects were analyzed utilizing the causal steps approach and executed with the R package “mets“ [23, 24].

Results

Study population

The research derived a cohort of 50,920 critically ill patients from the MIMIC IV database. In the cohort, there were 271 patients with co-morbid SLE, 513 patients with rheumatoid arthritis (RA), 105 patients with primary Sjögren’s syndrome (pSS), 73 patients with systemic sclerosis (SS), 13 patients with dermatomyositis (DM), 14 patients with polymyositis (PM), 56 patients with antiphospholipid syndrome (APS), 31 patients with connective tissue disease (CTD), 79 patients with spondyloarthritis (SpA), 2976 patients with gout, 8 patients with Behcet’s disease (BD) and 3 patients with adult Still’s disease (ASD). Notably, the patient counts for BD and ASD fell below 10, thereby precluding their inclusion in subsequent survival analyses.

The effect of SLE on the survival of critically ill patients

The impact of ten rheumatic diseases on the survival rates of critically ill patients at 7, 30, 90, and 180 days was investigated, with detailed findings presented in supplementary file 1. While patients with comorbid SLE, SS, and DM exhibited a noticeable decline in survival in comparison to the controls, statistically significant differences were only discerned in patients with SLE and DM at the 180-day mark. Meanwhile, Cox regression only showed significance for SLE (HR = 1.485, P = 0.015). Figure 1A illustrated the Kaplan-Meier (KM) curve for SLE, while the KM curves for other diseases were provided in supplementary file 2.

Fig. 1.

(A) Survival curves of patients with comorbid SLE and controls (B) Survival curves of patients with comorbid SLE and controls after adjusting for diagnosis

Furthermore, we found that the distribution of the first diagnosis at admission differed between patients with SLE and those without SLE (Supplementary file 3). Therefore, we performed matching adjustment based on the first diagnosis. The matching logic prioritized alignment on the first diagnosis; subsequently, age and sex matching were conducted within patient groups sharing the same first diagnosis. Results showed that even when matched on first diagnosis, patients with SLE exhibited significantly lower survival compared to patients without SLE (P = 0.032; Fig. 1B).

Dysfunction in patients with comorbid SLE

The mean SOFA score among SLE patients stood at 4.79 ± 3.68, significantly surpassing that of the control group (3.97 ± 3.26), indicating a greater severity of organ dysfunction among SLE patients. Notably, impairment predominantly manifested in coagulation and renal function, as evidenced by significantly higher scores in these domains among patients with comorbid SLE compared to controls (Supplementary file 4).

Patients with comorbid SLE exhibited notable abnormalities in the hematological system, with an average hemoglobin level of 10.20 ± 1.80 mg/dL, platelet count of 189.50 ± 106.09 × 10⁹/L, and white blood cell count of 10.92 ± 5.76 × 10⁹/L, all registering significantly lower than the mean values observed in the control group. While the disparities in means might not reach clinically significant thresholds, further analysis delved into the proportion of patients displaying abnormal test results. It was found that the prevalence of anemia (85.45%), thrombocytopenia (16.04%), and leukopenia (9.70%) among patients with comorbid SLE significantly exceeded that of the control group (Table 1).

Table 1.

Characteristics of SLE and control group (abnormal percentage)

		SLE		Control		P
		n	percentage	n	percentage
Hemoglobin (mg/dL)	< 12	229	85.45%	380	70.90%	< 0.001*
	>=12	39	14.55%	156	29.10%
Platelets (×109/L)	< 100	43	16.04%	54	10.06%	0.014*
	>=100	225	83.96%	483	89.94%
WBC (×109/L)	< 4	26	9.70%	16	2.99%	< 0.001*
	>=4	242	90.30%	520	97.01%
Creatinine (mg/dL)	> 1.5	90	33.33%	81	15.08%	< 0.001*
	<=1.5	180	66.67%	456	84.92%
Urineoutput (ml/24 h)	< 400	32	12.36%	34	6.45%	0.005*
	>=400	227	87.64%	493	93.55%
GCS	< 9	9	3.32%	37	6.88%	0.039*
	>=9	262	96.68%	501	93.12%
SOFA	>=2	221	81.55%	403	74.35%	0.022*
	< 2	50	18.45%	139	25.65%

*P < 0.05

Furthermore, patients with comorbid SLE were found to grapple with severe renal dysfunction, characterized by markedly elevated serum creatinine levels (2.07 ± 3.05 mg/dL) and diminished first-day urine output (1619.70 ± 1211.90 ml/24 h) compared to controls. The proportion of SLE patients exhibiting abnormal renal function (33.33%) and experiencing oliguria (12.36%) was also significantly higher in contrast to the control group.

Factors affecting survival

Due to the practical clinical significance, each measure was assessed with two variables: a continuous variable reflecting the measured value and a binary variable indicating the presence of an abnormality (detailed criteria were outlined in the Methods section). Each variable first underwent confirmation for compliance with the proportional hazards assumption via the Schoenfeld residual test. SOFA, GCS, GCS(binary), and Urine-output(binary) did not meet the criteria and were consequently excluded from the Cox regression analysis. Moreover, Creatinine(binary) in the subgroup analysis of SLE patients and Platelets in the subgroup analysis of controls deviated from the proportional hazards assumption, precluding Cox regression analysis in these instances.

Univariate Cox regression analysis revealed that hemoglobin served as a significant protective factor against mortality (HR = 0.88, P = 0.002). Correspondingly, a hemoglobin level lower than normal was associated with an elevated risk of death (HR = 1.61. P = 0.025), although this finding did not attain significance in both subgroup analyses (SLE patients and controls). Platelet counts lower than 100 × 10⁹/L emerged as a significant factor in univariate Cox regression (HR = 1.66, P = 0.017), whereas the platelet count itself did not exhibit significance (P = 0.518). Intriguingly, white blood cell count was a significant risk factor for mortality (HR = 1.01, P = 0.017), indicating that elevated levels of white blood cells increase the risk of death. Meanwhile, white blood cell counts lower than 4 × 10⁹/L, common in SLE patients, also heightened the risk of death (HR = 1.76, P = 0.044).

Measures of kidney function, specifically creatinine and urine output, also exerted a significant impact on survival. Notably, it’s important to highlight that elevated creatinine level was a significant factor (HR = 1.85, P < 0.001), whereas the measured value of creatinine itself was not significant (P = 0.413). This observation might suggest that fluctuations in creatinine levels within the normal range do not significantly influence survival.

A SOFA score greater than 1, indicating damage to one or more systems, exerted a significant effect on mortality (HR = 4.92, P < 0.001). This significant effect was further supported by two subgroup analyses. Results of the univariate Cox regression were depicted in Fig. 2A and provided in supplementary file 5.

Fig. 2.

Univariate and multivariate Cox regression (A) Univariate Cox regression (B) Multivariate Cox regression

These factors were further investigated through multivariate Cox regression. Before establishing regression formulas, the biological significance of these factors was evaluated to ensure a more rational analysis. Given that SOFA serve as a comprehensive assessment of system dysfunction, encompassing variables such as creatinine, urine output, and platelets, SOFA was not included in the multivariate Cox regression analysis alongside other factors. Additionally, Urine-output and Creatinine(binary), as indicators of kidney function, exhibited a strong correlation. Consequently, these two factors were disassembled to construct regression models incorporating Hemoglobin(binary), Platelets(binary), and WBC(binary), respectively. Multivariate regression revealed that Urine-output (P < 0.001) and Creatinine(binary) (P = 0.002) emerged as significant factors for survival, supporting the critical role of renal function. However, Hemoglobin(binary), Platelets(binary), and WBC(binary) showed no significance in either multivariate regression model. The results of the multivariate Cox regression were presented in Fig. 2B.

Infection in patients with comorbid SLE

Given that infection is a common cause of mortality among SLE patients, we analyzed both the infection rate at admission (defined by positive culture results within 48 h of admission) and the hospital-acquired infection rate (defined by positive culture results occurring after the initial 48-hour admission period, with no positive cultures within the first 48 h). Specific data are presented in Table 2. Analysis demonstrated that the admission infection rate was 8.12% in patients with SLE compared to 7.32% in those without SLE, with no statistically significant difference between the two groups (P = 0.617).

Table 2.

Comparison of infection rates between SLE and non-SLE patients

Group	Positive culture after 48 h			Positive culture within 48 h
	With, n(%)	Without, n(%)	P	With, n(%)	Without, n(%)	P
SLE	25 (9.23)	246 (90.77)	< 0.001*	22 (8.12)	249 (91.88)	0.617
Without SLE	2133 (4.21)	48,516 (95.79)		3710 (7.32)	46,939 (92.68)

*P < 0.05

However, the hospital-acquired infection (HAI) rate was markedly higher in patients with SLE (9.23%) compared to those without SLE (4.21%) (P < 0.001), indicating that SLE patients appear more susceptible to nosocomial infections.

Further analysis, conducted after matching for age and gender, revealed a significant association between SLE status and the overall infection rate (P = 0.002). However, the association between infection and mortality did not reach statistical significance (P = 0.082). Consequently, mediation analysis was not pursued.

Taken together, these analyses suggest that while SLE appears to increase the risk of hospital-acquired infections, this factor does not primarily explain the elevated mortality risk observed among SLE patients in the ICU.

Mediating effect of renal function

Previous analyses indicated that renal function significantly influences survival, with renal function being significantly more impaired in SLE patients compared to controls. This suggested the possibility that SLE impact survival by compromising kidney function in patients. Therefore, in this section, we investigated the mediating role of renal function (represented by Creatinine(binary)).

The mediating effect was analyzed using the causal steps approach. Figure 3 illustrated the relationship of the mediation analysis and the corresponding coefficients. The significance of regression coefficient “k” (HR = 1.49, 95%CI:1.08–2.04, P = 0.015) meant that SLE impacted survival significantly, consistent with the results in the first part. Meanwhile, the significance of regression coefficient “a” (OR:2.79, 95%CI: 1.98–3.94, P < 0.001) suggested that SLE was significantly associated with renal dysfunction. Additionally, the significance of regression coefficient “b” (RR: 1.14, 95%CI: 1.04–1.24, P = 0.004) indicated the significant mediating effect of creatinine. The regression coefficient “c”, representing the direct effect, was 0.082, greater than 0.05, suggesting that the effect of SLE on survival rate was mainly mediated by renal function.

Fig. 3.

Mediation analysis

However, given that 66.67% of the SLE patients in this study still exhibited normal renal function, we conducted further analysis to determine whether SLE affected the survival in patients with normal renal function. Interestingly, SLE patients with normal renal function demonstrated significantly lower survival rates (Survival_SLE=0.78) compared to controls (Survival_Control=0.87), with a P-value of 0.035 (as illustrated in the Supplementary file 6). This finding suggested that although the mediation analysis indicated a complete mediation, there may still be other factors contributing to the decreased survival among SLE patients.

Glucocorticoids improve survival in SLE patients

Glucocorticoids (GC) were crucial and commonly prescribed in the treatment of SLE. The effect of glucocorticoid administration on survival in SLE patients was analyzed in this section.

The results indicated no significant difference in survival between the group treated with GC and the group treated without GC at 7 days compared to their respective matched controls. However, a distinct phenomenon was observed over an extended period. At 30 days, there was no significant difference in survival between the group treated with GC and the matched control group, but the group treated without GC experienced a significant decrease in survival compared with the matched control group. A similar pattern persisted at 90 and 180 days (Table 3). In addition, the KM curves of the three groups (Group A: Controls, Group B: SLE patients treated without GC; Group C: SLE patients treated with GC) were compared in Fig. 4. It was observed that the survival of SLE patients who did not receive glucocorticoid treatment was the lowest, and there were significant differences in survival rates among the three groups (P = 0.024).

Table 3.

Survival difference between SLE patients treated with/without GC and corresponding controls

Survival	With GC			Without GC
	SLE	Control	P	SLE	Control	P
7d survival	0.98	0.96	0.265	0.93	0.96	0.183
30d survival	0.90	0.91	0.571	0.83	0.92	0.008*
90d survival	0.83	0.88	0.172	0.77	0.86	0.018*
180d survival	0.78	0.85	0.093	0.73	0.83	0.025*

*P < 0.05

Fig. 4.

Survival curves of SLE patients treated with GC, SLE patients treated without GC and controls

Discussion

Leveraging a vast cohort encompassing 50,920 critically ill patients, this study examined the impact of SLE on 180-day survival among critically ill patients and explored potential underlying factors. The findings revealed that patients with SLE exhibited significant hematological and renal dysfunction, with these abnormalities evidently influencing survival outcomes. Both renal function indicators (Creatinine(binary) and Urine-output) emerged as significant factors in multivariate Cox regression analysis, strongly suggesting the pivotal role of renal function in determining survival. Meanwhile, mediation analysis bolstered the notion that SLE primarily impacts survival through its effects on renal function. Additionally, the administration of glucocorticoids showed promise in enhancing the survival of critically ill patients with comorbid SLE to some extent. Subsequently, this section will delve into intriguing supplementary insights beyond the primary findings and discuss the implications thereof.

In the first part, alongside SLE, this study also presented data on various other rheumatic diseases for comparison. Interestingly, not all rheumatic diseases exhibited reduced short-term mortality rates in critically ill patients. Patients with arthritis (e.g., RA, SpA, Gout) exhibited higher short-term survival, with the 7-day survival in patients with gout being notably higher than that in controls, despite these diseases carrying a higher all-cause mortality rate than the general population [25–27]. We consider this could potentially be attributed to the substantial sample size of patients with gout, facilitating the detection of statistical differences. This discrepancy suggested that the impact of rheumatic diseases on short-term mortality of severely ill patients may not have consistently aligned with general mortality trends. One plausible explanation was that 180-day mortality strongly correlated with treatment response.

In addition to SLE, the potential threat posed by DM and SS to survival was noteworthy, although SS might not have achieved statistical significance due to the small sample size. Furthermore, DM and SS are all linked to specific autoantibody formation, leading them to be categorized as autoimmune diseases rather than autoinflammatory diseases at present, akin to SLE [28, 29]. However, the organ involvements differ from those in SLE; for instance, lung involvement is more common in DM and SS. This underscored the necessity for further analysis to understand how SLE, particularly due to organ function impairment, contributed to reduced survival.

Multi-organ involvement in SLE was common yet highly heterogeneous, posing challenges for evaluation and treatment. The 2019 diagnostic criteria encompassed hematologic, neuropsychiatric, mucocutaneous, serosal, musculoskeletal, and renal abnormalities, all frequently observed in SLE patients [30]. Considering the clinical characteristics of critically ill patients, this study primarily focused on hematologic and renal abnormalities. Parameters such as leukocyte count, platelet count, and hemolytic anemia were included not only in the diagnostic criteria but also in the SLE Disease Activity Score (SLE-DAS), utilized to assess SLE activity [31].

Anemia affected approximately 50% of SLE patients. Although autoimmune hemolytic anemia (AIHA) is the representative type and is included in the diagnostic criteria, anaemia of chronic disease (ACD) is also very common, accounting for up to 35.6% of cases [32]. Hemolytic anemia has been reported to be associated with damage accrual, but not with mortality [33]. Another study supported the notion that mild, moderate, and marked anemia were strongly associated with disease activity, while moderate and marked anemia were associated with damage accrual [34].

The incidence of immune thrombocytopenia ranged from 7% to 30% in SLE patients, potentially attributable to the presence of anti-platelet autoantibodies in up to 60% of SLE patients [35]. These autoantibodies bind to platelets, triggering platelet phagocytosis in the spleen and consequent thrombocytopenia. While platelet counts exceeding 30 × 10⁹/L typically do not result in serious consequences, severe thrombocytopenia (< 20 × 10⁹/L), which can lead to fatal bleeding, is less frequently observed in SLE patients. Nevertheless, thrombocytopenia acted as a significant independent risk factor for mortality in SLE [36, 37].

The prevalence of leukopenia was reported in 22-41.8% of cases, with lymphopenia cumulatively reported in 15% to 82% of patients, while neutropenia is described in 20–40% of cases [38]. Some studies have suggested a significant risk of infection in patients with leukopenia; however, this significance existed only in univariate models and not in multivariate models [39, 40].

In addition to hematological involvement, kidney injury emerged as another critical factor influencing survival. Lupus nephritis (LN), a form of glomerulonephritis, stands out as one of the most severe organ manifestations of SLE and is included in the 2019 ACR/EULAR diagnostic criteria. The intrarenal etiology of LN involves antibody binding to multiple intrarenal autoantigens and the histopathological feature of LN is deposition of immune complexes [41].

Among unselected SLE patients, approximately 25–50% exhibited signs or symptoms of kidney disease at the onset of SLE, with up to 60% of adult SLE patients experiencing these manifestations over the course of the disease [42]. Mortality associated with lupus was significantly higher in patients with LN compared to those without LN. Within 5 years, 5% to 25% of patients with proliferative LN (class III, IV, or III/IV + V) succumbed directly to kidney disease. Additionally, 10% to 30% of LN patients eventually progressed to kidney failure and necessitated kidney replacement therapy (KRT), with those having proliferative LN facing the highest risk of requiring KRT [43]. The response of LN patients to treatment significantly influences survival. According to a 10-year study, patient survival at 10 years was 95% for complete remission, 76% for partial remission, and 46% for no remission, while renal survival at 10 years was 94% for complete remission, 45% for partial remission, and 19% for no remission [44]. LN is typically managed with glucocorticoids and immunosuppressive drugs such as cyclophosphamide, or mycophenolate mofetil (MMF). However, traditional immunosuppressive therapy is not uniformly effective, and even among responders, 35% experienced relapse. Furthermore, 5% to 20% of LN patients developed end-stage kidney disease (ESKD) within 10 years [45–47]. These clinical experiences underscores the significant burden of kidney disease in SLE.

Previous studies primarily focused on natural all-cause mortality, while our study highlighted that SLE reduced 180-day survival following admission to the ICU. This suggested that the decreased survival among SLE patients might have partly stemmed from an inadequate response to intensive care treatment. Hematologic abnormalities and renal dysfunction were identified as potential contributing factors, with further multivariate Cox regression and mediation analysis suggesting that renal function might have exerted a more significant influence than hematologic abnormalities. However, it is crucial to note that SLE patients with normal kidney function exhibited significantly lower survival than controls with normal kidney function, despite mediating analyses demonstrating a fully mediating effect of renal function. These findings, alongside previous evidence, emphasized the necessity for heightened attention to patients with comorbid SLE in the ICU.

While glucocorticoids are indispensable in the treatment of SLE, their usage is constrained due to potential side effects, particularly with prolonged high-dose administration. The 2023 update of EULAR management recommendations for SLE underlined the pivotal role of glucocorticoids. Specifically, they were recommended for active neuropsychiatric disease attributed to SLE and for acute treatment of severe autoimmune thrombocytopenia [48]. Additionally, in cases of active proliferative LN, initial (induction) treatment with mycophenolate mofetil or low-dose intravenous cyclophosphamide, both combined with glucocorticoids, was advised. For pure membranous LN characterized by nephrotic-range proteinuria or proteinuria exceeding 1 g/24 hours despite renin-angiotensin-aldosterone blockade, MMF in conjunction with glucocorticoids was preferred [49]. Based on this premise, our study emphasized the critical importance of timely glucocorticoid administration in critically ill patients with comorbid SLE.

The present study was subject to certain unavoidable limitations that warrant clarification. Primarily, it was a retrospective analysis drawn from historical data, where despite a substantial sample size, the presence of confounding bias and selection bias cannot be overlooked. While we employed optimal matching to construct the control group in an attempt to mitigate known factors affecting mortality, such as age, sex, and race, it was imperative to acknowledge that certain critical factors, such as the primary diagnosis upon admission to the ICU, couldn’t be entirely balanced. Additionally, the routine absence of immunological indicators like complement and immunoglobulin assessments in ICU settings meant that certain potential causal factors remained unanalyzed. This indicated the importance of subsequent prospective cohort studies within specific departments to investigate these aspects comprehensively.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ACD

Anaemia of chronic disease

AIHA

Autoimmune hemolytic anemia

APS

Antiphospholipid syndrome

ASD

Adult Still’s disease

BD

Behcet’s disease

BIDMC

Beth Israel Deaconess Medical Center

CTD

Connective tissue disease

DM

Dermatomyositis

ESKD

End-stage kidney disease

GC

Glucocorticoids

GCS

Glasgow Coma Scale

HIPAA

Health Insurance Portability and Accountability Act

ICD

International classification of diseases

ICU

Intensive care unit

KM

Kaplan-Meier

KRT

Kidney replacement therapy

LN

Lupus nephritis

MICE

Multivariate Imputation by Chained Equations

MIMIC

Multiparameter Critical Care Intelligent Monitoring

MMF

Mycophenolate mofetil

PM

Polymyositis

pSS

Primary Sjögren’s syndrome

RA

Rheumatoid arthritis

SOFA

Sequential Organ Failure Assessment

SpA

Spondyloarthritis

SS

Systemic sclerosis

WBC

White blood cell

Author contributions

Hongjing Zhang, Yimei Ding: Data curation; Formal Analysis; Methodology; Writing –original draft. Hong Zhang, Jia Zhou: Validation; Writing - review & editing. Wenxiong Zhou: Funding acquisition; Supervision; Writing - original draft.

Funding

1. Shanghai Acupuncture Clinical Research Center of Medicine Project (No. 20MC1920500). 2. National Administration of Traditional Chinese Medicine high-Level Disciplines Construction Project (No. ZYYZDXK-2023068). 3. Three Year Action Plan for Shanghai to Further Accelerate the Inheritance, Innovation and Development of Traditional Chinese Medicine (2025–2027) (No. 1-1-2).

Data availability

The data in this study were derived from MIMIC IV database. The database can be accessed via PhysioNet.

Declarations

Ethics approval and consent to participate

Review and/or approval by an ethics committee was not needed for this study because all data in this study were derived from public databases. Appropriate ethics approval has been provided at the publication of original articles. All methods were performed in accordance with the guidelines and regulations and the relevant qualification certificate and Data Use Agreement have been obtained before data acquisition.

Consent for publication

Not applicable.

Patient and public involvement

Patients and the public were not involved in this study.

Competing interests

The authors declare no competing interests.