This retrospective propensity score matched cohort study was based on the MIMIC-IV (Johnson et al., 2023), a publicly available database containing de-identified health information of patients admitted to the intensive care units of the Beth Israel Deaconess Medical Center (BIDMC) in Boston, Massachusetts, between 2008 and 2019. The database contained a variety of clinical data for each patient, such as demographics, vital signs, laboratory results, hospital/ICU admission and discharge time, ICU transfer date, prescriptions, nursing records, and other information. Informed consent was waived by the Institutional Review Board at BIDMC due to the de-identified nature of this database. The author, Li, passed the Examination of Protection of Human Research Participants and was granted access to the database for data extraction (record ID: 33047414). The study was reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (von Elm et al., 2007).

All consecutive adult patients of both sexes were screened according to the following inclusion criteria: (1) Diagnosed with sepsis (Sepsis-3)(1) upon hospital admission. (2) Aged 18 years or older. (3) For patients with multiple ICU stays, only data related to the first ICU stay were included. Patients with an ICU stay of less than 24 hours were also excluded.

Medication exposure was identified using the prescription table. Vitamin D exposure was defined simply as vitamin D supplementation during the ICU stay, with no other restrictions. The route of vitamin D administration could vary, either intravenous administration (IV), oral administration (PO) or nasogastric feeding (NG). The primary outcome was 28-day all-cause mortality. The secondary outcomes included ICU mortality, in-hospital mortality, length of ICU stay, length of hospital stay, duration of MV and duration of continuous renal replacement therapy (CRRT). According to vitamin D exposure during ICU stay, all eligible patients were categorized into either the vitamin D group or the no vitamin D group to assess the effect of vitamin D supplementation on clinical outcomes.

All data were extracted using Structured Query Language (SQL). The SQL script codes for data extraction were available on GitHub (https://github.com/MIT-LCP/mimic-iv). The following data were collected: demographics, including age, gender, race and body mass index (BMI); vital signs, such as respiratory rate (RR), heart rate, temperature and mean blood pressure (MBP); comorbidities, including congestive heart failure, cerebrovascular disease, chronic pulmonary disease, diabetes, renal disease, cancer, severe liver disease; severity scores, including acute physiology score III (APS III), charlson comorbidity index (CCI), logistic organ dysfunction system (LODS), oxford acute severity of illness score (OASIS), sequential organ failure assessment (SOFA), glascow coma scale (GCS); laboratory tests, including 25-hydroxyvitamin D, hemoglobin, platelets, white blood cell (WBC), blood urea nitrogen (BUN), creatinine, alanine aminotransferase (ALT), aspartate aminotransferase (AST), total bilirubin, glucose, potential of hydrogen (pH), partial pressure of oxygen (pO2), partial pressure of carbon dioxide (pCO2), partial pressure of arterial oxygen to fraction of inspired oxygen ratio (PaO2/FiO2 ratio), base excess, lactate, calcium, sodium, potassium, chloride, anion gap, international normalized ratio (INR); clinical measures, including antibiotic lag, first-day vasopressor, duration of MV, duration of CRRT. Comorbidities were assessed on admission. Initial vital signs and clinical indices obtained within the first 24 hours after ICU admission were used as baseline characteristics. Variables with missing values of more than 50% were excluded, while the rest variables were included in the study. The missing rate for each variable was presented in Supplementary Table S1 and Supplementary Figure S1 . Missing values for each variable were imputed using the missForest method (Sterne et al., 2009).

No prior statistical power calculation was performed, as the sample size for this retrospective study was determined only by the number of eligible patients in the MIMIC-IV database. Continuous variables were expressed as either mean (standard deviation [SD]) or median (interquartile range [IQR]) and analyzed using either the Students’ t-test or the Mann-Whitney’s U-test, depending on their distribution. Categorical variables were presented as numbers and percentages and examined using either the chi-square test or Fisher’s exact test. For the primary outcome, the Cox proportional hazards model was employed to calculate the hazard ratio (HR) and 95% confidence interval (CI). The Kaplan-Meier method and the log-rank test were used to estimate and compare the cumulative incidence of 28-day all-cause mortality. For dichotomous secondary outcomes, the logistic regression model was applied to calculate the odds ratio (OR) and 95% CI. For continuous secondary outcomes, the Hodgese-Lehmann method was utilized to calculate the median difference (MD) and 95% CI. Potential multicollinearity between variables was evaluated using the variance inflation factor (VIF), with VIF below 5 indicating the absence of multicollinearity ( Supplementary Table S2 , Supplementary Figure S2 ; Supplementary Table S3 , Supplementary Figure S3 ). For all analyses, a two-tailed p < 0.05 was deemed statistically significant. All statistical analyses and visualizations were conducted using R software (version 4.2.3; R Foundation for Statistical Computing, Vienna, Austria).

To reduce the impact of potential confounding factors and selection bias inherent in observational studies, PSM was performed following the methodological guidelines proposed by Lonjon and colleagues (Lonjon et al., 2017). According to a consensus statement (Cecconi et al., 2018), the following variables were included in the propensity score model: age, gender, race, BMI, diabetes, renal disease, severe liver disease, OASIS, GCS, MBP, heart rate, first care unit, hemoglobin, BUN, creatinine, glucose, pO2, PaO2/FiO2 ratio, base excess, calcium, chloride, anion gap, INR, antibiotic lag and first-day vasopressor. The propensity score, the predicted probability of receiving vitamin D supplementation, was calculated by using baseline covariates in a logistic regression model. Patients were matched using the 1:4 nearest neighbor method with no replacement and a caliper width of 0.05. After PSM, a matched cohort of patients with similar baseline characteristics was assembled. Covariate balance between groups was assessed using standardized mean differences (SMD) before and after matching, with SMD < 0.1 indicating insignificant differences (Austin, 2009). Stepwise Cox regression analyses were also employed to screen independent prognostic factors and adjust for confounding factors. In the matched cohort dataset, variables with p-value < 0.1 in the univariable analysis were included as candidate variables in the multivariable analysis by stepwise selection. The independent variables included in the final model were age, race, BMI, APS-III, Charlson Comorbidity Index, LODS, OASIS, SOFA, GCS, MBP, respiratory rate, heart rate, temperature, hemoglobin, WBC, BUN, creatinine, AST, total bilirubin, glucose, pH, pO2, PaO2/FiO2 ratio, base excess, lactate, potassium, chloride, anion gap, INR, antibiotic lag, first-day vasopressor, vitamin D supplementation ( Supplementary Table S4 ).

To determine the effect of different variables on 28-day all-cause mortality in patients with sepsis, we conducted a subgroup analysis within the matched cohort according to age (>60 vs. <=60 years), gender (female vs. male), race (white, black, unknown, other), BMI (obesity, overweight, normal, underweight), Charlson Comorbidity Index (<6 vs. >=6) and serum 25-hydroxyvitamin D (deficiency with 25-hydroxyvitamin D levels < 20 ng/mL, insufficiency with 25-hydroxyvitamin D levels between 20 and 29 ng/mL vs. sufficiency with 25-hydroxyvitamin D levels >= 30 ng/mL).

To verify the reliability of the findings from the matched cohort, we performed a sensitivity analysis in the unmatched cohort. Stepwise Cox regression analyses were employed to identify independent prognostic factors and adjust for potential confounders. Variables with p-value < 0.1 in the univariable analysis were integrated into the multivariable analysis to adjust for potential confounding factors. The independent variables included in the final model were: age, gender, race, BMI, APS-III, Charlson Comorbidity Index, LODS, OASIS, SOFA, GCS, MBP, respiratory rate, heart rate, temperature, hemoglobin, WBC, BUN, creatinine, ALT, AST, total bilirubin, pH, pO2, pCO2, PaO2/FiO2 ratio, base excess, lactate, Calcium, Sodium, potassium, chloride, anion gap, INR, antibiotic lag, first-day vasopressor, vitamin D supplementation ( Supplementary Table S5 ).

The process of patient selection is depicted in Figure 1 . A total of 30133 adult patients with sepsis were identified during the study period. Following the removal of ineligible records, 20230 patients were included in the entire unmatched cohort, among which 1783 (8.81%) received vitamin D supplementation during their ICU stay. After matching, 8710 patients were included in the matched cohort, with 6930 in the no vitamin D group and 1780 in the vitamin D group.

The baseline characteristics of the unmatched and matched cohort are shown in Table 1 . In the unmatched cohort, patients in the vitamin D group tended to be older and more likely to be female, exhibited higher APS-III and Charlson Comorbidity Index, lower serum 25-hydroxyvitamin D levels and longer antibiotic lag. In the matched cohort, propensity score-matched variables were well balanced, with an SMD < 0.10, indicating that no imbalances remained for variables included in the PSM ( Figure 2 ). The distributional balance of propensity scores for both groups before and after PSM is presented in Figure 3 .

The relationship between serum 25-hydroxyvitamin D and 28-day mortality in septic patients was examined through the RCS curves in the matched cohort ( Supplementary Figure S4 ). The RCS curve showed nearly linear relationship. between 28-day mortality and serum 25-hydroxyvitamin D on a continuous scale, suggesting that excessively low serum 25-hydroxyvitamin D increase the risk of mortality in septic patients.

The prevalence of vitamin D deficiency in the unmatched cohort was 1.12% (226/20230), with 7.35% (131/1783) in the vitamin D group and 0.51% (95/18447) in the no vitamin D group (p < 0.001). Vitamin D insufficiency was observed in 0.63% (127/20230) of the unmatched cohort, including 3.20% (57/1783) in the vitamin D group and 0.38% (70/18447) in the no vitamin D group (p < 0.001). The prevalence of vitamin D sufficiency was 0.69% (139/20230) in the unmatched cohort, comprising 2.24% (40/1783) in the vitamin D group and 0.54% (99/18447) in the no vitamin D group (p < 0.001).

The prevalence of vitamin D deficiency in the matched cohort was 2.07% (180/8710), with 7.36% (131/1780) in the vitamin D group and 0.71% (49/6930) in the no vitamin D group (p < 0.001). Vitamin D insufficiency was observed in 1.09% (95/8710) of the matched cohort, including 3.20% (57/1780) in the vitamin D group and 0.55% (38/6930) in the no vitamin D group (p < 0.001). The prevalence of vitamin D sufficiency was 1.18% (103/8710) in the matched cohort, comprising 2.25% (40/1780) in the vitamin D group and 0.91% (63/6930) in the no vitamin D group (p < 0.001).

Clinical indications for initiating and discontinuing vitamin D therapy were unavailable in the database.

In recent years, the relationship between vitamin D and all-cause mortality has become a spotlight topic of growing research interest. Vitamin D deficiency in critically ill hospitalized patients is considered to be a risk factor for sepsis and is associated with higher mortality rates (Braun et al., 2011; Malinverni et al., 2023; Seok et al., 2023; Vanichkulbodee et al., 2023). Specifically, Moromizato et al. reported that patients with pre-admission 25-hydroxyvitamin D levels below 15 ng/mL were more susceptible to sepsis and experienced higher 90-day mortality (Moromizato et al., 2014). In 2018, Zhou and colleagues conducted a meta-analysis of 24 studies to investigate the association between serum vitamin D levels and sepsis (Zhou et al., 2018). They found that patients with sepsis exhibited lower vitamin D levels than those without sepsis, but vitamin D deficiency was not significantly associated with increased sepsis-related mortality (Zhou et al., 2018). However, a recent systematic review of 8 studies showed that lower 25-hydroxyvitamin D levels were independently related to greater mortality in septic patients (Li and Ding, 2020). Subsequent subgroup analyses found that only severe vitamin D deficiency (<10 ng/ml) was associated with higher mortality in sepsis, whereas no such association was observed in individuals with higher 25-hydroxyvitamin D levels (Li and Ding, 2020). Further research is therefore needed to determine whether vitamin D deficiency is involved in the pathogenesis of infection and sepsis, or whether vitamin D status can be used as a marker of severity in patients with sepsis.

Although further research is needed to better understand the specific mechanisms, several possible explanations for the beneficial effects of vitamin D supplementation on sepsis have been proposed. The beneficial effects of vitamin D on infectious diseases may be attributed to its potential immunomodulatory properties (Murdaca and Gangemi, 2022; Miao and Goltzman, 2023). First, in innate immunity, by binding to its receptor, vitamin D promotes the transcription of the antibacterial peptide gene and increases the expression of LL-37 (Liu et al., 2006). In addition, vitamin D suppresses the expression of Toll-like receptor (TLR) on the surface of monocytes, thereby preventing an exaggerated immune response to pathogen-associated molecular patterns and contributing to the alleviation of sepsis (Sadeghi et al., 2006). Second, in terms of adaptive immunity, researchers have found that intravenous usage of calcitriol after sepsis could modulate the homeostasis of CD4+ T-cells and reduce sepsis-induced kidney injury in obese mice (Yeh et al., 2022). Vitamin D also inhibits pokeweed mitogen-stimulated human B-cell activation (Shiozawa et al., 1985). Therefore, vitamin D plays a crucial role in maintaining immune homeostasis and protecting organs from damage caused by excessive immune responses in sepsis. Third, vitamin D can suppress systemic inflammation and enhance antioxidant capacity in patients with sepsis. Reports suggest that vitamin D can ameliorate endotoxemia and improve the prognosis in lipopolysaccharide (LPS) treated mice by regulating free radicals and thromboxane A2 (TXA2) formation (Horiuchi et al., 1991). Furthermore, vitamin D has been demonstrated to protect mice against oxidative damage by activating the Nrf2-HO-1 signaling pathway and inhibiting the phosphorylation of NF-kappaB (Zhang et al., 2021). Notably, the association between high-dose vitamin D supplementation and improved clinical outcomes in patients with COVID-19 has been established through the suppression of hyperinflammatory cytokine storms (Sarhan et al., 2022).

For many years, vitamin D has been recommended to maintain calcium balance and enhance musculoskeletal health. Recently, beyond its importance in endocrine function, a growing body of evidence showed that vitamin D plays an important role in critical care settings. According to a comprehensive review, vitamin D may lower all-cause mortality in individuals with COVID-19 or liver disease, as well as cause-specific mortality in patients with respiratory cancer (Cao et al., 2023). Vitamin D plays a key role in immunomodulation, particularly in the context of autoimmunity (Triantos et al., 2022), by attenuating the inflammatory response, inducing phagocytosis, and promoting lymphocyte proliferation (Vanichkulbodee et al., 2023). Animal studies also showed that vitamin D might mitigate sepsis-induced acute lung injury by inhibiting endoplasmic reticulum stress (Ahmad et al., 2022). Therefore, vitamin D supplementation is a promising adjunctive therapy for sepsis in clinical settings.

Several observational studies have shown that vitamin D deficiency is common in patients admitted to the ICU, with reported rates ranging from 40% to 70% (Amrein et al., 2018). Despite the high prevalence of vitamin D deficiency in critically ill patients, routine vitamin D supplementation strategy is not practiced in ICU. Our study showed that vitamin D could improve clinical outcomes, highlighting the need for closer monitoring and proactive management of vitamin D in critically ill patients, especially those with sepsis. As the current study is based on a retrospective database, the cause-and-effect relationship is uncertain. Patients with both vitamin D sufficiency and insufficiency were included in the current study, but we believe that future studies should analyze these subpopulations separately. To better clarify the role of vitamin D supplementation in different critically ill patient populations, further validation studies are also needed. A paradoxical finding in this study is that vitamin D supplementation is associated with reduced mortality but a prolonged hospital stay. Such a discrepancy may be due to the competition between mortality and hospital stay.

This study has some limitations. First, due to the retrospective observational design of our study, the results are susceptible to residual bias and unmeasured confounders despite the use of rigorous methodologies such as PSM and multivariable analyses. Second, the study covers a long period, during which substantial improvements have occurred in sepsis management, thus exposing our study to bias. Third, due to the missing data in the MIMIC-IV database, we were not able to obtain 25-hydroxyvitamin D concentration measurements at different time points for all participants, and the relationship between 25-hydroxyvitamin D levels, supplementation practices, and patient outcomes was not thoroughly evaluated. Fourth, the indications for initiation and discontinuation of vitamin D were not available in the MIMIC-IV database, so we did not evaluate the effect of timing, dose and route of vitamin D administration on the prognosis of patients with sepsis. Finally, we did not evaluate the safety of vitamin D because adverse effects of vitamin D supplementation outside the ICU are very rare.