Sepsis is a clinical syndrome characterised by a severe disorder of pathophysiology caused by infection of pathogenic micro-organisms. To facilitate early identification of this disease, Sepsis-3 defines it as life-threatening organ dysfunction resulting from a dys-regulated host response to infection. This means that the diagnosis is considered to be met when the patient has a suspected or confirmed infection combined with the Sequential Organ Failure Assessment ≥2(Reference Singer, Deutschman and Seymour1). Despite the lack of global epidemiological data on sepsis(Reference Finfer and Machado2), current studies estimate that it affects more than 30 million people annually and may cause 6 million deaths(Reference Bauer, Gerlach and Vogelmann3), making it a major worldwide public health problem(Reference Cecconi, Evans and Levy4). Currently, various antibiotics and organ support therapies are widely used in the clinical treatment of sepsis, but the mortality rate remains high(Reference Font, Thyagarajan and Khanna5). The mechanism of sepsis is very complex, and further research is needed.
The primary causative factors of sepsis or the alterations in the internal environment and haemodynamics that occur during disease progression can lead to massive cytokine release, oxidative stress imbalance and mitochondrial dysfunction in the organism. When antioxidant defences are overwhelmed, reactive oxygen species cause cellular damage, leading to organ dysfunction and tissues becoming hypoxic, along with an increase in free radical production and oxidative stress. Mitochondria are both sites of reactive oxygen species production and targets of reactive oxygen species-mediated damage, therefore playing a prominent role in the pathogenesis of sepsis(Reference Suliman, Welty-Wolf and Carraway6,Reference Vanasco, Cimolai and Evelson7) . For example, it was found that mitochondrial dysfunction plays a potential role in sepsis-induced acute kidney injury (AKI)(Reference Sun, Zhang and Tian8). Linear stereoscopic dysfunction and biological failure are also important causes of cardiac insufficiency in patients with sepsis(Reference Stanzani, Duchen and Singer9). The restoration of mitochondrial function, also known as metabolic resuscitation, may help relieve organ dysfunction in septic patients and improve their prognosis. Thiamine (vitamin B1) is a water-soluble vitamin that plays an essential role in mitochondrial energy metabolism and is involved in multiple cellular metabolic processes in mitochondria and peroxisomes in the body. Moreover, it provides a certain resistance to oxidative stress of cells(Reference Kartal and Palabiyik10). Thiamine deficiency occurs in most acute and critically ill patients and may be related to the accumulation of lactate(Reference Leite and de Lima11). A retrospective study revealed that in the early stages of sepsis, intravenous vitamin C along with corticosteroids and thiamine can reduce patient mortality and prevent organ failure(Reference Marik, Khangoora and Rivera12). A randomised, double-blind, controlled study showed that intravenous thiamine administration reduces lactate production in a thiamine-deficient subgroup of patients with septic shock(Reference Donnino, Andersen and Chase13). Thiamine supplementation has also been shown to reduce the risk of sepsis-related kidney injury(Reference Moskowitz, Andersen and Cocchi14).
Mitochondrial dysfunction is commonly observed in critically ill patients, especially in those experiencing sepsis. The addition of antioxidant micronutrient therapies such as ascorbic acid and thiamine to sepsis treatment remains controversial(Reference Berger15,Reference Belsky, Wira and Jacob16) . Here, we conducted a retrospective cohort study based on the Medical Information Mart for Intensive Care IV, a sizeable critical disease database, to investigate the effect of thiamine on the prognosis of patients with sepsis.
Methods
Data source
The Medical Information Mart for Intensive Care (MIMIC) IV(Reference Zhou, Zeng and Wei17) is a free open-access database(Reference Johnson, Pollard and Shen18) containing comprehensive information on tens of thousands of critically ill patients from 2008 to 2019, including comprehensive information on the patient’s hospital period, laboratory tests, medication and vital signs. The data used in this study were from MIMIC-IV version 0.4. The database was approved by the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center, and consent was obtained for the original data collection. Researchers need to complete the corresponding courses and obtain certificates before they can access and carry out data extraction (Record ID: 38601114). Considering that the database hides information about the identity of the patient, informed consent is not required.
Population
Patients who met the diagnosis of Sepsis-3 in the Medical Information Mart for Intensive Care IV database were included, whereas patients who died within 24 h after admission to the intensive care unit (ICU) and patients younger than 18 years were excluded.
Data extraction
Structured query language was used to extract data from the database. The data collected included age, sex, weight, ethnicity, first care unit, the severity of the disease (assessed using Sequential Organ Failure Assessment and Acute Physiological Score III), ventilator use, vasopressor use, continuous renal replacement therapy use, Charlson co-morbidity index and co-morbidities including myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatic disease, peptic ulcer disease, mild liver disease, severe liver disease, diabetes uncomplicated, diabetes complicated, paraplegia, renal disease, malignant cancer, metastatic solid tumour and AIDS. Results of the first laboratory examination after admission to the ICU included leucocyte count, Hb, platelet count, lactate, creatinine, urea nitrogen, glucose, PaCO2, pH and PaO2. The mean values of vital signs within 24 h of ICU admission include heart rate, mean arterial pressure, respiratory rate, temperature, pulse oximetry-derived oxygen saturation (SpO2) and total urine output in the first 24 h.
The primary outcome included ICU mortality, while the secondary outcomes included 28-d mortality, ventilation-free days in 28 d, vasopressor-free days in 28 d and incidence of AKI within 7 d after diagnosis of sepsis.
Statistical analyses
The study population was divided into two groups, the thiamine received group (TR) and the thiamine unreceived group (TUR), based on whether thiamin was supplemented intravenously while in the ICU.
Missing data in public databases is a common phenomenon. In this study, we only included variables with missing rates less than 20 %, and the missing parts were filled using multiple imputation. This function can be implemented by the ‘mice’ package of the R programme. Data for continuous variables were expressed as median (interquartile range), and categorical data were presented as frequency. Mann–Whitney U test was used for continuous variables, and the χ 2 test or Fisher’s exact test was used for categorical variables. The relationship between thiamine and ICU mortality was measured using Cox proportional risk models, with estimated hazard ratios (HR) and 95 % CI. The effects of thiamine on the ventilation-free days and vasopressor-free days in 28 d were analysed by linear regression, expressed as regression coefficients (Coefs) with 95 % CI. The association of thiamine with the occurrence of AKI after sepsis diagnosis was performed by logistic regression analysis and rendered as OR with 95 % CI. Multivariate analysis was used to control for confounders. Confounders included age, sex, weight, ethnicity, first care unit, Sequential Organ Failure Assessment, Acute Physiological Score III, ventilator use, vasopressor use, continuous renal replacement therapy use, Charlson co-morbidity index, myocardial infarct, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatic disease, peptic ulcer disease, mild liver disease, severe liver disease, diabetes uncomplicated, diabetes complicated, paraplegia, renal disease, malignant cancer, metastatic solid tumour, AIDS, leucocyte, Hb, platelet, lactate, creatinine, urea nitrogen, glucose, PaCO2, pH, PaO2, heart rate, mean arterial pressure, respiratory rate, temperature, SpO2 and urine output.
Propensity score matching (PSM) and inverse probability of treatment weighting methods were conducted to balance the baseline characteristics of the two groups of patients. Standard mean differences are commonly used to make quantitative comparisons between the mean (continuous covariate) or rate (categorical covariate) of covariates dealing with groups. Some researchers suggest that the baseline covariate is considered to have reached an acceptable equilibrium level when the normalisation difference is less than 0·2(Reference Austin19,Reference Heinze and Jüni20) . In this study, PSM applied one-to-one matching of the nearest neighbours, while inverse probability of treatment weighting used the generalised boosted model(Reference McCaffrey, Griffin and Almirall21) to estimate the necessary tendency of the weighted score. The generalised boosted model estimation method estimates the use of a flexible dual propensity score for the treatment of indicators. It involves an iterative process with multiple regression trees to capture the treatment allocation and complex nonlinear relationship between pretreatment covariates without excessive fitting data. Studies have shown that among various propensity score estimation methods, generalised boosted model provides the estimated weight, which achieves the best balance among processing variables(Reference Harder, Stuart and Anthony22). When imbalance still exists after weighting, the doubly robust estimation combines the multivariate regression model with the IPTW model, which can be used to eliminate residual confounding factors(Reference Nguyen, Collins and Spence23). This results in more accurate estimates of therapeutic effects(Reference Cefalu, Dominici and Arvold24).
A two-sided P < 0·05 was considered statistically significant. R (version 4.0.3) was used for all statistical analyses.
Results
Subject characteristics
A total of 11 553 patients meeting the selection criteria were included in this study (Fig. 1). The patients with sepsis were divided into two groups according to whether they received intravenous thiamine or not. The TUR group included 10 017 patients and the TR group included 1536 patients. Table 1 summarises the characteristics of the patients, including general conditions, degree of illness, co-morbidities and laboratory parameters. In the original population, the age of the TR group was 59 (49, 68), which is lower than that of the TUR group 68 (56, 79). The Acute Physiological Score III (65 (49, 89)) and Sequential Organ Failure Assessment score (3 (2, 5)) in the TR group were higher than those in the TUR group (54 (41, 72) and 3 (2, 4), respectively). The length of ICU stay in the TR group was longer than that in the TUR group (8·48 (3·76, 15·88) v. 4·15 (2·02, 9·61)).
Flow chart of study cohort selection.
Baseline characteristics of the study population
(median (interquartile range [IQR]); numbers and percentages)
TR, thiamine received group; TUR, thiamine unreceived group; PSM, propensity score matching; MICU, medical intensive care unit; SICU, surgical intensive care unit; TSICU, trauma surgical intensive care unit; CCU, coronary care unit; SOFA, Sequential Organ Failure Assessment; APS III, Acute Physiological Score III; CRRT, continuous renal replacement therapy; SpO2, pulse oximetry-derived oxygen saturation.
Cox proportional hazards regression model
After adjusting for confounders, the TR group showed a reduced risk of all-cause sepsis mortality rates. In comparison with the TUR group, the TR group had a statistically significantly lower risk of ICU mortality. The HR of ICU mortality for the TR group was 0·80 (95 % CI 0·70, 0·93), indicating that, compared with the TUR group, the TR group had 0·80 times the risk for ICU mortality (Table 2).
Analysis of the associations between outcomes and thiamine received
(Hazard ratios; odds ratios; 95 % confidence intervals)
PSM, propensity score matching; IPTW, inverse probability of treatment weighting.
* Cox proportional hazards regression models were used to calculate hazard ratios (HR) with 95 % CI.
† Logistic regression models were used to calculate OR with 95 % (CI).
‡ Linear regression models were used to calculate coefficients (Coefs) with 95 % (CI).
§ Confounders for the doubly robust with unbalanced covariates included age, sex, ventilator use, vasopressor use, CRRT use, first care unit, APS III, myocardial infarct, mild liver disease, severe liver disease, PaCO2, pH, glucose, platelet and SpO2. Confounders of other models are consistent with multivariate analysis.
Propensity score matching and inverse probability of treatment weighting
To reduce confounding bias, we performed PSM based on whether thiamine was used. A total of 1563 pairs were successfully matched (online Supplementary Fig. S1). After matching, the imbalance was controlled based on a comparison of the patients’ clinical data (standard mean difference < 0·2, Table 1, online Supplementary Fig. S2). After inverse probability of treatment weighting, the standard mean difference and the baseline characteristics between the two groups of the virtual population can be seen in online Supplementary Table S1.
Although the baseline differences between the two groups were well controlled, there were still variables that were statistically different. We separately included variables with differences in the matched and weighted populations in the Cox regression models, yielding results with consistent trends in the original population. And by including all of the above variables in the multivariate Cox regression models, the results remained stable (Table 2).
Secondary outcome studies with the whole population
Based on the original population, we conducted a series of secondary outcome analyses to confirm the benefits of thiamine on the prognosis of patients. Multivariate Cox regression showed that the 28-d mortality of the TR group was improved (HR: 0·82; 95 % CI (0·71, 0·94)). The ventilation-free days in 28 d of the TR group were 1·09 d longer than those in the TUR group (Coef: 1·09; 95 % CI (0·52, 1·67)), and vasopressor-free days in 28 d of the TR group were 1·61 d longer than those in the TUR group (Coef: 1·61; 95 % CI (1·10, 2·12)). There was no significant difference in the probability of AKI after the diagnosis of sepsis between the two groups (Table 2).
Subgroup analyses
We performed a subgroup analysis of the primary outcomes by using clinically significant scores and several complications (Fig. 2, online Supplementary Table S2). A significant interaction was observed in MI complication. The HR of people without MI was 0·88 (95 % CI 0·76, 1·01), and the HR of people with MI was 0·61 (95 % CI 0·44, 0·87). The use of thiamine has a significant protective effect on patients with MI.
Subgroup analysis of the associations between ICU all-cause mortality and thiamine received, *P < 0·05. Confounders were consistent with the multivariate model in Table 2.
Discussion
In the present study, we investigated the link between thiamine supplementation and sepsis mortality. After controlling for potential confounders using Cox regression models, we found that the use of thiamine is associated with the improvement of ICU all-cause mortality in patients with sepsis. This observation was also supported by the double robustness validation on the matched and weighted populations. Secondary outcome analyses demonstrated that the use of thiamine was associated with the improvement of 28-d mortality, ventilation-free days in 28 d and vasopressor-free days in 28 d in patients with sepsis, while there was no obvious association with the occurrence of AKI.
The biologically active form of thiamine, thiamine pyrophosphate, is an essential co-enzyme for glucose metabolism, which provides energy through the regulation of mitochondrial production of NADPH and ATP(Reference Manzanares and Hardy25). Additionally, thiamine levels are positively correlated with the activity of glutathione peroxidase, a major component of the cellular antioxidant system, which is also critical in scavenging oxygen-free radicals(Reference Costa, Gut and de Souza Dorna26). Thiamine’s important role in cellular energy metabolism, oxidative stress and maintenance of mitochondrial function provides the basis for its therapeutic application in sepsis. The human body cannot synthesise thiamine by itself and has limited storage, so it must rely on external sources to avoid deficiency. Patients with sepsis have an increased need for thiamine due to infection, but the accompanying decrease in nutritional intake makes its deficiency common(Reference Sedhai, Shrestha and Budhathoki27). Moreover, the combination of multi-organ dysfunction in sepsis patients may impede the transport and utilisation of thiamine, and operations such as haemodialysis may also increase its excretion(Reference Polegato, Pereira and Azevedo28). Animal tests show that sepsis inhibits the intestinal absorption of thiamine, resulting in less ATP production and affecting energy metabolism(Reference Sassoon, Zhu and Fang29). In one study, the prevalence of thiamine deficiency in patients with sepsis was 10 % upon initial admission, increasing to 20 % within the next 72 h(Reference Donnino, Carney and Cocchi30). Thiamine deficiency is common in critically ill adults and children(Reference Costa, Gut and de Souza Dorna26,Reference Raj, Baranwal and Attri31) . A retrospective analysis of critically ill patients showed that thiamine levels were significantly higher in patients who survived than in those who died, and thiamine deficiency is associated with higher mortality(Reference Cruickshank, Telfer and Shenkin32), in agreement with our results. As such, thiamine supplementation is of great interest to improve the prognosis of patients. Although intravenous thiamine supplementation showed no statistically significant relationship with the occurrence of AKI after sepsis, it does reduce the time that patients are treated with a ventilator and vasopressors, which improves cardiopulmonary function and general condition. The effects of hypoxia, hormones and inflammatory mediators on mitochondria in critical conditions are thought to be functional changes rather than structural changes and may be reversible(Reference Singer, De Santis and Vitale33). Therefore, exogenous supplementation with thiamine could promote the recovery of mitochondrial function, prevent progression of the disease and reduce the mortality of patients with sepsis.
Subgroup analysis showed a beneficial effect of thiamine in patients with MI, with an HR of 0·61 (95 % CI 0·44, 0·87). The cardiovascular system is essential for maintaining adequate organ perfusion, and its failure to function can affect the progression of sepsis. Cardiac muscle is rich in mitochondria that enable it to provide a large amount of energy for the systolic activity of the heart(Reference Barth, Stämmler and Speiser34). In patients with MI, the presence of necrotic and apoptotic myocardial cells leads to a reduction in the number of mitochondria, despite the ability of the rest of the healthy cells to compensate for the energy needed for myocardial activity. Sepsis under autonomic nervous system disorders, reduced blood volume and the release of endogenous cytokines all damage cardiac muscle(Reference Snow, Dickey and Tapp35,Reference Dhainaut, Huyghebaert and Monsallier36) . In addition, the electron transport chain(Reference Lee and Hüttemann37), abnormal oxidative phosphorylation and various mechanisms mentioned above in sepsis can impair mitochondrial function(Reference Arulkumaran, Deutschman and Pinsky38). Thus, intravenous thiamine supplementation may protect mitochondrial function in patients with MI, thereby reducing their risk of death.
Strengths and limitations of the study
The advantage of this study is the use of the Medical Information Mart for Intensive Care IV database, which has a large sample size and contains relatively recent data, thus providing strong evidence for our conclusion. For further verification, we obtained the same results after adjusting for baseline levels by using PSM and inverse probability of treatment weighting. However, this study also has some limitations. First, this study is a single-centre clinical study. Second, we included only patients who received thiamine intravenously in consideration of gastrointestinal dysfunction in critical conditions. The prognostic impact of different routes of administration on patients could be studied in the future. Third, we did not consider the effects of specific injection doses and timing. Fourth, because patients have a large number of missing values of lactate levels 24 h after the diagnosis of sepsis, we were unable to determine whether thiamine improves the prognosis of patients by reducing the concentration of lactate. This can be verified in future studies. One more point was the lack of test results for thiamine levels in patients in the database, so we could not take it into account in this study.
Conclusion
Supplementation with thiamine has a beneficial effect on the prognosis of patients with sepsis. More randomised controlled trials are needed to confirm the effectiveness of thiamine supplementation in the treatment of sepsis.
Acknowledgements
This study received financial support from National Natural Science Foundation of China (No. 82072232; No. 81871585), Science and Technology Program of Guangzhou, China (No. 202201020028), Special Projects in Key Areas of General Colleges and Universities in Guangdong Province (No. 2022ZDZX2003), the 2021 Annual Medical Teaching and Education Management Reform Research Project of Jinan University (No. 2021YXJG029), Clinical Frontier Technology Program of the First Affiliated Hospital of Jinan University, China (No. JNU1AF-CFTP-2022-a01235) and the Science and Technology Projects in Guangzhou, China (No. 202201020054).
L. Z. created the study protocol, performed the statistical analyses and wrote the first manuscript draft. F. Z. conceived the study and critically revised the manuscript. S. L. and F. X. assisted with the study design and performed data collection. X. Z. assisted with data collection and manuscript editing. T. H. assisted the analysis and explained statistical methods. J. L. assisted with manuscript revision and data confirmation. H. Y. contributed to data interpretation and manuscript revision. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Supplementary material
For supplementary material referred to in this article, please visit https://doi.org/10.1017/S0007114522003373
Open access
