It is a systematic review and meta-analysis assessing risk factors for sepsis-associated encephalopathy. Before the study started, a proposal was developed and registered in PROSPERO (CRD42023483721) (see Supplementary Table S1 for PROSPERO protocol). This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines (20).

We based search terms on Medical Subject Headings (MeSH) and other standard (controlled vocabulary) terms. A concept-based approach was used, including terms related to “Sepsis,” “Encephalopathy,” “Delirium,” “Risk Factors,” and others. Supplementary Table S2 details the entire search strategy. Two reviewers (ZZ and LJ) independently searched the Web of Science, Cochrane Library, Embase, and PubMed databases from inception to August 10, 2023. Our research team then searched the references of relevant studies to identify possible additional eligible studies. We used Endnote documentation software as a literature screening tool. Our research team repeated this process until we found no new relevant papers. Any discrepancies or differences were resolved through discussion or consultation with a third reviewer to reach a consensus.

Data extraction was independently and systematically conducted by two reviewers (ZZ and LG) using a pre-specified data extraction form; any disputes were resolved through consensus or adjudicated by a third reviewer (HZ) until the disagreement was resolved. Data collection included study characteristics: first author, publication year, study design, total number of patients, age, gender, incidence of SAE, sepsis diagnostic criteria, and SAE diagnostic criteria. The baseline characteristics of the included studies are illustrated in Table 1.

When only 1 study provided data on the association between a potential risk factor and SAE, we reported the estimate extracted from the original article in the table (i.e., no meta-analysis was performed). When two or more studies provided data on the association between a potential risk factor and SAE, we calculated a meta-analyzed estimate of that possible association. When encountering studies that provided unadjusted and adjusted estimates, we preferred to use adjusted effect estimates because they represent effect estimates closer to the actual value (i.e., less biased). If forecasts from different overlapping cohorts were available, we used data from the report with the largest sample size to avoid duplicating data in the meta-analysis.

The diagnostic criteria for sepsis are constantly being updated. There are also no uniform diagnostic criteria for sepsis-associated encephalopathy. In each study we included, the diagnosis of both conditions varied. If it does not violate general principles, we respect the diagnostic criteria for sepsis and sepsis-associated encephalopathy used by researchers in each study and directly extract patient information from the data published in each study. The primary outcome of this study was the risk factors of sepsis-associated encephalopathy.

We will group the studies included based on the type of study design (case-control studies or cohort studies). Additionally, due to the lack of uniform diagnostic criteria for sepsis-associated encephalopathy, we will also group the studies based on whether the diagnosis in the included studies was made using objective scoring (GCS <15 or CAM-ICU) or based on the subjective evaluation of medical personnel. If heterogeneity analysis is required, we will determine the source of heterogeneity by employing a study-by-study exclusion approach for the included studies.

Two independent authors assessed the included studies’ methodological strength and risk of bias using the Newcastle–Ottawa Scale (NOS) (22). We used this assessment tool to evaluate the design quality of nonrandomized case-control and cohort studies. Scores were assigned based on selection criteria, comparability, and outcome (cohort studies) or exposure (case-control studies). The maximum score of 9 reflects the highest quality.

We used the Grades of Recommendations, Assessment, Development, and Evaluation (GRADE) evidence rating system to assess the credibility of the available evidence for the relevant factors associated with this meta-analysis (23, 24). We initially regarded observational studies as being of poor quality. Depending on the GRADE criteria, the level of credibility may be reduced (with five domains, including the risk of bias, inconsistency, circumstantial evidence, uncertainty, or publication bias) or increased (with three fields, including larger effect sizes, dose-response relationships, or confounders). We ultimately categorized the credibility of the evidence as high, moderate, low, or very low.

We used Review Manager (version 5.3; Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) (25) to perform meta-analysis. Continuous variables were pooled using standard mean difference (SMD), and We calculated a 95% confidence interval (CI). In contrast, binary variables were pooled using odds ratio (OR), and a 95% confidence interval (CI) was calculated. A fixed effects model was used when statistical heterogeneity was low (I² ≤ 50%). When statistical heterogeneity was high (I² > 50%), meta-analysis was performed using a random-effects model, and sensitivity and subgroup analyses were performed to investigate potential sources of heterogeneity (26). p < 0.05 was considered statistically significant.

A flow diagram for study selection is shown in Figure 1. Through database searching from inception to August 10, 2023, the date of our final search, we identified 2,041 original literature records. After the removal of duplicates, 1,718 records remained. Of these, 1,645 were excluded after screening titles and abstracts as they met our exclusion criteria. Excluding one article that could not be retrieved, we assessed the full texts of the remaining 72 articles. Ultimately, a total of 13 studies met the eligibility criteria for a full-text review (27–39).

Of the 13 studies included, there were six cohort studies and seven case-control studies. All these studies were published in English. The total sample size of these studies was 1906, with a cumulative case group of 864 and a control group of 1,042. The prevalence of SAE ranged from 17.67 to 83.59%. The studies were published between 2011 and 2023. The included studies were conducted in multiple countries, including China (n = 10), South Korea (n = 1), Finland (n = 1), and Belgium (n = 1). Thirteen studies were conducted in single-center settings. One study exclusively utilized the Glasgow Coma Scale (GCS) criteria for assessing SAE. Four studies solely employed the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) criteria for evaluating SAE. Three studies incorporated both GCS and CAM-ICU criteria for the assessment of SAE. Additionally, five studies assessed SAE through clinical records maintained by healthcare professionals (more details are seen in Table 1).

According to the NOS scores, 12 were high-quality studies, and one was of moderate quality (Supplementary Table S3). When performing multivariate analyses, explanations of the factors that accounted for most of the lost points were made more evident.

In our study, observational cohort studies were initially considered low quality. However, there was no high-quality evidence for any of the risk factors regarding deterministic assessment. Only APACHE II, SOFA, age, dependent activities, high care needs, low level of consciousness, and ALT showed moderate-quality evidence (Table 2). For all other risk factors, the certainty of evidence was categorized as low. The reduced quality of evidence for some risk factors was attributed to variations in study results and imprecision.

The main strength of this study is that it is the first meta-analysis of risk factors for SAE, where multiple factors affecting SAE were analyzed separately. However, like all studies, this study has some limitations. First, the diagnostic criteria for sepsis and sepsis-associated encephalopathy differed between studies. The diagnostic criteria for sepsis 1.0 have high sensitivity but low specificity, which may lead to overdiagnosis while missing some immunosuppressed patients. Sepsis 2.0 is a transitional criterion. However, the latest definition of sepsis 3.0 remains controversial. When a patient’s SOFA score changes by ≥2 points, the patient’s condition is exacerbated, caused by an infection that may adversely affect the early recognition and treatment of sepsis and lead to a delayed diagnosis of the disease. In addition, non-infectious conditions in critically ill patients may also lead to organ damage that can achieve a SOFA score of ≥2, leading to overdiagnosis. The diagnostic criteria for burn infection are expressly limited to burn patients (21). All four diagnostic criteria for sepsis are deficient. Among the currently used diagnostic criteria for SAE, GCS <15 and CAM-ICU are relatively objective diagnostic methods. At the same time, data recorded by medical staff is somewhat subjective; all three diagnostic procedures are based on consistent clinical symptoms of SAE. Second, only English-language databases were searched, possibly excluding relevant studies published in other languages. Third, only a few studies were included for certain factors such as tau protein, IL-6, and albumin, resulting in small sample sizes for these studies. Fourth, the exclusion of minors under the age of 18 limits the generalizability of the results. In addition, most of the studies were conducted in Asia and Europe, limiting the applicability of generalizing the results to other parts of the world.

Therefore, future research needs to address these limitations and provide a more comprehensive investigation of the risk factors associated with SAE. Efforts should include studies from different geographic regions and populations, consider a broader range of languages, and use standardized assessment methods. By overcoming these limitations, we can further advance our understanding of SAE and provide more substantial evidence for clinical practice.