Low circulatory levels of total cholesterol, HDL-C and LDL-C are associated with death of patients with sepsis and critical illness: systematic review, meta-analysis, and perspective of observational studies

Search strategy

Embase, Ovid Medline, Cumulative Index to Nursing and Allied Health Literature (CINAHL), Cochrane Central Register of Controlled Trials, Web of Science, Pubmed and SCOPUS databases were searched from inception to 03 October 2023, to identify prospective observational studies which investigated associations between serum cholesterol (total cholesterol and/or HDL-C and/or LDL-C) on day of ICU admission and mortality in patients with critical illness. Reference lists of included studies were hand-searched to identify additional studies. An attempt was made to contact all authors via the listed correspondence email address to receive patient level data and no unpublished studies were used. A combination of key word search terms related to sepsis, cholesterol and intensive care were combined with relevant MeSH terms to identify potentially relevant articles from databases. Full search strategies and logic are available in Supplementary material.

Study selection

Duplicates were removed and articles were screened for potential relevance based on their title and abstracts. Full texts of relevant studies were retrieved and assessed for suitability of inclusion by whether they met inclusion and exclusion criteria, conference abstracts with no full-text available were not included. Inclusion criteria were: [1] prospective study, [2] adult human subjects (≥16 years of age), [3] intensive care population, [4] in-hospital, ICU or short-term mortality (<30 day) recorded, [5] serum total cholesterol and/or HDL-cholesterol and/or LDL-cholesterol measured on day of ICU admission, [6] authors measured an association between serum cholesterol and hospital mortality. Exclusion criteria were: [1] interventional study assessing efficacy and/or safety of a therapeutic intervention in addition to standard of care. Non-English articles were considered if full text was available. The database search and study selection process were carried out by two independent reviewers (RT, CZ) with any discrepancies resolved by discussion and mutual consensus.

Data extraction

The following data were extracted from the final list of articles to enable assessment of whether serum cholesterol was associated with mortality in prospective observational ICU cohorts: [1] first author surname, [2] year of publication, [3] country and setting where study was performed, [4] study design and characteristics, [5] number of participants, [6] inclusion and exclusion criteria of study, [7] participant age and sex, [8] serum cholesterol mg/dL or mmol/L (total, LDL-C or HDL-C) assessed within 24 h of ICU admission, [9] in-hospital mortality, ICU mortality or 28/30 day mortality recorded, [10] key results relating to serum cholesterol and mortality.

Risk of bias assessment

A modified version of the Newcastle–Ottawa Scale (NOS) for cohort studies was used to assess study quality.²⁸ As part of this process, we identified potential confounders of the relationship between HDL-C, LDL-C and total cholesterol and mortality. We reviewed each paper for evidence that the authors measured and/or adjusted for these variables (Supplementary Table S2).

Statistics (meta-analysis and meta-regression)

Unadjusted serum cholesterol concentration on the first day of ICU admission was reported by included studies as either mean (standard deviation) or median (interquartile range). We converted median and interquartile range to mean and standard deviation using previously outlined methods.²⁹ To compare risk factors across non-survivor and survivor study groups we combined data between studies for each of three independent variables: Total cholesterol (22 studies, n = 2419), LDL cholesterol (16 studies, n = 1386), and HDL cholesterol (17 studies, n = 1542). Mean difference (95% CI) between surviving and non-surviving patients was calculated for each independent variable. Unadjusted serum cholesterol levels as reported in studies were used for this analysis. When presented in mmol/L serum cholesterol levels were converted to mg/dL using a multiplication factor of 38.66976. All meta-analyses estimated overall effects under a Random Effects Model. We did not assess for skewness in our meta-analysis due to a lack of availability of individual patient data in the included studies. Heterogeneity was calculated using the I² statistic and I² values greater than 50% and 75% were considered to indicate moderate and high heterogeneity, respectively. An effect size p-value of <0.05 for the t-distribution in the meta-analyses and meta-regressions was considered significant. Publication bias was assessed visually using funnel plots and also using Egger's regression test to quantify asymmetry with p-value <0.05 as the threshold for statistical significance. Additionally, a meta-regression was conducted for all three variables (TC, HDL-C, LDL-C) to allow adjustment for age and the proportion of males in each study that had this data available, with the adjusted forest plots produced. All meta-analyses and regressions were conducted using R package Metafor 3.8-1.

Role of funders

The funders had no role in the design of the study; the collection, analysis, and interpretation of data; or approval of the final manuscript.

Ethics

Ethics committee approval was not required for this systematic review and meta-analysis of anonymous publicly available datasets.

Study selection

Fig. 1 shows a flow diagram of database searches. The search (Medline, Embase, CINAHL, Cochrane Central Register of Controlled Trials Database, Pubmed, Scopus, Web of Science from inception to 03 Oct 2023) identified 12,777 records. One additional record was identified from handsearching the references of included studies.³⁰ After duplicates were removed there was 6469 unique records. The titles and abstracts of these records were screened for potential inclusion. This led to 48 full text articles being assessed for eligibility, with 29 articles included in systematic review. Twenty-four studies were included in quantitative meta-analysis, primarily using numerical data from manuscripts and supplemental data, with one study included using unpublished mortality stratified data obtained from contacting the authors.³¹ Our study selection process is outlined in Fig. 1 and excluded studies with reasons for exclusion are summarised in Supplementary Table S3.

Study characteristics

A total of 29 studies published between 2000 and 2022 were included in the review^17,19, 20, 21, 22^,30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53 (Table 1). All studies were prospective observational cohort studies with mixed sex adult participants. There was a total of 3003 participants across all 29 studies with study size ranging from 17 to 502. Age of participants were reported in all but one study³⁰ as either median or mean age and this ranged from 34 to 75. Twenty eight of the studies assessed serum total cholesterol on day of admission to critical care,^17,19, 20, 21, 22^,30, 31, 32, 33, 34, 35, 36, 37, 38, 39^,41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53 twenty two assessed HDL cholesterol^17,19,20,22,31, 32, 33, 34^,36, 37, 38^,40, 41, 42, 43^,45, 46, 47, 48, 49, 50, 51 and 20 assessed LDL cholesterol.^17,19,20,22,32, 33, 34^,36, 37, 38^,41,43,45, 46, 47, 48, 49, 50, 51 Characteristics of included studies are shown in Table 1. Twenty-four of these studies were synthesised in quantitative meta-analysis with 2542 participants.^17,20, 21, 22^,30, 31, 32^,34,36,37,39, 40, 41^,43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53 All studies assessed patients admitted to critical care. One study⁵¹ assessed patients with critical illness and sepsis in the emergency department with a high proportion (>90%) admitted directly to the ICU. As patient level data was available, we were able to analyse the cohort who were admitted to ICU as part of our meta-analysis. Thirteen of the 24 studies included in the meta-analysis recruited patients with a diagnosis of sepsis on ICU admission.^{17,20,21,34,40,41,44,46,}48, 49, 50, 51^,53 The remaining studies included in the meta-analysis assessed participants with: acute kidney injury,³⁹ acute liver failure,³² acute heart failure,²² severe acute pancreatitis,⁴³ multiple organ failure,³⁰ systemic inflammatory response syndrome (SIRS),³⁶ surgical ICU admissions,⁴⁷ a mixture of patients with sepsis and non-septic controls,³¹ general ICU cohort,⁵² severe community acquired pneumonia³⁷ and COVID-19 pneumonia.⁴⁵

Risk of bias assessment

Study quality ranged from 5 to 7 out of a total possible score of 9 (Supplementary Table S2). The greatest area of potential bias which was uniform across all studies, was the poor comparability of the survivor and non-survivor cohorts as a result of undetected (and/or unadjusted) confounding factors which were unaccounted for. Several studies reported on important baseline confounding factors such as age, body mass index (BMI) and co-morbidity, and performed some degree of multivariable regression analysis, however only six performed multivariable regression analysis using demographic and/or co-morbidity factors.^{20,31,32,38,41,53} These studies also failed to account for all the important confounding factors which we identified in our modification of the NOS tool (Supplementary Table S2).

Meta-analysis

Data from 24 studies with 2542 patients were pooled in quantitative meta-analyses.^17,20, 21, 22^,30, 31, 32^,34,36,37,39, 40, 41^,43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53 Pooled weighted mean differences were used to quantify differences between serum total, HDL cholesterol and LDL cholesterol levels in survivors and non-survivors.

A comparison with other systematic reviews

A recent small scale systematic review focusing on HDL-C on admission to ICU, involving 791 participants provides further support for an association of HDL-C with sepsis outcome.²³ More recently another systematic review of cholesterol levels and sepsis by Hofmaenner et al.,²⁴ involving 1283 participants, provides additional agreement with our findings regarding sepsis. The key differences between this review and that of Hofmaenner appears to relate to the slightly more restrictive inclusion criteria we applied and hence the following studies were excluded from our analyses: Yamano et al.⁷¹ on the basis that this was considered a retrospective investigation, Guirgis et al., 2020⁷² as it was a phase 1 interventional trial of a lipid emulsion, Sharma et al., 2017 and 2019^73,74 as patient recruitment and/or lipid sampling in those studies was within 48 hours and not 24 hours of ICU admission and Guirgis et al., 2021⁷⁵ as we lacked patient level data to analyse the subgroup who were admitted to ICU from that study. While our review does not include an analysis of triglyceride levels Hofmaenner et al.²⁴ found no association between mortality and admission triglyceride levels. However, in contrast to Hofmaenner et al.²⁴ we provide a study quality assessment using the Newcastle Ottawa scale, assess heterogeneity using meta-regression and assessed for evidence of publication bias. Most notably our meta-analysis extends the observed suppression to a much larger sample size (2542 participants) and unlike previous published meta-analysis studies include non-survivors across non-sepsis ICU cohorts, indicating a more general nonspecific host response link of cholesterol homeostasis with inflammation and thus is not exclusive to an infection response. Key differences between this review and Hofmaenner et al. are summarised in Supplementary Table S4.

Thus, sepsis and acute illness in non-survivors leads to a coordinate lowering (around 7–14%) in all major measures of circulatory cholesterol, namely TC, HDL-C and LDL-C. This is consistent with a regulated acute phase inflammatory reaction that can lead to a maladaptive-setpoint of cholesterol metabolism. Key upstream mediators are inflammatory cytokines such as IL-6 and IL-1. Our results of lower levels in non-survivors supports the notion for a greater amplitude and duration of this inflammatory reaction that leads to deaths⁷⁶

Cholesterol homeostasis, immunometabolism and inflammation

Structural and functional changes of HDL and LDL occur during inflammation and acute phase response (Fig. 5a). During inflammation the most prominent changes take place for HDL and associated lipids,^77,78 although changes in LDL can also occur and include increased oxidation of LDL to a more pro-inflammatory phenotype and an increase in levels of the pro-inflammatory Lipoprotein (a).77, 78, 79, 80 The measurements of HDL-C and LDL-C in our analyses does not account for these morphological or functional changes. While the biological role for these changes is not precisely known it is clear they are under inflammatory control, where IL-6 signalling has been shown recently to explain the reduction in small HDL particle numbers.⁶⁸

Cellular cholesterol biosynthesis has a central rate determining role in cholesterol homeostasis, and flux activity and products of the pathway are integral to host-protection governing inflammatory and adaptive immune cell functions.^4,5 Notably, the cross regulation between inflammation and cholesterol metabolism in macrophages and other immune cells (termed immunometabolism) is well established and where in acute infection Toll Like Receptor and interferon signalling coordinately suppresses flux in cholesterol biosynthesis and stimulates the production of oxysterols that directly control innate and adaptive immune reactions, including the switching of adaptive immune cell functions from inflammatory (IFN gamma) to resolving (IL-10) phenotypes.^4,5 In this regard small HDL particles have a vital contribution in controlling the efflux of cholesterol⁸¹ and hence is tightly linked to feedback regulation of flux in cholesterol biosynthesis in immune cells (Fig. 5b). Furthermore, oxidised LDL can further promote pro-inflammatory effects (Fig. 5b). In particular, flux control in macrophage cholesterol biosynthesis is central to determining the immunophenotype of macrophages and activation of the inflammatory cascade.^4,5 Accordingly, from an immunometabolism perspective prolonged hypocholesteremia has the risk of potentially driving macrophage states from a pro-inflammatory to anti-inflammatory phenotype. Understanding how systemic changes are mechanistically interlinked with immunometabolism of innate and adaptive cellular effector functions is incomplete and is an area that warrants further research.

Prognostic metabolic biomarkers in sepsis and critical illness—a potential role for HDL associated biomarkers

Nuclear magnetic resonance (NMR) measurement of HDL particle count and particle size, indicate that pre-existing low number of small HDL particles increases the risk of death from infectious disease.^68,82 However, while small HDL particles are the most relevant form in terms of function in a very recent study, enumeration of HDL particle size by NMR, has shown not to have any causal association with death from sepsis⁶⁸ and may point to a “red herring” for diagnostics and therapeutic intervention. However, the efficacy of HDL particles is strongly dependent on the local environment partially driven by Apolipoprotein A1 (ApoA1), Paraoxonase 1 (PON1), and lecithin-cholesterol acyltransferase (LCAT) activity.^83,84 Therefore, the functionality of HDL particles is important in addition to HDL quantity or size. The accurate measurement of HDL particle size is not trivial and instrumentation such as NMR present significant challenges (both technical and financial) for a near term diagnostic platform. However, this does not rule out a role for other HDL associated biomarkers. In this connection, studies of the HDL lipidome have revealed that cholesteryl esters (30–40% of total lipid), rather than free cholesterol (5–10%) predominate and account for the major proportion of cholesterol in the small HDL particles⁸⁴ and could potentially offer a more readily implementable HDL-biomarker. Circulatory levels of esterified cholesterol are also an excellent functional marker of LCAT enzyme activity that is responsible for producing lysophospholipids on small HDL particles.⁸⁴ Inflammatory induced phospholipases are associated with reduced HDL cholesterol and are associated with changes in lysophospholipid composition which are further imbued with pro- and anti-inflammatory activity.⁸⁴ A more complete characterisation of the HDL lipidome in chronic and acute inflammation is warranted. As a cautionary note, HDL-cholesterol levels and to a lesser extent NMR measured HDL are not completely accurate indicators of the concentrations of various functional and physical subtypes of HDL.⁸⁵ Nevertheless, comparisons within a given methodology consistently reveals an association of low HDL with death from sepsis and critical illness. Therefore, future clinical studies investigating the prognostic utility of HDL-based biomarkers, in combination with other host-based biomarkers, is strongly warranted.

Therapeutic potential of HDL

Evidence for a direct causal role of HDL-C in contributing toward a maladaptive host response is more controversial. As mentioned above a recent large population study of 270,000 participants with NMR based data showed that low levels of small HDL particles have no causal role with sepsis and sepsis deaths.⁶⁸ Critically, this study reveals that low levels of small HDL particles are causally linked to IL-6 signalling (Fig. 5). As a corollary, in an acute infection a large IL-6 driven inflammatory response would lead to a strong suppression of HDL and is suggestive that low levels in sepsis is a bystander effect of high inflammatory reaction. A possible protective role of HDL-C in sepsis cannot be ruled out. In this connection clinical genetics and murine models that demonstrate that inhibition of the cholesteryl ester transfer protein (CETP), whose function is associated with decreases in cholesterol esters and circulating HDL-C, is protective against mortality in acute sepsis.⁸⁶ In contrast gain-of-function mutations affecting CETP, leading to decreased HDL-C, are associated with increased risk of mortality due to sepsis.⁸⁶ Gaining more precise mechanistic insight into the multi-faceted relationship between sepsis and HDL and host immunometabolism and whether this represents a protective inflammatory response or pathological change may identify novel therapeutic targets and interventions for use in critical illness.

The notion of increasing circulatory levels of HDL in patients with sepsis was first piloted in preclinical models using reconstituted HDL⁸⁷ but failed clinical translation due to endotoxin contamination, low purity and pharmacokinetic properties. More recently, the increasing number of clinical and pre-clinical studies involving the administration of synthetic HDL particles, that show high clinical grade purity and safety, in acute coronary syndrome has rekindled interests in sepsis.^11,26,88, 89, 90, 91, 92 These studies showed not only an attenuating effect of synthetic HDL on endotoxin shock but also the suppression of inflammatory signalling in macrophages and endothelial cells, raising potential for a clinical translatable therapy. However, we propose that it would be prudent before embarking on therapeutic trials for sepsis, to precisely delineate if causality is associated with specific HDL functional class(es) and risk of mortality. As caveat emptor, trial failure risks curtailing key research to uncover critical mechanistic links.

Limitations

A significant limitation relates to the inability to fully control for patient characteristics that could simultaneously affect both outcome variable, mortality, and the serum concentrations of HDL-C, LDL-C, and total cholesterol. Whilst we are aware of, and actively sought evidence for the effects of potential confounders whilst assessing study quality, we could not assess this at an individual patient level and adjust for factors other than age and sex. Similarly we were only able to utilise mean age and male sex as covariables in our meta-regression. There was also variability amongst the 24 studies synthesised for meta-analysis in accounting for the use of cholesterol lowering drugs such as statins. Of these only seven excluded patients using statins or excluded hyperlipidemia as a comorbidity and had no statin use.^{17,32,40,41,43,44,46} Four additional studies^22,31,48,50 reported on statin use in their cohort, of which two of them^22,31 assessed whether there was a relationship between statin use and survival and concluded there was no significant difference.

Conclusions and future research

We provide a strong basis to support that a drop in HDL-C levels either prior to or early during sepsis and critical illness has prognostic value. We suggest that subclass specific HDL-biomarkers could have clinical utility when used alongside existing metabolic markers such as lactate in the early detection of sepsis and development of septic shock. We propose future investigations should consider circulatory levels of cholesterol esters as a possible readily implementable HDL associated biomarker. Future work should also aim to precisely define HDL functional subclasses and associated biomarkers that could be used as a reference range below which serum HDL cholesterol is significantly suppressed and likely to be associated with severe illness. In this regard, with increasing interest in exploring synthetic HDL and lipid supplementation as therapeutic interventions for sepsis, consideration of the available evidence for both association and a causation in sepsis is called for. Hence, in addition to quantitative lipoprotein analyses, functional assays for HDL particles should also be investigated. Our analysis was restricted to adults, and although limited studies in neonates can be found,⁹³ the question remains for future studies of whether similar observations are evident in early life. We also highlight areas in which further research can be more rigorous, through careful selection of patient cohorts and adjustment for relevant co-morbidities which may affect both lipid profile and ICU outcome and measuring more precisely the various functional and structural subclasses of HDL.