Abstract
Importance
Whether low LDL-C levels are associated with increased risk of sepsis and poorer outcomes is unknown.
Objective
To examine the association between low-density lipoprotein cholesterol (LDL-C) levels and risk of sepsis among patients admitted to the hospital with infection.
Design, Setting, and Participants
Using de-identified electronic health records (EHRs) we defined a cohort of patients admitted to Vanderbilt University Medical Center with infection. Patients were white adults, had an International Classification of Diseases (ICD) code indicating infection, and received an antibiotic within 1 day of hospital admission (N=61502).
Interventions
We extracted (1) clinically measured LDL-C levels (excluding measurements <1 year before hospital admission and those associated with acute illness) (N=3961, female 2288 and male 1673), and (2) a genetic risk score (GRS) for LDL-C (N=7804, female 3590 and male 4214).
Main Outcomes and Measures
The primary outcome was sepsis; secondary outcomes were admission to an intensive care unit (ICU), and in-hospital death.
Results:
Lower measured LDL-C levels were significantly associated with increased risk of sepsis (odds ratio [OR] 0.86, 95% confidence interval [CI] (0.79 – 0.94), p=0.001) and ICU admission (OR 0.85, 95% CI (0.76 – 0.96), p=0.008), but not with in-hospital mortality (OR 0.80, 95% CI (0.63 – 1.00), p=0.055); however, none of these associations were statistically significant after adjustment for age, sex and comorbidity variables (risk of sepsis, OR 0.96, 95% CI (0.88 – 1.06); ICU admission, OR 0.94, 95% CI (0.83 – 1.06); in-hospital death, OR 0.97, 95% CI (0.76 – 1.22); all p values > 0.05). The LDL GRS correlated with measured LDL-C levels (p < 2.2 × 10−16) but was not significantly associated with any of the outcomes.
Conclusion:
Lower measured LDL-C levels were significantly associated with increased risk of sepsis and admission to ICU in patients admitted to the hospital with infection; however, this association was due to comorbidities since both the clinical models adjusted for confounders, and the genetic model showed no increased risk. LDL-C levels do not directly alter the risk of sepsis or poor outcomes in patients hospitalized with infection.
Introduction
In the United States, sepsis is a common cause for admission to an intensive care unit (ICU) and contributes to 1 in every 2 to 3 deaths in hospital.1–4 Sepsis is a complication of infection and is characterized by an uncontrolled systemic inflammatory response, organ failure, poor clinical outcomes, and a high mortality rate. There are no specific effective treatments for sepsis; thus, there is great interest in new approaches to prevent sepsis and treat patients.
The effect of lipoproteins on sepsis and its outcomes is one such area of interest. Lipoproteins, including low-density lipoproteins (LDL), bind toxic bacterial products such as lipopolysaccharide (LPS) that mediate many of the manifestations of sepsis such as vasodilation, increased capillary permeability, and decreased peripheral vascular resistance.5,6 Studies that manipulated LDL levels in animals suggested that LDL protected against LPS-induced mortality. 5,6
Studies in humans also suggest that LDL is protective against sepsis but are less clear. Patients with low LDL cholesterol (LDL-C) levels have an increased risk of sepsis and worse outcomes;7–9 however, these studies do not answer the question whether LDL modifies the risk of sepsis and poor outcomes directly, or if it does so indirectly through the effects of comorbid illness. A study performed in community-dwelling adults showed that low LDL-C levels at entry to the cohort were associated with increased risk of future sepsis,10 suggesting that LDL-C levels could affect the risk of sepsis directly.
Understanding the link between LDL-C levels and sepsis is important because newer lipid-lowering medications (i.e., proprotein convertase subtilisin/kexin type 9 (PCSK9) inhibitors) can reduce LDL-C concentrations to very low levels.11,12 Also, administering lipoproteins to patients at risk for sepsis is a therapeutic strategy that could potentially prevent or ameliorate the disease. Thus, we tested the hypothesis that low LDL-C levels are directly associated with increased risk of sepsis and poorer outcomes using an epidemiological approach with careful attention to confounders, and a genetic approach that is less prone to confounding. We used a de-identified electronic health record (EHR) repository linked to a DNA biobank to define baseline measured LDL-C levels and an LDL-C genetic risk score (GRS). To define the association between LDL-C and sepsis we examined the association between measured LDL-C levels as well as the LDL-C GRS and three outcomes in patients admitted to the hospital with infection: (1) sepsis, (2) admission to an ICU, and (3) in-hospital mortality.
Methods
Data sources
Data were obtained from the Synthetic Derivative, a database that contains a de-identified copy of the EHR for every patient in the Vanderbilt University Medical Center (VUMC) system (N>2.5 million).13–15 ICD9 and ICD10 codes were used for outcomes and ICD9 codes for cohort construction and covariates.
Cohort identification
The study was approved by Vanderbilt University Medical Center Institutional Review Board (IRB# 150137). We identified white patients (≥ 18 years) who had been admitted to Vanderbilt hospital with an infection from 1993 to 2017 (Figure 1 and Supplementary Figure 1). The day of hospital admission was designated day 0. Infection was defined as having a billing code indicating infection (Supplementary Table 1) and receiving an antibiotic within one day of hospital admission (i.e., on days −1, 0, or +1). We selected ICD-9-CM codes for infection based on the criteria of Angus et al.16,17 excluding viral, mycobacterial, fungal and spirochetal infections. The qualifying antibiotics are shown in Supplementary Table 2. If more than one qualifying episode of infection was present in a patient’s EHR, only the first was included; in the sub-cohort with measured LDL-C levels the first episode of infection occurring after a qualifying LDL measurement was included. We further identified two sub-cohorts:
Figure 1. Overview of clinical and genetic approaches.
VUMC, Vanderbilt University Medical Center; LDL-C, Low-density Lipoprotein Cholesterol; GLGC, Global Lipid Genetics Consortium; ICU, Intensive care unit
(1) Sub-cohort with baseline LDL-C measured
To reduce the confounding effects of illness we excluded individuals with evidence of severe chronic illness in the year before the index hospital admission (Supplementary Figure 1). We excluded individuals with conditions such as HIV infection, chronic kidney disease, and liver disease, as well as those who had received cancer chemotherapy (Supplementary Table 3). We defined the baseline LDL-C for each individual as the median value of qualifying measurements. Qualifying LDL-C measurements were those performed > 1 year before the index hospital admission and excluded those performed (1) in hospital; (2) before 18 years of age; (3) within 30 days of a serum albumin level <3 g/dL; and (4) after the first mention of a statin in the EHR.
(2) Sub-cohort with genome wide genotyping available
We identified a sub-cohort in whom genome-wide genotyping had been performed (including the Infinium Multi-Ethnic Genotyping Array, Affymetrix Axiom Biobank Array, Illumina OMNI-Quad, HumanOmni5-Quad, Illumina 660W, Illumina 1M).
Outcomes
Primary outcome:
The primary outcome was sepsis—we used the Sepsis-3 definition and defined sepsis as concurrent infection and organ dysfunction identified using a validated algorithm to detect sepsis in the EHR with minor modifications (Figure 2; Supplementary Methods).1 The algorithm uses a combination of billing codes and clinical criteria and had a sensitivity of 69.7% and specificity of 98.1% in a previous study.1 To minimize the contribution of sepsis occurring after surgery or as a complication of events in the hospital, we studied sepsis occurring within one day of hospital admission (days −1, 0, and +1). Surveillance studies have shown that most sepsis cases (86.8%) are present on admission to the hospital.1 In brief, individuals with infection met the definition of sepsis if they had either septic shock or severe sepsis or met any organ dysfunction criterion. Septic shock and severe sepsis were defined by ICD codes that are highly specific (99.3%).1 Criteria for organ dysfunction included: (1) cardiovascular failure: use of the vasopressor levophed (norepinephrine), or use of the vasopressors dobutamine or dopamine and a billing code for administration of a vasopressor; (2) respiratory failure: use of codes for ventilation and admission to an ICU; (3) renal failure: doubling or greater increase of baseline creatinine (baseline creatinine: the lowest creatinine between 1 year before admission and hospital discharge); (4) hepatic failure: Total bilirubin ≥ 34.2 umol/L and doubled from baseline (the baseline value : the lowest total bilirubin occurring between 1 year before admission and hospital discharge) (5) hematologic failure: Platelet count <100,000 /microL and ≥ 50% decline from a baseline that must have been ≥100,000 (the baseline value : the highest platelet count occurring between 1 year before admission and hospital discharge)(see Supplementary Table 4 for codes to identify organ failure).
Figure 2. Algorithm to identify sepsis within the infection cohort.
ICD, International Classification of Disease; ICU, Intensive care unit
Secondary outcomes:
There were two secondary outcomes: admission to ICU and in-hospital mortality.
Covariates
We calculated median values for all BMI, HDL-C and triglyceride (Trig) values in the EHR for each individual; age was ascertained at the time of the index hospital admission. We also established the presence or absence of the comorbidities that comprise the Charlson/Deyo comorbidity index, with modifications. 18–20 For each patient, relevant diagnostic codes in the year before the index hospital admission were grouped (Supplementary Table 5) into PheCodes20 using the PheWAS package.21 The PheCodes were further grouped into the 16 modified Charlson/Deyo comorbidity categories; the diabetes and diabetes with complications categories were combined into a single variable: diabetes. Four of the comorbidity categories (mild liver disease, moderate and severe liver disease, AIDS/HIV and metastatic solid tumor) were not present in any individual in the sub-cohort in whom baseline LDL-C had been measured because of the exclusions applied to that group. Each comorbidity group was included as an independent covariate in the analysis.
Manual Chart Review
We manually reviewed a sample of EHRs to confirm phenotyping (Supplementary Methods, Supplementary Table 6).
Genetic risk score for LDL-C
We generated a GRS for LDL-C using 81 SNPs independently associated (p<5×10−8)with LDL-C in the meta-analysis of genome wide studies performed by the Global Lipids Genetics Consortium (GLGC) (Supplementary Methods, Supplementary Table 7).22 Specifically, the GRS was calculated for each individual by adding the number of minor alleles (0, 1, or 2) weighted for the effect size (β) of the SNP-LDL association. The association between the LDL GRS and measured LDL-C levels was validated in an independent cohort of 4313 whites, none of whom were included in the LDL GRS cohort.
Statistical Analyses
We used logistic regression to test the associations between LDL-C (measured values and GRS) and outcomes. We estimated odds ratios (ORs) per increase of one standard deviation (SD) and 95% confidence intervals (CIs) for the association between baseline LDL-C levels or LDL GRS and sepsis outcomes (sepsis, ICU admission and in-hospital mortality) adjusting for age, sex, and comorbidities in the measured LDL-C analyses, and in the GRS analyses for age and sex. Sensitivity analyses in the sub-cohort with measured LDL-C included: 1) additionally adjusting for HDL-C, Trig, EHR length and BMI; 2) excluding individuals (n= 473) who had a hospital admission for infection before a qualifying LDL measurement; 3) imposing no exclusions on the LDL-C measurement and using the value closest to admission and imposing no exclusions for co-morbidities (n=12334). In secondary analyses, we estimated ORs and 95% CIs for quartiles of LDL-C levels and LDL GRS using the highest quartile as reference and adjusting as in the primary analyses. Because we tested 3 outcomes, we considered p<0.0167 as statistically significant.
Results
Measured baseline LDL-C levels
There were 61502 patients who met the definition of infection. Of these 3961 had qualifying baseline LDL-C measurements (Figure 2). In this subgroup, 594 patients developed sepsis, 323 were admitted to the ICU, and 82 died during hospitalization. There were more females (2288, 57.8%) than males (1673, 42.2%) with an average of 14.2 ± 6.3 years of follow up in the EHR (Table 1).
Table 1.
Demographic Characteristics
| Demographic Variables | Measured baseline LDL-C (N=3961) |
Genetically predicted LDL-C (N=7804) |
||
|---|---|---|---|---|
| Sex | Female, No (%) | 2288 (57.8%) | 3590 (46.0%) | |
| Male, No (%) | 1673 (42.2%) | 4214 (54.0%) | ||
| Age (years), mean ± SD | 64.1 ± 15.9 | 59.75 ± 15.2 | ||
| BMI (kg/m2), mean ± SD | 30.32 ± 7.7 | - | ||
| Lipid panel | LDL (mg/dL), mean ± SD | 103.4 ± 32.6 | - | |
| Trigs (mg/dL), mean ± SD | 161.3 ± 108.3 | - | ||
| HDL (mg/dL), mean ± SD | 50.84 ± 17.1 | - | ||
| Myocardial infarction, No (%) | 518 (13.1%) | 1225 (15.7%) | ||
| Congestive heart failure, No (%) | 854 (21.6%) | 1829 (23.4%) | ||
| Peripheral vascular disease, No (%) | 349 (8.8%) | 746 (9.6%) | ||
| Cerebrovascular disease, No (%) | 742 (18.7%) | 1342 (17.2%) | ||
| Dementia, No (%) | 310 (7.8%) | 180 (2.3%) | ||
| Chronic pulmonary disease, No (%) | 1069 (27.0%) | 1987 (25.5%) | ||
| Rheumatic disease, No (%) | 243 (6.1%) | 429 (5.5%) | ||
| Peptic ulcer disease, No (%) | 110 (2.8%) | 296 (3.8%) | ||
| Mild liver disease, No (%) | NA | 615 (7.9%) | ||
| Diabetes, No (%) | 1241 (31.3%) | 2554 (32.7%) | ||
| Hemiplegia or paraplegia, No (%) | 63 (1.6%) | 218 (2.8%) | ||
| Renal disease*, No (%) | 144 (3.6%) | 647 (8.3%) | ||
| Any malignancy, including lymphoma and leukemia, except malignant, No (%) neoplasm of skin | 484 (12.2%) | 1717 (22.0%) | ||
| Moderate or severe liver disease , No (%) | NA | 571 (7.3%) | ||
| Metastatic solid tumor , No (%) | NA | 957 (12.3%) | ||
| AIDS/HIV, No (%) | NA | 105 (1.3%) | ||
NA= excluded for this cohort
Of the 61502 patients, 7804 had been genotyped: of these, 2520 developed sepsis, 1194 were admitted to ICU, and 243 died during hospitalization. Compared to the cohort with measured baseline LDL-C levels, they were younger (59.8 ± 15.2 years), more likely to be male (4214, 54.0%) and had more renal, hepatic, cancer and HIV co-morbidity because these conditions were exclusion criteria for the measured LDL-C cohort but not the GRS cohort (Table 1). There were 802 patients who were in both cohorts.
Measured baseline LDL-C and (1) sepsis, (2) ICU admission and (3) in-hospital mortality
LDL-C levels were significantly inversely associated with risk of sepsis (odds ratio [OR] 0.86, 95% CI 0.79–0.94; p=0.00127) and ICU admission (OR 0.85, 95% CI 0.76–0.96; p=0.00795), but not in-hospital mortality (OR 0.80, 95% CI 0.63–1.00; p=0.055). These associations were not significant after adjustment for age, sex, and comorbidity variables (sepsis: OR 0.96, 95% CI 0.88–1.06, p=0.42; ICU: OR 0.94, 95% CI 0.83–1.06, p=0.32; in-hospital death: OR 0.97, 95% CI 0.76–1.22, p=0.79, Table 2). Sensitivity analyses that additionally adjusted for HDL-C, Trigs, EHR length and BMI yielded similar results (sepsis: OR 0.95, 95% CI 0.86–1.04, p=0.25; ICU: OR 0.97, 95% CI 0.86–1.10, p=0.61; in-hospital death: OR 0.98, 95% CI 0.76–1.24, p=0.85). An analysis that excluded individuals with a previous admission to hospital for infection (sepsis: OR 0.95, 95% CI 0.86–1.05, p=0.34; ICU: OR 0.97, 95% CI 0.85–1.11, p=0.69; in-hospital death: OR 1.01, 95% CI 0.78–1.29, p=0.95) was consistent with the main analysis (Supplementary Table 8). Similarly, the sensitivity analysis that used the LDL-C value closest to admission and had no exclusions for comorbidities yielded results similar to those of the primary analysis (with full adjustment, sepsis: OR 0.98, 95% CI 0.94–1.02, p=0.24; ICU: OR 1.04, 95% CI 0.98–1.09, p=0.19; in-hospital death: OR 0.92, 95% CI 0.83–1.01, p=0.078, Supplementary Table 9).
Table 2.
Associations between LDL-C and sepsis-related adverse outcomes
| Cohorts | Phenotypes | Unadjusted | adjusted | ||
|---|---|---|---|---|---|
| Odds Ratio | P-value | Odds Ratio | P-value | ||
| (A) Measured baseline LDL-C, N=3961 | Sepsis | 0.86 (0.79 - 0.94) | 0.00127 | 0.96 (0.88 - 1.06)* | 0.42* |
| ICU admission | 0.85 (0.76 - 0.96) | 0.00795 | 0.94 (0.83 - 1.06)* | 0.32* | |
| in-hospital death | 0.80 (0.63 - 1.00) | 0.0549 | 0.97 (0.76 - 1.22)* | 0.79* | |
| (B) GLGC-based LDL-C GRS, N=7804 | Sepsis | 1.02 (0.97 - 1.07) | 0.41 | 1.02 (0.97 - 1.07)# | 0.37# |
| ICU admission | 1.01 (0.95 - 1.07) | 0.84 | 1.01 (0.95 - 1.07)# | 0.78# | |
| in-hospital death | 0.92 (0.81 - 1.04) | 0.17 | 0.92 (0.81 - 1.05)# | 0.22# | |
LDL-C, Low-density Lipoprotein Cholesterol; GLGC, Global Lipid Genetics Consortium; ICU, Intensive care unit; GRS, Genetic risk score
= adjusted for age, sex and comorbidity covariates;
= adjusted for age and sex
GLGC-based LDL-C GRS and (1) sepsis, (2) ICU admission and (3) in-hospital mortality
The GRS was significantly associated with LDL-C levels (p<2.2 × 10−16), accounted for 5.8 % of the variability in LDL-C, and was associated with hyperlipidemia (OR 1.18, 95% CI 1.15–1.22, p=1.4 × 10−34) and coronary atherosclerosis (OR 1.07, 95% CI 1.04–1.10, p=4.3 × 10−7). However, the GRS was not significantly associated with sepsis, ICU admission, or in-hospital mortality (Table 2), (1) without adjustment (sepsis: OR 1.02, 95% CI 0.97–1.07, p=0.41; ICU: OR 1.01, 95% CI 0.95–1.07, p=0.84; in-hospital death: OR 0.92, 95% CI 0.81–1.04, p=0.17, respectively); (2) adjusted for age and sex (sepsis: OR 1.02, 95% CI 0.97–1.07, p=0.37; ICU: OR 1.01, 95% CI 0.95–1.07, p=0.78; in-hospital death: OR 0.92, 95% CI 0.81–1.05, p=0.22, respectively).
We also compared the risk of sepsis, ICU admission, and in-hospital mortality within different LDL-C quartiles for measured LDL-C levels and the GRS. Patients in the lowest quartile of measured LDL-C concentrations had a higher risk of sepsis and ICU admission compared to those in the highest quartile of measured LDL-C; however, after adjustment for age, sex, and comorbidity variables, none of measured LDL-C quartiles differed significantly (Figure 3 A and B). The lowest LDL-C GRS quartile did not have different risks of sepsis, ICU admission or in-hospital death compared to the highest quartile (Figure 3 C and D).
Figure 3. Association between LDL-C quartiles and sepsis, ICU admission and death.
The associations between sepsis and its adverse outcomes and measured LDL-C quartiles (A,B) and genetic risk score quartiles (C,D). Analyses A and C are unadjusted, B adjusted for age, sex, and comorbidity variables, and D for age and sex.
Discussion
We used epidemiological and genetic approaches to define the association between LDL-C levels and sepsis and found that lower LDL-C levels measured in patients at least a year before they were admitted to the hospital with infection were associated with increased risk of sepsis, ICU admission, and in-hospital mortality. However, when co-morbidities were considered, there was no association between LDL-C levels and sepsis or its outcomes. Concordant with this finding, genetic predictors of LDL-C were not associated with sepsis or its outcomes.
Sepsis is a life-threatening and costly condition, which is difficult to predict and treat. There is no specific treatment for sepsis other than early antibiotic therapy and supportive care. Strategies to predict which patients with infection are likely to develop sepsis are important because these patients can be targeted for early intensive monitoring. Also, new therapies to prevent or treat sepsis are an important but elusive goal. Thus, LDL-C represents an interesting candidate, both for prevention and treatment of sepsis.
In vitro studies and studies in animals showed that LDL binds toxic bacterial products such as LPS, and has other immunologic effects that prevent or ameliorate many of the manifestations of sepsis.5,6 In murine models, high endogenous LDL levels (induced by deleting the LDL receptor) protected against the lethal effects of LPS and Gram-negative infection.5 In contrast, mice that were rendered hypolipidemic had increased LPS-induced mortality and this could be reversed by administering exogenous lipoproteins that increased serum lipids back to within the physiologic range.6 Despite the animal studies suggesting that LDL could affect the outcomes of sepsis directly, few studies have defined this association in patients.
Low LDL-C levels were associated with increased risk of death in patients with community acquired pneumonia 9 and with increased risk of sepsis in hospitalized patients;7 also, outcomes from sepsis were worse than in those with high LDL-C levels.8,10 However, these studies do not necessarily implicate LDL-C directly as a modifier of sepsis because the confounding effects of co-morbid illness could account for the findings (i.e., sicker patients could have both lower LDL-C levels and worse outcomes). An epidemiologic study of approximately 30,000 community-dwelling adults was less prone to confounding and showed that lower LDL-C levels at entry to the cohort were associated with increased risk of future sepsis.10 This finding suggested that LDL levels could affect the risk of sepsis directly; however, individuals in the lowest LDL-C quartile were more likely to have pre-existing comorbidities and the possibility of residual confounding remained a concern.10
To define the association between LDL-C and sepsis more clearly, we took two approaches to minimize the confounding effects of comorbid illness. The first used an epidemiological approach that accounted for comorbidities, and the second used a genetic risk score that should be independent of the confounding effects of illness. Lower measured LDL-C levels were indeed associated with higher risk of sepsis and its complications, even though we took steps to exclude LDL-C levels obtained when patients were acutely ill and excluded patients with several comorbid illnesses associated with low LDL-C levels and increased risk of infection. This unadjusted analysis was concordant with previous reports.8,29 However, there was no association between LDL-C and risk of sepsis after statistical adjustment that included the comorbidity variables.
The study design sought to reduce the confounding effects of preceding and concomitant illness and thus only included LDL-C measurements obtained more than a year before the index admission and excluded individuals with many co-morbidities that could lower LDL and increase the risk of infection. However, this design could have introduced bias and also reduced the generalizability of the findings. Two findings suggest that the findings were robust. First, the GRS, which does not anchor on LDL-C measurement and thus provides a way of assessing the validity of the measured LDL-C approach, yielded concordant findings. Second, the sensitivity analysis that imposed no time exclusion on the LDL-C and used the value closest to admission and imposed no exclusions for co-morbidities yielded results (Supplementary Table 9) very similar to those of the primary analysis, showing a strong association between low LDL-C and risk of sepsis in the unadjusted analysis and a marked attenuation after adjustment for baseline co-morbidities. Our finding suggests that although LDL-C is indeed associated with increased risk of sepsis, this is not a direct association but is due to comorbid illness.
Newer LDL lowering agents, such as PCSK9 inhibitors, can reduce LDL-C levels to extremely low levels. Although LDL-C lowering with PCSK9 inhibitors has been shown to reduce CVD events in statin-treated patients with CAD and acute coronary syndrome,23 there is little information about other effects of long-term low LDL-C levels. Traditional post-marketing drug safety approaches to determine these long-term effect will require years of study. By applying a clinical and genetic approach, we observed no association between low LDL-C and sepsis. Therefore, LDL-C lowering therapy is unlikely to increase sepsis risk for patients with serious infection.
The genetic approaches further supported the interpretation that LDL-C levels are not directly associated with sepsis. The genetic architecture of circulating lipid levels is well understood; the GLGC has published a series of large-scale genome-wide association studies,24,25 and these observations have been widely used to construct GRSs for lipids levels. A GRS combined with Mendelian randomization has been a powerful tool for testing the potential causal relationships between lipid levels and specific diseases. Studies using this genetic approach suggested a direct relationship between high LDL-C and risk of cardiovascular disease22 and the presence of aortic valve calcium and incidence of aortic stenosis;26 between high triglyceride levels (but not HDL-C levels) and risk of cardiovascular disease;22 and between low LDL-C levels and risk of diabetes.25,27 Similarly, an LDL-C GRS approach suggested that low LDL-C was not associated with Alzheimer’s disease, although this had been a clinical association of concern.28 In the current study, the GLGC-based LDL-C GRS was significantly associated with LDL-C levels extracted from the EHR in patients who had had them measured, and with hyperlipidemia and coronary atherosclerosis, but was not significantly associated with risk of sepsis or its adverse outcomes.
The current study has several strengths. It is difficult to perform large clinical trials in patients with sepsis; we were able to use existing clinical and genetic information from a large EHR system linked to a DNA bank that provided longitudinal information, including information about drugs, diagnoses, and laboratory findings in a large number of patients to address the questions of interest. The extensive clinical data available allowed us to restrict the measured LDL-C cohort to individuals who had their LDL-C measured at least a year before the index admission to hospital. This not only minimized the effect of the current illness on LDL-C levels but also provided clinical diagnoses in the preceding year to adjust for comorbidities. Additionally, using a clinical and a genetic approach maximized our ability to distinguish whether any association between LDL-C levels and sepsis was likely to be direct or indirect.
However, there were also several limitations. First, the EHR provides information about a range of variables that facilitates statistical adjustment; however, we did not perform a randomized controlled study. Thus, results could be influenced by unmeasured confounders. This seems unlikely to be an important consideration since the confounders included accounted for much of the association between LDL-C levels and sepsis. Second, patients could have received care outside of the VUMC system. Thus, covariates could be incomplete, and although we set out to capture the first hospitalization due to infection, a patient could have been treated elsewhere previously for a serious infection that was not captured. Third, although the GRS was significantly associated with measured LDL-C levels, it explained only a small percent of overall LDL-C variation in the population. This is the case even for large scale genetic analyses of circulating LDL-C levels;24 however, the genetic component has been strong enough for the reliable detection of several disease associations.22,25–27 Nevertheless, those associations were between LDL GRSs and chronic diseases; the association with acute illness may be more difficult to detect. Fourth, the LDL GRS did not capture LDL changes due to environmental factors, such as diet and life-style. Fifth, we cannot exclude the possibility of unrelated competing effects of LDL-C and comorbidities on sepsis risk, although there is there is little evidence to support it. Sixth, we only studied LDL-C and the relatively common variants reported to be associated with it in GWAS studies. Concentrations of other lipids such as HDL-C 9 and the presence of other variants in genes such as PCSK9 that affect LDL-C but were not identified in GWAS studies could be important since they may affect the outcomes of sepsis.29 Last, we studied white patients admitted to a tertiary-level US hospital with infection who had LDL-C measured at least a year before admission or had been genotyped, which limits the generalizability of the findings.
In conclusion, lower measured LDL-C levels were significantly associated with increased risk of sepsis and admission to ICU in patients admitted to hospital with infection, but this was due to comorbidities since adjusted clinical models and genetic models showed no increased risk. LDL-C levels do not directly alter the risk of sepsis or poor outcomes in patients hospitalized with infection.
Supplementary Material
Key Points.
Question: To understand the association between low LDL-C levels and risk of sepsis in patients admitted to the hospital due to serious infection.
Findings: Neither measured LDL-C levels nor an LDL-C genetic risk score was associated with increased risk of sepsis after adjusting for comorbidities.
Meaning: LDL-C levels are not directly associated with the risk of sepsis or poor outcomes in patients hospitalized with infection.
Acknowledgments
Funding/Support: This study was supported by GM120523, GM109145, HL133786, K23AR064768, Rheumatology Research Foundation (K-supplement), American Heart Association (16SDG27490014 and 15MCPRP25620006), and Vanderbilt Faculty Research Scholar Fund. The dataset used for the analyses described were obtained from Vanderbilt University Medical Center’s resources, BioVU and the Synthetic Derivative, which are supported by institutional funding, the 1S10RR025141 instrumentation award, and by the Vanderbilt National Center for Advancing Translational Science grant UL1TR000445 from NCATS/NIH. Existing genotypes in BioVU were funded by NIH grants RC2GM092618 from NIGMS/OD and U01HG004603 from NHGRI/NIGMS. The authors wish to acknowledge the expert technical support of the VANTAGE and VANGARD core facilities, supported in part by the Vanderbilt-Ingram Cancer Center (P30 CA068485) and Vanderbilt Vision Center (P30EY08126).
Footnotes
Conflict of Interest Disclosures:
Linton, MF: Research support Amgen, Sanofi Aventis, Regeneron, Regenxbio
Role of the Funder/Sponsor: The funders had no role in design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
References
- 1.Rhee C, Dantes R, Epstein L, et al. Incidence and Trends of Sepsis in US Hospitals Using Clinical vs Claims Data, 2009-2014. JAMA. 2017;318(13):1241–1249. doi: 10.1001/jama.2017.13836 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Liu V, Escobar GJ, Greene JD, et al. Hospital Deaths in Patients With Sepsis From 2 Independent Cohorts. JAMA. 2014;312(1):90–92. doi: 10.1001/jama.2014.5804 [DOI] [PubMed] [Google Scholar]
- 3.Perner A, Gordon AC, De Backer D, et al. Sepsis: frontiers in diagnosis, resuscitation and antibiotic therapy. Intensive Care Med. 2016;42(12):1958–1969. doi: 10.1007/s00134-016-4577-z [DOI] [PubMed] [Google Scholar]
- 4.Genga KR, Russell JA. Update of Sepsis in the Intensive Care Unit. J Innate Immun. 2017;9(5):441–455. doi: 10.1159/000477419 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Netea MG, Demacker PN, Kullberg BJ, et al. Low-density lipoprotein receptor-deficient mice are protected against lethal endotoxemia and severe gram-negative infections. J Clin Invest. 1996;97(6):1366–1372. doi: 10.1172/JCI118556 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Feingold KR, Funk JL, Moser AH, Shigenaga JK, Rapp JH, Grunfeld C. Role for circulating lipoproteins in protection from endotoxin toxicity. Infect Immun. 1995;63(5):2041–2046. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Shor R, Wainstein J, Oz D, et al. Low serum LDL cholesterol levels and the risk of fever, sepsis, and malignancy. Ann Clin Lab Sci. 2007;37(4):343–348. [PubMed] [Google Scholar]
- 8.Lagrost L, Girard C, Grosjean S, et al. Low Preoperative Cholesterol Level Is a Risk Factor of Sepsis and Poor Clinical Outcome in Patients Undergoing Cardiac Surgery With Cardiopulmonary Bypass*. Crit Care Med. 2014;42(5):1065–1073. doi: 10.1097/CCM.0000000000000165 [DOI] [PubMed] [Google Scholar]
- 9.Chien Y-F, Chen C-Y, Hsu C-L, Chen K-Y, Yu C-J. Decreased serum level of lipoprotein cholesterol is a poor prognostic factor for patients with severe community-acquired pneumonia that required intensive care unit admission. J Crit Care. 2015;30(3):506–510. doi: 10.1016/j.jcrc.2015.01.001 [DOI] [PubMed] [Google Scholar]
- 10.Guirgis FW, Donnelly JP, Dodani S, et al. Cholesterol levels and long-term rates of community-acquired sepsis. Crit Care. 2016;20:408. doi: 10.1186/s13054-016-1579-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Sabatine MS, Giugliano RP, Wiviott SD, et al. Efficacy and Safety of Evolocumab in Reducing Lipids and Cardiovascular Events. N Engl J Med. 2015;0(0):null. doi: 10.1056/NEJMoa1500858 [DOI] [PubMed] [Google Scholar]
- 12.Sabatine MS, Giugliano RP, Keech AC, et al. Evolocumab and Clinical Outcomes in Patients with Cardiovascular Disease. N Engl J Med. 2017;376(18):1713–1722. doi: 10.1056/NEJMoa1615664 [DOI] [PubMed] [Google Scholar]
- 13.Wei W-Q, Denny JC. Extracting research-quality phenotypes from electronic health records to support precision medicine. Genome Med. 2015;7(1):41. doi: 10.1186/s13073-015-0166-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Xu H, Stenner SP, Doan S, Johnson KB, Waitman LR, Denny JC. MedEx: a medication information extraction system for clinical narratives. J Am Med Inf Assoc. 2010;17:19–24. doi:17/1/19 [pii] 10.1197/jamia.M3378 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Denny JC, Bastarache L, Ritchie MD, et al. Systematic comparison of phenome-wide association study of electronic medical record data and genome-wide association study data. Nat Biotechnol. 2013;31(12):1102–1110. doi: 10.1038/nbt.2749 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Angus DC, Linde-Zwirble WT, Lidicker J, Clermont G, Carcillo J, Pinsky MR. Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care. Crit Care Med. 2001;29(7):1303–1310. [DOI] [PubMed] [Google Scholar]
- 17.Donnelly JP, Safford MM, Shapiro NI, Baddley JW, Wang HE. Application of the Third International Consensus Definitions for Sepsis (Sepsis-3) Classification: a retrospective population-based cohort study. Lancet Infect Dis. 2017;17(6):661–670. doi: 10.1016/S1473-3099(17)30117-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992;45(6):613–619. doi: 10.1016/0895-4356(92)90133-8 [DOI] [PubMed] [Google Scholar]
- 19.Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–383. [DOI] [PubMed] [Google Scholar]
- 20.Quan H, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43(11):1130–1139. [DOI] [PubMed] [Google Scholar]
- 21.Carroll RJ, Bastarache L, Denny JC. R PheWAS: data analysis and plotting tools for phenome-wide association studies in the R environment. Bioinforma Oxf Engl. 2014;30(16):2375–2376. doi: 10.1093/bioinformatics/btu197 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Do R, Willer CJ, Schmidt EM, et al. Common variants associated with plasma triglycerides and risk for coronary artery disease. Nat Genet. 2013;45(11):1345–1352. doi: 10.1038/ng.2795 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Schwartz GG, Steg PG, Szarek M, et al. Alirocumab and Cardiovascular Outcomes after Acute Coronary Syndrome. N Engl J Med. 2018;0(0):null. doi: 10.1056/NEJMoa1801174 [DOI] [PubMed] [Google Scholar]
- 24.Global Lipids Genetics Consortium. Discovery and refinement of loci associated with lipid levels. Nat Genet. 2013;45(11):1274–1283. doi: 10.1038/ng.2797 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Teslovich TM, Musunuru K, Smith AV, et al. Biological, clinical and population relevance of 95 loci for blood lipids. Nature. 2010;466(7307):707–713. doi:nature09270 [pii] 10.1038/nature09270 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Smith JG, Luk K, Schulz C-A, et al. Association of Low-Density Lipoprotein Cholesterol–Related Genetic Variants With Aortic Valve Calcium and Incident Aortic Stenosis. JAMA. 2014;312(17):1764–1771. doi: 10.1001/jama.2014.13959 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Andersson C, Lyass A, Larson MG, Robins SJ, Vasan RS. Low-density-lipoprotein cholesterol concentrations and risk of incident diabetes: epidemiological and genetic insights from the Framingham Heart Study. Diabetologia. 2015;58(12):2774–2780. doi: 10.1007/s00125-015-3762-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Proitsi P, Lupton MK, Velayudhan L, et al. Genetic Predisposition to Increased Blood Cholesterol and Triglyceride Lipid Levels and Risk of Alzheimer Disease: A Mendelian Randomization Analysis. PLOS Med. 2014;11(9):e1001713. doi: 10.1371/journal.pmed.1001713 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Walley KR, Thain KR, Russell JA, et al. PCSK9 is a critical regulator of the innate immune response and septic shock outcome. Sci Transl Med. 2014;6(258):258ra143–258ra143. doi: 10.1126/scitranslmed.3008782 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.