So far, the hospital mortality rate of sepsis remains high. Elucidating the risk factors or protective factors related to its mortality can provide key clues for improving its prognosis[21]. There is growing evidence that some laboratory indicators of blood lipids play a key role in the prognosis of sepsis. However, the relationship between TG, HDL and TG/HDL in the prognosis of sepsis is still not fully understood and further study is needed.
In this study, we first used univariate COX regression to analyze which factors might be associated with poor prognosis of sepsis, and then used multivariate COX regression to analyze the correlation between TG, HDL, TG/HDL and hospitalized mortality of sepsis. Univariate COX regression analysis showed that HDL was a protective factor for poor prognosis of sepsis, with an effect value and 95CI% confidence interval of 0.86 (0.64,1.15), P = 0.309. Although P value > 0.05 was not considered as statistically significant, it was still clinically significant due to its significant effect value. This result is consistent with some previous studies in which patients with higher HDL levels had better sepsis outcomes than patients with lower HDL levels. An animal experiment conducted by Morin et al. showed that the mice with HDL deficiency were prone to infection and death, while the mice with increased HDL level due to increased ApolA1 expression had a lower mortality rate[18]. This was also well demonstrated in the study conducted by Dai et al.[22], in which recombinant HDL infusion reduced endotoxins, reduced inflammation, battled hypotensive responses, and improved survival in sepsis mice. This result theoretically supports the results of our univariate analysis. At the same time, our univariate analysis showed that TG and TG/HDL were not correlated with sepsis mortality, and their effect values (HR) and 95% confidence intervals (CI) were 0.95(0.87,1.03) and 0.98(0.95,1.02), respectively, with P > 0.05 for both.
The results of our multivariate COX regression are not consistent with those of univariate COX regression. Under the premise of controlling the potential confounding factors, multivariate COX regression analysis of the relationship between TG, HDL, TG/HDL and poor prognosis of sepsis showed that TG and TG/HDL were negatively correlated with poor prognosis of sepsis, and their effect values HR and 95% confidence interval were 0.90 (0.80-0.998), P = 0.036 and 0.95 (0.91–0.99), P = 0.049, respectively. There was a positive correlation between HDL and poor prognosis of sepsis, and its effect value and 95% confidence interval were 1.14 (0.8–1.63) and P = 0.469, respectively. Although the P > 0.05 showed no statistical significance, we still believed that HDL was positively correlated with poor prognosis of sepsis due to its significant clinical effect value. This is in stark contrast to previous findings that higher HDL levels are more likely to lead to an increased in-hospital sepsis mortality rate. In view of this, we conducted a survivability analysis of the results. We also found this in the survival analysis (Fig. 5–7). We divided TG, HDL, and TG/HDL into two groups according to the number of people. The results showed that the mortality of patients in the group with higher HDL was higher, while that in the group with lower TG and TG/HDL was higher. A prospective, multicentre, randomized controlled phase II study by Dellinger RP et al was also disappointing: HDL infusion in patients with gram-negative bacteria-induced sepsis did not reduce the incidence of 28-day all-cause mortality or new organ failure, and the phase III trial was discontinued because of adverse reactions[23]. In 2020, Liu et al.[24] conducted a meta-analysis of the correlation between HDL and sepsis prognosis, which found that low HDL level was associated with higher mortality in adult patients with sepsis (OR and 95%CI were 2.00, 1.23–3.24, P = 0.005.). This conclusion supports the negative correlation between HDL levels and sepsis mixtures, but it is worth noting that the sample sizes of sepsis patients in this study were 11,28,30,51,65,68,108, respectively[25–31]. Compared with our samples, the average sample size is too small. Our sample size is relatively more representative.
In this study, we found that high levels of TG may be a protective factor in patients with sepsis. Survival analysis showed that patients with higher TG levels had a better prognosis. Luan et al. found that growth and differentiation factor 15(GDF15) is a factor essential for the survival of bacterial and viral infections and sepsis[32]. The protective effect of this factor is largely independent of the degree of pathogen control or inflammatory response, and the regulation of this factor requires the participation of TG. This explains the relationship between TG and sepsis prognosis to some extent. In this study, we found that the association between TG/HDL and poor prognosis of sepsis was not as significant as the association between single TG or HDL and poor prognosis of sepsis. This suggests that TG or HDL alone is better than TG/HDL ratios in predicting poor prognosis in patients with sepsis.
In subgroup analysis, we found that TG and TG/HDL were stable in all subgroups. It is worth mentioning that in sepsis patients with liver disease, the association between HDL level and sepsis was strengthened, with an effect value and 95%CI of 14.08(3.54–55.97). HDL is known to be synthesized mainly in the liver and small intestine[33]. Maintaining high levels of HDL increases the burden on the liver. We hypothesize that in sepsis patients with liver disease, a high level of liver burden reduces immunity and ultimately leads to an increase in in-hospital mortality.
In this study, there are still some deficiencies: First, this study is a retrospective study, which inevitably has some selective bias. Secondly, the time span of this study is large, from 2008 to 2019. The detection method of TG or HDL may change during this period, but the detection method cannot be identified by the MIMIC-IV database at present. But given that the samples in the database are all from the same medical center, we think the effect is likely to be small. In this study, some of our work is also worthy of affirmation: compared with previous studies, our sample size was larger and more confounders were excluded. Compared with previous studies, our results better reflected the risk of TG, HDL and TG/HDL and sepsis hospitalization mortality. At the same time, we carried out multiple interpolation for the missing data to maximize the statistical capability and minimize the error. We also performed more subgroup analyses to make our results more robust.