Relationship Between Serial Lactate Testing and Clinical Outcome of Critically Ill Patients With Sepsis

Objective Our study was to investigate the relationship between serial lactate testing (SLT) and the clinical outcome of critically ill patients with sepsis. Materials and Methods This was a retrospective cohort study. We extracted the clinical data of patients with sepsis from the Medical Information Mart for Intensive Care version III, 1.4 database. Kaplan-Meier survival curves and multivariate logistic regression models were used to predict the relationship between clinical indicators and 28-day mortality. Receiver operating characteristic curves (ROCs) were used to analyze the impact of the maximum, minimum, and initial values of lactate on 28-day mortality; calculate the Youden index to take the best cutoff value; and then perform a subgroup analysis. Results In total, 2367 patients with sepsis were enrolled, including 1961 in the SLT group and 406 in the non-SLT group. Clinical indicators seemed to be more serious in the SLT group than in the non-SLT group. ROC analysis results showed that the initial value, maximum value and minimum value of blood lactate had a signicant impact on the 28-day mortality of patients, P<0.001; multivariate logistic regression analysis indicated that SLT had a negative impact on 28-day mortality [OR=2.03, 95% CI (1.67, 3.60)], p<0.05. Conclusion We found that SLT was correlated with sepsis severity but might not improve the clinical outcome of critically ill patients with sepsis. Clinicians should pay attention to differential diagnosis of the source of lactate elevation; our study results need randomized controlled trials for further verication.


Introduction
Sepsis is a life-threatening organ dysfunction disease caused by the host's unregulated response to infection 1 and has the characteristics of high morbidity and mortality. [2][3][4] Serial lactate testing (SLT) is essential for the risk strati cation and management of sepsis. 5 In addition, SLT is also an important indicator for evaluating the prognosis of patients with sepsis. 6 Although recent guidelines recommend that patients with sepsis should have their blood lactate levels assessed every 2-4 hours until the lactate value is normal, 5 whether SLT can predict the prognosis of patients with sepsis is still controversial. [7][8] Some researchers have found that SLT can be used as a target for uid resuscitation in patients with emergency sepsis, reducing pulmonary complications 9 and 28-day mortality, 10 shorting the ICU length of stay (ICULOS), and reducing vasoactive drugs and ventilator use. 11 In addition, veterinary studies have shown that SLT is a more useful prognostic indicator than single lactate monitoring for adult horses diagnosed with septic shock. 12 However, some studies have found that SLT does not improve the clinical outcomes of patients, including hospital mortality, because SLT is not the only factor in the sepsis bundle and may be affected by other measures. 13 Some scholars believe that SLT performed within 6 hours in patients with sepsis is highly correlated with hospital mortality, but the patient's initial lactate value should be greater than 4 mmol/L. 14 Therefore, the purpose of our study was to investigate the relationship between SLT and the clinical outcomes of patients with sepsis based on the MIMIC-III database. The primary observation endpoint was 28-day mortality, and the secondary observation endpoints were hospital mortality, ICU mortality, ICU length of stay (ICULOS), hospital length of stay (HLOS) and other outcome indicators, which provided clinicians with reliable medical decision-making.
Materials And Methods

Data Sources
This was a retrospective cohort study. All sepsis patient data were obtained from the MIMIC-III 1.4 database.MIMIC is a large, single-center, public critical care database of more than 40,000 ICU inpatients who were admitted to the intensive care unit of the Massachusetts Institute of Technology A liated Hospital from 2001 to 2012. 15 We had applied for access to the database, and the quali cation certi cate number was 9730539. The sepsis diagnostic codes are 99591 and 99592.The diagnostic criteria for sepsis were based on meeting the third international consensus de nition of sepsis and septic shock. 1 Data on patients with recurrent sepsis are only recorded on their rst admission. We extracted clinical data through structured query language (SQL). Our research was approved by the Institutional Review Board of the Beth Israel Deaconess Medical Center (Boston, MA) and the Massachusetts Institute of Technology.
We collected baseline data of patients within 24 hours of admission to the ICU, including age, sex, sequential organ failure assessment (SOFA) score, systemic in ammatory response syndrome (SIRS) score, comorbidities (congestive heart failure, paralysis, chronic pulmonary disease, diabetes, hypothyroidism, renal failure, liver disease, rheumatoid arthritis, coagulopathy, obesity, de ciency anemias, alcohol abuse, drug abuse and depression), laboratory tests (the maximum/minimum value of serum creatinine, red blood cell distribution width [RDW], serum lactate [including the initial value], white blood cells [WBCs], platelets, hemoglobin, blood glucose, anion gap), and hemodynamic parameters (the maximum/minimum of the mean arterial pressure [MAP], dobutamine duration, norepinephrine duration and vasopressin duration). SLT was de ned as more than 2 measurements within 24 hours of the patient's admission to the ICU. The primary endpoint of the study was the 28-day mortality of patients, and the secondary endpoints were hospital mortality, ICU mortality, HLOS, ICULOS, ventilator and continuous renal replacement therapy (CRRT) use within 24 hours of admission to the ICU.

Inclusion and exclusion criteria
According to ICD_9 codes (99591 and 99592), patients with sepsis were screened out. Exclusion criteria: only the rst hospitalization data for patients who have been admitted to the ICU repeatedly; age <18 years or >75 years; ICULOS <24 hours; acquired immunode ciency syndrome, lymphoma, metastatic cancer and solid tumors.

Statistical methods
We used Stata 16.0 statistical software. Continuous variables and normally distributed data are represented by the mean ± standard deviation, nonnormally distributed data are represented by the median (interquartile range), and categorical variable data are represented by the percentage (rate). The Kaplan-Meier survival analysis method was used to predict the relationship between SLT and 28-day mortality. The receiver operating characteristic curve (ROC) was used to analyze the impact of the maximum, minimum and initial value of lactate on 28-day mortality, and the Yoden index was calculated to take the best cutoff value, and then a subgroup analysis was performed. The multivariate logistics regression model was used to predict the relationship between the dependent variables and 28-day mortality. The processing methods used for outliers included the median method, average method, missing value conversion method, and winsorization. Variables with missing values greater than 20% were directly deleted. Variables with missing values less than 20% were processed by the regression lling method and multiple interpolation method.

General characteristics
A total of 5185 sepsis patients were enrolled, and only the rst hospitalization data were retained for each patient (720 cases were excluded); other excluded cases included patient age <18 years or >75 years (1373 cases), ICULOS< 24 hours (64 cases), acquired immunode ciency syndrome (39 cases), lymphoma (109 cases), metastatic cancer (327 cases) and solid tumors (186 cases). Therefore, 2367 patients were enrolled in the group, 1961 in the SLT group, and 406 in the non-SLT group ( Figure 1). The SOFA score, SIRS score, serum creatinine, RDW, serum lactate, WBC, hemoglobin, blood glucose and anion gap in the SLT group were signi cantly higher than those in the non-SLT group (P <0.05). Compared with the non-SLT group, the MAP of the SLT group was signi cantly lower, and the durations of dopamine, norepinephrine and vasopressin were signi cantly prolonged (P<0.05). However, there were no signi cant differences in age, sex, comorbidities or platelets between the two groups (p>0.05) ( Table 1).   Figure 2). Survival analysis of subgroup variables showed that the initial value of serum lactate < 1.9 mmol/L was signi cantly better than that of serum lactate >= 1.9 mmol/L, the initial value of serum lactate < 2.7 mmol/L was signi cantly better than that of serum lactate >= 2.7 mmol/L, the initial value of serum lactate <1.3 mmol/L was signi cantly better than that of serum lactate >=1.3 mmol/L, and non-SLT was signi cantly better than SLT, P<0.05 (Figure 3). In addition, further survival analysis based on hospital and ICU mortality revealed that there was no signi cant bene t for SLT patient survival (Figure 4).  OR, odds ratio; CI,con dence interval; SOFA, sequential organ failure assessment; RDW, red blood cell distribution width;Wbc, white blood cell ; SLT, serial lactate testing;CRRT,continuous renal replacement therapy.

Discussion
Few researchers have studied the effect of SLT on the clinical outcomes of critically ill patients with sepsis. The results of several randomized controlled trials are also contradictory. Some researchers believe that hyperlactic acidemia is related to the severity of sepsis in patients, mainly due to impaired tissue oxygen utilization. 16 Therefore, other factors that lead to increased serum lactate need to be considered when formulating a serum lactate-oriented uid resuscitation strategy. In addition, when comparing the effects of serum lactate and peripheral capillary perfusion-oriented uid resuscitation strategies on the 28-day mortality of patients with sepsis, it was found that there was no signi cant difference between the two approaches. 17 Other researchers have also found that serum lactate monitoring within 1 hour of admission can reduce the length of hospitalization and mortality of patients with sepsis, but this requires complete life monitoring and timely antibiotic infusion within 1 hour. 18 A serum lactate kinetics-oriented uid resuscitation strategy might improve the 90-day mortality of patients with high lactate-associated sepsis. 19 Our study found that SLT was related to sepsis severity and did not signi cantly improve the clinical outcomes of patients with sepsis.
Previous studies found that the rate of SLT within 6 hours in patients with sepsis increased from 23-69%, and the rate of SLT within 24 hours increased from 59-94%, 7 which was generally consistent with our research results (1961/2367, 82.8%) from 2003 to 2013. We also found that the disease severity score, laboratory indicators and hemodynamic parameters of the SLT group seemed to be worse than those of the non-SLT group. In fact, the initial, maximum, and minimum values of serum lactate in the SLT group were signi cantly higher than those in the non-SLT group. This seemed to indicate that SLT, similar to hyperlactemia, was related to the severity of sepsis. We believed that SLT would not improve the clinical outcomes of patients regardless of whether the lactate value was reduced to a normal level. This result might be inconsistent with the conclusions of some researchers. [20][21][22][23] However, we maintain that SLT of critically ill patients must be combined with effective treatment so that critically ill patients might have a bene cial prognosis.
We believe that hyperlactemia can occur in anaerobic glycolysis or in the absence of hypoxia. The latter might be affected by the adrenergic state and the use of beta agonists. [24][25] Hyperlactemia can also be seen in liver cirrhosis or end-stage liver disease, 26 with metformin use 27 or propofol use, 28 and in toxic diseases. [29][30] For critically ill patients with sepsis, the serum lactate-oriented uid resuscitation program should clarify the reason for the initial value increase in serum lactate and facilitate the adoption of a targeted treatment strategy. The treatment of lactic acidosis caused by insu cient tissue perfusion should focus on restoring local or overall tissue perfusion. Therefore, uid resuscitation is the core principle, and hemodynamics are supported by vasoactive drugs and source control. 31 The increasing serum lactate caused by other factors usually needs to be treated by removing harmful substances and correcting the primary metabolic defect as much as possible. Usually, the source of the elevated serum lactate is di cult to determine, but we believe that it is necessary to continually reassess the patient's volume status after the initial guided volume resuscitation to prevent uid overload. A study con rmed that for every liter of uid added by uid resuscitation, the patient's mortality will increase by 2.3%. 32 Therefore, if serum lactate cannot be eliminated, clinicians should pay attention to the differential diagnosis of the source of insidious serum lactate elevation.
Our study has several limitations. First, it is a retrospective observational study and thus is subject to the limitations of this type of study. However, we believe that the MIMIC-III database contains a large amount of data for this entire group, so it is a good representation of the characteristics of patients with sepsis. In addition, our research is not perfect because we only studied the impact of categorical variables on patient outcomes. Although the results are representative, the investigation does not include parameters such as the number of lactate monitoring events within 24 hours or within 6 hours; ultimately, we believe that prospective studies are still needed to further verify our results.
In summary, we found that SLT was correlated with the severity of sepsis. SLT might not improve the clinical outcomes of critically ill patients with sepsis. Clinicians should pay attention to the differential diagnosis of the source of lactate elevation. Our study results need to be veri ed further in randomized controlled trials.

Competing interests
GL, and YW contributed to determining variables, extracting variables from the trauma registry, providing input, and nally approving the manuscript. Besides ,YL was a contributor to the research design, responsible for providing input data and nalizing the draft.

Acknowledgements
Authors would like to show our deepest gratitude to our supervisor, Dr. Liu, a respectable, responsible and resourceful scholar, who has provided us with valuable guidance in every stage of the writing of this paper. Without his enlightening instruction, impressive kindness and patience, We could not have completed our paper. His keen and vigorous academic observation enlightens us not only in this study but also in our future study and work.    See image above for gure legend.