Risk Stratification of Patients with Acute Respiratory Distress Syndrome Complicated with Sepsis Using Lactate Trajectories

DOI: https://doi.org/10.21203/rs.3.rs-1586299/v2

Abstract

Background: No consensus has been reached on an optimal blood lactate evaluation system although several approaches have been reported in the literature in recent years. A group-based trajectory modeling (GBTM) method could better stratify patients with acute respiratory distress syndrome (ARDS) complicated with sepsis in the intensive care unit (ICU).

Methods: 762 patients from the comprehensive ICU of Tianjin Medical University General Hospital with ARDS complicated with sepsis were eligible for analysis. Serial serum lactate levels were measured within 48 h of admission. In addition to the GBTM lactate groups, the initial lactate, peak lactate level, the area under the curve of serial lactate (lactate AUC), and lactate clearance were also considered for comparison. The short- and long-term outcomes were the 30- and 90-day mortality rates, respectively.

Results: Three lactate groups were identified based on GBTM, with group 3 exhibiting the worse short- (hazard ratio (HR): for 30-days mortality: 2.26, 95% CI: 1.75 to 2.92, P <0.001) and long term (HR for 90-day mortality: 3.06, 95% CI: 1.78 to 5.27, P <0.001) outcomes followed by group 2 [HR for 30-day mortality: 1.79, 95% confidence interval (CI): 1.30 to 2.47, P <0.001] and HR for 90-days mortality: 1.73, 95% CI: 1.30 to 2.31, P <0.001). GBTM lactate groups exhibited significantly improved diagnostic performance of initial lactate + SOFA scores/APACHE II scores models. Based on the multivariable fractional polynomial interaction (MFPI) approach, GBTM lactate groups could better differentiate high-risk patients than the initial lactate groups in short- and long-term outcomes.

Conclusions: To the best of our knowledge, this is the first report that GBTM-based serial blood lactate evaluations significantly improve the diagnostic capacity of traditional critical care evaluation systems and bring many advantages over previously documented lactate evaluation systems.

1 Background

Elevated lactate levels are common in the setting of sepsis and usually represent an imbalance between increased production in the presence or absence of tissue hypoxia and changes in clearance [1]. Given their well-established association with multiple organ dysfunction [2], blood lactate levels and their trend over time are increasingly recognized as reliable markers of illness severity and mortality [37]. Border and Weil et al. observed that lactate levels ≥ 4 mmol/L were associated with a 50% mortality in patients with shock during a study period of 36 hours [8]. An increasing body of evidence suggests that lactate measurements represent a valuable and straightforward indicator for the early identification of severe sepsis and are associated with poor outcomes [913]. Interestingly, current evidence suggests that compared with initial lactate levels at presentation or other single measured lactate values, serial blood lactate level measurements can better characterize the dynamic process of lactate production and elimination in vivo [2]. Serial lactate levels can also help define the patient's trajectory and encourage a review of the therapeutic strategy [14]. Over the years, the prognostic value of serial measurements of multiple types of lactate levels or calculations from lactate levels has been investigated, for example, the duration of hyperlactatemia [15], the area under the curve of serial lactate measurement (lactate AUC) [16], and lactate clearance [17]. Importantly, decreased lactate levels were associated with improved outcomes in almost all critically ill patients [5]. Moreover, the later the lactate level is reduced, the worse the clinical prognosis is [18].

Despite the long-standing predictive power and the prominent place of lactate levels in early goal-directed therapy for critically ill patients, there is still a lack of a dynamic and flexible method to identify high-risk patients with hyperlactatemia. Given the lag in blood lactate concentration observed during testing, it remains unclear how lactate levels affect clinical management [19]. Notwithstanding that it is well-recognized that poor outcomes in ARDS patients are often attributed to sepsis/multiple organ failure rather than respiratory failure per se [20], few studies have hitherto focused on the predictive value of lactate for the prognosis of acute respiratory distress syndrome (ARDS) patients, along with the hypoxia process [1]. Indeed, a better understanding of the association between lactate and poor prognosis in ARDS patients complicated with sepsis upon admission is required.

We validated that lactate levels are elevated in septic patients with ARDS based on cohort data, including serial measurements of blood lactate over a period. A recently developed statistical method, also known as group-based trajectory modeling (GBTM), was harnessed to identify clusters of individuals exhibiting similar progression of a specific measurement over time and to identify the association between lactate trajectory within 48 hours and short- and long-term outcomes of ARDS patients complicated with sepsis. Meanwhile, although several lactate evaluation systems have been documented, including initial serum lactate levels, we sought to validate that novel subtypes of lactate based on GBTM algorithm yield better diagnostic value.

2 Methods

2.1 Study design, patients

This was a single-center, retrospective observational study where patients discharged from the comprehensive intensive care unit of Tianjin Medical University General Hospital from January 2019 to December 2021 were enrolled. Patients with the following conditions were excluded: (1) patients discharged within 48 hours of admission; (2) patients without admission arterial blood gas analysis measurements; (3) patients not diagnosed with ARDS and sepsis; and (4) patients with repeated admissions. The diagnosis of ARDS was made according to the criteria stated by the American-European Consensus Conference (AECC) on ARDS at admission [21]. The diagnostic criteria of sepsis were based on the consensus Sepsis-3 definitions at admission [9]. The plasma lactate concentration was measured by a COBAS B-123 POC system (Roche Diagnostics, Rotkreuz, Switzerland) every 6 hours from admission for 48 hours using arterial catheterization. Besides the initial lactate measurement, missing values were allowed for the other time points. The patients were followed up through outpatient visits, telephone calls or hospital readmission records. This study was approved by the Ethics Committee of Tianjin Medical University General Hospital (IRB2022-YX-041-01) and conducted in accordance with the ethical standards of the Declaration of Helsinki. The need for informed consent was waived due to the retrospective nature of our study. Reporting of the study followed the guidelines from the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [22].

2.2 Study Variables

The study variables used for analysis included patient demographics (age, sex), past medical history (hypertension, diabetes, coronary artery disease, stroke, renal failure, cancer), clinical characteristics upon admission [levels of systolic and diastolic blood pressure (SBP and DBP), heart rate, estimated glomerular filtration rate (eGFR), fluid intake per hour, fluid outtake per hour, red blood cell (RBC), white blood cell (WBC), platelet, hemoglobin, alanine aminotransferase (ALT), aspartate aminotransferase (AST), albumin, Glasgow Coma Scale (GCS) [23], Acute Physiology And Chronic Health Evaluation Ⅱ (APACHE Ⅱ) score [24] and Sequential Organ Failure Assessment (SOFA) score [25], ARDS severity (classification by oxygenation index), emergency surgical operation, invasive mechanical ventilation, and other inpatient treatment within 24 hours of admission [antibiotics, sedative/analgesic, glucocorticoid, continuous renal replacement therapy (CRRT), low molecular weight heparin (LMWH), sodium bicarbonate, vasoactive medications, and blood component transfusion]. The eGFR was calculated according to the Chronic Kidney Disease Epidemiology Collaboration equation [26]. The definition of the above study variables is listed in Supplemental Table 1. Except for the APACHE II and SOFA scores, all the above parameters were included as covariables in the multivariate Cox proportional-hazards model. In addition, the initial lactate was also included during model adjustment.

In addition to the lactate groups based on the GBTM algorithm (GBTM lactate groups), the following lactate evaluation criteria were also considered in this study for comparison :(1) initial lactate level; (2) peak lactate level; (3) lactate AUC; (4) lactate clearance. The calculation of lactate AUC is shown in Supplemental Fig. 1.

For variables with missing values, we imputed the missing data using MissForest, a random forest imputation algorithm for missing data implemented in R (version 4.0.0) [27]. The missing rates of the study variables are shown in Supplemental Table 2.

2.3 Study Outcomes

The short- (30-day mortality) and long-term (90-day mortality) outcomes of the study were defined as death within 30 days and 90 days of admission.

2.4 Statistical Analysis

STATA 17.0 (StataCorp, College Station, TX) was used for statistical analysis. A two-tailed P < 0.05 was statistically significant. To capture the pattern changes of each blood lactate parameter, we used GBTM to identify distinct subgroups that shared a similar trajectory. Since the blood lactate levels were not normally distributed, the data were first logarithmically transformed. We fitted models with up to three latent classes and up to cubic polynomials. The optimal number of trajectory classes and trajectory shapes were selected based on the following criteria [28]: (1) a censored normal model for continuous tendency toward a parsimonious model, (2) the higher value of Bayesian information criterion (BIC) and Akaike information criterion (AIC), (3) each trajectory group had the odds of correct classification by weighted posterior proportions > 5, (4) each trajectory group had an average posterior probability of membership > 0.80 and (4) the expected number based on the sums of the posterior probabilities should be the same as in the proportion in each group based on the assignments for the maximum posterior probability.

To determine whether GBTM lactate groups could refine the predictive ability for short- and long-term outcomes, we compared the changes in AUC, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) after incorporating the identified GBTM lactate groups, peak lactate value, lactate clearance and lactate AUC into the established models (initial lactate + SOFA score/APACHE II score).

The baseline characteristics between groups were reported as the means (± SD) for normally distributed data and medians with 25th and 75th percentiles for non-parametric data. Categorical variables were summarized using percentages. Differences between groups were compared using Fisher's exact test for data with a binomial distribution. The Mann-Whitney U test and unpaired t-test were used for group comparisons of continuous variables with skewed and normal distributions, respectively. The associations between GBTM lactate groups and short- and long-term risk were estimated using multivariable Cox proportional hazard analyses. Box-Whisker plots and 3D scatter plots were used to describe the GBTM lactate group characteristics by prior lactate evaluation systems.

Finally, the GBTM groups and initial lactate groups were compared. The initial lactate groups were defined as group 1: ≤2 mmol/L; group 2: 2–4 mmol/L; and group 3: ≥4mmol/L. We used the multivariable fractional polynomial interaction (MFPI) approach [29] to investigate potential interactions between group assignment (group 2 vs. 1 and group 3 vs. 1) and continuous variables such as age, eGFR, SOFA score, and APACHE II score for short- and long-term outcomes. All MFPI analyses were adjusted for covariates mentioned above. We decided to use only fractional polynomial degree 2 functions to analyze interactions. MFPI was used to estimate a fractional polynomial function for each group representing the prognostic effect of the continuous covariate of interest, optionally adjusting for other covariates. The covariables involved in adjustments are described in the section "2.2 Study Variables". A P-value < 0.05 was statistically significant for covariate selection. To quantify the magnitude of effects, we estimated hazard ratio (HR) with pointwise 95% confidence intervals (CI) as a continuous treatment effect. We adopted the 10th to the 90th quantile of distribution for the four continuous indicators.

3 Results

3.1 Patient characteristics

From January 2019 to December 2021, 1,437 patients were discharged from the comprehensive ICU of Tianjin Medical University General Hospital. As shown in Fig. 1, 762 patients were included in the final analysis. 254 and 306 patients experienced mortality at 30 and 90-days, respectively. There were no cases of loss to follow-up. Comparisons of short- and long-term outcomes are shown in Supplemental Table 3. For both short- and long-term outcomes, lactate levels at all time points within 48 hours, peak lactate levels, and lactate AUC were significantly higher in the group with adverse events.

3.2 Trajectories of serial blood lactate within 48 hours after admission to the ICU based on the GBTM algorithm

As shown in Fig. 2, based on the GBTM algorithm, after considering parsimony and assessing the relative changes in model fitting at each step, we selected a model with three latent groups and linear, quadratic, and linear polynomials (see Supplemental Table 4 for parameters for models fitted with different numbers of latent groups and different degrees of polynomials). GBTM lactate groups 1, 2, and 3 accounted for 51.7% (n = 394), 41.2% (n = 314) and 7.09% (n = 54) of patients, respectively. The baseline characteristics of the groups are shown in Table 1. The blood pressure, platelet count, albumin, and GCS score were significantly reduced. In contrast, heart rate, fluid intake per hour, WBC, ALT, AST, SOFA scores were significantly increased from group 1 to group 3 and required more treatment, including emergency surgical operation, invasive mechanical ventilation support, glucocorticoid, CRRT, intravenous sodium bicarbonate, vasoactive medications, and blood component transfusion.

Table 1

Baseline characteristics

Characteristics

Total n = 762

GBTM lactate groups

P value

Group 1 n = 394

Group 2 n = 314

Group 3 n = 54

Demographics

         

Age, year

62.5 ± 16.7

61.9 ± 17.9

63.8 ± 15.4

60.3 ± 14.7

0.21

Male, n (%)

468 (61.4)

250 (63.5)

189 (60.2)

29 (53.7)

0.32

Previous history

         

Hypertension, n (%)

409 (53.7)

219 (55.6)

168 (53.5)

22 (40.7)

0.12

Diabetes, n (%)

242 (31.8)

141 (35.8)

89 (28.3)

12 (22.2)

0.034

CAD, n (%)

180 (23.6)

103 (26.1)

67 (21.3)

10 (18.5)

0.23

Stroke, n (%)

167 (21.9)

96 (24.4)

65 (20.7)

6 (11.1)

0.069

Renal failure, n (%)

84 (11.0)

55 (14.0)

47 (15.0)

13 (24.1)

0.003

Cancer, n (%)

96 (12.6)

36 (9.10)

47 (15.0)

13 (24.1)

0.003

On-admission clinical characteristics

         

SBP, mmHg

137.6 ± 35.5

145.9 (34.8)

131.4 (33.5)

112.9 (34.0)

< 0.001

DBP, mmHg

70.4 ± 18.9

74.7 (18.6)

67.2 (17.8)

57.8 (17.6)

< 0.001

Heart rate, bpm

101.4 ± 24.4

97.2 (23.0)

104.8 (25.1)

111.3 (24.7)

< 0.001

eGFR, mL/min/1.73m2

57.4 ± 41.6

60.3 (44.1)

55.3 (38.6)

48.5 (39.1)

0.073

Fluid intake, mL/hour

161.2 (109.1 to 240.8)

150.7 (105.1 to 224.1)

164.5 (112.6 to 233.4)

259.8 (170.1 to 445.3)

< 0.001

Fluid outtake, mL/hour

136.1 (82.5 to 235.1)

143.4 (84.0 to 228.0)

122.4 (76.3 to 232.9)

150.9 (100.4 to 325.5)

0.12

RBC, 109/L

3.36 (2.63 to 4.11)

3.37 (2.63 to 4.12)

3.37 (2.66 to 4.09)

3.07 (2.47 to 4.09)

0.32

WBC, 109/L

10.8 (7.66 to 15.8)

10.3 (7.51 to 14.1)

11.9 (8.26 to 17.6)

12.0 (6.94 to 19.3)

0.002

Platelet, 109/L

165.0 (93.0 to 227.0)

177.5 (122.0 to 246.0)

151.5 (77.0 to 208.0)

99.0 (37.0 to 161.0)

< 0.001

Hemoglobin, g/L

101.5 (80.0 to 124.0)

101.0 (80.0 to 124.0)

104.0 (81.0 to 124.0)

94.5 (74.0 to 125.0)

0.44

ALT, U/L

36.0 (22.0 to 78.0)

29.0 (20.0 to 55.0)

44.5 (24.0 to 98.0)

66.0 (37.0 to 254.0)

< 0.001

AST, U/L

54.0 (32.0 to 119.0)

42.5 (27.0 to 76.0)

67.5 (36.0 to 176.0)

153.0 (63.0 to 472.0)

< 0.001

Albumin, g/L

30.0 (26.0 to 34.0)

31.0 (28.0 to 35.0)

30.0 (26.0 to 33.0)

25.0 (22.0 to 30.0)

< 0.001

GCS Score

13 (7 to 15)

14 (10 to 15)

11 (5 to 15)

13 (5 to 15)

< 0.001

APACHE II Score

18 (13 to 24)

17 (12 to 22)

21 (16 to 26)

19 (15 to 31)

< 0.001

SOFA Score

11 (8 to 14)

10 (7 to 12)

12 (10 to 16)

17 (13 to 20)

< 0.001

ARDS Severity, n (%)

       

< 0.001

Mild

305 (40.0)

196 (49.7)

93 (29.6)

16 (29.6)

 

Moderate

326 (42.8)

160 (40.6)

139 (44.3)

27 (50.0)

 

Severe

131 (17.2)

38 (9.60)

82 (26.1)

11 (20.4)

 

Emergency surgical operation, n (%)

 

89 (11.7)

30 (7.60)

47 (15.0)

12 (22.2)

< 0.001

Invasive mechanical ventilation, n (%)

 

479 (62.9)

197 (50.0)

236 (75.2)

46 (85.2)

< 0.001

Other in-hospital treatment within 24 hours of admission

Antibiotics, n (%)

674 (88.5)

339 (86.0)

287 (91.4)

48 (88.9)

0.08

Sedative and analgesic, n (%)

463 (60.8)

225 (57.1)

200 (63.7)

38 (70.4)

0.068

Glucocorticoid, n (%)

336 (44.1)

127 (32.2)

178 (56.7)

31 (57.4)

< 0.001

CRRT, n (%)

262 (34.4)

132 (33.5)

100 (31.8)

30 (55.6)

0.004

LWMH, n (%)

101 (13.3)

58 (14.7)

40 (12.7)

3 (5.60)

0.16

Sodium bicarbonate, n (%)

361 (47.4)

155 (39.3)

167 (53.2)

39 (72.2)

< 0.001

Vasoactive medications, n (%)

249 (32.7)

77 (19.5)

132 (42.0)

40 (74.1)

< 0.001

Blood component transfusion, n (%)

267 (35.0)

108 (27.4)

122 (38.9)

37 (68.5)

< 0.001

Abbreviations: ALT, alanine aminotransferase; APACHE II, Acute Physiology And Chronic Health Evaluation Ⅱ; AST, aspartate aminotransferase; CRRT, continuous renal replacement therapy, DBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; GBTM, group-based trajectory modeling; GCS, Glasgow Coma Scale; LMWH, low molecular weight heparin; RBC, red blood cell; SBP, systolic blood pressure; SOFA, Sequential Organ Failure Assessment; WBC, white blood cell.

As shown in Supplemental Fig. 2A, compared with group 1, groups 2 and 3 were characterized by significantly increased initial lactate, peak lactate level, lactate clearance, and lactate AUC (Supplemental Table 5 for details). A "funnel shape" 3D scatter plot was observed when three dimensions (initial lactate, peak lactate value, and lactate AUC) were considered (Supplemental Fig. 2B).

3.3 Associations between GBTM lactate groups and short- and long-term outcomes

The GBTM group 3 exhibited the worst (HR for 30-day mortality: 2.26, 95% CI: 1.75 to 2.92, P < 0.001; HR for 90-day mortality: 3.06, 95% CI: 1.78 to 5.27, P < 0.001 outcomes followed by the GBTM lactate group 2 (HR for 30-day mortality: 1.79, 95% CI: 1.30 to 2.47, P < 0.001; HR for 90-day mortality: HR: 1.73, 95% CI: 1.30 to 2.31, P < 0.001). The Cox hazard curves are shown in Fig. 3. When the GBTM lactate groups were added to the initial lactate models + SOFA score, there were significant increases in the established models' AUC value for short- (0.721 vs. 0.694, P < 0.001) and long-term outcomes (0.708 vs. 0.679, P < 0.001) with enhanced risk of reclassification capacity (positive and significant continuous NRI values), as well as IDI values. Furthermore, the GBTM lactate groups + initial lactate + SOFA score yielded the highest predictive performance compared with the models where the lactate AUC, lactate clearance and the highest lactate values were added to the initial lactate + SOFA score (Fig. 4). These results were consistent in initial lactate + APACHE II score models (Supplemental Fig. 3).

3.4 Comparison of the nonlinearity on age, eGFR, SOFA score, and APACHE II score between GBTM lactate groups and initial lactate groups assignment with respect to short- and long-term outcomes

The baseline characteristics based on initial lactate levels are shown in Supplemental Table 6. The initial lactate groups 1, 2, and 3 accounted for 54.1% (n = 412), 26.8% (n = 204) and 19.2% (n = 146) of the cases, respectively. Compared with initial lactate group 1, group 2 experienced a 1.34- (HR: 1.34, 95% CI: 0.98 to 1.84, P = 0.068) and 1.51-fold (HR: 1.67, 95% CI: 1.15 to 2.41, P = 0.007) greater risk of short- and long-term mortality, respectively. Group 3 were susceptible to the greatest increase in short-(HR: 1.51, 95% CI: 1.14 to 2.00, P = 0.004) and long-term (HR: 1.59, 95% CI: 1.12 to 2.24, P = 0.009) risk (Supplemental Fig. 4).

Using the MFPI procedure, non-linearities in survival were found for the GBTM lactate groups/initial lactate groups combined with age, eGFR, SOFA score, and APACHE II score. As seen in Fig. 5, GBTM lactate group 3 experienced higher 30-day mortality than group 2 and was clearly separated and consistent across multiple continuous indicators (despite the overlap of CI). However, based on the initial lactate groups, the risk of groups 3 and 2 could not be well discriminated. GBTM lactate groups can better identify high-risk patients on four continuous scales. Compared with the GBTM lactate group 1, age 35–80, eGFR (10–110 mL/min/1.73m2). SOFA score (9–17) and APACHE II score (10–26) were statistically significant in 85.2% (46/54), 81.5% (44/54), 48.1% (26/54), and 59.3% (32/54) of GBTM lactate group 3 patients [with the lower 95% CI above the line of equivalence (HR: 1)]. In the GBTM lactate group 2, the age, eGFR, SOFA score and APACHE II achieved statistical significance in 80.3% (252/314), 81.2% (255/314), 65.3% (205/314), and 68.5% (215/314) of the total population in group 2. In contrast, no significant difference in the four continuous parameters were found in the initial lactate groups. Hence, high-risk patients could not be identified. No significant interaction was observed across the four continuous parameters. The 90-day mortality was also roughly consistent with the above results (Supplemental Fig. 5).

4 Discussion

In this retrospective cohort study of ARDS patients complicated with sepsis in the ICU, we validated the presence of elevated lactate levels in septic patients with ARDS. Importantly, this is the first study to document the value of GBTM-based serial blood lactate evaluations to significantly improve the diagnostic capacity of traditional critical care evaluation systems compared to previously described lactate approaches. Interestingly, we used the MFPI algorithm to demonstrate that GBTM lactate groups yield superior performance in identifying high-risk patients than initial lactate groups for both short- and long-term outcomes.

Sepsis/multiple organ failure is well-established as a common cause of mortality in ARDS patients, and improvement in clinical outcomes may depend more on the management of sepsis/multiple organ failure than on oxygenation measures alone [20]. It is widely acknowledged that blood lactate level is a useful marker of the severity of sepsis or multiple organ failure, with higher levels predictive of higher mortality [9]. The peak lactate level has been initially applied in various blood lactate evaluation systems [8]. However, it should be borne in mind that tissue hypoperfusion and hypoxia may be present with elevated lactate levels before routine hemodynamic measurements change [30]. Therefore, initial measurements of lactate levels in the ICU are valuable for early identification. Mikkelsen et al. found that the initial serum lactate was associated with mortality independent of clinically apparent organ dysfunction and shock in patients admitted to the emergency department with severe sepsis [10]. Early initial lactate-guided therapy can significantly reduce ICU mortality [11]. Overwhelming evidence substantiates that initial blood lactate levels ranging from 2.0 to 4.0 mmol/L are associated with poor outcomes [9, 10, 12, 13]. The initial lactate is simple to calculate and readily available in all institutions. However, a single measurement of the lactate concentration can be challenging to interpret, especially in patients with ARDS, since it reflects a balance between lactate production and elimination. In this study, we divided the initial lactate into three groups with 2.0 mmol/L and 4.0 mmol/L as cut-off values for comparison with the GBTM lactate groups. The high initial lactate group exhibited lower mortality rates than the corresponding GBTM-based lactate group. The initial lactate groups exhibited lower survival during survival curve analysis than the GBTM lactate groups (Supplemental Fig. 4). More importantly, we specifically compared GBTM lactate groups with the initial lactate groups using the MFPI algorithm. Both high (group 3) and low (group 2) initial lactate groups did not correspond well to relative high-risk and low-risk clinical outcomes on the pre-specified continuous age, eGFR, SOFA score, and APACHE II score, nor could it obtain the high-risk interval with statistical significance. Taken together, our findings substantiate that GBTM lactate groups are more advantageous than the initial lactate groups.

Since continuous blood lactate levels assessment may provide more information than individual values, serial blood lactate measurements have been proposed [31]. Subsequently, the concepts of hyperlactatemia duration [15], lactate clearance [17], and lactate AUC [16] have been recommended. In our study, GBTM-based lactate groups could roughly cover patients with high initial blood lactate, peak lactate level and lactate AUC values (Supplemental Fig. 2B), with group 3 more tightly associated with the three parameters than group 2 or 1. Accordingly, the GBTM groups could account for findings in the above four lactate evaluation systems. Findings in GBTM lactate groups were not driven by individual parameters such as initial lactate, peak lactate value, lactate AUC, etc., but by a combination of these. In addition, GBTM lactate groups exhibited significantly enhanced diagnostic performance of the established model of SOFA score/APACHE II score with the addition of initial lactate. Meanwhile, the AUC, NRI and IDI obtained by adding GBTM lactate groups were the largest compared with established models where lactate AUC, lactate clearance and the peak lactate values were added, suggesting better diagnostic efficacy.

Several limitations and shortcomings were present in this study. Given its retrospective design, our study may have been subject to systematic errors and biases with missing values for serial lactate measurements. Nevertheless, the random forest data interpolation method based on machine learning [27], known to outperform all other data interpolation methods [32], was used to handle the missing data. Moreover, as an observational study, we could not establish a causal relationship between lactate groups and clinical outcomes and exclude the impact of potential confounders. Indeed, we could not include these potential confounders during multivariate analysis due to the lack of data. Given the strong association between lactate groups and short- or long-term outcomes and the consistency of trends for continuous indicators during MFPI, it is highly unlikely that these factors can affect the conclusions of this study. Due to a lack of information (Such as bilateral infiltration on chest roentgenogram), we could not use the Berlin definition [33] to define patients with ARDS. Moreover, due to the study's retrospective nature, we could not determine the causal relationship between ARDS and sepsis. Given that the sample size in the study was small, our findings may have been biased in favor of effect size inflation, and the statistical ability may be insufficient to detect the correct number of tracks when using GBTM. Hence, future GBTM-based interventional studies with larger sample sizes are required to validate our findings.

5 Conclusions

The GBTM-based serial lactate evaluation method significantly improved the diagnostic capacity of traditional critical care evaluation systems and possessed many advantages over previously described lactate evaluation systems, especially over groups based on initial lactate levels. Importantly, GBTM-based serial lactate evaluation improved the identification of individuals at high risk of short- and long-term mortality in our retrospective cohort study of ARDS patients complicated with sepsis in the ICU.

Abbreviations

AIC = Akaike information criterion

ALT = alanine aminotransferase

APACHE II = Acute Physiology And Chronic Health Evaluation Ⅱ

ARDS = acute respiratory distress syndrome

AST = aspartate aminotransferase

AUC = area under the curve

BIC = Bayesian information criterion

CI = confidence interval

CRRT = continuous renal replacement therapy

DBP = diastolic blood pressure

eGFR = estimated glomerular filtration rate

GBTM = group-based trajectory modeling

GCS = Glasgow Coma Scale

HR = hazard ratio

ICU = intensive care units

IDI = integrated discrimination improvement

MFPI = multivariable fractional polynomial interaction

NRI = net reclassification improvement

RBC = red blood cell

SBP = systolic blood pressure

STROBE = Strengthening the Reporting of Observational Studies in Epidemiology

SOFA = Sequential Organ Failure Assessment

WBC = white blood cell

Declarations

Ethical approval and consent for participation

This study was approved by the Ethical Committee of Tianjin Medical University General Hospital (IRB2022-YX-041-01) and waived the need for informed consent. Also, this study performed in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Availability of data and material

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interest.

Funding

Not applicable.

Authors' contributions

KX and YY conceived and designed the study, supervised the analysis process, interpreted the data, and revised the manuscript. HZ, ZL and LZ analyzed the data and drafted the manuscript. WZ and TY helped to analyze the data and generate the tables and figures. All authors read and approved the final manuscript.

Acknowledgements

Not applicable.

References

  1. Bakker J, Nijsten MW and Jansen TC. Clinical use of lactate monitoring in critically ill patients. Ann Intensive Care. 2013;3:12.
  2. Bakker J, Gris P, Coffernils M, Kahn RJ and Vincent JL. Serial blood lactate levels can predict the development of multiple organ failure following septic shock. Am J Surg. 1996;171:221–6.
  3. Weil MH and Afifi AA. Experimental and clinical studies on lactate and pyruvate as indicators of the severity of acute circulatory failure (shock). Circulation. 1970;41:989–1001.
  4. Bakker J, Coffernils M, Leon M, Gris P and Vincent JL. Blood lactate levels are superior to oxygen-derived variables in predicting outcome in human septic shock. Chest. 1991;99:956–62.
  5. Vincent JL, Quintairos ESA, Couto L, Jr. and Taccone FS. The value of blood lactate kinetics in critically ill patients: a systematic review. Crit Care. 2016;20:257.
  6. Shapiro NI, Howell MD, Talmor D, Nathanson LA, Lisbon A, Wolfe RE, et al. Serum lactate as a predictor of mortality in emergency department patients with infection. Ann Emerg Med. 2005;45:524–8.
  7. Nichol A, Bailey M, Egi M, Pettila V, French C, Stachowski E, et al. Dynamic lactate indices as predictors of outcome in critically ill patients. Crit Care. 2011;15:R242.
  8. Broder G and Weil MH. EXCESS LACTATE: AN INDEX OF REVERSIBILITY OF SHOCK IN HUMAN PATIENTS. Science. 1964;143:1457–9.
  9. Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). Jama. 2016;315:801–10.
  10. Mikkelsen ME, Miltiades AN, Gaieski DF, Goyal M, Fuchs BD, Shah CV, et al. Serum lactate is associated with mortality in severe sepsis independent of organ failure and shock. Crit Care Med. 2009;37:1670–7.
  11. Jansen TC, Van Bommel J, Schoonderbeek FJ, Sleeswijk Visser SJ, Van Der Klooster JM, Lima AP, et al. Early lactate-guided therapy in intensive care unit patients: a multicenter, open-label, randomized controlled trial. Am J Respir Crit Care Med. 2010;182:752–61.
  12. Howell MD, Donnino M, Clardy P, Talmor D and Shapiro NI. Occult hypoperfusion and mortality in patients with suspected infection. Intensive Care Med. 2007;33:1892–9.
  13. Wacharasint P, Nakada TA, Boyd JH, Russell JA and Walley KR. Normal-range blood lactate concentration in septic shock is prognostic and predictive. Shock. 2012;38:4–10.
  14. Vincent JL and Bakker J. Blood lactate levels in sepsis: in 8 questions. Curr Opin Crit Care. 2021;27:298–302.
  15. Jansen TC, Van Bommel J, Woodward R, Mulder PG and Bakker J. Association between blood lactate levels, Sequential Organ Failure Assessment subscores, and 28-day mortality during early and late intensive care unit stay: a retrospective observational study. Crit Care Med. 2009;37:2369–74.
  16. Van Beest PA, Brander L, Jansen SP, Rommes JH, Kuiper MA and Spronk PE. Cumulative lactate and hospital mortality in ICU patients. Ann Intensive Care. 2013;3:6.
  17. Zhang Z and Xu X. Lactate clearance is a useful biomarker for the prediction of all-cause mortality in critically ill patients: a systematic review and meta-analysis*. Crit Care Med. 2014;42:2118–25.
  18. Kliegel A, Losert H, Sterz F, Holzer M, Zeiner A, Havel C, et al. Serial lactate determinations for prediction of outcome after cardiac arrest. Medicine (Baltimore). 2004;83:274–279.
  19. Vink EE and Bakker J. Practical Use of Lactate Levels in the Intensive Care. J Intensive Care Med. 2018;33:159–165.
  20. Ferring M and Vincent JL. Is outcome from ARDS related to the severity of respiratory failure? Eur Respir J. 1997;10:1297–300.
  21. Bernard GR, Artigas A, Brigham KL, Carlet J, Falke K, Hudson L, et al. Report of the American-European Consensus conference on acute respiratory distress syndrome: definitions, mechanisms, relevant outcomes, and clinical trial coordination. Consensus Committee. J Crit Care. 1994;9:72–81.
  22. Vandenbroucke JP, Von Elm E, Altman DG, Gøtzsche PC, Mulrow CD, Pocock SJ, et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and elaboration. Ann Intern Med. 2007;147:W163-94.
  23. Teasdale G and Jennett B. Assessment of coma and impaired consciousness. A practical scale. Lancet. 1974;2:81–4.
  24. Knaus WA, Draper EA, Wagner DP and Zimmerman JE. APACHE II: a severity of disease classification system. Crit Care Med. 1985;13:818–29.
  25. Moreno R, Vincent JL, Matos R, Mendonça A, Cantraine F, Thijs L, et al. The use of maximum SOFA score to quantify organ dysfunction/failure in intensive care. Results of a prospective, multicentre study. Working Group on Sepsis related Problems of the ESICM. Intensive Care Med. 1999;25:686–96.
  26. Levey AS, Stevens LA, Schmid CH, Zhang YL, Castro AF, 3rd, Feldman HI, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150:604–12.
  27. Stekhoven DJ and Bühlmann P. MissForest–non-parametric missing value imputation for mixed-type data. Bioinformatics. 2012;28:112–8.
  28. Anderson-Cook, Christine and M. Group-Based Modeling of Development. Journal of the American Statistical Association. 2006.
  29. Royston P and Sauerbrei W. A new approach to modelling interactions between treatment and continuous covariates in clinical trials by using fractional polynomials. Stat Med. 2004;23:2509–25.
  30. Rady MY, Rivers EP and Nowak RM. Resuscitation of the critically ill in the ED: responses of blood pressure, heart rate, shock index, central venous oxygen saturation, and lactate. Am J Emerg Med. 1996;14:218–25.
  31. Vincent JL, Dufaye P, Berré J, Leeman M, Degaute JP and Kahn RJ. Serial lactate determinations during circulatory shock. Crit Care Med. 1983;11:449–51.
  32. Waljee AK, Mukherjee A, Singal AG, Zhang Y, Warren J, Balis U, et al. Comparison of imputation methods for missing laboratory data in medicine. BMJ Open. 2013;3.
  33. Ranieri VM, Rubenfeld GD, Thompson BT, Ferguson ND, Caldwell E, Fan E, et al. Acute respiratory distress syndrome: the Berlin Definition. Jama. 2012;307:2526–33.