Saturday, September 23, 2023

Evaluating clinical heterogeneity and predicting mortality in severely burned patients through unsupervised clustering and latent class analysis – Scientific Reports

Study population

The present study analyzed the clinical characteristics and outcomes of 2579 patients admitted to the BICU between January 2010 and December 2021. A total of 483 patients were excluded because their admission occurred more than 24 h after the injury. Additionally, 369 patients were excluded because fluid resuscitation was not completed for them. Consequently, 1727 eligible patients meeting the criteria for this study were included in the analysis. The patients were systematically divided into four distinct groups based on their follow-up period in the ICU. These groups were categorized as follows: week 1, encompassing admission to 1 week; week 2, including those surviving for more than 1 week up to 2 weeks; week 3, for patients surviving longer than 2 weeks up to 3 weeks; and week 4, for those surviving longer than 3 weeks up to 4 weeks. The enrollment data for these patients are illustrated in Fig. 1, and the follow-up data corresponding to each group are detailed in Fig. S2.

Figure 1

Flowchart illustrating the enrollment process for the study participants.

The results of the study demonstrated that 22.5% of the patients (389 patients) died in week 1. The overall median age of the patients was 50 years (interquartile range [IQR] 40–60 years), and the majority were male (81.2%). The median TBSA was 30%, and inhalation injury was present in 44.2% of patients. The median length of stay in the BICU was 15 days (IQR 7–30 days), and the median APACHE IV, Abbreviated Burn Severity Index, rBaux, Hangang, and SOFA scores were 27, 8, 91, 132, and 3, respectively. Comorbidities included hypertension (16.5%), diabetes mellitus (7.5%), hyperlipidemia (2.6%), and cardiovascular disease (2.3%). The mortality rate was lower in week 4 (16.9%) than that in the other periods, and the length of ICU stay increased over the study period. The use of ventilators was highest in week 4 (70.1%), and the use of continuous renal replacement therapy (CRRT) was highest in week 1 (25.8%). The severity scores, except for the SOFA score, showed statistically significant differences between the groups (Table 1). The detailed results for each group are presented in Table S1. The mean prevalences of sepsis and AKI during the study period were 24.9% and 17.5%, respectively. The prevalence of AKI was high in the early stages of week 1, with the incidence of sepsis peaking thereafter and gradually declining over time (Fig. 2).

Table 1 Characteristics of enrolled patients in each of the four study groups.
Figure 2
figure 2

Prevalence of acute kidney injury (AKI) and sepsis during the study period.

Prediction performance

In the present study, we evaluated 22 biomarkers that were checked at least every 4 days. These biomarkers included white blood cell count, hematocrit, platelets, red cell distribution width (RDW), neutrophils, lymphocytes, blood urea nitrogen, creatinine, aspartate aminotransferase, alanine aminotransferase, total bilirubin, albumin, glucose, creatine kinase, lactate dehydrogenase, pH, partial pressure of carbon dioxide, partial pressure of oxygen, bicarbonate, lactate, sodium, and potassium levels. Of these biomarkers, five (RDW, bicarbonate, pH, platelets, and lymphocytes) showed statistical significance for at least 3 weeks (Table S2). The odds ratios (ORs) for RDW were significant, except in cluster B at week 4. The overall p-value for bicarbonate was significant at week 1; however, the categorical p-values were not significant (Fig. 3 and Table S2). The OR plots for all the biomarkers are presented in Fig. 3. Considering CRRT and ventilation, only ventilation was significant at weeks 1 and 2. CRRT was significant at week 1 (Fig. 3).

Figure 3
figure 3

Multiple logistic analysis of mortality by time period (A Week 1, B Week 2, C Week 3, D Week 4).


The highest utilization of CRRT was observed in Class 1, which was designated as a low-risk group during week 1, and in Class 3, which was characterized as a high-risk group during weeks 2–4. The highest usage of ventilators was recorded in Class 3 throughout the entire study period (Tables S3S6). During week 1, class 3 was identified as a high-risk group for mortality (72.3%), which had a large proportion of cluster C individuals with high lactate levels (85.6%) and pH (84.7%) (Table S3). In week 2, class 3 individuals had a mortality rate of 82.1%, and a high proportion of cluster C individuals had low pH (86.4%), and platelets (88.6%) and lymphocyte (73.9%) levels (Table S4). In week 3, class 3 had a mortality rate of 63.9% and a high proportion of cluster C individuals had low platelets (82.3%) and lymphocytes (67.5%) (Table S5). In week 4, class 3 individuals had a mortality rate of 67.5%, and a high proportion of cluster C individuals had low pH (79.2%), platelets (90.0%), and albumin (80.0%) (Table S6). A Sankey diagram depicting the association between the predictors, latent class, and mortality is shown in Fig. 4. Additionally, an interactive Sankey diagram representing the same relationship is available on the website11.

Figure 4
figure 4

Sankey Diagram to explore the predictors, clusters, latent class, and mortality.

Longitudinal profile

Our study found that pH, bicarbonate, RDW, albumin, and lactate displayed similar patterns overall, with slight variations in their levels (Figs. S3S6). In particular, cluster B in albumin showed a sharp decrease to 3 mg/dL or lower on the first day of week 1, indicating a less favorable outlook compared with cluster A. In addition, lactate levels displayed a sharp decrease from 6 mg/dL in cluster C during week 1 (Fig. 5). The biomarkers of lymphocytes and platelets showed different patterns, with cluster C’s high mortality rate in week 1 demonstrating a sharp decrease in the lymphocyte levels (Fig. 6), and cluster B’s middle mortality rate showing the highest levels of platelets in week 1 (Fig. 7). The initial clusters, defined using the kmlShape package for these biomarkers, are shown in Figs. S7S13, and the characteristics and level changes over time are listed in Tables S7S13.

Figure 5
figure 5

Longitudinal profile of the lactate levels.

Figure 6
figure 6

Longitudinal profile of the lymphocyte counts.

Figure 7
figure 7

Longitudinal profile of the platelet counts.

Source link

Related Articles

Leave a Reply

Stay Connected

- Advertisement -spot_img

Latest Articles

%d bloggers like this: