Usefulness of KL-6 for Predicting Clinical Outcomes in Hospitalized COVID-19 Patients

Abstract Background: Krebs von den Lungen 6 (KL-6) is a novel biomarker for interstitial lung disease, and it reflects acute lung injury. We explored the usefulness of KL-6 to predict clinical outcomes in hospitalized coronavirus disease 2019 (COVID-19) patients. Methods: In a total of 48 hospitalized COVID-19 patients, KL-6 levels were measured using the HISCL KL-6 assay (Sysmex, Kobe, Japan) with the HISCL 5000 automated analyzer (Sysmex). Clinical outcomes (intensive care unit [ICU] admission, ventilator use, extracorporeal membrane oxygenation [ECMO] use, and 30-day mortality) were analyzed according to KL-6 percentiles. Age, initial KL-6 level, Charlson comorbidity index (CCI), and critical disease were compared using the receiver operating characteristic (ROC) curve and Kaplan-Meier methods for clinical outcomes. Results: KL-6 quartiles were associated with ICU admission, ventilator use, and ECMO use (all p < 0.05), except 30-day mortality (p = 0.187). On ROC curve analysis, initial KL-6 level predicted ICU admission, ventilator use, and ECMO use significantly better than age, CCI, and critical disease (all p < 0.05); age, initial KL-6 level, CCI, and critical disease predicted 30-day mortality comparably. On Kaplan–Meier survival analysis, hazard ratios (95% confidence interval) were 4.8 (1.2–19.3) for age, 4.7 (1.1–21.6) for initial KL-6 level, 3.9 (0.9–16.2) for CCI, and 2.1 (0.5–10.3) for critical disease. Conclusions: This study demonstrated that KL-6 could be a useful biomarker to predict clinical outcomes in hospitalized COVID-19 patients. KL-6 may contribute to identifying COVID-19 patients requiring critical care, including ICU admission and ventilator and/or ECMO use.

To the best of our knowledge, no studies have evaluated the predictive performance of KL-6 compared with age, comorbidities, and disease severity in COVID-19 simultaneously. We hypothesized that KL-6 might be beneficial in predicting clinical outcomes in COVID-19. We explored the usefulness of KL-6 for predicting clinical outcomes in hospitalized COVID-19 patients compared with age, CCI, and WHO disease severity. We also analyzed serial KL-6 levels according to the 30-day mortality.

Study Population
The enrollment and clinical outcomes of the study population are presented in Figure 1. From February to June 2020, a total of 48 COVID-19 patients derived from a previous study were enrolled [2]. The enrollment criteria were as follows: hospitalized adult patients over 20 years of age with available serial residual ethylene-diamine-tetraacetic acid (EDTA) plasma samples after routine blood tests and known 30-day status. Medical records were reviewed thoroughly to assess demographic, clinical, and laboratory data. CCI, the sequential organ failure assessment (SOFA), and WHO disease severity were assessed at enrollment as described previously [3,33,34]. Clinical outcomes included ICU admission, ventilator use, ECMO use, and 30-day mortality.
In the total study population, moderate and critical disease was observed in 27.1% (n = 13) and 72.9% (n = 35), respectively, with no mild or severe disease. In critical disease patients, sepsis and septic shock were 82.9% (n = 29) and 17.1% (n = 6), respectively. Among the 13 patients with moderate disease, 1 patient was admitted to ICU. Among the 35 patients with critical disease, 12 patients were admitted to ICU. Among them, ventilators were applied to 11 patients (ARDS [n = 9]), and ECMO was applied to 6 ventilated ARDS patients. The 30-day mortality was 16.7% (n = 8).

KL-6 Assay
A total of 332 residual EDTA plasma samples were consecutively collected from 48 patients from enrollment to discharge or death. The collected residual samples were aliquoted into 200 µL and were stored at -70 • C until measurement. Frozen samples were thawed at room temperature and gently mixed immediately before measuring KL-6 levels. KL-6 levels were measured using the HISCL KL-6 assay (Sysmex, Kobe, Japan) with the HISCL 5000 automated analyzer (Sysmex) based on a two-step sandwich chemiluminescence enzyme immunoassay.

KL-6 Assay
A total of 332 residual EDTA plasma samples were consecutively collected from 48 patients from enrollment to discharge or death. The collected residual samples were aliquoted into 200 μL and were stored at -70 °C until measurement. Frozen samples were thawed at room temperature and gently mixed immediately before measuring KL-6 levels. KL-6 levels were measured using the HISCL KL-6 assay (Sysmex, Kobe, Japan) with the HISCL 5000 automated analyzer (Sysmex) based on a two-step sandwich chemiluminescence enzyme immunoassay.
The manufacturer's reference interval was up to 398.0 U/mL. Analytical measurement intervals were from 10 to 6000 U/mL. KL-6 levels were measured according to the manufacturer's instructions. Finally, 96 consecutive KL-6 levels from 48 patients were included for statistical analysis. Initial KL-6 (n = 48) indicated KL-6 level at enrollment. Follow-up (F/U) KL-6 (n = 48) indicated KL-6 level at discharge in survivors or at death in non-survivors.

Statistical Analysis
Data were presented as a number (percentage) or median (interquartile range, IQR). The Shapiro-Wilk test was used for assessing the normality of data distribution. The Kruskal-Wallis test and chi-squared test were used to compare the four groups according to KL-6 quartiles (from Q1 to Q4). KL-6 quartiles were as follows; Q1 < 160.0 U/mL, 160.0 U/mL ≤ Q2 < 234.5 U/mL, 234.5 U/mL ≤ Q3 < 449.0 U/mL, and Q4 ≥ 449.0 U/mL. The manufacturer's reference interval was up to 398.0 U/mL. Analytical measurement intervals were from 10 to 6000 U/mL. KL-6 levels were measured according to the manufacturer's instructions. Finally, 96 consecutive KL-6 levels from 48 patients were included for statistical analysis. Initial KL-6 (n = 48) indicated KL-6 level at enrollment. Follow-up (F/U) KL-6 (n = 48) indicated KL-6 level at discharge in survivors or at death in non-survivors.

Statistical Analysis
Data were presented as a number (percentage) or median (interquartile range, IQR). The Shapiro-Wilk test was used for assessing the normality of data distribution. The Kruskal-Wallis test and chi-squared test were used to compare the four groups according to KL-6 quartiles (from Q1 to Q4). KL-6 quartiles were as follows; Q1 < 160.0 U/mL, 160.0 U/mL ≤ Q2 < 234.5 U/mL, 234.5 U/mL ≤ Q3 < 449.0 U/mL, and Q4 ≥ 449.0 U/mL.
With the receiver operating characteristic (ROC) curve analysis, the optimal cut-off value of age, CCI, and WHO disease severity was 70 years, 3, and critical disease, respectively. The optimal cut-off value of initial KL-6 was 412 U/mL for ICU admission, ventilator use, and ECMO use, which was the same for all, and 322 U/mL for 30-day mortality. The distribution of age > 70 years, initial KL-6 level > 412 U/mL, CCI > 3, and critical disease was compared according to ICU admission, ventilator use, and ECMO use, using the chi-squared test or Fisher's exact test. For 30-day mortality, initial KL-6 level > 322 U/mL was applied. Mann-Whitney test or Wilcoxon signed-rank test were used to compare KL-6 levels between moderate and critical diseases, between sepsis and septic shock, and between survivors and non-survivors according to the 30-day mortality.
In the ROC curve analysis, AUC, sensitivity, and specificity of age > 70 years, initial KL-6 level > 412 U/mL, CCI > 3, and critical disease were obtained to predict ICU admission, ventilator use, and ECMO use. Initial KL-6 level > 322 U/mL was applied to predict 30-day mortality. Kaplan-Meier survival analysis was used to estimate the HR with a 95% confidence interval (CI) for 30-day mortality for age > 70 years, initial KL-6 level > 322 U/mL, CCI > 3, and critical disease. The sample size for the Kaplan-Meier survival analysis was estimated based on our previous study, and the inputs were identical to those described in our previous study, except for the alternative survival probability [2]. The alternative survival probability was set to set to S 1 (t) = 0.167 based on the 30-day mortality of this study. The sample size of 48 was considered sufficient to perform the Kaplan-Meier survival analysis. MedCalc Software (version 20.111, MedCalc Software, Ostend, Belgium) was used for statistical analysis. p value < 0.05 was considered statistically significant.

Results
Basic characteristics of the study population are summarized in Table 1. The median age (IQR) was 72.0 years (63.0-79.0), and males were 58.4% (n = 23). Among four patients with pulmonary disease, one patient had ILD with ARDS and sepsis. In laboratory data, white blood cells (WBC), neutrophils, total bilirubin, lactate dehydrogenase (LDH), and C-reactive protein (CRP) levels differed significantly according to initial KL-6 quartiles (all p < 0.05). In the severity assessment, the SOFA score differed significantly according to initial KL-6 quartiles (p = 0.005). The distribution of WHO disease severity did not differ significantly according to initial KL-6 quartiles, but the distribution of ARDS and septic shock differed significantly according to initial KL-6 quartiles. In clinical outcomes, ICU admission, ventilator use, and ECMO use differed significantly according to initial KL-6 quartiles, but not the 30-day mortality. Table 2 shows the comparison of age, initial KL-6 level, CCI, and critical disease according to clinical outcomes. The proportion of age > 70 years differed significantly according to the 30-day mortality; there was no significant difference according to ICU admission, ventilator use, and ECMO use. The proportion of initial KL-6 level > 322 U/mL (or >412 U/mL) differed significantly according to ICU admission, ventilator use, and ECMO use, except for the 30-day mortality. The proportion of CCI > 3 did not differ significantly according to all clinical outcomes. The proportion of critical disease differed significantly according to ventilator use.

Discussion
This is the first study that explored KL-6 levels, age, CCI, and critical disease simultaneously in hospitalized COVID-19 patients. Our data showed that the KL-6 level reflected COVID-19 disease severity. Initial KL-6 quartiles showed a stepwise increase in critical disease, ARDS, and septic shock (Table 1). Initial KL-6 level was higher in critical

Discussion
This is the first study that explored KL-6 levels, age, CCI, and critical disease simultaneously in hospitalized COVID-19 patients. Our data showed that the KL-6 level reflected COVID-19 disease severity. Initial KL-6 quartiles showed a stepwise increase in critical disease, ARDS, and septic shock (Table 1). Initial KL-6 level was higher in critical disease than in moderate disease without significance, but it was significantly higher in septic shock than in sepsis (Figure 2). The present findings are in line with previous findings [20,21,24,26,28].
In this study, most initial KL-6 levels were lower than 500 U/mL, the cut-off value for ILD [18]. Similar to our study, previous studies demonstrated that the median KL-6 level in COVID-19 was lower than 500 U/mL regardless of disease severity or prognosis [20,22,28,29]. In this study, only one 86-year-old female patient had ILD. She was admitted to ICU for ventilator use with ARDS and sepsis at the time of COVID-19 diagnosis. ECMO was not used. Her initial KL-6 level at ICU admission was 1103 U/mL, and it was the second highest level. She died 29 days after admission.
It has been reported that KL-6 levels significantly correlated with other inflammatory biomarkers, such as CRP, neutrophils, and IL-6 levels in COVID-19 [8,27]. KL-6 levels could increase in ARDS and could be affected by alveolar epithelial damage and the activity of TNF-α [18,20,25,29,32]. If COVID-19-induced ARDS persists for longer than three weeks, it can develop into PF [13]. In this study, eight patients presented with COVID-19-induced ARDS for longer than three weeks, but none of them underwent F/U chest tomography. Therefore, we could not confirm whether post-COVID-19 PF developed or not in these patients. Further studies on the relationship among KL-6, conventional inflammatory biomarkers, or organ-specific biomarkers are needed to assess the mechanism of the increased KL-6 levels in severe/critical COVID- 19. This study showed that initial KL-6 level was significantly associated with ICU admission, ventilator use, and ECMO use (Tables 1 and 2 and Figure 2). Regarding 30-day mortality, non-survivors had a higher proportion of initial KL-6 levels above the optimal cutoff value without significance. Age > 70 years was only significantly associated with 30-day mortality. CCI > 3 was not associated with all clinical outcomes. Critical disease was only significantly associated with ventilator use. The number of ICU admissions, ventilator use, ECMO use, and 30-day mortality was the highest in Q4 (initial KL-6 level ≥ 449.0 U/mL). Each ICU admission, ventilator use, ECMO use, and 30-day mortality rate had a higher proportion of initial KL-6 levels above their respective cut-off values than each control group. Both initial and F/U KL-6 levels were higher in non-survivors than in survivors. In addition, this study showed a substantial increase in the median KL-6 level in non-survivors in serial KL-6 measurements. Although the time points of serial KL-6 level measurements varied across previous studies [22,25,29,30], based on our data and previous studies, KL-6 levels seem to reflect COVID-19 disease progression.
Initial KL-6 level significantly predicted ICU admission, ventilator use, and ECMO use better than age, CCI, and critical disease (Figure 3). Although age and CCI were the only significant predictors of 30-day mortality, the predictive performance of KL-6 was comparable to that of age and CCI. HRs of both age and initial KL-6 level were significantly high, which were almost the same. COVID-19 prognosis is unpredictable, and patients could progress rapidly to a deteriorating state requiring critical care [3,15]. In Japan, COVID-19 patients with critical care demand accounted for 106.3% of designated medical institutions during the peak period of the COVID-19 epidemic [35]. Patients requiring a ventilator or ECMO were 88.9% and 17.7% of designated medical institutions, respectively [35]. Therefore, it is important for hospitals to have both an early prediction system and clinical care capacity for prompt management [3,35,36]. The WHO recommends monitoring signs of clinical deterioration with vital signs, clinical scores, laboratory data, electrocardiogram, or chest imaging [3]. If an ideal biomarker is available, that would be a simpler and more objective parameter compared with the aforementioned signs. Based on our data, KL-6 could be utilized as a parameter for the risk assessment requiring ICU, ventilator, or ECMO, and for predicting 30-day survival. In this study, most ARDS patients received ventilator and ECMO therapy. KL-6 may be a parameter for a tailored approach or prognostication in COVID-19-induced ARDS.
This study is limited in that it was a small-sized, single-center study that was conducted during the early COVID-19 pandemic. Although the sample size (n = 48) was enough for statistical analyses, it may not be enough to have strong implications for clinical insights into the usefulness of KL-6. In addition, our data may have been affected by the study population being biased toward critical diseases and may not be representative of COVID-19 by other variants. Second, we could not conduct blood sampling at a defined exact time, and thus, our data may have been influenced by the clinical course of COVID-19 [2]. Third, we assessed KL-6 levels only to predict ICU admission, ventilator use, ECMO use, and 30-day mortality. The prediction of post-COVID-19 PF was not within the scope of this study due to a lack of information on it.
In conclusion, this is the first study that investigated the usefulness of KL-6 for predicting clinical outcomes in hospitalized COVID-19 patients compared with age, CCI, and critical disease. KL-6 was superior to age, CCI, and critical disease for predicting ICU admission, ventilator use, and ECMO use; for predicting 30-day mortality, they showed comparable performance. KL-6 could be an objective biomarker for predicting COVID-19 patients with critical care demand. Further studies are needed to implement KL-6 as a predictive biomarker for COVID-19 in routine clinical practice. Despite these limitations, this study provides basic data on the predictive power of KL-6 for future pandemics of respiratory viruses other than COVID-19.