Associations of Serum Total Bilirubin with Survival Outcomes in Patients with Cancer Cachexia: A Retrospective, Multicenter Cohort Study

Background: Cancer cachexia is a systemic paraneoplastic phenomenon involving multiple organs, including the liver. Total bilirubin (TBIL) is an easily obtained blood biomarker that reects liver homeostasis. This work evaluated the prognostic value of serum TBIL in patients with cancer cachexia. Methods: This study included 2,282 patients from a multicenter research database who were diagnosed with cancer cachexia between June 2012 and December 2019. The hazard ratio (HR) for all-cause mortality was analyzed using Cox proportional hazards regression models. The association of serum TBIL with all-cause mortality was modeled with restricted cubic splines. The optimal cutoff value for TBIL was calculated with maximally selected rank statistics. Results: Among the participants, there were 1,327 (58.2%) males and 955 (41.8%) females. The mean patient age was 60.4±1.5 years. The 12-month all-cause mortality rate for patients with cancer cachexia was 29.5% (95% CI: 27.6%-31.3%), resulting in a rate of 209.58 events per 1000 patient-years. An inverted L-shaped association between TBIL and all-cause mortality was observed. The cutoff point for TBIL for the prediction of the time to mortality was < 21.7 µmol/L. A high TBIL level but not the direct bilirubin (DBIL) or indirect bilirubin (IBIL) level was identied as an independent prognostic factor (HR, 1.60, 95% CI, 1.32-1.93). For patients with digestive system tumors, a high serum TBIL level ( ≥ 21.7 µmol/L) was signicantly associated with mortality. Conclusion: High TBIL levels are associated with increased all-cause mortality in patients and might be a promising prognostic indicator in patients with cancer cachexia. and interquartile ranges (IQRs) for data with TBIL log-transformed before A Cox proportional hazards model was used for univariate and multivariate OS analyses. Hazard ratios (HRs) and their 95% condence intervals (CIs) reported. Correlations between TBIL and nutritional parameters explored using pairwise Spearman’s rank correlation coecients. A possible nonlinear relationship between TBIL and HR was analyzed by restricted cubic spline regression. TBIL categorized based on an optimal cutoff point of 21.7 obtained by maximizing the sensitivity and specicity. Survival endpoints were assessed by Kaplan-Meier survival curves and compared by the log-rank test. Separate Cox models were applied to each subgroup in stratied covariate subgroup analyses. The Wald test were used to tested for trends across the median value for each quartile. Interaction tests were conducted with cross-classied multilevel interaction terms in the multivariable models. Kaplan–Meier survival curves are shown for the subgroups with signicant (P for interaction < 0.1) interaction terms. All the statistical analysis was performed using R software (version 4.0.2).


Introduction
Cancer cachexia is a systemic syndrome involving progressive body weight and muscle reductions, systemic in ammation as well as adipose tissue wasting that seriously affects patient quality of life and shortens the lifespan 1,2 . Cachexia affects 70% of cancer patients and accounts for at least 22% cancerrelated death cases 3 .Currently, there is no available standard treatment to fully reverse cancer cachexia 4 .
Cachexia Experts have reached an agreement on the de nition and classi cation of cancer cachexia 1 .
Cancer cachexia was divided into three stages: precachexia, cachexia, and refractory cachexia. Based on this model, Blum et al improved the classi cation system by further developing a four-group model that improve the accuracy of ability of discrimination 5 . Importantly, high heterogeneity in the prognosis of patients with cancer cachexia has been observed 6 . However, there is no effective method of assessing the prognosis of patients with different stages of cachexia according to these classi cation systems. Notably, prognostic biomarkers help not only identify patients with poor clinical outcomes but also stratify patients based on the optimal treatment or eligibility for clinical trials. Therefore, there exists a signi cant demand for easily determinable and readily available prognostic biomarkers to guide the assessment of life expectancy.
Cancer cachexia can affect and/or be in uenced by many organs, such as the liver 7 . It has long been known that the liver is a major contributor to cancer cachexia due to the presence of systemic in ammation and hepatic metabolism alterations 8, 9 . Bilirubin (BLB), as the nal product of heme degradation, is essential to re ect liver function. For many years, BLB has been regarded as an inauspicious sign of liver dysfunction with very limited physiological bene ts 10 . Temme et al found high serum total bilirubin (TBIL) levels, albeit within the normal range, was associated with reduced cancer mortality, especially nonlung cancer mortality (P for trend < 0.02) among 5,460 men and 4,843 women during a 10-year follow-up 11 . Consistent with this nding, recent studies showed that mildly elevated TBIL levels may be associated with reduced prevalences of cardiovascular diseases (CVDs), cancer, and metabolic syndrome 12,13 . However, the highest serum levels of TBIL were found in patients with cardiac cachexia, and they were correlated with reduction in systolic right ventricle (RV) function and increased right atrial pressure 14 . Cancer cachexia is complex, with a multifactorial etiology, and the association between serum TBIL levels and mortality in patients with cancer cachexia remains unclear. Blood is easily accessible and obtaining a blood sample is a minimally invasive procedure, which makes blood biomarkers appealing for use in clinical practice. TBIL has great potential for the prediction of individual clinical outcomes and the strati cation of patients into different prognostic subgroups.
Herein, we established the cutoff point for TBIL for the prediction of all-cause mortality in Chinese cancer patients with cachexia. In addition, a summary analysis of the clinical parameters of the patients was conducted to evaluate the potential associations of those parameters with the serum TBIL level. The aim of this study was to evaluate the potential prognostic value of the serum TBIL level in patients with cancer cachexia.

Study Participants
Brie y, this retrospective cohort study included 12,792 patients age 22 to 94 years who were diagnosed with malignant cancer based on pathology and were enrolled at more than 40 clinical centers throughout China from June 1, 2012, through December 31, 2019. Patients with multiple hospitalizations were regarded as one case. All patients provided written informed consent. No other special exclusion criteria were imposed except evidence of hemolytic disease and refusal to participate in the study. Patients with insu cient data concerning body mass index (BMI), blood test results, European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ-C30 Version 3.0) responses, and survival duration were excluded, leaving 2,282 patients with cancer cachexia for inclusion in the current analysis (Fig. 1). The average age of recipients was 60.4 ± 1.5 years with 41.8% (955/2282) of them were female. The study was approved by the institutional review board and complied with current ethics guidelines.

Data Collection and De nitions
Diagnoses of malignant cancers were based on histological evidence after operation or needle biopsy.

De nition of Cachexia
The international consensus criterion for cancer cachexia was weight loss more than 5% over the past 6 months (in the absence of simple starvation) or weight loss greater than 2% in individuals already underweight according to current BMI (BMI < 20 kg/m²) or skeletal muscle mass (sarcopenia). Skeletal muscle depletion was assessed based on mid-upper-arm muscle area according to anthropometry (men < 32 cm², women < 18 cm²) 1 . Cancer cachexia was diagnosed when any one of the criteria was satis ed.

Statistical Analyses
Overall survival (OS) was de ned as the interval between the rst assessment in the clinic until the date of death, or end of follow-up (September 31, 2019), or time of last contact, whichever came rst. The Kolmogorov-Smirnov test was used to test the normality of continuous data. Normally distributed data Data are presented as the means ± SD or as the medians and interquartile ranges (IQRs) for data with skewed distribution. Serum TBIL was log-transformed before analysis. A Cox proportional hazards model was used for univariate and multivariate OS analyses. Hazard ratios (HRs) and their 95% con dence intervals (CIs) were reported. Correlations between TBIL and nutritional parameters were explored using pairwise Spearman's rank correlation coe cients. A possible nonlinear relationship between TBIL and HR was analyzed by restricted cubic spline regression. TBIL was categorized based on an optimal cutoff point of 21.7 µmol/L obtained by maximizing the sensitivity and speci city. Survival endpoints were assessed by Kaplan-Meier survival curves and compared by the log-rank test. Separate Cox models were applied to each subgroup in strati ed covariate subgroup analyses. The Wald test were used to tested for trends across the median value for each quartile. Interaction tests were conducted with cross-classi ed multilevel interaction terms in the multivariable models. Kaplan-Meier survival curves are shown for the subgroups with signi cant (P for interaction < 0.1) interaction terms. All the statistical analysis was performed using R software (version 4.0.2).

Characteristics of Serum TBIL in Patients with Cancer Cachexia
A total of 2,282 patients of cancer-related cachexia were diagnosed among 12,279 participants. By the end of the study, (December 31, 2019), 1,245 (5%) death cases were identi ed. The serum TBIL levels were higher in liver cancer patients and lower in patients with nasopharyngeal cancer and colorectal cancer (CRC) (Fig. 2). No association between the serum TBIL concentration and tumor stage was observed ( Supplementary Fig. 1). Further potential associations between serum TBIL levels and covariates in patients with cancer cachexia are presented in Supplementary Fig. 2. The serum TBIL levels were positively correlated with AST and ALT levels. Weak correlations were found between serum TBIL levels and other clinical indexes (age, BMI, albumin, platelet count, RBC count, HGS, Karnofsky Performance Status (KPS) score, EORTC QLQ-C30 score, and PG-SGA score).

Association of Serum TBIL with Overall Survival
The results of the univariate and multivariate Cox regression analyses of the associations of serum TBIL levels with OS in cancer cachexia patients are shown in Supplementary Table 1. Most baseline characteristics were associated with an increased risk of mortality in the univariate analyses. Age, tumor type, TNM stage, albumin, TBIL, neutrophil count, AST, ALT, RBC count, platelet count, HGS, sarcopenia, and EORTC QLQ-C30 score were independent prognostic factors in multivariate analysis. However, DBIL and IBIL were not independent prognostic markers. Restricted cubic spline models revealed that the risk of mortality was positive correlated with the serum TBIL level (Fig. 3). In addition, TBIL was associated with a poor prognosis in patients with cancer cachexia after adjustment for sex, age, tumor type, TNM stage, alcohol intake, albumin, neutrophil count, AST, ALT, RBC count, platelet count, HGS, sarcopenia, and EORTC QLQ-C30 score (Table 1). Each increase of one standard deviation of serum Tbil was signi cantly associated with a 13% increase in the risk of mortality in Model b (HR, 1.13; 95% CI, 1.08-1.18; P < 0.001). When the serum TBIL level was divided into tertiles, tertiles 2 (HR, 1.16; 95% CI, 1.04-1.39) and 3 (HR, 1.32; 95% CI, 1.14-1.54) were associated with a higher risk of mortality (P for trend < 0.001).

Demographics and Disease Characteristics Dichotomized by TBIL in Patients with Cancer Cachexia
Patients was further strati ed according to the cutoff point of 21.7 µmol/L that was the best for the prediction of OS ( Supplementary Fig. 3A). Kaplan-Meier curves revealed that patients with high TBIL levels had poorer OS ( Supplementary Fig. 3B). Table 2 presents the characteristics of 234 patients with cancer cachexia strati ed by high and non-high prediagnosis levels of TBIL. Patients with high TBIL levels tended to be older and with higher percent of male, anorexia, hepatobiliary disease, and receive radiotherapy and chemotherapy. Higher TBIL accompanied by other biochemical criteria: higher neutrophil count, AST, ALT, and higher DBIL level, lower albumin, lymphocyte count, RBC count, and lower platelet count. Furthermore, a higher TBIL level was associated with a higher PG-SGA score, and EORTC QLQ-C30 score (Supplementary Table 2), lower KPS score and nutrition support. Although the distribution of TBIL levels differed among cancer populations, TBIL levels were signi cantly associated with the OS of patients with digestive system tumors, especially those with CRC. (data not shown) (Fig. 4).

Subgroup Analyses of Effect Modi ers of the Association of TBIL with Survival
Strati ed analyses were conducted to compare the relationship between TBIL and OS in several subgroups ( Fig. 4 and Supplementary Table 3). Evidence of the association of a high TBIL level with an increased mortality risk was observed in the majority of subgroups. However, the similarassociations were not found among patients over 65 years old with abnormal BMI values, TNM stage III disease, lung cancer, radiotherapy, and abnormal neutrophil counts. Interestingly, interactions were observed between high TBIL levels and BMI, TNM stage, and tumor type (all P for interaction < 0.1). There were no other moderating effects of variables on the association between TBIL (< 27.1 µmol/L vs ≥ 27.1 µmol/L) and all-cause mortality. Moreover, the differential effects of each variable were assessed (Supplementary Table 4). Three covariates, namely, low BMI, advanced tumor stage, and lung cancer, combined with elevated TBIL were associated with the worst prognosis in patients with cancer cachexia ( Supplementary  Fig. 4).

Sensitivity analysis
To test the robustness of our results, a sensitivity analysis was conducted by examining whether the association would change if individuals who died less than 6 months after the diagnosis of cancer cachexia or patients with hepatobiliary disease and hepatobiliary tumors were removed at baseline (

Discussion
To our knowledge, this study is the largest to directly evaluate the association of serum TBIL with allcause mortality in patients with cachexia. The distribution of TBIL concentration was associated with several types of tumors in this cohort. Furthermore, TBIL was an independent prognostic marker in multivariate analysis. We then calculated the TBIL cutoff point for the prediction of mortality in Chinese cancer patients with cachexia. Using this cutoff point, we found that a high TBIL level was strongly associated with all-cause mortality. Further analysis showed that the mortality risk was especially elevated in patients with high TBIL levels who had low BMI values, advanced cancer stages, and digestive system cancers. In the sensitivity analysis, the exclusion of patients with evidence of hepatobiliary disease and hepatobiliary tumors did not change the positive association of TBIL with the risk of allcause mortality. In general, serum TBIL can serve as an indicator for the daily assessment of the prognosis of patients with cancer cachexia.
Cancer cachexia is a multi-organ syndrome that involves more than just skeletal muscle pathogenesis 7,16 . Hepatic function can directly or indirectly cause whole-body energy expenditure and increased mortality of patients with cancer cachexia 17,18 . BLB (TBIL, DBIL, and IBIL), albumin, ALT, and AST are frequently used as indicators of liver function. Elevated physiological marker levels re ect liver disease and hepatocyte injury. Similar to previous studies [19][20][21] , these liver parameters, except DBIL and IBIL, were independent prognostic factors in this population-based cohort study. However, there was no association between serum TBIL and all-cause mortality in the subgroup of patients with hepatocellular carcinoma (HCC) and cachexia (HR, 1.40; 95% CI, 0.68-2.87, P = 0.641, data not shown). It is conceivable that this is attributable to the complexity of pathological changes in the liver, and other mechanisms should be investigated. For example, the different effects of the various forms (alpha-to delta-) of serum TBIL should be explored. Another explanation might be the limited number of HCC participants in our cohort.
To date, a few studies have investigated the associations of BLB with other cancer risks. Moderately elevated serum BLB levels were found to be associated with a markedly reduced prevalence of CRC 22 .
Meanwhile, two previous studies reported thatserum TBIL was inversely correlated with lung cancer mortality. One study proposed that relatively higher levels of BLB may protect people against lung cancers 23 . Laura et al also identi ed an association between elevated levels of serum BLB due to genetic factors and lower rates of lung cancer 24 . Reports concerning breast cancer are consistent with these ndings 25,26 . Unfortunately, several studies have reported con icting results regarding the relationship between the BLB concentration and cancer. Two cohort studies con rmed the absence of an association between BLB and CRC 27,28 . This may be due to the fact that BLB has antioxidant and antineoplastic effects 29 , and a number of studies have demonstrated that BLB inhibits the activity of cytotoxic T lymphocytes and promotes T regulatory cell (T reg ) expansion 30,31 . Overall, consistent evidence that plasma BLB has a protective effect against cancer is lacking. In our study, in patients with digestive system tumors, there was a signi cant negative correlation between the TBIL level and overall mortality. We conjecture that BLB signaling to the gut microbiota and its regulation of the liver-gut axis may have led to the above results. This emphasizes the importance of further research to elucidate the current results and con rm whether TBIL can be used to predict survival in patients with cachexia. Further investigation of the genetic factors affecting BLB is also warranted. UDP-glucuronosyltransferase (UGT1A1) is a key enzyme involved in the metabolism of BLB, and its mutations cause BLB metabolism disorders, such as Gilbert's syndrome 32,33 . The clinical impacts of UGT1A1 promoter gene variation have been thoroughly studied in CRC patients 34 . UGT1A1*28 allele carrier status was found to be inversely associated with the development of CRC in males. Nevertheless, the limitation in most analyses was the lack of BLB measurements. In the future study, we will focus on combining genetic analyses with BLB analyses to further elucidate the mechanism.
Patients who are malnourished have a poor response to antitumor therapy and an elevated risk of cachexia-related mortality 35 .Our interaction analyses established an association between TBIL and BMI, which is a nutrition status indicator. In the subgroup analysis, TBIL was an effective indicator of prognosis in patients with normal BMI values. Patients with high TBIL levels and low BMI values had shorter survival durations, according to the KM analysis. The mechanisms underlying this association are unclear. Interestingly, patients with high BMI tended to have a better prognosis, indicating that obesity might have a protective role against cancer cachexia. Consistent with our study, several meta-analyses reported that women with higher BMI values had a lower lung cancer risk and better prognosis 36,37 .
Some limitations of our study need to be considered. Serum TBIL concentrations change dynamically in healthy people, especially depending on the fasting state. The changes in TBIL in patients with cancer cachexia should be taken into consideration. Follow-up studies are warranted to detect the impacts of dynamic changes in TBIL on mortality in patients with cancer cachexia. Another limitation is that some unmeasured indicators, such as alkaline phosphatase (ALP) and lactate dehydrogenase (LDH), which also re ect liver function, or other measured confounding factors may have affected our results.

Conclusion
In conclusion, we established for the rst time the cutoff point of TBIL for the prediction of mortality in patients with cancer cachexia. A lower level of serum TBIL was associated with a decreased risk of allcause mortality. This is important, as serum TBIL is an easily assessable, highly reproducible, and inexpensive biomarker and might be a promising clinical tool to nd patients with cancer cachexia at risk of mortality. Long-term studies on the development and progression of cancer cachexia are needed to evaluate the causal associations between serum TBIL levels and cancer cachexia. A comprehensive understanding of these mechanisms may guide the selection of appropriate clinical treatments. The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Prior to data collection, all study participants signed an informed consent form This study was approved by the institutional review board of Beijing Shijitan Hospital (ChiCTR1800020329). The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013). The study was approved by institutional review board of Beijing Shijitan Hospital (ChiCTR1800020329) and individual consent for this retrospective analysis was waived. The authors acquired written informed consent from the patient.

Consent for publication
Consent for publication was obtained from all co-author.
Availability of supporting data Data will be made available after request application and approval.

Competing interests
The authors have no con icts of interest to declare. Figure 1 Diagram of participants included in the analysis