FAN score comprising fibrosis-4 index, albumin–bilirubin score and neutrophil–lymphocyte ratio is a prognostic marker of urothelial carcinoma patients treated with pembrolizumab

It is important to identify prognostic and predictive markers of metastatic urothelial carcinoma (mUC) treated with immunocheckpoint inhibitors. We sought to establish a prognostic marker for patients with mUC treated with pembrolizumab based on only blood test results. We included 165 patients with mUC in the discovery cohort and 103 with mUC who were treated with pembrolizumab in the validation cohort. Multivariate and Cox regression analyses were used to analyse the data. In the discovery cohort, the fibrosis-4 index (hazard ratio [HR]: 2.13, 95% confidence interval [CI] 1.20–3.76, p = 0.010), albumin–bilirubin score (HR 1.91, 95% CI 1.27–2.88, p = 0.002), and neutrophil–lymphocyte ratio (HR: 1.84, 95% CI 1.22–2.79, p = 0.004) were independent significant prognostic factors. We established a ‘FAN score’ that included these three aforementioned items, which were assigned one point each. We divided patients into the 0–1 point (n = 116) and 2–3 points (n = 49) groups. The FAN score was a significant prognostic marker for cancer-specific survival (CSS) (HR 1.48, 95% CI 1.19–1.83, p < 0.001) along with the Eastern Cooperative Oncology Group Performance Status. The FAN score was also a prognostic factor of progression-free survival (PFS) (HR: 1.25, 95% CI 1.01–1.54, p = 0.036) along with the presence of liver metastasis. In the validation cohort, the FAN score was a significant prognostic factor for CSS (HR: 1.48, 95% CI 1.19–1.85, p = 0.001) and PFS (HR: 1.29, 95% CI 1.02–1.62, p = 0.034). We established the FAN score as a prognostic marker for patients with mUC treated with pembrolizumab.

Recently, several studies have reported attempts to establish a diagnostic and prognostic model based on blood test results in daily clinical practice using machine learning techniques with the artificial intelligence 13 . However, the rationale is usually unknown because of the black box problem. Therefore, it is still necessary to develop a clear clinical model based on evaluations of biological function by human intelligence.
It well known the presence of liver metastasis to become a poor prognostic factor of mUC patients treated with chemotherapy and ICI therapy commonly. The aforementioned poor prognostic factors of mUC patients are associated with cancer cachexia related to liver dysfunction. So, we supposed that scoring system reflecting liver function could be useful for establishing a prognostic model using blood test.
So, we sought to establish a prognostic marker for patients with mUC treated with pembrolizumab based on only blood test results, including the scoring system reflecting liver function in this study.

Materials and methods
Study design and population. In this study, we aimed to establish a significant prognostic model using only blood tests through two independent cohorts. First, we collected and analysed blood test results of 165 patients with mUC from six institutions as a discovery cohort. After establishing a significant prognostic model in multivariate analysis, we blindly collected clinical data and evaluated its validity in 103 patients with mUC from seven institutions as a validation cohort (Supplemental Fig. 1).
Totally, we retrospectively analysed a database comprising 268 patients treated from April 2016 to December 2020 with pembrolizumab as second-line or later treatment for mUC at Osaka University Graduate School of Medicine and its affiliated hospitals listed in the acknowledgements.
This study was approved by the institutional review board of Osaka University, which provided the necessary institutional data-sharing agreements before initiation of the study (#19083), and the study was conducted in accordance with the Declaration of Helsinki.
Data collection. The patient characteristics including laboratory findings were evaluated at the time of drug administration. Clinical features evaluated were age, sex, the Eastern Cooperative Oncology Group Performance Status (ECOG PS), levels of haemoglobin, serum albumin, serum bilirubin, serum aspartate aminotransferase (AST), serum alanine aminotransferase (ALT), estimated glomerular filtration rate (eGFR) 14 , and C-reactive protein (CRP), the presence of radical surgery, sites and number of organs involved in metastasis, and counts of leukocytes, monocytes, and platelets. The fibrosis-4 (Fib-4) index is reported as a non-invasive liver fibrosis marker and was calculated using the formula: age (years) × AST (U/L)/(platelet count [103/ μL] × (ALT [U/L]0.5) 15 . The albumin-bilirubin (ALBI) score has been proposed as a validated index of liver dysfunction and was calculated using the formula: log10 (T-bil [mg/dL] × 17.1) × 0.66 + albumin [mg/dL] × 10(− 0.085) 16 . The neutrophil-lymphocyte ratio (NLR) was calculated using the formula: neutrophil count/ lymphocyte count. The following cut-off values were defined according to previous reports: Fib-4 > 3.5 17 , ALBI score > − 2.6 16 , and NLR > 5.0 8 .
The proximal efficacy of pembrolizumab was evaluated based on RECIST 1.1 18 . Cancer-specific survival (CSS) time was calculated from the date of initiation of pembrolizumab until death or the date of the patient's last follow-up visit. Progression-free survival (PFS) was calculated from the date of initiation of pembrolizumab until radiological and clinical disease progression or death. The CSS and PFS rates were calculated using the Kaplan-Meier method.
To establish the prognostic prediction model for CSS, only blood test results were evaluated using Cox regression analysis with stepwise regression backward and forward selection with p < 0.05 as the criterion for model entry or continued inclusion in multivariate analysis.

Statistical analysis.
Clinical items in the multivariate analysis were sex (male or female), the primary site (upper urinary tract involvement or others), surgical removal of the primary site (yes or no), previous chemotherapy regimen (cisplatin based or non-cisplatin based), time from the last chemotherapy (< 90 days or ≥ 90 days), ECOG PS (0, 1, or ≥ 2), organs involved in metastasis (only lymph nodes, liver metastasis, or other organs), discontinuation of pembrolizumab due to adverse events (yes or no), presence of anaemia, and presence of renal dysfunction. We used the Cox regression model to calculate the hazard ratios (HRs) in univariate and multivariate analyses. Prognostic factors related to CSS and PFS were analysed using Cox regression analysis, and statistical significance was set as p < 0.05. Statistical analyses were performed using the Statistical Package for the Social Sciences software, version 20.0 (IBM Corp., Armonk, NY, USA).

Results
Patient characteristics. The clinical characteristics of the 268 patients with mUC enrolled in this study are shown in Table 1. Median patient age was 73 (range, 28-93) years, and primary cancer sites were the bladder (106 patients), upper urinary tract (157 patients), and both (5 patients). The numbers of patients with a high Fib-4 index, ALBI score, and NLR were 22 (8.2%), 134 (50%), and 83 (31%), respectively. Only the rate of surgical removal at the primary site was significantly different between the two cohorts (p < 0.001). The median duration of treatment with pembrolizumab was 3.22 (0.26-35.9 months), and the median follow-up period was 6.69 (0.26-37.0) months.
Treatment results in the two independent cohorts. In the discovery cohort, the median PFS and CSS were 2.60 months (95% confidence interval [CI] 2.32-2.88) and 8.58 months (95% CI 5.62-11.54) (Fig. 1a,b). Complete response (CR), partial response (PR), and stable disease (SD) were achieved in 14 patients (8.5%), 22 patients (13.3%), and 28 patients (17.0%), respectively. The objective response rate (ORR) was 21.8%, and the disease control rate (DCR) was 38.8% (Supplemental Table 2). In the validation cohort, the median PFS and CSS were 3.55 months (95% CI 1.79-5.31) and 9.60 months (95% CI 7.07-12.13) (Fig. 1c,d). CR, PR, and SD were achieved in four patients (3.9%), 27 patients (26.2%), and 20 patients (19.4%), respectively. The ORR was 30.1%, and the DCR was 49.5% (Supplemental Table 2). Establishment of the FAN score. Cox regression analysis identified three independent significant prognostic factors for CSS by stepwise backward analysis. The prognosis of patients with a high Fib-4 index (HR: 2.13, 95% CI 1.20-3.76, p = 0.010), high ALBI score (HR: 1.91, 95% CI 1.27-2.88, p = 0.002), and high NLR (HR:  Table 2). Because the HRs of these three prognostic factors were almost the same, we defined the FAN score (Fib-4 index, ALBI score, and NLR) as a prognostic factor by assigning one point to each item and compared the prognosis between the patients grouped by points. Although the prognosis became poorer as the score increased, there was no significant difference between scores of 0 and 1 and scores of 2 and 3 after multiple comparisons (Fig. 2d). Therefore, we classified scores of 0 and 1 as the low-risk group and scores of 2 and 3 as the high-risk group, and then evaluated its usefulness as a prognostic factor. The median PFS durations of the low-risk and high-risk groups were 2.76 months (95% CI 2.24-3.28) and 2.07 months (95% CI 1.89-2.25) (p < 0.001) (Fig. 3a). Moreover, the median CSS durations of the low-risk and high-risk groups were 12.4 months (95% CI 10.6-14.2) and 3.91 months (95% CI 1.9-5.91) (p < 0.001) (Fig. 3b).  ORR was 25.9% and 12.2% in low-risk and high-risk group based on the FAN score, respectively. Also, DCR was 48.3% and 16.3% in low-risk and high-risk group based on the FAN score, respectively and DCR was significantly higher in low-risk group base on the FAN score (p < 0.001) (Supplemental Table 3). Regarding PFS, short time from the last chemotherapy, anaemia, the presence of liver metastasis, and high-risk group based on the FAN score were associated with poor prognosis in univariate analysis (Table 3). In multivariate analysis, the presence of liver metastasis (HR: 1.93; 95% CI 1.13-3.23, p = 0.018) and high-risk group based on the FAN score (HR: 1.25, 95% CI 1.01-1.54, p = 0.036) were significant prognostic factors for predicting an inferior PFS (Table 3).
Regarding CSS, short time from the last chemotherapy, anaemia, poor ECOG PS, the presence of liver metastasis, and high-risk group based on the FAN score were associated with poor prognosis in univariate analysis (  (Table 4).
We examined the usefulness of the FAN score stratified by ECOG PS score. In 80 patients with ECOG PS of 1, the prognosis of high risk group of FAN score was significantly poorer about PFS (p = 0.018) and CSS (p < 0.001) (Supplemental Fig. 2c,d). Also, in 27 patients with ECOG PS of 2, the prognosis of high risk group of FAN score was significantly poorer about PFS (p = 0.049) (Supplemental Fig. 2e).
About clinical therapeutic response, ORR was 28.8% and 33.3% in low-risk and high-risk group based on the FAN score, respectively. Also, DCR was 53.5% and 40.0% in low-risk and high-risk group based on the FAN score, respectively and there were no significant differences about ORR (p = 0.644) and DCR (p = 0.279) between low-risk and high-risk group based on FAN score (Supplemental Table 3).
We examined the usefulness of the FAN score stratified by ECOG PS score. In 23 patients with ECOG PS of 2, the prognosis of high risk group of FAN score was significantly poorer about PFS (p = 0.018) and CSS (p < 0.001) (Supplemental Fig. 3e,f).
Although ECOG PS is well established prognostic factor, the FAN score could predict prognosis more accurately in some patient groups.

Discussion
Various prognostic factors of patients with mUC have been reported. They were mainly blood test results related to systemic inflammation [8][9][10][11] . Additionally, clinical items reflecting cancer-related cachexia have been mainly reported. Further, the presence of liver metastasis has been reported as a poor prognostic factor for patients with mUC treated with either chemotherapy 19 or immunotherapy [8][9][10][11][12] . Although these previous studies used the same blood test results, the cut-off values of each item were different according to each cohort and their versatility was low. The FAN score established herein was a significant prognostic factor of CSS and PFS in two independent cohorts, and it was based on only blood test results using the previous established cut-off points.
Among the elements comprising our established FAN score, the Fib-4 index was established as a scoring system for predicting liver fibrosis in patients with viral hepatitis and non-alcoholic fatty liver 15 . A score of ≥ 3.5 is used as a non-invasive indicator of liver cirrhosis 17 . Median Fib-4 in the discovery cohort in our study was www.nature.com/scientificreports/ 1.673, and it was significantly lower than cut-off value compared to the other several reports. We adopted to use 3.5 as a cut-off point because it was the value of 90th percentile in the discovery cohort and it could stratify the prognosis. The ALBI score is as useful as the Child-Pugh grade as a liver reserve scoring system for patients with hepatocellular carcinoma (HCC) 16 . The score is based on the serum albumin and total bilirubin levels, and the cut-off of − 2.6 is widely accepted. Since the median value of ALBI was − 2.56 in the discovery cohort of this study, we adopted − 2.6 as cut-off value, which was established for HCC. The NLR has been often reported as a prognostic factor for various types of cancer, but its cut-off value has not been established. For UC, cut-off values of 3.0 10,11 and 5.0 8 have been reported, but in our study, we used 5.0 because this value was used in a previous study with a large number of cases. In our study, we were able to show that a combination of the aforementioned three items is the most suitable scoring system for CSS and PFS in mUC, even when compared with other items such as the CRP level, which has already been reported 11 . The fact that the FAN score could be validated in two independent cohorts supports the versatility of this scoring system.
The Fib-4 index and ALBI score used in the present study are not only scoring systems for hepatic function 15,17 , but they have also recently been reported as prognostic factors for cancer immunotherapy for hepatocellular carcinoma [20][21][22] , cardiac disease 23,24 and coronavirus disease 25,26 . Additionally, they are presumed to be scores that reflect general conditions related to cachexia. Cancer-related cachexia is thought to be caused by changes in the systemic metabolic environment due to elevated inflammatory cytokines such as IL-6 and TNF 27,28 in adipose tissue 29,30 , tumour cells 31 , and hepatocytes 32 ; changes in protein synthesis and degradation in skeletal muscle; and insulin resistance 33 , which together lead to weight loss, anorexia, decreased systemic function, and increased adiposity 34 .
We could not find a significant correlation between the presence of liver metastasis and the Fib-4 index or ALBI score. The main reason for this may be the low diagnostic accuracy of simple computed tomography used in daily clinical practice for evaluating liver metastasis. In our cohort, it significantly correlated both ALBI score and NLR with cachexia related blood tests such as anaemia (p < 0.001) and CRP level (p < 0.001). Whereas, the Fib-4 index was significantly correlated with sex (p = 0.013) and age (p = 0.028) and it was not significantly correlated with cachexia related blood tests. There were no significant correlations between the Fib-4 index and ALBI score or NLR and these results suggest FAN score can include 3 scoring systems reflecting different liver dysfunctions. In the future, it will be important to elucidate the mechanism of changes in the Fib-4 index and www.nature.com/scientificreports/ ALBI score related to liver function decline caused by aging and cancer-associated cachexia, and to develop novel treatments to improve the therapeutic effect of ICI therapy.
In the present study, there was almost no difference in the CSS and PFS between the two cohorts. These results were similar to the outcomes of a phase III clinical trial 2 and other large cohort studies 10 . Herein, the median CSS durations of the patients classified in the high-risk group based on the FAN score were 3.9 months and 3.6 months, which were significantly worse than those of the patients classified in the low-risk group. Of course, about 10% of patients in the high-risk group also had long-term survival, so we cannot conclude that physicians should not treat patients with a high FAN score. However, it is necessary to develop biomarkers for each patient group based on this score to provide more appropriate medication to patients in the future.
In the current study, we did not examine the association between the incidence of adverse drug reactions and the FAN score. At the very least, there was no correlation between the FAN score and cases of drug discontinuation due to side effects.
There are several limitations to our study. These include the fact that this was a retrospective analysis, the number of cases was small, and the blood collection items were limited to those used in daily clinical practice.

Conclusions
In conclusion, we created the FAN score consisting of the Fib-4 index, ALBI score, and NLR based on blood test results. Using two independent cohorts, we found that the FAN score was a significant prognostic predictor in patients with mUC treated with pembrolizumab.

Data availability
Detailed data would remain confidential and was not allowed to be shared in our institutional review board.