Relationship of tumor fraction in circulating tumor DNA with prognosis in patients with advanced urothelial cancer

Abstract

Background

Advanced urothelial cancer (aUC) is often characterized by rapid symptomatic disease progression, making early disease control of particular importance. Circulating tumor DNA (ctDNA) is a promising prognostic biomarker and emerging evidence shows that a single time point/measurement of ctDNA tumor fraction (TF) is prognostic in several other tumor types but has not yet been investigated in aUC. This study aimed to evaluate the prognostic value of ctDNA TF in aUC.

Patients and methods

This study used a cohort of patients from a de-identified nationwide (US-based) urothelial clinico-genomic database who underwent ctDNA testing using FoundationOne®Liquid or FoundationOne®Liquid CDx as part of routine care. Patient/disease characteristics and laboratory and treatment data were captured from the electronic health record. Real-world progression-free survival (rwPFS) and overall survival (rwOS) were evaluated by ctDNA TF while controlling for relevant covariates.

Results

Eighty-three patients with aUC were included. High ctDNA TF (≥1%) was associated with poor prognostic clinical features. High ctDNA TF also correlated with significantly reduced rwPFS in univariable analysis [hazard ratio (HR) 2.01, 95% CI (1.14-3.56), P = 0.01] and after correction for covariates [HR 2.37, 95% CI (0.99-5.66), P = 0.05]. There was a trend toward shorter rwOS in univariable [HR 1.71, 95% CI (0.87-3.36)] and multivariable [HR 1.55, (0.57-4.23)] analysis.

Conclusions

ctDNA TF is a prognostic biomarker in aUC with potential to inform expected longevity of patients. Uniform cohorts, with regard to treatments given and line of therapy, would help further evaluate the ability of ctDNA TF to identify patients with aggressive disease and inform the design of future studies to personalize therapeutic decision making.

Highlights

• •ctDNA TF is assessed in real-world data in urothelial carcinoma.
• •ctDNA TF in UC is associated with important clinical and pathological features.
• •ctDNA TF is associated with rwPFS.
• •ctDNA TF is prognostic across a range of TF cut points.
• •More work is needed to optimize and validate a specific cut point for prognosis.

Introduction

Advanced urothelial cancer (aUC) presents significant challenges in terms of clinical management and clinical trial design due to the aggressive nature of the disease which is often characterized by rapid rates of progression. Platinum chemotherapy regimens provide control of disease progression for most patients (∼75%) eligible to receive them but responses can be short-lived¹ and tolerability is often an issue for many patients.² This is in part due to the fact that urothelial carcinoma is known to be a very heterogeneous disease with high rates of intra- and interpatient heterogeneity and expansion of chemotherapy-resistant clones.³ In contrast to platinum chemotherapy, immune checkpoint inhibitors (ICIs), specifically those targeting programmed cell death protein 1/programmed death-ligand 1, can provide durable responses but only in a fraction of the patients treated.⁴ Despite significant research efforts, the development of new treatment modalities to replace or augment platinum chemotherapy with ICIs has proven to be exceptionally challenging due to aforementioned limited disease control rates in unselected patients. Identification and validation of biomarkers to personalize therapy represent a critical unmet need in aUC. Specifically, a validated biomarker to accurately characterize tumor growth kinetics and prognosis would be of significant clinical utility for prioritization between available therapies and to inform treatment intensification.

Evaluation of circulating tumor DNA (ctDNA) is established as a useful predictive biomarker platform for treatment selection in multiple tumor types including aUC. Furthermore, recent work shows that the quantified levels of total ctDNA (as opposed to the presence of specific genetic markers) can serve as a powerful prognostic biomarker. It has been demonstrated that a single time point/measurement of ctDNA tumor fraction (TF) predicts prognosis in several other tumor types but this has not yet been investigated in aUC.⁵ This study aimed to evaluate the prognostic value of ctDNA TF in aUC.

Patients and methods

Study design and patients

We investigated a nationwide (US-based) de-identified Flatiron Health-Foundation Medicine advanced urothelial clinico-genomic database (FH-FMI CGDB). De-identified data originated from ∼280 US cancer clinics (∼800 sites of care). Retrospective longitudinal clinical data were derived from electronic health record (EHR) data, comprising patient-level structured and unstructured data, curated via technology-enabled abstraction, and were linked to genomic data derived from comprehensive genomic profiling (CGP) tests in the FH-FMI CGDB by de-identified, deterministic matching.⁶ The data are de-identified and subject to obligations to prevent re-identification and protect patient confidentiality. Institutional review board (IRB) approval of the study protocol by the WCG IRB (registration number 1342690) was obtained before study conduct and included a waiver of informed consent. The cut-off date for data was 31 March 2023.

Patients in the advanced urothelial CGDB had a chart-confirmed diagnosis of stage IV or node-positive cancer of the bladder or urinary tract. Eligibility for inclusion in this study among the CGDB multi-tumor cohort is outlined in the Consolidated Standards of Reporting Trials (CONSORT) diagram (Figure 1). Patients in the final study cohort required CGP using either FoundationOne®Liquid or FoundationOne®Liquid CDx (Foundation Medicine, Cambridge, MA) on a specimen collected within 60 days before the start of a systemic line of therapy and this test must have resulted in a quantifiable (i.e. a single, discrete number, which may be zero) ctDNA TF result. Since most patients in the CGDB completed tissue-based Foundation Medicine testing, and patients typically receive only one CGP test, only 8.2% (190/2325) of total patients had a liquid biopsy compatible with ctDNA TF calculation. Lines of therapy were derived from structured and unstructured EHR treatment data using an oncologist-defined, rule-based approach⁷ and counted only systemic lines of therapy within the advanced setting. For patients with multiple eligible ctDNA TF results across different lines of therapy, the earliest line of therapy and the temporally closest ctDNA TF to the therapy start date were used.

Figure 1.

Diagrams illustrating patient selection for inclusion in this study.

(A) CONSORT diagram. (B) CGP blood draw timing requirement. CGDB, clinico-genomic database; CGP, comprehensive genomic profiling; CONSORT, Consolidated Standards of Reporting Trials.

Comprehensive genomic profiling

Hybrid capture-based next-generation sequencing was carried out as part of routine clinical care (Foundation Medicine) with either the 70-gene FoundationOne®Liquid assay or the 324-gene FoundationOne®Liquid CDx assay,⁸ based on commercial availability at the time of testing. Both assays assess base substitutions, short insertions/deletions, rearrangements/fusions, and copy number variations.

The level of ctDNA shed for each specimen was quantified by calculating an investigational measure that merges two methods for estimation of ctDNA TF.⁹ When aneuploidy is detectable, ctDNA TF is estimated based on a computational model of tumor purity, ploidy, and copy number alterations, excluding potential clonal hematopoiesis (CH)-derived aneuploid genomic segments.¹⁰ This aneuploidy-based approach avoids erroneously inferring elevated ctDNA TF due to the presence of germline variants detected at high variant allele frequency. For samples lacking detectable tumor aneuploidy, a variant-based calculation is made by identifying the highest allele fraction non-germline variant, excluding alterations that are determined to be derived from CH based on a multiomic analysis of the alterations. The primary analyses of this study treated ctDNA TF as a binary variable, distinguishing only whether a specimen had ctDNA TF ≥1% or ctDNA TF <1%, as this cut point had previously shown utility in informing negative liquid biopsy results.¹¹ Exploratory analyses assessed the effect of varying this cut point, including 10%, a cut point previously shown to have prognostic value in four cancer types.⁵

General statistical considerations

Missing values in variables with <20% missingness were imputed from random forest models trained on the observed covariates using the R package ‘missForest’. Variables with greater degrees of missingness had missing values treated as a separate category. Real-world overall survival (rwOS) was calculated from the start of treatment to death from any cause, and patients alive at last observation were right censored. The CGDB is built on a dataset that is at least 85% sensitive for detecting patient deaths when benchmarked against the National Death Index.¹² Issuance of an FMI liquid biopsy CGP report was an inclusion criterion for the database used in this study and this can potentially occur after the start of therapy. As a result, this dataset is left truncated for the purposes of rwOS analyses since any patient who died before getting CGP would be excluded from the dataset. To account for this, risk-set adjustment was carried out, including at each time point only patients who have met all inclusion criteria before that time point as at risk in Kaplan–Meier and Cox model for rwOS analyses.¹³ The assumption of independent left truncation was verified by univariable modeling of the effect of delayed entry time on the survival outcome. Real-world progression-free survival (rwPFS) was calculated from the start of treatment to progression or death from any cause. Real-world progression was abstracted from EHR data using a clinician-anchored approach.^14,15 This dataset is not left truncated for the purposes of rwPFS since a progression event does not prevent patients from subsequently getting CGP. R version 4.3.1 software (R Foundation for Statistical, Vienna, Austria) was used for all statistical analyses.

Statistical analyses

An a priori statistical analysis plan was developed and executed. The inclusion criteria, exclusion criteria, potential biases, primary outcome measures, exploratory outcome measures, handling of missing data, and all statistical methods were specified before analysis execution.

The prespecified analysis assessed the prognostic value of ctDNA TF in both univariable and multivariable contexts. The primary univariable analysis consisted of Kaplan–Meier plots of rwOS for each tumor type stratified by ctDNA TF at a cut point of 1% and corresponding Cox proportional hazards models. Median rwOS and hazard ratios (HRs), with their respective 95% confidence intervals (CIs), along with log-rank P value are all reported. The primary multivariable analysis consisted of a single multivariable Cox proportional hazards model that incorporates ctDNA TF at a cut point of 1% and a list of variables selected by expert consultation from a larger set of variables available in the CGDB on the basis of documented prognostic relevance in UC: age, Eastern Cooperative Oncology Group (ECOG) performance status (PS), self-reported race, line number, therapy type initiated, sex, body mass index (BMI), geographic area socioeconomic status (based on the approach of Yost et al.¹⁶; lowest three quintiles versus highest two quintiles), histology, grade, primary site (upper versus lower tract), albumin [<lower limit of normal (LLN)], creatinine [>1.5 × upper limit of normal (ULN)], hemoglobin (<10 g/dl), neutrophil-to-lymphocyte ratio (>5),¹⁷ aspartate aminotransferase (>2.5 × ULN), alanine aminotransferase (>2.5 × ULN), bilirubin (>1.5 × ULN), and smoking status (ever versus never smoker). All variables were measured within the same 60-day pre-therapy window as ctDNA TF. Sites of metastasis were not available in this dataset. From this, a forest plot showing HRs and P values for each variable level was generated.

The aforementioned univariable and multivariable analyses were repeated using rwPFS as a secondary endpoint.

Given the large number of potentially prognostic variables selected by our experts, we repeated each multivariable analysis using a more targeted variable list obtained by applying LASSO regression (with 10-fold cross-validation to optimize the regularization parameter).

An exploratory analysis was conducted to assess the robustness of the prognostic associations by varying the ctDNA TF cut point used in the primary and secondary analyses. ctDNA TF values between 1% and 20% were tested at 1% increments and the HR for the univariable Cox proportional hazards model of rwOS and rwPFS was plotted for each cut point. Additionally, the primary and secondary analyses were repeated at the 10% cut point that had previously shown prognostic value in four cancer types.⁵

Results

Patient population

The study included 83 patients (Figure 1), representing 44% of the patients who underwent FMI CGP testing on a liquid biopsy compatible with ctDNA TF calculation. The liquid biopsy specimen collection date range was 27 January, 2020 to 9 February, 2023. The patient characteristics (aggregated and separated by ctDNA TF status) reflect disease traits which are known to be of prognostic or clinical importance based on prior studies^17,18 or in practice (Table 1, Supplementary Table S1, available at https://doi.org/10.1016/j.esmorw.2023.100004). None of the variables of interest for multivariable modeling had >20% missingness (Supplementary Table S2, available at https://doi.org/10.1016/j.esmorw.2023.100004).

Table 1.

Patient clinical characteristics

	Total	ctDNA TF <1%	ctDNA TF ≥1%	P value
	n = 83	n = 22	n = 61
Age at start of therapy	72.0 [67.0-79.0]	73.5 [69.2-79.8]	72.0 [66.0-79.0]	0.252
ECOG performance score, n (%)				0.007
0	29 (34.9)	14 (63.6)	15 (24.6)
1	33 (39.8)	<5 (<22.7)	29 (47.5)
2+	16 (19.3)	<5 (<22.7)	12 (19.7)
Unknown	5 (6.02)	0 (0.00)	5 (8.20)
Race (self-reported), n (%)				0.547
White	61 (73.5)	18 (81.8)	43 (70.5)
Other	14 (16.9)	2 (9.09)	12 (19.7)
Unknown/not documented	8 (9.64)	2 (9.09)	6 (9.84)
Line number, n (%)				0.289
1	51 (61.4)	11 (50.0)	40 (65.6)
2	20 (24.1)	8 (36.4)	12 (19.7)
3+	12 (14.5)	3 (13.6)	9 (14.8)
Type of systemic therapy initiated n (%)				0.016
Chemotherapy	29 (34.9)	3 (13.6)	26 (42.6)
ADC or targeted therapy	13 (15.7)	7 (31.8)	6 (9.84)
Chemo–ICI	3 (3.61)	0 (0.00)	3 (4.92)
ICI only	32 (38.6)	9 (40.9)	23 (37.7)
Other	6 (7.23)	3 (13.6)	3 (4.92)
Sex, n (%)				0.509
Female	31 (37.3)	10 (45.5)	21 (34.4)
Male	52 (62.7)	12 (54.5)	40 (65.6)
Body mass index	25.6 [23.4-29.7]	25.2 [23.5-29.8]	26.2 [23.1-29.5]	0.945
Areal socioeconomic status (grouped quintiles), n (%)				0.809
1-3: Lower SES	36 (43.4)	9 (40.9)	27 (44.3)
4-5: Higher SES	40 (48.2)	12 (54.5)	28 (45.9)
Unknown	7 (8.43)	1 (4.55)	6 (9.84)
Histology, n (%)				0.711
Urothelial	77 (92.8)	20 (90.9)	57 (93.4)
Other	5 (6.02)	2 (9.09)	3 (4.92)
Unknown/not documented	1 (1.20)	0 (0.00)	1 (1.64)
Disease grade, n (%)				0.051
High grade (G2/G3/G4)	61 (73.5)	12 (54.5)	49 (80.3)
Low grade (G1)	7 (8.43)	3 (13.6)	4 (6.56)
Unknown/not documented	15 (18.1)	7 (31.8)	8 (13.1)
Primary site, n (%)				0.706
Lower tract	61 (73.5)	15 (68.2)	46 (75.4)
Upper tract	22 (26.5)	7 (31.8)	15 (24.6)
Albumin <LLN, n (%)				0.051
WNL	57 (68.7)	19 (86.4)	38 (62.3)
<LLN	23 (27.7)	2 (9.09)	21 (34.4)
Unknown	3 (3.61)	1 (4.55)	2 (3.28)
Creatinine >1.5 × ULN, n (%)				0.835
≤1.5 × ULN	62 (74.7)	17 (77.3)	45 (73.8)
>1.5 × ULN	14 (16.9)	4 (18.2)	10 (16.4)
Unknown	7 (8.43)	1 (4.55)	6 (9.84)
Hemoglobin <10 g/dl, n (%)				0.831
≥10 g/dl	62 (74.7)	17 (77.3)	45 (73.8)
<10 g/dl	15 (18.1)	3 (13.6)	12 (19.7)
Unknown	6 (7.23)	2 (9.09)	4 (6.56)
Neutrophil-to-lymphocyte ratio >5, n (%)				0.185
≤5	39 (52.7)	13 (68.4)	26 (47.3)
>5	35 (47.3)	6 (31.6)	29 (52.7)
Aspartate aminotransferase >2.5 × ULN, n (%)				0.433
≤2.5 × ULN	75 (90.4)	21 (95.5)	54 (88.5)
>2.5 × ULN	5 (6.02)	0 (0.00)	5 (8.20)
Unknown	3 (3.61)	1 (4.55)	2 (3.28)
Alanine aminotransferase >2.5 × ULN, n (%)				1.000
≤2.5 × ULN	78 (94.0)	21 (95.5)	57 (93.4)
>2.5 × ULN	2 (2.41)	0 (0.00)	2 (3.28)
Unknown	3 (3.61)	1 (4.55)	2 (3.28)
Bilirubin >1.5 × ULN, n (%)				0.475
≤1.5 × ULN	76 (91.6)	19 (86.4)	57 (93.4)
>1.5 × ULN	2 (2.41)	1 (4.55)	1 (1.64)
Unknown	5 (6.02)	2 (9.09)	3 (4.92)
Smoking status, n (%)				0.532
History of smoking	59 (71.1)	14 (63.6)	45 (73.8)
No history of smoking	24 (28.9)	8 (36.4)	16 (26.2)

Patient clinical characteristics of the analytical cohort as a whole and between TF groups dichotomized at 1%. Numerical variables are presented as median [IQR].

ADC, antibody drug conjugate; ctDNA, circulating tumor DNA; ECOG, Eastern Cooperative Oncology Group; ICI, immune checkpoint inhibitor; IQR, interquartile range; LLN, lower limit of normal; SES, socioeconomic status; TF, tumor fraction; ULN, upper limit of normal; WNL, within normal limit.

ctDNA TF is associated with important clinical and pathological features

In the study cohort, elevated ctDNA TF of at least 1% was associated with worse ECOG PS (P = 0.007). ctDNA TF ≥1% was also marginally associated with higher pathological tumor grade (≥grade 2, P = 0.051) and with laboratory values associated with poor prognosis including low serum albumin (P = 0.051). Notably, ctDNA TF ≥1% was not associated significantly with sex, line number of systemic therapy for advanced disease, or primary tumor site (upper versus lower tract) (Table 1).

ctDNA TF is associated with real-world progression-free survival

In the study cohort, patients with elevated ctDNA TF ≥1% when compared to those with ctDNA TF <1% had a greater hazard of death but the difference was not statistically significant (HR 1.71, 95% CI 0.87-3.36, P = 0.11) (Figure 2A).

Figure 2.

Elevated ctDNA TF is prognostic for worse rwOS and rwPFS in patients with advanced urothelial cancer.

Kaplan–Meier plots of (A) rwOS and (B) rwPFS from therapy start, stratified by ctDNA TF at a cut-off of 1% as measured within 60 days prior. Exploratory analyses repeating the above at a ctDNA TF cut-off of 10%: (C) rwOS and (D) rwPFS. In all panels, adjusted P values and Cox HRs are given for the multivariable models described in the text and presented in Supplementary Figures S1 and S2, available at https://doi.org/10.1016/j.esmorw.2023.100004. ∗P < .05 and ∗∗P < .01. CI, confidence interval; ctDNA, circulating tumor DNA; HR, hazard ratio; NR, not reached; rwOS, real-world overall survival; rwPFS, real-world progression-free survival; TF, tumor fraction.

We also sought to further evaluate if the presence of at least 1% ctDNA TF had independent prognostic value to standard clinical and pathological features utilized for assessment of patient prognosis. The rwOS HR for ctDNA TF ≥1% adjusted for selected baseline characteristics was 1.55 [95% CI (0.57-4.23), P = 0.4, Figure 3A]. After refinement of the set of adjustment characteristics, the adjusted HR of rwOS for ctDNA TF ≥1% was 1.96 [95% CI (0.84-4.61), P = 0.12, Supplementary Figure S1A, available at https://doi.org/10.1016/j.esmorw.2023.100004].

Figure 3.

Multivariable modeling shows that ctDNA tumor fraction ≥1% remains prognostic for worse progression-free survival even after adjusting for established prognostic markers.

Forest plots showing hazard ratios for each variable used in multivariable Cox proportional hazards modeling for (A) rwOS and (B) rwPFS from the start of therapy. Adjustment covariates were selected by expert consultation and included age, ECOG PS, self-reported race, line number, therapy type, sex, BMI, socioeconomic status, histology, grade, primary site (upper tract versus lower tract), albumin, creatinine, hemoglobin, neutrophil-to-lymphocyte ratio, aspartate aminotransferase, alanine aminotransferase, bilirubin, and smoking status. ctDNA, circulating tumor DNA; ECOG, Eastern Cooperative Oncology Group; ICI, immune checkpoint inhibitor; LLN, lower limit of normal; PS, performance status; rwOS, real-world overall survival; rwPFS, real-world progression-free survival; SES, socioeconomic status; TF, tumor fraction; ULN, upper limit of normal; WNL, within normal limit.

However, analysis of the secondary endpoint of rwPFS showed an association between ctDNA TF ≥1% and worse rwPFS. In univariable analysis, those with ctDNA TF ≥1% had a greater hazard of progression compared to those with ctDNA TF <1% [HR 2.01, 95% CI (1.14-3.56), P = 0.01] (Figure 2B). After adjusting for selected patient characteristics, the adjusted HR of rwPFS for ctDNA TF ≥1% was 2.37 [95% CI (0.99-5.66), P = 0.05, Figure 3B]. After refinement of the set of adjustment characteristics, the adjusted HR of rwPFS for ctDNA TF ≥1% was 2.02 [95% CI (1.02-3.98), P = 0.04, Supplementary Figure S1B, available at https://doi.org/10.1016/j.esmorw.2023.100004].

As part of a further exploratory analysis discussed in the following section, we looked at ctDNA TF with a cut point of 10%, which was previously found to be robustly prognostic across multiple tumor types.⁵ In univariable analysis, TF ≥10% showed significantly worse rwOS and rwPFS [HR 2.24, 95% CI 1.26-3.96, P = 0.005 and HR 2.04, 95% CI (1.23-3.38), P = 0.005, respectively, Figure 2C and D]. ctDNA TF with a 10% cut point also showed independent prognostic value in rwOS and rwPFS after adjustment for baseline characteristics [HR 4.02, 95% CI 1.53-10.6, P = 0.005 and HR 2.13, 95% CI (1.02-4.46), P = 0.04, respectively, Supplementary Figures S2 and S3, available at https://doi.org/10.1016/j.esmorw.2023.100004].

ctDNA TF is prognostic across a range of cut points

ctDNA TF displayed a heavily right-skewed distribution in this aUC cohort with a median of 7.4% (interquartile range 0.59%-19%), similar to the distributions previously observed in other cancer types⁵ and not unlike the general characteristics of distributions of circulating tumor cell enumerations.19, 20, 21 We found a consistent prognostic association with ctDNA TF across a wide range of potential cut points from ctDNA TF ± 1% to ctDNA TF ± 20% with overlapping CIs (Figure 4). This suggests that ctDNA TF has some prognostic value across cut points.

Figure 4.

Exploratory analysis of varying the ctDNA TF cut point shows TF remains prognostic for rwOS and rwPFS regardless of the cut point.

Exploratory analysis of the effect of varying the ctDNA TF cut point on the hazard ratio for the ctDNA TF high versus ctDNA TF low groups in univariable Cox proportional hazards models of (A) rwOS and (B) rwPFS. Cut points between 1% and 20% are tested in increments of 1%. The dashed line shows a hazard ratio of 1. Behind each plot of hazard ratios is a histogram of ctDNA TF values between 1% and 20%, expressed as a percentage of all patients. ctDNA, circulating tumor DNA; rwOS, real-world overall survival; rwPFS, real-world progression-free survival; TF, tumor fraction.

Discussion

Our study demonstrates a clear association between high ctDNA TF and shorter rwPFS, but based on our prespecified analysis plan with a cut point of 1%, we did not find a significant difference in rwOS in patients with aUC. There could be multiple reasons for this discrepancy and we hypothesize that the heterogeneous nature and relatively small size of this cohort may have contributed. Furthermore, we have shown that high ctDNA TF is associated with clinical features that have previously been linked to poor prognosis¹⁸ (ECOG PS, high tumor grade, and low serum albumin). Importantly, trends in differences in rwPFS and rwOS remained after adjusting for covariates and expert-selected patient characteristics. These findings are consistent with previously described observations in other metastatic cancers including castration-resistant prostate cancer, breast cancer, non-small-cell lung cancer, and colorectal cancer suggesting the generalizability of our results.⁵

For this study, we set out to validate the prognostic value of ctDNA TF at a cut point of 1% specifically because that cut point has been shown to be useful in informing the need for reflex to tissue testing when sequencing of a liquid biopsy fails to detect actionable tumor alterations. In our exploratory analyses, we saw that ctDNA TF was prognostic across a wide range of cut points, which was similar to a previous study across four common cancer types.²² Moreover, the previously studied cut point of 10% demonstrated superior performance characteristics in this aUC cohort compared to a cut point of 1%. It is possible that the ideal cut point for informing negative liquid biopsy results and the need to reflex to tissue testing could be different from the ideal cut point for prognosis. More work will be needed to optimize and validate a specific cut point for prognosis.

Taken in totality, these results suggest that ctDNA TF is a promising prognostic biomarker which likely mirrors the growth kinetics of several metastatic tumors including aUC. As such, the implication of these findings suggests that ctDNA TF may be useful clinically for personalization of cancer treatment, especially those dependent upon patient prognosis and anticipated aggressiveness of disease. Although not a direct inference from these data, we speculate that patients who have low ctDNA TF may be better candidates for chemotherapy-free regimens with ICIs that may be characterized by lower rates of response but responses that are more durable and are accompanied by relatively lower treatment-related toxicity. Conversely, it may be that patients with high ctDNA TF likely require prioritization of therapeutic agents that offer higher rates of disease control or other strategies for intensification of therapy. While risk stratification and prognosis is currently informed by a number of clinical and pathological features, a strength of algorithmically derived ctDNA TF use is the lack of interobserver variability of measurement that can be challenging with radiological assessments. Moreover, ctDNA TF may be useful in validating genomic sequencing results obtained from liquid biopsy samples.²² Cases with low ctDNA TF are particularly vulnerable to false-negative mutation reporting due to inadequate ctDNA levels in the patient’s blood sample.¹¹ For patients with aUC, this can be useful for evaluating the validity of negative mutation calls in fibroblast growth factor receptor 2/3 given the availability of the approved targeted therapy erdafitinib for patients with susceptible alterations in these genes.²³ In light of previous reports of the association between detection of ctDNA and tumor burden across a variety of tumor types,24, 25, 26, 27, 28, 29, 30 we anticipate that, with further validation, ctDNA TF may be useful not only as a single time point prognostic tool but that eventually serial measurements of TF may also be useful in aUC as a tissue-naïve liquid biopsy-based analog of RECIST for monitoring response and progression.

When considering these findings, several limitations of this study should be acknowledged. Firstly, any non-randomized study can have unmeasured confounders, and independent validation of results would strengthen confidence in results. Additionally, the sample size in our study was relatively small, warranting caution in generalizing our findings to larger populations. We also acknowledge that the use of a single prospectively selected cut point for ctDNA TF may not represent the optimal threshold for prognostic assessments in all clinical scenarios. This is additionally confounded by the variety of prior treatments, each of which may have a different impact on ctDNA TF. Furthermore, the lack of available information on sites of metastasis as a prognostic covariate and the limited availability of TNM (tumor–node–metastasis) staging at therapy start rather than just at initial diagnosis pose additional limitations. Specifically, we were also unable to distinguish between node-only metastasis and other distant metastases. Lastly, this study focused on the value of a single measurement of ctDNA TF before initiation of therapy; however, future investigation may also explore what the addition of serial measurements of ctDNA TF may add in terms of prognostic and predictive significance. Despite these limitations, our study provides valuable insights into the prognostic significance of ctDNA TF as a biomarker and highlights its potential clinical utility in UC. Future research efforts should focus on replicating our findings in larger cohorts and optimizing the determination of ctDNA TF to further enhance its prognostic and predictive value with the goal of being implemented in future prospective interventional studies.