The current study consisted of two phases with independent cohorts: the algorithm development phase in a case-control cohort (n = 382) and the clinical validation phase in a double-blinded, multicenter, prospective trial (n = 947; Fig. 1). To our knowledge, this is the largest study of UC diagnosis that covers both BCa and UTUC. Participant enrollment and clinical and laboratory procedures are detailed in Supplementary Methods (Additional file 1).

Flow diagram of algorithm development and clinical validation of the mt-utDNA test. A total of 1,329 participants were enrolled in this study. The algorithm development cohort included 382 individuals involved in the construction of the integrated prediction model and the determination of appropriate clinical cutoffs of the mt-utDNA test. The performance of the test was validated in the prospective validation cohort (n = 947) for UC detection according to the clinical diagnosis results. Sanger sequencing and postsurgical follow-up were performed to verify the results of the test. UC: urothelial carcinoma; mt-utDNA test: multitarget urine tumor DNA test

The algorithm for the mt-utDNA test was established prior to the clinical trial (Additional file 3: Table S1). First, the cutoff values for the reference gene and single target were determined (Additional file 1: Supplementary Methods; Additional file 3: Table S2). Subsequently, to improve the accuracy and convenience in clinical applications, an integrated prediction model was constructed with the UC-score as the output. The model was trained (n = 252), and the cutoff was determined (i.e., UC-score ≥ 0 was considered positive) to balance the sensitivity and false-positive rate. Then, the established cutoffs were validated in the testing set (n = 130). The UC-score had an AUC of 0.9342, a sensitivity of 89.23%, and a specificity of 92.31% with the prespecified cutoff (Additional file 2: Fig. S2; Additional file 3: Table S3).

In the clinical validation, 947 (94.7%) participants were ultimately included, consisting of 417 pathology-confirmed UC patients, 391 participants with benign urologic diseases, and 139 with non-UC cancers (Fig. 1; Additional file 3: Table S4).

UC-scores were significantly greater in UC patients than in those with benign diseases or non-UC cancers (p < 0.001; Fig. 2A). UC-score increased with tumor grade and stage, but varied little among tumors located in the bladder, ureter, or renal pelvis (Fig. 2A; Additional file 2: Fig. S3). The area under the ROC curve (AUC) of the test was 0.9583 (Fig. 2B). The mt-utDNA test exhibited a sensitivity of 91.37% (381/417) in UC patients. The overall specificity was 95.09% (504/530), while it was 94.88% (371/391) for benign urologic diseases (AUC = 0.9596; Fig. 2C). The test demonstrated an outstanding consistency with the pathological diagnosis (κ = 0.8668). Additionally, the performance was stable among the three independent centers suggesting the robustness of the test (Additional file 3: Table S5 and S6).

Performance of the mt-utDNA test in the clinical validation study. (A) Distribution of the UC-scores across pathological groups. The median values are depicted as black dots with the line range of the interquartile range. Statistical significance was assessed by the Wilcoxon rank sum test between the UCs and non-UCs. “benign” represented the urological benign disease group and “non-UC” represented the non-UC cancer group. (B) and (C) ROC curve plots of the mt-utDNA test for distinguishing UC patients from non-UC individuals, including those with benign urologic diseases or non-UC cancers (B), or from the subgroup of patients with benign urologic diseases (C). (D) Sensitivity of the mt-utDNA test for the indicated stage (left) or location (right) of UC tumors. Tumors of Ta stage were stratified as Ta-PUNLMP, Ta-LG and Ta-HG. (E) Specificity of the mt-utDNA test for the indicated types of benign urologic diseases (left) and non-UC cancers (right). (F) Sensitivity of the mt-utDNA test in UC patients with the indicated stage (left), grade (center), and location (right), in comparison with urine cytology. Statistical significance was assessed by Pearson’s Chi-squared test. (G) Specificity of the mt-utDNA test in non-UC participants with the indicated type of benign urologic diseases or non-UC cancers, in comparison with urine cytology. Statistical significance was assessed by Pearson’s Chi-squared test. Tumors combined with carcinoma in situ (CIS) were defined as Tis; UC, urothelial carcinoma; PUNLMP, papillary urothelial neoplasm of low malignant potential; LG, low grade; HG, high grade; AUC, area under the ROC curve; CI, confidence interval; BCa, bladder cancer; UTUC, upper tract urothelial carcinoma; *P < 0.05; **P < 0.01; ***P < 0.001; n.s., no significance

Among UC patients, the detection rates increased with tumor stage and grade (p < 0.001; Additional file 3: Table S7), with sensitivities of 78.22%, 94.47%, and 100%, respectively, for Ta, T1, and T2 or above tumors. Notably, among Ta tumors, 75.81% of low-grade and 93.94% of high-grade tumors were identified (Fig. 2D), indicating that the test outperformed the single-marker assays reported in Oh’s and Deng’s studies [11, 12]. Given the high heterogeneity of UC tumors, these advantages in sensitivity, especially for early-stage tumors, were presumably attributed to the synergistic effect of multiple targets (Additional file 3: Table S8). Besides, the test enabled simultaneous detection of BCa and UTUC with a sensitivity exceeding 90% for both (Fig. 2D). The test showed widely low sensitivity to urological benign lesions and non-UC cancers (Fig. 2E; Additional file 3: Table S9), indicating strong resistance to common confounding factors. Overall, the mt-utDNA test was proven to be accurate and robust across a variety of patients suspected of UC in clinical settings.

The need for invasive, costly, and time-intensive examinations (cystoscopy, contrast-enhanced CT scans, etc.) to achieve a diagnostic conclusion is a significant issue associated with screening for UC. Impressively, the mt-utDNA test exhibited a positive predictive value (PPV) and negative predictive value (NPV) of 93.61% and 93.33%, respectively. Given the potential enrollment bias of this case-control trial, estimations of microscopic and gross hematuria populations (with an incidence of UC of 3.3% and 17%, respectively) were performed, and the NPVs were 99.69% and 98.17%, respectively (Additional file 2: Fig. S4A-B) [4]. Woldu et al. stratified hematuria patients into American Urological Association risk strata (low, intermediate, or high risk) based on sex, age, degree of hematuria, and smoking history [13], and reported the incidence of malignancy as 0.4%, 1.0%, and 6.3% in each group [14]. Correspondingly, the estimated post-test probability of UI-Seek positive subjects raised to 6.96%, 15.83%, and 55.60%, respectively (Additional file 2: Fig. S4C-E). The combination of the mt-utDNA test and other risk factors was expected to provide more accurate risk stratification. This could eventually increase clinical benefits by facilitating clinical decision-making and accelerating the referral of suspected UC patients.

Methodology comparisons of the mt-utDNA test with current urinary tests were conducted via head-to-head analysis. The sensitivity of the mt-utDNA test was remarkably superior to that of urine cytology (p < 0.001), the NMP22 test (p < 0.001), and UroVysion FISH (p = 0.039), while the specificity was comparable (Additional file 3: Table S10). A 2-fold improvement in sensitivity was observed for Ta stage (p = 0.012) and low-grade tumors (p = 0.011) compared with cytology (Fig. 2F). Indeed, the advantage was prevalent across subgroups of tumors of the indicated stages, grades, and locations (Fig. 2F). Comparisons of specificity showed similar tendencies, but no statistical significance was observed owing to the insufficiency of paired results (Fig. 2G). Overall, the mt-utDNA exhibited marked advantages in terms of sensitivity, particularly for low-grade and early-stage tumors.

The accuracy of each target was verified by Sanger sequencing, with a consistency ranging from 98.45 to 100% (Additional file 3: Table S11). Furthermore, in the post-surgery study conducted in randomly selected UC patients, 97.7% (43/44) of the subjects presented decreased UC-scores (p < 0.001; Additional file 3: Table S12; Additional file 2: Fig. S5), and 82.1% (32/39) experienced positive-to-negative conversion after resection, strongly suggesting that the positive results were specifically due to the signals released by the resected tumors. These results indicated that the UC-score might correlate with tumor burden and could imply the presence or residual of UC, which needed further confirmation in a long-term follow-up study.

Limitations existed in this study. First, the number of patients enrolled in the methodology comparisons was limited, hence sufficient power for full assessment in a representative population was lacking. Second, since the enrollment was based on a symptomatic population, the feasibility of the mt-utDNA test for screening in the asymptomatic population still requires in-depth evaluations in large-scale prospective studies. Other limitations included: (1) the clinical effectiveness of combining mt-utDNA test with additional risk factors (age, sex, hematuria, etc.) still needed to be validated in subsequent studies; (2) the long-term follow-up data was lacking; (3) the potential impact of random voided urine on the test needed further investigation; (4) since the study was performed in China, the generalizability for other populations remained to be assessed.