The impact of biomarkers in multivariate algorithms for bladder cancer diagnosis in patients with hematuria


  • The authors acknowledge the patients and consultants from the Departments of Urology at Craigavon Area and Belfast City Hospitals, Northern Ireland, who participated in or facilitated this study.

  • See editorial on pages 2566-7, this issue.



We appraised 23 biomarkers previously associated with urothelial cancer in a case-control study. Our aim was to determine whether single biomarkers and/or multivariate algorithms significantly improved on the predictive power of an algorithm based on demographics for prediction of urothelial cancer in patients presenting with hematuria.


Twenty-two biomarkers in urine and carcinoembryonic antigen (CEA) in serum were evaluated using enzyme-linked immunosorbent assays (ELISAs) and biochip array technology in 2 patient cohorts: 80 patients with urothelial cancer, and 77 controls with confounding pathologies. We used Forward Wald binary logistic regression analyses to create algorithms based on demographic variables designated prior predicted probability (PPP) and multivariate algorithms, which included PPP as a single variable. Areas under the curve (AUC) were determined after receiver-operator characteristic (ROC) analysis for single biomarkers and algorithms.


After univariate analysis, 9 biomarkers were differentially expressed (t test; P < .05). CEA AUC 0.74; bladder tumor antigen (BTA) AUC 0.74; and nuclear matrix protein (NMP22) 0.79. PPP included age and smoking years; AUC 0.76. An algorithm including PPP, NMP22, and epidermal growth factor (EGF) significantly improved AUC to 0.90 when compared with PPP. The algorithm including PPP, BTA, CEA, and thrombomodulin (TM) increased AUC to 0.86. Sensitivities = 91%, 91%; and specificities = 80%, 71%, respectively, for the algorithms.


Addition of biomarkers representing diverse carcinogenic pathways can significantly impact on the ROC statistic based on demographics. Benign prostate hyperplasia was a significant confounding pathology and identification of nonmuscle invasive urothelial cancer remains a challenge. Cancer 2011. © 2011 American Cancer Society.


Patients presenting with hematuria require investigations, including cystoscopy and imaging of their upper urinary tracts, to identify the source of bleeding. This is a significant health care burden, which is set to increase because of the aging population.

Cystoscopy incurs false-negatives,1 and is a costly procedure. Further, it is invasive, painful, and potentially embarrassing for the patient. Cytology, has high specificity but poor sensitivity, and hence, cannot act alone for diagnosis of urothelial cancer. Less than 20% of patients with macroscopic, and <5% with microscopic hematuria, have urothelial cancer. There is an urgent need to reduce the number of costly invasive procedures undertaken on hematuria patients because the service will not be able to cope with the patient numbers. Urologists may need to redesign the hematuria clinic to more effectively address the needs of all patients.2 Perhaps also those designing clinical studies on biomarkers should emphasize more that the goal must be a predictive biomarker or algorithm that can readily be translated into improved medical decision making or patient outcomes.3

Many biomarkers for bladder cancer have low specificity and are positive in large proportions of patients with pathologies other than bladder cancer, particularly inflammatory and prostate disease. In addition, interpretation of the vast literature is difficult because of the lack of a systematic approach to study design and reporting. To make interstudy comparisons of sensitivities and specificities more worthwhile a working group reviewed the Standards for Reporting of Diagnostic Accuracy (STARD) in 2003. We adopted their key recommendations in this study.4, 5

The aim of this study was to determine whether single biomarkers and/or multivariate algorithms could significantly improve on the predictive power of an algorithm based on demographics, for the prediction of urothelial cancer in patients presenting with hematuria.

The usefulness of a diagnostic test is dependent on the number of true positives, that is, sensitivity; against the number of true negatives, that is, specificity. Realistically urologists would require sensitivity and specificity >90% before they would consider novel algorithms as adjuncts for stratification of low-risk hematuria patients. In addition, to be of clinical use, algorithms would have to significantly increase the predictive power that could be achieved using algorithms based on demographics. This reasoning has been already been applied to predict recurrence and progression.3

Bladder tumor antigen (BTA)6 and nuclear matrix protein (NMP22)7 are both US Food and Drug Administration (FDA) approved for surveillance and detection of bladder cancer, respectively. We included these biomarkers as benchmarks for the biomarkers we assessed.8-28 (Table 1). Because 85% of circulatory creatinine crosses the glomerular basement membrane into urine, we deemed that creatinine would be an accurate measure of filtration rate and patient hydration status.29-31

Table 1. Biomarker Pathways
  1. Abbreviations: BTA, bladder tumor antigen; CEA, carcino-embryonic antigen; CK18, cytoKeratin 18; IL, interLeukin; CRP, C-reactive protein; EGF, epidermal growth factor; HA, hyaluronidase; MCP-1, monocyte chemoattractant protein-1; MMP-9, matrix metalloproteinase-9; NGAL, neutrophil-associated gelatinase lipocalin; NMP22, nuclear matrix protein 22; NSE, neuron-specific enolase; TNF-α, tumor necrosis factor-alpha; TM, thrombomodulin; VEGF, vascular endothelial growth factor; vWF, von Willebrand factor.

DifferentiationCEA, CK18
InflammationCRP, HA, IL-6, IL-8, NGAL, MCP-1, TNFα
Coagulationd-Dimer, TM
GrowthEGF, HA, IL-1α, IL-1β, IL-8, NMP22
ApoptosisFAS, NMP22
AngiogenesisHA, IL-1α, IL-1β, IL-8, VEGF, vWF
MetastasesCEA, HA, IL-8, MMP-9, MMP-9 NGAL complex, NGAL
ImmuneBTA, IL-2, IL-4, IL-8


We designed a prospective case-control study to explore the contributions of demographic and clinical factors (Table 2) to a diagnostic algorithm that we designated prior predicted probability (PPP). We appraised 23 biomarkers, representing proteins from diverse pathways involved in bladder cancer carcinogenesis (Table 1).

Table 2. Patient Characteristics
 nAgeCyst +veM:FSmokersDrink AlcoholChemical ExposureAH MedicationOccupational RiskDipstickCytologyTumor
  1. Abbreviations: AH, antihypertensive medication; Cyst, cystoscopy; inf, inflammation; inflammatory, inflammatory cells present in cytology; inflamm, inflammation present in tumor.

  2. Figures in brackets are percentages.

Controls77540/7755:2241/77 (53)46/7713/77 (17)22/77 (29)4721949/77 (64)22/7717/773/65 (5)26/65 (40)N/A
Ca6650/66: 05/6 (83)3/60/6 (0)2/6 (33)4205/6 (83)5/63/62/5 (40)4/5 (80)N/A
BPH12670/1212: 07/12 (58)8/122/12 (17)5/12 (42)7505/12 (42)2/122/120/12 (0)6/12 (50)N/A
Stones/inf17550/1711: 68/17 (47)8/171/17 (6)5/17 (29)115116/17 (94)8/174/171/14 (7)7/14 (50)N/A
Benign6460/64: 22/6 (33)3/61/6 (17)2/6 (33)4115/6 (83)2/62/60/5 (0)2/5 (40)N/A
No diagnosis36490/3622:1419/36 (53)24/369/36 (25)8/36 (22)218718/36 (50)5/366/360/29 (0)7/29 (24)N/A
Urothelial ca806979/8065:1560/80 (75)42/8014/80 (18)51/79 (65)4331665/80 (81)47/8032/8033/74 (45)54/74 (73)46/67 (69)
pTaG14694/43:14/4 (100)2/40/4 (0)4/4 (100)1213/4 (75)2/41/40/3 (0)2/3 (33)1/4 (25)
pTaG2357035/3522:1329/35 (83)17/352/35 (6)23/35 (66)2410125/35 (71)14/3510/354/32 (13)17/32 (53)13/28 (46)
pTaG38687/88:08/8 (100)6/83/8 (38)6/8 (75)3505/8 (63)4/83/84/7 (57)5/7 (71)6/7 (86)
pT1G23763/33:03/3 (100)2/31/3 (33)2/3 (67)2103/3 (100)3/32/32/3 (33)3/3 (100)2/2 (100)
pT1G3106410/109:17/10 (70)4/105/10 (50)5/10 (50)4429/10 (90)7/103/108/10 (80)8/10 (80)8/8 (100)
pT2aG21701/11:00/1 (0)0/10/1 (0)1/1 (100)0101/1 (100)1/11/11/1 (100)1/1 (100)1/1 (100)
pT2aG39729/99:04/9 (44)6/90/9 (0)4/9 (44)7209/9 (100)8/96/98/9 (89)9/9 (100)5/7 (71)
pT2bG32662/22:02/2 (100)1/20/2 (0)2/2 (100)1102/2 (100)2/22/20/1 (0)1/1 (100)2/2 (100)
pT3aG31701/11:01/1 (100)1/11/1 (100)0/1 (0)0101/1 (100)1/11/11/1 (100)1/1 (100)1/1 (100)
pT3bG33693/33:02/3 (67)1/31/3 (33)2/2 (100)0303/3 (100)3/32/33/3 (100)3/3 (100)3/3 (100)
pT4aG32562/22:01/2 (50)1/21/2 (50)1/2 (50)0112/2 (100)1/20/21/2 (50)2/2 (100)2/2 (100)
CIS2782/22:0  0/2 (0)1/2  0/2 (0)  1/2 (50)101  2/2 (100)1/21/2  1/2 (50)  2/2 (100)2/2 (100)

This study was approved by the regional Research Ethics committee (ORECNI 80/04) and hospital review boards and conducted according to STARD guidelines.4, 5 Patients presenting with hematuria with planned cystoscopy were recruited between November 2006 and October 2008 (Fig. 1 and Table 2). After written informed consent, we collected urine (50 mL) and serum (2 mL) samples from each patient. Samples were stored at −80°C until biomarker analyses (undertaken within 12 months of collection).

Figure 1.

This shows a flow chart of study. Patients presenting with hematuria (n = 181) were recruited. Biomarker analyses were completed on 157 patients.

Point of Care Assays and Investigations

Aution Sticks 10EA used for dipstick analyses were interpreted using PocketChem (Arkray factory, Inc., Japan). NMP22 was assessed qualitatively (<10 U/mL negative) (Matritech Inc, Newton, Mass). Cytology was assessed on Papanicolaou and Giemsa-stained preparations.

Diagnostic Algorithm

Clinicopathological data were recorded at the time of recruitment (Table 2). Three of the authors classified each patient's occupational history as low risk (score = 1), moderate risk (score = 2), or high risk (score = 3). Scores were averaged. Occupations classed as high risk included painters, wood lathe operators, and dye mixers. Current medications were grouped into 14 categories: antihypertensives (AH), anticholesterol, antiplatelets, antiulcer, benign prostate hyperplasia (BPH) therapy, that is, α-blocker and 5 α-reductase inhibitor, antiasthma, analgesics, antidepressants, anti-inflammatory, antidiabetes, anxiolytics, anticoagulants, and vitamins. After investigations patients were classified as “no diagnosis,” “benign pathologies,” “stones and inflammation,” “BPH,” “other cancers,” or “urothelial cancer.”

Biomarker Analyses

Scientists, blinded to patient data, completed analyses of biomarkers at Randox Laboratories Ltd. (County Antrim, Northern Ireland). Sixteen biomarkers in urine and 3 in serum: carcinoembryonic antigen (CEA), free prostate-specific antigen (FPSA), and total PSA (TPSA) were measured in triplicate using Randox biochip array technology (Randox Evidence and Investigator), which is multiplex system for protein analysis32 (Table 3). PSA analyses were undertaken for diagnostic confirmation and quality control purposes only.

Table 3. Univariate Analyses of Biomarker Levels in Urothelial Ca and Controls
 SensitivitiesUrothelial Ca median (IQR)Controls median (IQR)P Value
  • Abbreviations: BTA, bladder tumor antigen; CEA, carcino-embryonic antigen; CK18, cytoKeratin 18; IL, interLeukin; CRP, C-reactive protein; EGF, epidermal growth factor; HA, hyaluronidase; MCP-1, monocyte chemoattractant protein-1; MMP-9, matrix metalloproteinase-9; NGAL, neutrophil-associated gelatinase lipocalin; NMP22, nuclear matrix protein 22; NSE, neuron-specific enolase; TNF-α, tumor necrosis factor-alpha; TM, thrombomodulin; VEGF, vascular endothelial growth factor.

  • Biomarkers measurements were transformed to achieve normal distributions, which were then compared using t test (P < 0.05, two tailed significance). Median levels and (IQR) interquartile ranges are quoted for ease of comparison with other published studies. NMP22 was assessed qualitatively.

  • The average biomarker levels were divided by average creatinine levels and then log transformed before univariate analyses.

  • *

    Biomarkers levels that were significantly different in Uro Ca patients compared to controls (P ≤ .05; t test).

  • Biomarkers for which only single analysis was done. All other biomarkers were measured in triplicate.

  • §

    Square root transformed data.

  • Biomarkers that were significantly different when nonmuscle invasive and muscle invasive urothelial cancers were compared.

  • +

    Biomarkers that were significantly different when grade 1 and 2 combined were compared to grade 3.

  • NB medians and P values are not quoted for ˆ and +.

Protein (mg/mL)N/A0.14 (0.06 to 0.45)0.08 (0.05 to 0.17)0.007*+
Creatinine (μmol/L) §N/A9607.52 (6010.87 to 13589.38)9235.65 (5709.57 to 13828.70)0.958
Osmolality (mOsm)N/A543.17 (347.75 to 684.33)553.67 (415.17 to 720.50)0.394
BTA (U/mL)0.65 U/mL59.40 (16.28 to 365.93)5.24 (1.27 to 24.76)0.001*+
CEA (serum) (ng/ml)0.29 ng/mL1.96 (1.36 to 3.20)1.18 (0.82 to 1.60)0.001*+
CK18 (ng/mL)0.12 ng/mL2.70 (1.00 to 4.74)2.59 (0.87 to 6.00)0.662
CRP (ng/mL)0.67 ng/mL0.80 (0.67 to 1.10)1.01 (0.81 to 1.28)0.052
d-Dimer (ng/ml)2.10 ng/mL7.71 (2.10 to 136.78)2.10 (2.10 to 6.47)0.001 *
EGF (pg/mL)2.90 pg/mL6187.36 (3579.00 to 9593.58)8233.66 (4396.57 to 17358.79)0.103
FAS (pg/mL)5.00 pg/mL104.00 (54.71 to 152.75)64.21 (40.43 to 94.29)0.019 *
HA (ng/mL)25.00 ng/mL872.33 (485.79 to 1353.63)735.71 (453.69 to 900.41)0.039 *+
IL-1α (pg/mL)0.80 pg/mL2.21 (0.90 to 11.71)0.90 (0.90 to 2.27)0.001 *
IL-1β (pg/mL)1.60 pg/mL1.60 (1.60 to 4.22)1.60 (1.60 to 1.60)0.145
IL-2 (pg/mL)4.80 pg/mL5.71 (5.29 to 6.81)5.54 (5.21 to 6.15)0.578
IL-4 (pg/mL)6.60 pg/mL6.60 (6.60 to 6.60)6.60 (6.60 to 6.60)0.967
IL-6 (pg/mL)1.20 pg/mL5.30 (1.20 to 41.28)1.23 (1.20 to 3.33)0.001 *+
IL-8 (pg/mL)7.90 pg/mL277.67 (25.03 to 2136.67)32.08 (7.90 to 159.42)0.001*+
MCP-1 (pg/mL)13.20 pg/mL136.04 (51.63 to 289.13)86.41 (38.47 to 200.06)0.157
MMP-9 (ng/mL)3.03 ng/mL3.03 (3.03 to 3.03)3.03 (3.03 to 3.03)0.803
MMP-9NGAL complexN/A0.10 (0.08 to 0.23)0.09 (0.07 to 0.11)0.072 +
NGAL (ng/mL)17.80 ng/mL142.83 (88.63 to 523.64)139.93 (88.08 to 264.00)0.858
NSE (ng/mL)0.26 ng/mL0.26 (0.26 to 0.56)0.26 (0.26 to 0.28)0.504
TM (ng/mL)0.50 ng/mL4.18 (2.71 to 5.88)4.29 (2.80 to 5.99)0.210
TNFα (pg/mL)4.40 pg/mL10.00 (8.30 to 13.24)9.99 (7.51 to 12.49)0.979
VEGF (pg/mL)14.60 pg/mL194.83 (61.51 to 683.81)78.23 (38.27 to 176.78)0.001*+

BTA was measured using BTA TRAK enzyme-linked immunosorbent assays (ELISAs) from Polymedco, Inc., Cortlandt Manor, NY; epidermal growth factor (EGF) and the MMP-9 NGAL complex were measured using in-house ELISAs. Due to cost restraints, single measurements were carried out for hyaluronidase (HA), FAS, and cytokeratin (CK 18) using ELISAs from Echelon Biosciences Inc. (Salt Lake City, Utah), Raybio, Inc. (Norcross, Ga), and USCNLIFE Science & Technology Co. Ltd. (China), respectively. All other biomarkers were measured in triplicate.

Creatinine levels (μmol/L) and Osmolarity (mOsm) were measured in triplicate using a Daytona RX Series Clinical Analyzer (Randox) and a Löser Micro-Osmometer (Type 15) (Löser Messtechnik, Germany), respectively. Total protein levels (mg/mL) in urine were determined using a Bradford assay A595nm (Hitachi U2800 spectrophotometer) using BSA as standard.

Statistical Analyses

Average measurements for each biomarker were divided by the average creatinine level measured in the same patient's urine sample and then log transformed before statistical analyses using SPSS v17. All P values reported in this study are 2 tailed.

Receiver operating characteristic (ROC) curves were created to rank area under the curves (AUCs). From these we determined the cutoff limit for each biomarker/algorithm that would delineate positive from negative test results. This point was taken as the measured level at the minimum distance from the top of the y-axis of the ROC curve, that is, the point of maximum specificity. We pasted the coordinate points of ROC curves into Excel and applied Pythagoras' theorem to determine the distance of each point from the top of the y-axis of the ROC curve, that is, distance = (1 − sensitivity)2 + (1 − specificity)2.

We entered demographic variables into a Forward Wald binary logistic regression analyses (cutoff probability for case classification = 0.5) to create a diagnostic algorithm that we designated as PPP. PPP was created because the baseline characteristics of the urothelial cancers and control groups were different (Table 2). PPP represents, in a single measure, the intrinsic contribution toward group membership that each subject commences with.

We then determined whether addition of single biomarkers or sets of biomarkers could significantly increase the AUC of the PPP algorithm. We undertook principle components analysis (PCA) (rotation method: Varimax with Kaiser normalization) to reduce the dimensionality of the data and then ran a series of regression analyses entering ≤5 biomarkers for each analysis. To determine the impact of biomarkers/algorithms, that is, their additional impact over demographics (PPP) we used the equation (new AUC − PPP AUC)/(1 − PPP AUC), where the new AUC is the biomarker/algorithm AUC and PPP AUC is the AUC for PPP, taking 0.6 as the threshold for a significant impact. The equation is in the format of a kappa or agreement statistic, which is thus subject to a similar interpretation.33 Predicted probabilities against final disease classifications were plotted as scatter charts.

To facilitate interstudy comparisons of biomarker levels we quote median and interquartile ranges (IQR) for data that had been transformed before statistical analyses using parametric methods.


Clinical and Demographic Characteristics of Study Population

The patients recruited (n = 181) were White Caucasians except for 1 of Black African origin (Fig. 1). Eighty patients had pathologically proven urothelial cancer, either recurrent (n = 34) (43%) or newly diagnosed (n = 46) (58%). Sixty-nine of 77 controls and 76 of 80 urothelial cancers presented with a history of macrohematuria. Sufficient cells were present for cytological assessment in 65 of 77 (84%) controls and 74 of 80 (93%) urothelial cancers. Two of the controls classified as malignant had prostate cancer and 1 a bladder stone (Table 2).


Antihypertensives were the most common medication (Table 2). Patients taking AH medication had significantly higher total protein levels (median = 0.15; IQR = 0.07-0.37) than patients not taking AH (median = 0.08; IQR = 0.05-0.17) mg/mL (P = .003; t test).

Univariate Analyses

Nine biomarkers were significantly higher in urothelial cancers compared with controls (Table 3). We observed normal distributions in frequency histograms plotted using log transformed data for all biomarkers except that of von Willebrand factor (vWF), FPSA, and TPSA. The latter 2 were statistically analyzed using nonparametric methods. vWF was excluded from subsequent statistical analyses. Interestingly, creatinine and osmolarity levels were significantly correlated in urine (r = 0.796, Pearson correlation). FPSA and TPSA levels (measured as controls) were significantly higher in males (n = 119) (median = 0.11; IQR = 0.05-0.18, and median = 0.88; IQR = 0.04-2.50, respectively) ng/mL than in females (n = 37) (median = 0.04 (IQR = 0.04-0.04) and median = 0.06; IQR = 0.06-0.06, respectively) ng/mL (Mann-Whitney; P < .001).


CEA behaved favorably against BTA and NMP22 as a single biomarker for urothelial cancer, contributed to 1 of the algorithms and was the most accurate predictor for BPH (83% sensitivity) (Tables 3 and 4). CEA was significantly elevated in smokers (median = 1.77; IQR = 1.18-2.65) compared with nonsmokers (median = 1.15; IQR = 0.76-1.80) ng/mL (P = .003, t test). When smoking quantity, smoking years and CEA were entered into the Forward Wald binary logistic regression analysis, both smoking years and CEA were retained in the equation, indicating that CEA acts independently of smoking as a biomarker for urothelial cancer.

Table 4. Sensitivities and Specificities of Biomarkers and Algorithms
Biomarker(s)Correctly Classified/totalSensitivitySpecificityAUC (95% CI)Impact
  • Abbreviations: AUC, area under the curve; BTA, bladder tumor antigen; CEA, carcino-embryonic antigen; IL, interLeukin; HA, hyaluronidase; NMP22, nuclear matrix protein 22; PPP, prior predicted probability; TM, thrombomodulin; VEGF, vascular endothelial growth factor.

  • Sensitivities and specificities of the individual biomarkers and algorithms were determined using the cutoff values determined from the receuver-operator characteristic (ROC) curves (AUC, area under the curve). The impact of the biomarker or algorithm was determined as its improvement on Prior Predicted Probability (PPP) as calculated by the equation (new AUC – PPP AUC)/(1 – PPP AUC) where the new AUC is the AUC of the biomarker or algorithm and PPP AUC is the Area under the Curve for PPP, taking 0.6 as the threshold for a significant impact.

  • Algorithms.

  • §NMP22 was not measured in the first 59 patients and therefore algorithms based on NMP22 as a single or contributing variable have reduced patient numbers.

BTA (U/mL)52/7759/790.73 (0.64 to 0.83)0.68 (0.57 to 0.78)0.74 (0.66 to 0.82)−0.08
CEA (serum) (ng/mL)58/7653/800.66 (0.56 to0.77)0.76 (0.67 to 0.86)0.74 (0.66 to 0.82)−0.08
d-Dimer (ng/mL)49/7749/790.62 (0.51 to 0.73)0.64 (0.53 to 0.74)0.67 (0.59 to 0.76)−0.38
FAS (pg/mL)37/7758/770.74 (0.65 to 0.84)0.48 (0.37 to 0.59)0.62 (0.53 to 0.71)−0.58
HA (ng/mL)49/7747/770.61 (0.50 to 0.72)0.65 (0.53 to 0.74)0.63 (0.54 to 0.72)−0.54
IL-1α (pg/mL)50/7643/790.54 (0.43 to0.65)0.66 (0.55 to 0.76)0.64 (0.56 to 0.73)−0.50
IL-6 (pg/mL)48/7653/790.67 (0.57 to 0.74)0.63 (0.52 to 0.74)0.68 (0.60 to 0.77)−0.33
IL-8 (pg/mL)48/7654/790.68 (0.58 to 0.79)0.63 (0.52 to 0.74)0.69 (0.61 to 0.77)−0.30
NMP2234/3939/650.60 (0.48 to 0.72)0.87 (0.77 to 0.98)0.79 (0.64 to 0.83)+0.13
VEGF(pg/mL)59/7645/790.57 (0.46 to 0.68)0.78 (0.68 to 0.87)0.68 (0.59 to 0.76)−0.33
Age and smoking years (PPP) Age (y) Smoking years (y)48/7771/800.89 (0.82 to 0.96)0.62 (0.52 to0.73)0.76 (0.69 to 0.84)0
PPP, BTA, CEA, TM BTA (U/mL) CEA (serum) (ng/mL) TM (ng/mL)54/7672/790.91 (0.85 to 0.97)0.71 (0.61 to 0.81)0.86 (0.80 to 0.92)+0.42
PPP, NMP22, EGF NMP22 EGF (pg/mL)31/3957/630.91 (0.83 to 0.98)0.79 (0.67 to 0.92)0.90 (0.83 to 0.97)+0.58

Biomarkers and Pathology

Levels of 9 biomarkers were significantly different when nonmuscle invasive and muscle invasive were compared and 7 when grades 1 and 2 combined were compared with grade 3 tumors (t test; P < .05) (Table 3). Urinary levels of interleukin (IL)-8 were significantly higher in urines from patients with tumors with an inflammatory infiltrate (n = 46) when compared with those without an inflammatory component (n = 21) (P = .015, t test).

Multivariate Algorithms Enhanced Prediction of Urothelial Cancers

Age and smoking years contributed to the PPP algorithm. We identified 2 algorithms with enhanced AUCs in comparison to PPP (Table 4). BPH was a significant confounding pathology for algorithms (Fig. 2A). Median predicted probability scores for recurrences and newly diagnosed were similar (Fig. 2B).

Figure 2.

This shows predicted probabilities of 4 algorithms. Each algorithm, created using Forward Wald binary logistic regression analyses, generated a predicted probability between 0 and 1 for each patient (represented by a circle). For (CON) CONtrols predicted probabilities <.05, that is, below the .5 predicted probability line indicate correctly classified cases. Conversely, for urothelial cancers correctly classified cases appear above this line. Predicted probabilities were generated for each patient using 4 algorithms according to their diagnostic classification as (A): ND, no diagnosis; benign, benign pathologies; INF, INFlammatory conditions; BPH, benign prostate hyperplasia; Cancers, cancers other than urothelial cancer; Sup, superficial Ur Ca; Inv, invasive Ur Ca and (B): as CON, NEW, NEWly diagnosed, or (RECUR), RECURrence; PPP, prior predicted probability; VEGF, vascular endothelial growth factor; AUC, area under the curve.


Our findings significantly add to the understanding and knowledge of the role of biomarkers in the diagnosis of bladder cancer in patients presenting with hematuria. We have shown that as a single biomarker, CEA, performs equally to BTA, the FDA approved biomarker for bladder cancer. CEA is 1 of the components of the ImmunoCyt assay.34 Surprisingly, thrombomodulin made a significant contribution to 1 algorithm, despite not being significantly different when urothelial cancer and control levels were compared using univariate analyses. If thrombomodulin had been assessed in a single biomarker study its potential as a contributory biomarker to an algorithm would have been missed. We have demonstrated that biomarkers representing diverse pathways, can significantly improve the ROC statistic based on demographics. Importantly, we have highlighted the technology to take this concept from bench to bedside by embracing the high-throughput capability of multianalyte biochip array technology,32 which can produce hundreds of results in 1 to 2 hours. This biochip technology could significantly inform medical decision making during stratification of low-risk patients with hematuria and impact on health economics.

In contrast to our diverse pathway rationale, Kramer et al35 focussed on 1 pathway hypothesizing that the functional synergy of HA-family molecules would have better synergy. We report that HA was significantly higher in urothelial cancer patients than in controls, and further, correctly classified 17 of 18 of the muscle invasive cancers.

CEA, an FDA approved biomarker for colorectal cancer, played an important role in 1 of the algorithms which we identified. CEA is a reexpressed oncofetal protein and also belongs to a group of glycoproteins that includes human epidermal growth factor receptor 2 (Her2), PSA, and CA19-9.36 Aberrant glycosylation may play a role in cancer progression. Interestingly, thrombomodulin that contributed to the same algorithm, is also a glycoprotein that has been associated with coagulation, inflammation, fibrinolysis, and cell proliferation.37

Unfortunately, we could not detect survivin despite using 3 different kits (Quantikine—DSV00 and Survivin Surveyor kits; R&D Systems, Minneapolis, MN) and EK0641 (Boster, Malden, Mass). Survivin may degrade rapidly.38 We did not appraise telomerase or fluorescence in situ hybridization (FISH) because their analyses require intact cells and complex specialized procedures that are costly and unsuitable for high-throughput analyses. NMP22 was not assessed in the first 59 samples, which limits our interpretation of NMP22′s performance. NMP22 has been shown to enhance predictive models based on clinical factors.39

We accrued extensive clinicopathological data that allowed us to probe the sensitivities of biomarkers for the confounding pathologies. Biomarkers with high specificity for BPH and high sensitivity for nonmuscle invasive disease should be targets for future validation as these were significant confounding pathologies in this study (Fig. 2A).

Patient numbers in this study precluded separate evaluation and test groups and were insufficient to exclude the potential of missing the effect of 1 or more biomarkers. Nonetheless, the numbers were sufficient to identify changes in ROC statistic of the order attained from 0.76 to 0.90. The mean age of controls was significantly lower than in urothelial cancer patients. To overcome the clear differences between the urothelial cancer and control groups, we included PPP in the multivariate algorithms, which meant that members of each group no longer started from the same point.

We have demonstrated that biomarkers can significantly improve the ROC statistic based on demographics and have highlighted the high-throughput technology to take this concept forward.32 The cost would be $60/per patient for biochip analyses. The Bladder Cancer Biochip could be used to reduce the number of cystoscopies through screening low-risk patients presenting with hematuria either at the GP surgery or at the hematuria clinic. This would undoubtedly have significant health care benefits.


No specific funding was disclosed.


We declare a conflict of interest in that the following authors on the paper. M.R., C.R., J.L., and K.W., are named inventors on British Patent No 0916193.6, which protects the biomarkers in the algorithms. In addition, M.R., C.R., and J.L. are employees of Randox Laboratories Ltd, who undertook the biomarker analyses using multianalyte biochip technology. Randox funded the salary of F.A., who recruited the patients over 2 years.