Introduction

Hematuria is one of the most common symptoms in urological practice, as up to 20% of all urological visits are for hematuria [1]. Hematuria in the adult population can have different causes, e.g. urinary tract infections, urolithiasis, benign prostate enlargement (BPH), and urologic malignancies. A urological cancer is found in approximately 5% of patients presenting with microscopic hematuria and in around 20% of patients with macroscopic hematuria [2,3]. However, in up to 60% of the patients no source of bleeding is found. Thus, most important in the evaluation of hematuria is the discrimination between a malignant and non-malignant cause. As a consequence, patients will be subject to an extensive examination, including cystoscopy, cytology and imaging of the upper urinary tract. The sensitivity of cystoscopy for the detection of bladder cancer (BC) is high, ranging from 68-83% [4,5]. Yet, it is an invasive procedure, causing pain and discomfort. Cytology has a high specificity, but a poor sensitivity especially for the detection of low-grade BC. Therefore cytology is only used in combination with cystoscopy.

The use of molecular markers in the assessment of hematuria could be of importance in order to reduce costs and to avoid invasive diagnostic procedures. However, none of the current investigated markers have a high enough sensitivity and specificity to accurately distinguish between malignant and non-malignant causes.

Recently, we developed a combination of methylation markers for urine-based follow-up of bladder cancer patients. The sensitivity of the combination of markers was 74% for the detection of bladder cancer recurrences [6]. The aim of the current study is to investigate whether these markers could also be used to predict the risk of urothelial cell carcinoma (UCC) in patients presenting with painless microscopic and macroscopic hematuria.

Material and Methods

Results

A total of 169 patients with painless microscopic or macroscopic hematuria were included in this study, comprising 104 men and 65 women. The mean age was 59 years (range 17-92). Patients and tumor characteristics are depicted in Table 1. In 54 cases, a bladder tumor was the cause of hematuria. In the remaining 115 patients, the cause of hematuria was non-malignant. Patients with UCC were significantly older and presented more frequent with macroscopic hematuria compared to patients with a non-malignant cause(p<0.001 and p=0.002). Cytology was performed in 81% of the patients. The sensitivity and specificity for cytology were 39% and 97, retrospectively.

Methylation analysis was performed for OSR1, SIM2, OTX1, MEIS1 and ONECUT2 and methylation percentages were calculated. Next, univariable logistic regression analysis was performed for the methylation percentages of the five different genes and the clinical variables age, gender, type of hematuria and cytology (Table 2). OSR1, SIM2, OTX1, MEIS1, ONECUT2, age, type of hematuria and cytology were all significant predictors for the presence of UCC. In order to calculate the combined effect, multivariable logistic regression analysis was performed. First a model was developed, based on the methylation percentages of the five genes and the clinical variables age, gender and type of hematuria. This resulted in an apparent AUC of 0.88 (Figure 1). In this model age, type of hematuria, ONECUT2, OSR1 and SIM2 were independent predictors for the presence of UCC (Table 3). After internal validation by bootstrapping, the optimism-corrected AUC was 0.84. To discriminate between patients at low-risk for UCC vs. patients at high-risk for UCC, a cut-off value with optimal sensitivity and specificity was determined. Based on a cut-off of 0.307, 44/54 of patients with UCC were in the high-risk group, resulting in a sensitivity of 82%. 94/115 (82%) of patients without a malignancy were in the low-risk group. In figure 2, the predictive values are shown according to stage (Figure 2A) and grade (Figure 2B) of the detected tumors. Tumors that were detected in patients in the low-risk group were mostly low stage and low grade. However, there was also one patient with a grade 3 tumor. On the other hand, all patients with a ≥pT2 tumor were in the high-risk group. In order to determine the additional value of cytology, a second model was developed. This model resulted in an AUC of 0.89 as shown in table 3 and an optimism-corrected AUC of 0.85. Based on a cut-off of 0.306, the sensitivity and specificity for the second model were 85% (39/46) and 87% (80/92), respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. ROC of the multivariable model (AUC 0.88).

https://doi.org/10.1371/journal.pone.0077657.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Scatterplot of risk values for the detection of urothelial cell carcinoma (UCC) in patients presenting with painless hematuria.

X-axis depicts stage (A) and grade (B) of the resected tumors. Y-axis depicts the risk value based on the developed logistic regression model. Patients with risk value > 0.307 were considered at high-risk for having UCC. Patients with a risk value < 0.307 were considered at low-risk. Each dot represents a urine sample.

https://doi.org/10.1371/journal.pone.0077657.g002

Discussion

Painless hematuria is a major problem in urological practice and the distinction between malignant and non-malignant causes is crucial. In this study, we developed a prediction model for the assessment of patients presenting with painless microscopic or macroscopic hematuria. With this newly developed model, the urologist will be able to adjust patient examination according to patient risk, resulting in a reduction of costs and patients discomfort. Previously, the five methylation markers were proven to be sensitive for the detection of recurrent UCC [6]. The methylation assay also appeared highly reproducible between different investigators. In the current study, the five markers discriminated with a high sensitivity between patients with and those without primary bladder cancer. Addition of the clinical variables age gender and type of hematuria increased the accuracy of the model. Age is one of the greatest risk factors for the development of bladder cancer and since men have a 3-4 times higher chance of developing UCC compared to women, gender also contributes significantly [7]. The use of cytology even more improved the diagnostic accuracy. Yet, we believe that the use of cytology as diagnostic test especially for the detection of low grade tumors is debatable. Therefore, we decided to calculate the fist prediction model without the addition of cytology. Inclusion of smoking history may even improve the predictive value further, since previous studies demonstrated that a history of smoking is an important independent predictor in the evaluation of hematuria [7-9]. However, smoking history was unavailable for the patients in this study.

Up to now, multiple studies have been performed on the use of molecular tests in the diagnosis of bladder cancer in patients presenting with hematuria and some of these assays are already FDA approved [10-13]. However, due to suboptimal sensitivities and specificities, the analyses are mostly performed in addition to cystoscopy. Abogunrin et al. investigated whether biomarkers were able to improve the predictive power of a risk model which was based on clinical variables [14]. They considered the additional predictive value of nine different biomarkers to the prior predictive probability (PPP) that was based on age and smoking. They concluded that the addition of nuclear matrix protein 22 and vascular endothelial growth factor to the PPP improved the diagnostic accuracy from 0.76 to 0.90. In another study by Cha et al, the authors also combined molecular tests and clinical features into a multivariable regression model in order to predict the likelihood of having bladder cancer. They developed a nomogram based on the commercially available immunocytology assay (uCyt/ImmunoCyt), in combination with conventional cytology and clinical variables, i.e. age, gender, smoker, hematuria (microscopic vs. macroscopic) [15]. The AUC of this multivariable model was 0.904. However, cytology and immunocytology are highly dependent on the skills and experience of the pathologist [16].

A limitation of the current study was the retrospective design. Since this was not a consecutive series of patients, the composition of the patient cohort does not reflect true clinical practice. In addition, this model was internally validated by using the bootstrap method [17]. Therefore we suggest this model should be externally validated in a large prospective patient cohort of patients presenting with painless microscopic or macroscopic hematuria.

In conclusion, we developed an accurate risk model for the evaluation of patients presenting with painless hematuria. Predicting the risk of bladder cancer in these patients could be of great value, resulting in less extensive examination of low risk patients and reduction of costs. Further validation in a large prospective patient cohort is necessary to prove the true clinical value of this newly developed model.

Acknowledgments

The authors wish to thank Prof. Dr. Borre from the department of Clinical Medicine, Aarhus University Hospital, Denmark for his contribution to sample collection.